<%BANNER%>

Empirical Essays in the Economics of Regulation

Permanent Link: http://ufdc.ufl.edu/UFE0021070/00001

Material Information

Title: Empirical Essays in the Economics of Regulation
Physical Description: 1 online resource (173 p.)
Language: english
Creator: Silva, Hamilton Caputo
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: analysis, aneel, benchmarking, brazil, cap, data, dea, distribution, economics, efficiency, electricity, envelopment, event, frontier, incentive, model, price, privatization, productivity, regulation, sfa, stochastic, study
Economics -- Dissertations, Academic -- UF
Genre: Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Reform processes undertaken on network industries have encompassed unbundling, privatization, introduction of market-oriented regimes for their competitive segments, and implementation of a new regulatory framework for the remaining segments with natural monopoly characteristics. The new regulatory framework has involved the creation of independent regulatory bodies and the incorporation of theoretical advances from the economics literature on incentive regulation, under the ultimate objective of providing the conditions and incentives for efficiency improvement and for the possible achievement of second best prices. This dissertation adds to the literature on the impacts of the restructuring measures implemented and contains three empirical essays on the reforms accomplished in the Brazilian electricity sector. In Chapter 2, information on stock market reactions to regulatory announcements are employed to investigate the level of consistency and predictability of decisions taken by the Brazilian electricity sector regulator and to identify the extent the aforementioned regulatory body has been able to act independently and to balance multiple interests in regulation. The findings indicate that the regulator?s decisions have not favored a single interest group. The evidenced unpredictability of the regulatory agency?s decisions suggests the need for improvements in the regulatory discussion process, with the adoption of measures to increase the transparency and to promote more substantive public hearings. Additionally, the study shows that the need to provide incentives for new investments has played a significant role in the regulatory process. The estimates indicate that regulatory decisions accounted for roughly 12% of the difference in performance of the sample securities with respect to the market index. The Chapter 3 examines the impact of privatization and incentive regulation on firms? performance in the period of 1998 to 2003. Evidence indicates performance improvement after the implementation of sector reforms, with both privatized and public companies reducing the efficiency gap with respect to companies that were privately owned before the reforms. The results show that privatized firms responded more aggressively than public firms to the new incentives brought by price-cap regulation, and suggest that the high performance improvement experienced by privatized firms did not come from mere reduction in costs brought by deterioration in the quality of service. The findings also indicate a possible strategic behavior associated with the periodic aspect of price-cap regulation, as well as to cost shifting implemented by companies that operate in the electricity generation segment. The Chapter 4 is motivated by the increasing use of the model company approach to determine electricity distribution tariffs in Latin America, despite the criticisms made to the method?s subjectivity and obscurity. The study examines whether the use of the engineering approach in the Brazilian electricity distribution sector periodic tariff review enabled the attainment of the welfare maximizer regulator?s rate setting objectives, by comparing the method?s implied performance scores to efficiency measures provided by economic benchmarking approaches. Results show that some firms, mainly the ones serving more affluent consumers, operating in more densely populated areas and having a lower proportion of electricity delivered to industrial customers, received substantially lower repositioning indexes than the economic benchmarking methods would recommend, pointing to a possible violation of firms? break-even constraints. The findings also reveal that significantly higher repositioning indexes might have been given to companies with the opposite characteristics, negatively affecting the incentives for further productivity improvements, as some of the possibly benefited companies do not appear in the top ten of the benchmarking efficiency rankings.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Hamilton Caputo Silva.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Berg, Sanford V.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021070:00001

Permanent Link: http://ufdc.ufl.edu/UFE0021070/00001

Material Information

Title: Empirical Essays in the Economics of Regulation
Physical Description: 1 online resource (173 p.)
Language: english
Creator: Silva, Hamilton Caputo
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: analysis, aneel, benchmarking, brazil, cap, data, dea, distribution, economics, efficiency, electricity, envelopment, event, frontier, incentive, model, price, privatization, productivity, regulation, sfa, stochastic, study
Economics -- Dissertations, Academic -- UF
Genre: Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Reform processes undertaken on network industries have encompassed unbundling, privatization, introduction of market-oriented regimes for their competitive segments, and implementation of a new regulatory framework for the remaining segments with natural monopoly characteristics. The new regulatory framework has involved the creation of independent regulatory bodies and the incorporation of theoretical advances from the economics literature on incentive regulation, under the ultimate objective of providing the conditions and incentives for efficiency improvement and for the possible achievement of second best prices. This dissertation adds to the literature on the impacts of the restructuring measures implemented and contains three empirical essays on the reforms accomplished in the Brazilian electricity sector. In Chapter 2, information on stock market reactions to regulatory announcements are employed to investigate the level of consistency and predictability of decisions taken by the Brazilian electricity sector regulator and to identify the extent the aforementioned regulatory body has been able to act independently and to balance multiple interests in regulation. The findings indicate that the regulator?s decisions have not favored a single interest group. The evidenced unpredictability of the regulatory agency?s decisions suggests the need for improvements in the regulatory discussion process, with the adoption of measures to increase the transparency and to promote more substantive public hearings. Additionally, the study shows that the need to provide incentives for new investments has played a significant role in the regulatory process. The estimates indicate that regulatory decisions accounted for roughly 12% of the difference in performance of the sample securities with respect to the market index. The Chapter 3 examines the impact of privatization and incentive regulation on firms? performance in the period of 1998 to 2003. Evidence indicates performance improvement after the implementation of sector reforms, with both privatized and public companies reducing the efficiency gap with respect to companies that were privately owned before the reforms. The results show that privatized firms responded more aggressively than public firms to the new incentives brought by price-cap regulation, and suggest that the high performance improvement experienced by privatized firms did not come from mere reduction in costs brought by deterioration in the quality of service. The findings also indicate a possible strategic behavior associated with the periodic aspect of price-cap regulation, as well as to cost shifting implemented by companies that operate in the electricity generation segment. The Chapter 4 is motivated by the increasing use of the model company approach to determine electricity distribution tariffs in Latin America, despite the criticisms made to the method?s subjectivity and obscurity. The study examines whether the use of the engineering approach in the Brazilian electricity distribution sector periodic tariff review enabled the attainment of the welfare maximizer regulator?s rate setting objectives, by comparing the method?s implied performance scores to efficiency measures provided by economic benchmarking approaches. Results show that some firms, mainly the ones serving more affluent consumers, operating in more densely populated areas and having a lower proportion of electricity delivered to industrial customers, received substantially lower repositioning indexes than the economic benchmarking methods would recommend, pointing to a possible violation of firms? break-even constraints. The findings also reveal that significantly higher repositioning indexes might have been given to companies with the opposite characteristics, negatively affecting the incentives for further productivity improvements, as some of the possibly benefited companies do not appear in the top ten of the benchmarking efficiency rankings.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Hamilton Caputo Silva.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Berg, Sanford V.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021070:00001


This item has the following downloads:


Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20101113_AAAAOC INGEST_TIME 2010-11-13T18:35:27Z PACKAGE UFE0021070_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 16225 DFID F20101113_AACXLH ORIGIN DEPOSITOR PATH silva_h_Page_058.QC.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
1b7ddca275ce6348cab920da6bbf42cf
SHA-1
b16ce52bb1dd844828b5c99241a3e58df211dfb3
18486 F20101113_AACXKT silva_h_Page_072.QC.jpg
3621c133cd1a1c412086f8e8359b0414
5451821c20483b1c739668f7896058296fce128d
78583 F20101113_AACWHR silva_h_Page_112.jpg
1bcbfcc14569c7c1bb74ef852ba414b8
53f21f71c827ea2ac6e12dbf3f44d3713bfeaf6f
88248 F20101113_AACWIF silva_h_Page_129.jpg
2c0b791964ec3406f0cc016003065909
1c3b1a8b14ef41796849b510e7010a9095aa48e3
3463 F20101113_AACXLI silva_h_Page_003.QC.jpg
0929349027421d2822c57f69639ad60c
0bf73ec1a4c07942f629bfbf0942ebc7e7d727f6
7454 F20101113_AACXKU silva_h_Page_122thm.jpg
d9d289b997189b4d73a9819d914f5bd9
88afab63dfb03a28095222688f9d844740b8b1a5
77193 F20101113_AACWHS silva_h_Page_113.jpg
a601e56c3797ea3d2cf6e471defb3333
c9f1ec04c62f057286195a54ca6d70c99bccf672
70514 F20101113_AACWIG silva_h_Page_130.jpg
b5788877c333a4223ea48a1cc1c56ae6
04b5c8686a689c73dd865390e2fdd632609f3245
3682 F20101113_AACXLJ silva_h_Page_104thm.jpg
dfc15a7c92ceec42324e9aa2e43c05cf
6cfa388be3a71dcb0ec1212ee6daa0e92553fe5d
16752 F20101113_AACXKV silva_h_Page_106.QC.jpg
1cd5a881c142a1f5e06f0714dd1bc5e3
777ed0e789e91aa9a04cac94dd4e407fa7abc5d6
74580 F20101113_AACWHT silva_h_Page_114.jpg
a1ef4b8495ff6efc93238cc3089cd5ab
c1dda81535ca076dec41c5d9e3fbbea602eb3798
72613 F20101113_AACWIH silva_h_Page_131.jpg
7e88721322c96435f84ef5101d5e640b
99fadab0b47176f0dc7b3d38a0669243b87f5704
9187 F20101113_AACXLK silva_h_Page_011.QC.jpg
7713d27cae64a3ef9bce46a0b49cf5c5
fd0313031197d786625b001de330efbf55a7e6bd
27976 F20101113_AACXKW silva_h_Page_094.QC.jpg
293a9c671dc0d90dd3b19326c525eeb4
c63ea9fa96c682c6574bb9fbc990e7cabb711b8e
71859 F20101113_AACWHU silva_h_Page_115.jpg
d5464dfe410939d073a8ec50629bde4b
3041ce4fdad8f5c5e06b351153bfc1a03382b37a
87545 F20101113_AACWII silva_h_Page_133.jpg
2c2ce608989d2f3a27efb8b3612bb497
a71ba846e130d2c6b20dc13d63a6cd597695d760
4177 F20101113_AACXLL silva_h_Page_153thm.jpg
7839d701649d3548d596975c13d1475b
61c537b6dc2dd6234e990b3b4559c5ca86444ad9
6439 F20101113_AACXKX silva_h_Page_024thm.jpg
b0c151dfb5b116856bdf3f651268aa34
c3426fe7a5dc938b34c568420247656c9379e115
78090 F20101113_AACWHV silva_h_Page_116.jpg
f16ac95b6156639624b2de2a3b4aa10b
d2773ae67a37de797ab679fb92619af9feb93858
90604 F20101113_AACWIJ silva_h_Page_134.jpg
d5130bffa601b4c1d9f60ea47a6d3a30
7c396d8042646fcf766c58650775798141ed120b
6700 F20101113_AACXMA silva_h_Page_020thm.jpg
b86ca1c6a135f1f37a45eb439f9d9966
5117712eb87ef3acb81ef6d8fee215bea15d539d
22592 F20101113_AACXLM silva_h_Page_061.QC.jpg
26db36795cbf75a93f861a48481cd5e1
d55e595ce47e3b53b51d4b7d4cd20a882e73b5ce
25102 F20101113_AACXKY silva_h_Page_083.QC.jpg
6fa590b0ee135b04457845ba65b0e937
c256ba1d66c8cb1d4093902a0834150f9b92a519
77070 F20101113_AACWHW silva_h_Page_117.jpg
53545bd289b113a28a8241ba96a97851
fd0b797a4dbf36de9ca6513471a41d3e0015ee6e
80202 F20101113_AACWIK silva_h_Page_135.jpg
012fc3f9aa743afa183c6abf818a8c3d
d4a79380cbfcdc217598267a200ca0f8cbdaaaee
28337 F20101113_AACXMB silva_h_Page_026.QC.jpg
b9b99fdb1ac43174eb4159714b407c8b
d8c139c3b624ec7ad60b42372b3313301fe3c099
23896 F20101113_AACXLN silva_h_Page_157.QC.jpg
19caca74c7463f84d43774277a52ae9c
c8a4f1b1fa3dfccf8338590ef3f1803d5881fd27
7216 F20101113_AACXKZ silva_h_Page_134thm.jpg
a6daec6de01f62e135df89fe5fb61ce5
3791826f5fcba6a86868a4b54fd5ccc3b75a9aeb
80048 F20101113_AACWHX silva_h_Page_119.jpg
3c0114d7890b702ed4fb61d62a1cffaa
e123cc2efce055e25f897aa1801d2737f1bec6e7
87693 F20101113_AACWIL silva_h_Page_136.jpg
46808ea07be36a171f01e7446b679277
bd4a031772bce0eb138ff663acfe8d7c6089ed01
6466 F20101113_AACXMC silva_h_Page_027thm.jpg
d1bc40783f898a5060a6445fee244d5a
a901c80ae964e07fc5bdd367228bfa5e5035d240
16109 F20101113_AACXLO silva_h_Page_048.QC.jpg
550b1015d17875f2a8a5df8df23aed19
28b7461c2bc5430edd6e23d2c7b7a4b5746cf47f
63184 F20101113_AACWHY silva_h_Page_120.jpg
d6630b3fe2d151da475fc76a26939c5c
b4f820de5f1850168353531c1a3c01fcae99e316
73987 F20101113_AACWJA silva_h_Page_158.jpg
e142b0938493ccf016b6d685c5fedf74
cf34c181b43495a58ed4f803525715a50a1f74ab
77348 F20101113_AACWIM silva_h_Page_137.jpg
61d6d9bcec1953181ee5dad3a9e44754
587331ca44f2fac1a0220f4c0af5044ed7bf2f9e
7143 F20101113_AACXMD silva_h_Page_029thm.jpg
63814b2837b5c31014f7dfae17153672
6278fbd851dc3c86d0f02947a65f834197f40492
6993 F20101113_AACXLP silva_h_Page_139thm.jpg
7ab8c1b7da4aea8ee62dadbf3e841c65
41644f4c2e0337e2bd0d3309400b9cff2354910b
22216 F20101113_AACWJB silva_h_Page_160.jpg
594390ff6f0082f1228b285d47cdc6c4
142354c68d0925b4291a7e82223bf94438a70d9e
79908 F20101113_AACWIN silva_h_Page_138.jpg
37eaba3560ae5c9b707a7851f2c6b117
028adbf739960296fdd332d7f4e6b2fa280b7abe
25835 F20101113_AACXME silva_h_Page_036.QC.jpg
4bde839670479b03f25d3b18f8bde766
875b5480ab56579ab9b49637013e2a7a106da81c
26223 F20101113_AACXLQ silva_h_Page_080.QC.jpg
f28c44a957cb8d5222b68b3ec4252e33
9464530e8b31b7a84d41a4faa185525dd650f3d1
69538 F20101113_AACWHZ silva_h_Page_121.jpg
5c70b139e71b0eea12006cef6c4709c6
88b27c764ff72f2e1bf0fb8525eaf3a2feec9996
11687 F20101113_AACWJC silva_h_Page_161.jpg
f679e4dfb24fb1aac6b8116e7f5b61fe
506ed13538c34caed9c2148c17a483c93db76d3d
75964 F20101113_AACWIO silva_h_Page_139.jpg
d16b5efc83890d50611a426b93369feb
6d6c634177b2d62303a94caa21709d0627aee6fc
25930 F20101113_AACXMF silva_h_Page_040.QC.jpg
dc3f8aa82f0cbafe186710d822439dfb
fd552ccc42668ad5696693ad532136ba044eda58
13586 F20101113_AACXLR silva_h_Page_052.QC.jpg
c4e12bdd618f7da02677684efa541661
b4a185c4da689204f64215c3fc9f21b2659d61d5
27154 F20101113_AACWJD silva_h_Page_162.jpg
e603714c46aee835f8c08a3a8cc79f19
dc59a02a6da9d78487162c4cff493aebb3181c0d
84065 F20101113_AACWIP silva_h_Page_142.jpg
70ab4f40305d4edb1608c078c1b49311
af2292ad1f0d6c8c5f046742e141f6c59c09514e
25951 F20101113_AACXMG silva_h_Page_044.QC.jpg
dd6f3050184ac47be38bf12b4a5060b2
303a7320820b904897b44674b7a6a18f3b71c9bc
7428 F20101113_AACXLS silva_h_Page_094thm.jpg
0e4cdfeb81fe0eb8e5e0f810ad7fae00
e9156a2a40e78750c70a8bc89e84006c562d0903
44704 F20101113_AACWJE silva_h_Page_163.jpg
c14a6365d048e169a29c9af775e49699
2a17e71d108c9bde7e47f98f7d43a35805d15870
71442 F20101113_AACWIQ silva_h_Page_143.jpg
b84631d08b8f5c67e64f845a0d38b168
288cf8cafa09906873464ac3d7e02de462d07640
10749 F20101113_AACXMH silva_h_Page_059.QC.jpg
e6d67815da7f30846ccc27a4a24aeb2c
4c7d9b7e617350fc8482fa82f050b99818b13334
12696 F20101113_AACXLT silva_h_Page_153.QC.jpg
f12c09b40bda15243721134aa4295fc9
fd00274e688cd6c01b9e85afbe16dbd7e4e79f81
93689 F20101113_AACWJF silva_h_Page_164.jpg
544537e451a03c466a530d2ba7e49543
ed2d469292bab5e64274f427a677c100fd5c1b88
79467 F20101113_AACWIR silva_h_Page_144.jpg
f4d11c27d4a13d5a3157bf09af61dcf6
e4e5f420fb548dc8cec5df0cbaab2bb64f44e11a
6681 F20101113_AACXMI silva_h_Page_067thm.jpg
8efb512868530e7373b523e41c8751ee
973f6032925a03b6fd29b6430b3464cde4fd2834
6128 F20101113_AACXLU silva_h_Page_064thm.jpg
ac908cb3d7b38ee502773dd03876cfd9
f7cea4e11414f5e6564978c40cad0ffd8bbdfb78
83695 F20101113_AACWJG silva_h_Page_166.jpg
f76fbd7ca9693c57acc1d9b3d6bb791a
b19687ea8b71b4a02e2c72141a20c03f185ae3d1
78210 F20101113_AACWIS silva_h_Page_146.jpg
a5c673dcbcc077b4cd53a4d86e42453c
346179e56fcd602fb86bc60137e641577aa360c4
6850 F20101113_AACXMJ silva_h_Page_071thm.jpg
f8e926a2a7d6ec3004d3b34997e160db
be103c258fd655dc253e4aaf0fb57c9baf660e80
256672 F20101113_AACXLV UFE0021070_00001.xml FULL
6d6298f265e7efffed3a3131f18577cb
06ef49d8a010d0f854c81eed56f092e01e7ea245
92492 F20101113_AACWJH silva_h_Page_168.jpg
de7bf8561d0f51d985ad900a57aa853e
f9950e6204616cf697cf1fba4ea5f169d0543dc8
53580 F20101113_AACWIT silva_h_Page_147.jpg
3d58504f6e6c53d6135bf9e04009ad7a
8988d96ff3187dfa27af96383eb4f8b7106bc6e2
5275 F20101113_AACXMK silva_h_Page_072thm.jpg
27eaf0a5726cfaeb584b5eecc18afa28
d5d1128a74538013b5f5df37cf73ce9b3b4b6c5b
6878 F20101113_AACXLW silva_h_Page_001.QC.jpg
6d3b88525f65c6cfe3856dbe5a552270
bf21e949398d62746dedb6dff9873f9b82c0ec98
89785 F20101113_AACWJI silva_h_Page_171.jpg
b8866fda0af2a5c27c8acd98551eb906
57ae310b82d0e8aeefab1dde22b3403de0f30ae2
52988 F20101113_AACWIU silva_h_Page_149.jpg
3ba85420b15a2cf5c1db2eab8d28cf08
7bca1ff78ea337a1119c63cf18af1976b7911c54
21429 F20101113_AACXML silva_h_Page_074.QC.jpg
fc68ed7e971c1827449f4b5ddbd2ed25
7937073b758ec3d92a5f7ee6e1335981eb0ff630
1410 F20101113_AACXLX silva_h_Page_003thm.jpg
e7c9b1667cab83dba385210fa0da52c9
18fba95ace7c3277d2cf9957efab0320bf20a6d6
60384 F20101113_AACWJJ silva_h_Page_172.jpg
18fb593b74f42d9d3a9f30595ce29504
6e583ac09c5dca9052a4efccfa3b930b260ebde1
53122 F20101113_AACWIV silva_h_Page_151.jpg
547b02f7763f0721ac96dd5ebd12f938
cdcd94a63aa3f4a8efe7c68b4cd025b343c07a3a
6614 F20101113_AACXNA silva_h_Page_114thm.jpg
b0265673bbfedebe6151d352e175afe3
beb002c3767d2be8d8a43bb68e44373bb7a5ec1e
6126 F20101113_AACXMM silva_h_Page_074thm.jpg
779a5c9d98e6bb4fc7e83a5158bd812b
2b5367db35614e7bbe54401b326a74beaa3f1f86
7553 F20101113_AACXLY silva_h_Page_018thm.jpg
cbda7b9b70ef34cd1205eaff9012577a
4ec192b82e164a01908d9d88b3c27f353eda4dde
40216 F20101113_AACWJK silva_h_Page_173.jpg
1ba87701624274654c0e11d1e62bdc20
6638dc9e01e26f9cbbabe58f8e0d1254c7a7b4f2
71479 F20101113_AACWIW silva_h_Page_152.jpg
229cdf825d3cd921348af418de5b3dc6
1b9991f2ba67f735fb6fc286e84dd38b3a4f873a
7153 F20101113_AACXNB silva_h_Page_119thm.jpg
fa8df535ffb69f434cbcd1af5b1be6d0
18ad208fd3ff61821f1b86734e004ae3accadf8f
23629 F20101113_AACXMN silva_h_Page_076.QC.jpg
62343cc0c3a05d67bf559d326259a6ac
f38437f0aab4fa5f5cc640c72ca73f56ab24f3a2
27512 F20101113_AACXLZ silva_h_Page_019.QC.jpg
5ad990c5725796c2888e6faa5dd54f21
72fce85ae47caf5d19f8bba6823b16a8abfa4e06
23589 F20101113_AACWJL silva_h_Page_001.jp2
11c3200caef8b424efccdb4eccd87ba1
8076a130ba1c7049007a3c252a868b60f2451f37
37870 F20101113_AACWIX silva_h_Page_153.jpg
54cb459b39269c6c6188b6f0fdaa40c0
746e8a5d86f500d744355d96b7364ea753a181b9
22257 F20101113_AACXNC silva_h_Page_121.QC.jpg
37b192c053e0ea8f7907258c89f2c427
5177a6d68a7cb665673ad1022539290d9dc3e79f
26902 F20101113_AACXMO silva_h_Page_077.QC.jpg
082a7a8650c356e80b4ba0760bdf88f4
1c6f59463d8741b54370ce07f975417e53e2a963
116028 F20101113_AACWKA silva_h_Page_020.jp2
4b7af1b8949751a23b19946993741795
9c1bac6826d9b32dbdf6591e484c2193aaafe957
6156 F20101113_AACWJM silva_h_Page_002.jp2
2876fb2f0e04e0b89b569ce9e945ce7d
42e61b2312b984cc0dcc95722cbabc2426f9510d
75056 F20101113_AACWIY silva_h_Page_155.jpg
8f4727814d94f3beaf021c04374339e2
0cd3aa85722c9f311e07318705805cbc615e620b
26940 F20101113_AACXND silva_h_Page_122.QC.jpg
6468db393a86ed5a6e61db183ff7f7f8
43ae045f7f7be53b1d29e6a6285f6598ea52a718
20585 F20101113_AACXMP silva_h_Page_079.QC.jpg
2256bedc67fb433ddc624e2d79a6fac7
e5ae64715d4dd74e9b3a563513eee56641802c53
112915 F20101113_AACWKB silva_h_Page_021.jp2
f9d33397804d4d9959a144e1a0e1dc8e
98c96221615dae5a074cc6da1423a0ff298ce1d9
90238 F20101113_AACWJN silva_h_Page_004.jp2
6a30fab56e61f35238db35c3c35100e4
ca5eb2604edd6c9025e7e8be6ba31d4adcfff1f1
74024 F20101113_AACWIZ silva_h_Page_157.jpg
3f6643b393c8cb113437d822d7735053
357db28e07fdedf2ed88b4c00a4fca8907ac9e0c
24383 F20101113_AACXNE silva_h_Page_123.QC.jpg
66de07e011c17bace5148a824ffcc50d
dda65fb6c7c7f95392bff79d01d996b4a57b2ac0
25010 F20101113_AACXMQ silva_h_Page_082.QC.jpg
bbe49bc048315a6b25cf71f8d5d139a5
f173eda3f3e439e111e65a391e28b44ad16f3de9
142927 F20101113_AACWKC silva_h_Page_022.jp2
efd58002ef2dac243da7b09fa7746063
96d20fb99f733fb5056cdc29416c0f44f90bb0e1
1051935 F20101113_AACWJO silva_h_Page_005.jp2
43d24cbf72208193c7573142255868f7
975cabb167d7f624382c52f0de0ab867dafaf788
25458 F20101113_AACXNF silva_h_Page_125.QC.jpg
db93d94ee33adcd95076df8a3e476859
409cc46c77d68bcb69b74e7799e58e2518817dc1
7235 F20101113_AACXMR silva_h_Page_084thm.jpg
0fabbaa2274a2ed746414ed8ffa861bd
f37e9bb0e62863ec54d4e3d6f752bfbd7709b738
111803 F20101113_AACWKD silva_h_Page_023.jp2
90b7ea26259b85aa0d3c9ccfa17c9893
77bf3402bd8a653f4fdf4ea2313338b7583404cb
1051986 F20101113_AACWJP silva_h_Page_007.jp2
99f9c6ce594961b76270a3bc35c76ee4
bd9208ba1a458ec70e1484ec45a9a4e6b9117ddc
26184 F20101113_AACXNG silva_h_Page_129.QC.jpg
4a3b72b891980cf4b518026fb869954f
bdf4e6713d708c939f12bd5b4a5817043e4532ff
23450 F20101113_AACXMS silva_h_Page_085.QC.jpg
939f40b9c284ac0f0578641aeb3fd522
ccc011e288fc22736b7d67e076757f35468f0007
109202 F20101113_AACWKE silva_h_Page_024.jp2
245196a3a65c47bffd5f146baae6c8e3
beb4ac07ec393a6376fbc9ae8783540d5568e9ad
1051983 F20101113_AACWJQ silva_h_Page_008.jp2
c857a553612fa1ace55d2ee772d99551
a68e7f15d8282669ea3d552a9b5eda32f7f50d33
7037 F20101113_AACXNH silva_h_Page_133thm.jpg
ffcc6f13ef7ad0b344d670c9d9e8187b
2680ce8ffe70f8a3a8416f1ef4c830c60fa95e6c
6782 F20101113_AACXMT silva_h_Page_086thm.jpg
9421e0c5a78955357b48deb3ed90d413
6e4f9f1710df6502f51166a79aa20fd583570086
128410 F20101113_AACWKF silva_h_Page_029.jp2
7301a7e8e994ee46954e005f8dd65fe8
715f9582923ad28bc1d097ce339f8a4f3a5c7940
98713 F20101113_AACWJR silva_h_Page_009.jp2
839f60cafd9f4eaeb7c050054a955a3e
3a0f52e448859ac93cd7bca5d8b95e2a18a653f8
26401 F20101113_AACXNI silva_h_Page_134.QC.jpg
922b2a814e5464c62e21c602a2845e85
916a854dcbd76d34e46b15b1b7df94f4105fc037
25842 F20101113_AACXMU silva_h_Page_091.QC.jpg
3c335c5f9d02e4751b35fe0ef07e814c
aa798f210ce408d6dc8b348befa879174a0def00
124128 F20101113_AACWKG silva_h_Page_030.jp2
fa24d65826c75a5f59f74a7d9ef06273
c7cfb1d33adfb0c9a3d13d89882b87fdff4642f6
119753 F20101113_AACWJS silva_h_Page_010.jp2
b28a5913920d70c3b0929610b22b2ae6
b4ce5a894345422b9268cd9f59b01b870534daf6
25965 F20101113_AACXNJ silva_h_Page_144.QC.jpg
804af37bfc5deb810b1508559b063270
349241d8ca9b5a5f88943fa23ac0c72f1f466a0c
24854 F20101113_AACXMV silva_h_Page_093.QC.jpg
b9b74770b3183cd4f23b49c28fefc895
68b4e4c8817d19f8561577eeec6ed0bb183ef3e6
133020 F20101113_AACWKH silva_h_Page_031.jp2
85502496b35940601de5aea54c3722c2
317e36d95915852b311b9ba27f8df11b29c347e5
35039 F20101113_AACWJT silva_h_Page_011.jp2
d753c677f81f26c1c95a9bcc4122eebb
de8b9e08b055760942ef7bc354e517ab7eee6c1a
7376 F20101113_AACXNK silva_h_Page_148thm.jpg
a768c89a565f3bb401a69c981548b27e
0514303bb6e9c98843216e3617f1bcfcddc4a8cf
7082 F20101113_AACXMW silva_h_Page_096thm.jpg
30ba4032dec1410b6ad9a9dabfb48653
909a52ca75b6d90c25ea1b47136d8fd34d1f6416
129791 F20101113_AACWKI silva_h_Page_036.jp2
bc27e735c9e4c6131b8f62fb3fa27798
e689c434de6a5c7a15f0445f9ed44fad2aa4005a
111397 F20101113_AACWJU silva_h_Page_012.jp2
3b9a762b65431e46836beae8b1e559e6
f922bc1e5a2e1657b8e0dee8b2ce8d3b8740743a
25470 F20101113_AACXNL silva_h_Page_154.QC.jpg
9c21b1aae1504435f82fc0aa7b40a624
13157a059c239d25b6f048bdc08ad0028d604f24
16717 F20101113_AACXMX silva_h_Page_102.QC.jpg
80e2cab81463f67d1a1994e2d7c9f895
084a563782db4f410e0b2beb4e56890db08e6ef7
133718 F20101113_AACWKJ silva_h_Page_037.jp2
d0069fe40e5817126d26117c8d3500a9
931b75ec3975db46430f64c248da9da9e35cf53f
118038 F20101113_AACWJV silva_h_Page_013.jp2
6fd150950490959171442afa8d61a543
e0ca28695242554bc70f6fffc11a00799901b530
25033 F20101113_AACXNM silva_h_Page_156.QC.jpg
690aa9bd958a72c139ecfb65d53495d5
e477023442f0a018ff1a721c71c2bce130c65277
5260 F20101113_AACXMY silva_h_Page_106thm.jpg
b659b8d5909ee84bbe3210613cf173db
e3b084b0096d21af574309bf9af6269715251e4c
122506 F20101113_AACWKK silva_h_Page_038.jp2
5ca4130033cef22135ab40db19aea8cd
ec22c620c4d4a10dd72664d8e2f8f2afc78d4aec
83529 F20101113_AACWJW silva_h_Page_015.jp2
9b5c224edc7636e21e9a605043d521ad
1338dda6b320c2a3fa4ea2b475ae799f7ec4178a
25430 F20101113_AACXNN silva_h_Page_164.QC.jpg
63d6f2f15c81c74628296e3828acac14
7e7914f512c8a05ff29f9c5947ba66fa6db81b21
4703 F20101113_AACXMZ silva_h_Page_107thm.jpg
c5d780b95684dc603cedd0d2eada5e2c
97eec95bab3747e7b6a2d7744e16df197a517ff1
128367 F20101113_AACWKL silva_h_Page_039.jp2
d7b00c77a5c0366126359dc3153596e7
c46c705d9545358beda63bdfe0ad0366cc76ddf0
136685 F20101113_AACWJX silva_h_Page_016.jp2
ed98deac75e5261629e02028b5883223
9375cc8077d1330a2931a64898adbf2e19049894
25162 F20101113_AACXNO silva_h_Page_166.QC.jpg
f802758be0db92cc5cacfcb674aff415
03dcd6c0464a5aa9093fda160e53940f1f47c28e
129748 F20101113_AACWKM silva_h_Page_040.jp2
2f29f79b21d84ddf2d59b004c557d230
3254818f71bdaeeda96f0b697599d475c3b8b8e0
116334 F20101113_AACWJY silva_h_Page_017.jp2
ec64acb1169e4e2684345fa25978ff09
90b1286c2ec5c4eace32a2a3912648c0753b345b
114161 F20101113_AACWLA silva_h_Page_060.jp2
c33afe4fedef4bb90834f69d8c42045a
e6a519e17b316805ec625b18531eccc895f80386
26641 F20101113_AACXNP silva_h_Page_171.QC.jpg
c4ea6e743b8c17ac970ca96116d0d3b7
30aff4057475b1909775d9d22de018eef1b03e94
128845 F20101113_AACWKN silva_h_Page_041.jp2
8c0db6202045b47c126e1cf7f0d37ddc
0571654a294e7cad11c26a135602be3d25fed59f
1051976 F20101113_AACWJZ silva_h_Page_018.jp2
5acb929d9aa5ade00994f5018a4c8d38
5034a2d067da04cfd5cc68be7274831c464905fa
103803 F20101113_AACWLB silva_h_Page_061.jp2
ebcfbac2956f06dd5d77e35e72583aa7
189fe512561d97b3efa9299680bfd58aa3fe61e5
131775 F20101113_AACWLC silva_h_Page_062.jp2
b01ee8e341e4466a725a75781bb78102
6f6a923eeceb434e2092d532c42bc27d0e4145d5
124164 F20101113_AACWKO silva_h_Page_042.jp2
6f3fbe4a82be99e11ab932febc8ffd25
d69393cc6312137851fbf52edf261ea95c3c616d
99487 F20101113_AACWLD silva_h_Page_064.jp2
6971492d992c8680a94e18eff8492b7c
419088adec51ec885195f72f04d01ebf313e9725
119359 F20101113_AACWKP silva_h_Page_043.jp2
f2896b30026b202f7d26ec2b09bfae65
847e29dede1148581ae895ace30ff064cab899eb
138385 F20101113_AACWLE silva_h_Page_065.jp2
89c3da2a227b03e322e421435471ac04
0f3540c260a885f7a73f935c9f4ea12df3b3a58e
81505 F20101113_AACWKQ silva_h_Page_045.jp2
591bbbd935fa1f8c58937017dea892b8
1f97a7d21e3f9861c6b29aeeef2c6f5617e7cf07
109367 F20101113_AACWLF silva_h_Page_066.jp2
3522cc5ec034ff164ec2a4b36a6e4aa5
52125a42fb375d005b7a9594c276b64f382667a7
898717 F20101113_AACWKR silva_h_Page_047.jp2
cf2e8bbe323bc22a6bcfa6b5cdeedcd2
ba4d85ae8a4eaed1334c1561163218895a50de72
113377 F20101113_AACWLG silva_h_Page_067.jp2
f68bfcd4e0fa4e23dfb95087de6beea0
1a1922eb15b53dc9f9dc484187dc520156c00976
57087 F20101113_AACWKS silva_h_Page_050.jp2
a56ada404df569c657195202f473bec3
5517323ed806408b3296cd1d43f25e618e962f5a
114204 F20101113_AACWLH silva_h_Page_068.jp2
8f8f680bebb831f91b738d7368756791
6a4143a2cb4bb7e8b3582ba3b6eca8b946a483df
551864 F20101113_AACWKT silva_h_Page_051.jp2
f430ff55b944aa8d877300cb5a3863e0
42d128ee205904b3432bf75b26caa4f3ee231127
110086 F20101113_AACWLI silva_h_Page_069.jp2
3260f78c7564e4388722d1acac1c5e00
8e9d0ca3a53cb77516ae1024176d47ef4b4c6965
115437 F20101113_AACWKU silva_h_Page_052.jp2
f3c854fe74750eedd8148db5d21a5741
09b5be0d78201ed2bed79c7dd16124974d551d59
124097 F20101113_AACWLJ silva_h_Page_071.jp2
eb69a298764a2af76ae95a01ff5b20a7
8f44295e318cdd25b0e8fa55d61b0993f07ff935
87537 F20101113_AACWKV silva_h_Page_053.jp2
e15be8e07aadbbac601b5f98d4fa465f
e0dc443d40ba385cb6574b3f282c9e3663189900
83653 F20101113_AACWLK silva_h_Page_072.jp2
b0cefb8d525b5d46767222c273d6cf6c
a71bc065526fadcd7a1f22c24f503295c8d83441
116003 F20101113_AACWKW silva_h_Page_055.jp2
345eb02786261d97c38c76742037cdea
e2789526a81f073b02adcf5408ff3f3ff32fae26
94669 F20101113_AACWLL silva_h_Page_073.jp2
a954fd8cbf334aede1301f18988fcfc5
f6be9367e92655ab26122c2fcd21266296c2df88
119652 F20101113_AACWKX silva_h_Page_056.jp2
01cbd856a756a1c396aabd1fe7223442
bfa291fcfabbb88678e2964ec6b080c4bd88fe2a
125792 F20101113_AACWMA silva_h_Page_089.jp2
fc2064e15366f4deeb48cc038ad59dc2
e60753ad99e1ef4186082d00e92523ca27be2d84
98187 F20101113_AACWLM silva_h_Page_074.jp2
c247c8599af0c76c4161f4b467a4b85d
0694893534a2855468495e7d6fde9b63d6988a5c
122880 F20101113_AACWKY silva_h_Page_057.jp2
721485f495fac782f4a5a55f81040752
8ea7a00b8bbbd0e8292a26be45482fd2b6dd656e
118551 F20101113_AACWMB silva_h_Page_090.jp2
b2ca1cdbc22ba4d101072fb03b6568fd
2f8fe31e704219b3e1f3dd28654a7a7a547151f7
1051904 F20101113_AACWLN silva_h_Page_075.jp2
bee80a03e41628527e4eaf73450ff7f7
b818d4eb52ea07f933c13ed643a97bbf8050b920
96065 F20101113_AACWKZ silva_h_Page_058.jp2
1a70039713335cba83c363ad9afc356e
5895420d0eb23a6ef2874c9ca14347a9e34f2b89
122372 F20101113_AACWMC silva_h_Page_091.jp2
46c12bccad8d074f96b158ea246bc5d7
0d604d2e4fc52f910e0b26cc2563b6a5a298fc24
112485 F20101113_AACWLO silva_h_Page_076.jp2
b0d41c75ad1e01c03ca2be1cecece749
a851a8d3ef92136f13443f0cb02a3c570e45efac
124031 F20101113_AACWMD silva_h_Page_092.jp2
b6399b88ad898e3474dfd11db7d9184b
9c58c787be6ce66c2be5f3de8a79566ce68412eb
133157 F20101113_AACWLP silva_h_Page_077.jp2
dd5e249a4cdc034cabe642fd8f0db7c6
7f39d2bc2ec39e0a1964e3ec35da5b045814f2fd
116999 F20101113_AACWME silva_h_Page_093.jp2
922a3de5a86c6e9fe04d80902a42b8f0
bc592b9b8d2548541772e30d04f51192adeeaa17
122095 F20101113_AACWLQ silva_h_Page_078.jp2
9dd304ab78ba7caf7c10b7ee49136f50
036596fe1244a33d868aa1a1456ada02447b36c3
143755 F20101113_AACWMF silva_h_Page_095.jp2
e1e1fb4083f43f591e8695a50534bf85
5741ca353e13e4d62dbc849463d85347f8cc26b9
122158 F20101113_AACWLR silva_h_Page_080.jp2
fc5aec514e12b70058787e1d417ded8e
7a6414e59b5dd6f3650ad6b0c3300cb0258ac316
131771 F20101113_AACWMG silva_h_Page_096.jp2
daf4783db55f3323098b9f40749490c5
c4162adc96544b66ada3bb609d123ee24f945e3b
122337 F20101113_AACWLS silva_h_Page_081.jp2
53152b39e83d7e72cc619dac080a0e45
5ef17b493c1db8d53e4310da666f30cfa5027592
128752 F20101113_AACWMH silva_h_Page_098.jp2
a4f1885220cf9f646df21f6e6e098005
fd372500dc548439b56e1efa862685e601cda044
115578 F20101113_AACWLT silva_h_Page_082.jp2
9338e4f235ce1a7040f8ef94d95573ca
7c04ce16f413dc67157c8362a5613be0cc086ba5
117976 F20101113_AACWMI silva_h_Page_099.jp2
c52ec744f81c846ed6632c4c3dd641f5
72b48542f844b9834d47d889273e1b3bc3fa8f9d
128284 F20101113_AACWLU silva_h_Page_083.jp2
05f04afdbf518a5bc1635ee3c3974e73
dd0e21a345c3658ac91e1dee5204f9763444fb2e
118466 F20101113_AACWMJ silva_h_Page_100.jp2
e6a71500ac888791c4c2d9b823ea0eba
f79926d5e7fbd365dfa5dc70a96c6f1385f8ae4a
128095 F20101113_AACWLV silva_h_Page_084.jp2
84f7dc538b495baecbebb0fccc7dc6e4
542062ef6cb6cdc6e19cf794aeba1802399798dc
66983 F20101113_AACWMK silva_h_Page_101.jp2
0e98b7f189db63394ec10e7ca65f5575
8ce95496b82ee92f3695342b4cf7fc52111f4b84
109482 F20101113_AACWLW silva_h_Page_085.jp2
66e32557921f2f0612a38a7fcf18e407
981483bca4108697fe200423aa4ffd6a0c6bd9c5
91416 F20101113_AACWML silva_h_Page_102.jp2
f933ac2ea80cf76b1cc7cb23b89575f0
76296556596b41b6aca64a929b24805bf8dac007
117453 F20101113_AACWLX silva_h_Page_086.jp2
ad25536539baa7fc2461f319179ce072
9741b803aec065b05f76173c3a8f949e8ebc7acc
84188 F20101113_AACWMM silva_h_Page_103.jp2
a057f80b90be52bd3c899ee40cae474f
9f4ad9c297adf9619143fb4da8e96848a22b3675
128436 F20101113_AACWLY silva_h_Page_087.jp2
22dd75a78a90963bef0825f0db45d8ef
2a23c495cfe9f919afa33c89b7f749924f52312b
118305 F20101113_AACWNA silva_h_Page_125.jp2
0fa5faa1538d2ce68fe87038f61b5229
1ffcd1535a68e3d79328f3d709e505cba7478b6a
46876 F20101113_AACWMN silva_h_Page_104.jp2
6abfbcaf4a6358aa17d21a1d48f6589f
c7a783e70aef795dabf6a165250e692ee3ee2913
113193 F20101113_AACWLZ silva_h_Page_088.jp2
ca988825241bb6b4fc727c45b3bc846a
691afe8734ba66ba759e1e2b1d79e20e2815c4f0
137402 F20101113_AACWNB silva_h_Page_126.jp2
9523b4cc253e4f8645070eed00d416af
66f14621fd74a7d2bf47be346427e07501cc8c67
111362 F20101113_AACWMO silva_h_Page_105.jp2
8db4ed496e739997589ebf04d4a22342
af0fbcb9ffafe9cf87704482645a1b8f1c2b6d69
123493 F20101113_AACWNC silva_h_Page_127.jp2
9ef4355e2b4b7b64548e92de33ac378a
f54e64a001429ca954efc71eebc834fcce4fb098
88478 F20101113_AACWMP silva_h_Page_107.jp2
81602acabf388ea3f27161c319bb8f1a
1c607fd19c8e4b08af73b7a04d13b1d3d3c0e549
143837 F20101113_AACWND silva_h_Page_128.jp2
734ac0250bb5b95d821be8a0e396f2df
a07c7b6b96ef6e79f35f66dd34e11185b422ad0f
114316 F20101113_AACWMQ silva_h_Page_108.jp2
cb42f3cb0a93bad0dd5bb7ef7de4ab25
f9b42bc21d9a4da812780155dd90f89ead9a9135
135657 F20101113_AACWNE silva_h_Page_129.jp2
8c5e6ac6dbed6217d79debac9dfc8cd0
e223d2df7f309d7b08ea9fbca048c2b380e7e8e6
116666 F20101113_AACWMR silva_h_Page_110.jp2
192e606135b6908e8ec4badbb8c6b5b4
6a1ce0b1e20c11007451838c953ac87ac73be23c
107281 F20101113_AACWNF silva_h_Page_130.jp2
699aca7205c0f1f2035ce0a8dea8d688
8664d0b12f207229083c9252bdfbb8d6c2147936
113082 F20101113_AACWMS silva_h_Page_111.jp2
e49d5216754f149470141d061d87a9cd
acd00c3fd4ba541cea10ccfe95d920acc10cbb67
128211 F20101113_AACWNG silva_h_Page_132.jp2
fae33b6ab4d14ccd9e65a7979eb73440
b379ffa812d5bd5db409c29cba8c9a5e754688e3
118241 F20101113_AACWMT silva_h_Page_113.jp2
23f165deb5391c19a48c3ce5e826c949
85d0f6f6fef35a553054a0611a3fb74d7098a6f6
136225 F20101113_AACWNH silva_h_Page_133.jp2
09fbcb5d82c0eecc72fe8a7991c3556d
ed6075a2df345359c1a9b0ff744bf699faf1b17a
113324 F20101113_AACWMU silva_h_Page_114.jp2
e92330ed50435e5acdac5680602e2dbe
4531bf9498238df83143ad5839c01a7c4c0542f5
144700 F20101113_AACWNI silva_h_Page_134.jp2
9305282b5246f8f49e181ffcafaf310d
b0719d5a959bfddfaa9f684ae36958c33e35648f
110606 F20101113_AACWMV silva_h_Page_115.jp2
7a603d4d7f28db15f4493fccdac338c7
33949474e722fd2b057fd48e0a2f0c703c185d79
123033 F20101113_AACWNJ silva_h_Page_135.jp2
1e5ca9eea4385cc5ba28eb60eabbd332
b7ef1ec7129dcaca272733c3347f9c8cc96121b8
118023 F20101113_AACWMW silva_h_Page_116.jp2
bb383271f5dd95d4a24b544eefab492d
d1aa3244992f4780a2141fbe3f42c20712b03d76
119316 F20101113_AACWNK silva_h_Page_137.jp2
7e560c9cbfad4b9403b9786ae3dfe6cb
85c2b440f9e73246625ac747c38cb8b5501a4d98
121252 F20101113_AACWMX silva_h_Page_119.jp2
f886bd22c4c349733fe109aae973f1cb
dcc3d8d2aadb72d849e7a71fcfcfff936c15f71c
123866 F20101113_AACWNL silva_h_Page_138.jp2
9c69da13118615798a9ef8ff16efd5ac
859135112652ed96ec1c03d1676eb420e802c08e
141671 F20101113_AACWMY silva_h_Page_122.jp2
a53d0dc4e2dc1f05dbc4499be7797d95
ea17e9793a8c168e10e587b28d5c4dae40118c3e
117084 F20101113_AACWOA silva_h_Page_156.jp2
778cec724ad211cbcf95d736c22d157a
2606a5114d4500b1b0516956c89a1e76823d300f
114216 F20101113_AACWNM silva_h_Page_139.jp2
88d3a83f60578a8886ff4c31959bc20a
ebe99f64bcf1914f73b706e66a8c4e6f19fd8f95
137482 F20101113_AACWMZ silva_h_Page_124.jp2
b19181e6b76dfd24826f40a0089efc98
92f6d83bfec272257e3a1177d62e8e49aa563206
111205 F20101113_AACWOB silva_h_Page_158.jp2
f39317f7b1f5bfdb3294224eb6a1dbe4
8376035f5b3006a4a6cacd9525cdebeefb2a07ba
121258 F20101113_AACWNN silva_h_Page_140.jp2
1149ac1b2e1fa3a9eb03a68855800d0d
77d07c86938366aa1c9fc60d4e5e5355ab28553f
117242 F20101113_AACWOC silva_h_Page_159.jp2
2a9369a0eeec6fcac729253a991fa054
34eef5931ec5bd343b8f20ff92301557d5a84893
96642 F20101113_AACWNO silva_h_Page_141.jp2
a09914bdf6b8a62b9e98bc9624171cba
67e92e4f7b8cc4cfb7ca84e2d9d407451b96d73f
7397 F20101113_AACWOD silva_h_Page_161.jp2
2f6458f7a2304b441d39bd274f961b65
5c0d7a4cd95fbe0575a68f4494be4c41d5ad82fb
129607 F20101113_AACWNP silva_h_Page_142.jp2
2e435816f0055c03617fcc7afac5838d
a396c672f3004122f4e015cb731ee430d260da38
130950 F20101113_AACWOE silva_h_Page_166.jp2
aec2ea5698bc14be87b31ddd9e852eb7
9c03fd0c253253bb47dc7e8338f74973f585ccc0
110042 F20101113_AACWNQ silva_h_Page_143.jp2
8ce46dc9d3fbe851c15baa5997a47f37
3283a2bb8527d99659203855bed13fb21545c2a7
133749 F20101113_AACWOF silva_h_Page_169.jp2
e217b34751359aae0b9a4788eab8ab42
3a734c7812d5a4e0051746d7424cd8e9f78e6ca4
120671 F20101113_AACWNR silva_h_Page_144.jp2
cf069f93dd8f0aa24cc1ce47720fa2c0
85733199859ce157c8f6387fd77fec9b9f281190
133585 F20101113_AACWOG silva_h_Page_170.jp2
27b16257d76c1b6d9d88c4251930b605
4194ce4d4fbcf866c8019ef5305aba4da2438ce3
78491 F20101113_AACWNS silva_h_Page_147.jp2
87278377152febb24903144e9ba589cc
7490d33b2dcc4c7210a7ba612c75e71a0af6b4e2
138104 F20101113_AACWOH silva_h_Page_171.jp2
8c1bcde53916357c4a7193a208b5a63b
8feb489a2b8f2a3f05f4b7670a705cb5978727d1
188635 F20101113_AACWNT silva_h_Page_148.jp2
7f741547106d44bbed6e8d4e325f6981
0b6e570194c43a73ad520036eda2db7a9704fb07
55783 F20101113_AACWOI silva_h_Page_173.jp2
6a27abd976717235c0593dfa1a0268be
4a6172682d3b081e0de5c96788b9d1637ccaa5f7
92202 F20101113_AACWNU silva_h_Page_149.jp2
3be3595c598639a66559d939c2710bb9
6175d0640bf7df8312c85ea1af120593297e5a5a
1053954 F20101113_AACWOJ silva_h_Page_001.tif
9c8350e26421e040d3dac6b890b7b35c
d9918df102c03dc1745e2bd9748a9e0fbcbb3f6d
33007 F20101113_AACWNV silva_h_Page_150.jp2
4424d210347dfa1ed1e4f21ba6d43b73
65efc25a007e483989346f40d457dc318864d37d
F20101113_AACWOK silva_h_Page_002.tif
206e1b5f0574d58871b1419a9e2bcf37
88bf86275b1f5b8af28d44f0a1f85e289413a6d4
101400 F20101113_AACWNW silva_h_Page_151.jp2
89b502f472869169b21a1f9c4d791b9f
d5c9501cf0a8e5a86f1c38ac90ac0c5ffc6f5260
F20101113_AACWPA silva_h_Page_023.tif
bdeae527cf6372ea6771df3520744a07
12d31f96ccba3d74332735964cef344aedd64c15
F20101113_AACWOL silva_h_Page_004.tif
e636df2d70f2cc76f0a4121ec3714e2f
89ad97c4e1a3abdfb73077a8508f2b3cc712ea0a
118298 F20101113_AACWNX silva_h_Page_152.jp2
7863760e5ebf7bf040d00bd7283a5d7b
aa40271f563f7cad34addb967d236ba58d90c60f
25271604 F20101113_AACWOM silva_h_Page_005.tif
985059ab1166bf4d2d4da027252be61a
f05488c4014c662c54e66d1ce8572969af70f5ca
117922 F20101113_AACWNY silva_h_Page_154.jp2
85a1fdf1faa613f7e22e6c356e811a79
3b2b29559eb06b3afa8e8ba0851a1de917add1c4
F20101113_AACWPB silva_h_Page_024.tif
dc08d141eed0308cd4fac99d33238c54
72c4213fa32b6d9c9a41f48348e6673c2b1b09ca
F20101113_AACWON silva_h_Page_007.tif
6220b11776023044278a806df7eb6350
74745d4dc97f4863bc7494fe4dc0834383d3860d
114428 F20101113_AACWNZ silva_h_Page_155.jp2
ef7018dbfd93a074f8c3647bb870d476
f1d3bd86444380727ed51cc7524d3b9084f38009
F20101113_AACWPC silva_h_Page_025.tif
22ac2dd724ee4c25fd8109cb35858541
f1aca297c00386bb6589364c3ce464bb9f2a40fe
F20101113_AACWOO silva_h_Page_008.tif
63afac3bab44d9f1f78beb79b6466a08
2c199428d8c74b52c3fad3288aa73c18f0ca41c2
F20101113_AACWPD silva_h_Page_026.tif
fbe655ae3bcadd89f44db1fec1e9d7eb
a763e53cfaf41f76c7465365b0cabbbdaa9ede53
F20101113_AACWOP silva_h_Page_009.tif
30d3ed14ad852e4b8be46be77467c099
5a06d68ada6b4f3c9374e9a230136b814b740332
F20101113_AACWPE silva_h_Page_027.tif
eadf0a1b5f0f2177ffb9edf0b4963e22
60c6a40710f14176e7fcec32e15c52a743a35c77
F20101113_AACWOQ silva_h_Page_010.tif
8b458dc4e2a8f779b2394f77d5fb666c
66694424dae4f6df219a6cf97408fc2e7f567c65
F20101113_AACWPF silva_h_Page_028.tif
f777c199a0cf3fa5af73b841cea15ece
06ecd2a152d8def7f649b4bca475feb313688f0f
F20101113_AACWOR silva_h_Page_011.tif
18074c49bfa117b9070623c939940b67
65cbd2a456a32b191fc68b93a38585a7aba481de
F20101113_AACWPG silva_h_Page_029.tif
d7e5e27ec4744f976dc8c170fb5c7416
a1421f75e68b20dfda7e9b953e3f359f117a02ba
F20101113_AACWOS silva_h_Page_012.tif
3224ec05a063132ee78019b8be88b394
9a251ec1aa0ad74695fcaabfd38c60680a52359d
F20101113_AACWPH silva_h_Page_032.tif
f9582986f70063e0405d5ac3d154a4f2
e2ce3a9f78599ca0267d0abc9ca7ec390013fd66
F20101113_AACWOT silva_h_Page_015.tif
d8151fbef30685d87c18748cfa8b5068
7e7c04777bbcaa6af6e504831491868c7dfe6553
F20101113_AACWPI silva_h_Page_035.tif
d0a90054dbdf9c7712a23cfe85fc4f42
faac4fc03956ad02af94e929b73d12db25cc06a0
F20101113_AACWOU silva_h_Page_017.tif
5d686e3aa286cca4bca4a42531620d5a
04c4203e48f09b97061336888a2fa86414b4ea4a
F20101113_AACWPJ silva_h_Page_036.tif
ea2aa9e07a46abcaf9ff0fac161a31c5
5e854a580210fbbab832669ec815e32e4302943e
F20101113_AACWOV silva_h_Page_018.tif
5596bcc69687e3f3d1184b997a1e3ffe
731dce9ab1edd77ddfb963a1320c0a199f7c1092
F20101113_AACWPK silva_h_Page_038.tif
ff2f0a4a3a39a59d3ae1b730daecba7f
8d71baf8162b5ccfe539abe6c71601870e4632a3
F20101113_AACWOW silva_h_Page_019.tif
5c8f9e3754487bc2b5261c73500f5288
10b0c2ec9b187be953545007764efecb1e7c4bf3
F20101113_AACWOX silva_h_Page_020.tif
23a3529cfef57a62735139c24bda5d2e
05a314582dcc507497db92014b0ca333684647c2
1054428 F20101113_AACWQA silva_h_Page_059.tif
689c305b8e7a1ecebf062faf5ed9aeee
a83534e85a4aaee5ca554c1b91b8cec9fd353e3c
F20101113_AACWPL silva_h_Page_039.tif
003d66bb1e038e6505ae0afaa495215c
236cdaa0a7573194b73ee28ba32f845092934cb8
F20101113_AACWOY silva_h_Page_021.tif
e9f81d2e7b15c7004a1bc3760515020d
5601e7f15e4afc86429f5341fc02f8a3608a7a09
F20101113_AACWQB silva_h_Page_060.tif
c90c5a5ab3a57059be226e7b6ad96995
b4295779d95ed88e39c68214cea6c6b2b3d0e20b
F20101113_AACWPM silva_h_Page_040.tif
3328c7d68aa0f3954dc55bfa89e0e975
a3bb3034ea1708eea72614981b38f782b1657a7a
F20101113_AACWOZ silva_h_Page_022.tif
fafaf996290a195459ac7adcbb440b95
2d3bf792678ed2bb835de1531a7e114f1e70c13c
F20101113_AACWPN silva_h_Page_042.tif
5e4002787576be9bf54e81861ea121a0
2b1f41974455abaad53ebdce4e969832ffd2c5a4
F20101113_AACWQC silva_h_Page_061.tif
de7712e72122c378c1a43c50006aa803
d6dfdd3de5749baa05c2b2cd435083dcf6c121db
F20101113_AACWPO silva_h_Page_043.tif
a8076b7a512fe606a4e423fa84aaad9a
bf889a88bc3d2ff63a44bb7abf2f22995df9022e
F20101113_AACWQD silva_h_Page_063.tif
7286dfac0b764efe40f0f881cce76de7
1597f5d62c386ee417997a7ce41e82d368792db7
F20101113_AACWPP silva_h_Page_044.tif
254f507470a6a3c2560a32058cac9bc2
8a698cb4c5a21b30a19cfc707d096c8881d69f11
F20101113_AACWQE silva_h_Page_064.tif
22c070cf55ed8fce5de41f66c7468e80
93ece184fb344cae8bc48ba3d3f818af2377b873
8423998 F20101113_AACWPQ silva_h_Page_046.tif
0ab96a8c79c6705674749e4822b53a93
6f6c88bc4368cd8c5acbc7e219d9ee424b0196f4
F20101113_AACWQF silva_h_Page_065.tif
32755253e15b14764e4a27b380b4fbdf
6d0b749d66b806fedbec50d23b875c90e09fe5a3
F20101113_AACWPR silva_h_Page_047.tif
4a4f3373f4acaefc0bb96927fbc541e6
ec376daa71069e8fa02a88d713cf02d7823a7284
F20101113_AACWQG silva_h_Page_067.tif
23cfe5b5eea46e7f7806bf3506ee205a
4f594ddc0ec1cfd51b4cf361004cb0f7ea4237c0
F20101113_AACWPS silva_h_Page_048.tif
014481e725c5870d189471dffc04574f
8393ab1c6e741deebe3817bd214ec6efbc30fdef
F20101113_AACWQH silva_h_Page_069.tif
3fef3ee2b828ee817f97227962836501
74b28b27e2e08359d7fdc6e18d5cf29fce53cc4d
F20101113_AACWPT silva_h_Page_049.tif
448269a399ad83af0d6e0f7cb6b850c0
12841211e60790718ad1105bcc7505a137247489
F20101113_AACWQI silva_h_Page_070.tif
0ad268ab552800c158dc59a312a14214
10c023103c5311c4ca836a354c3dff3f67708c67
F20101113_AACWPU silva_h_Page_051.tif
a37f203a971f63cdc78369cbb34f73dd
90a6a145afc9ff6a2d9867fb5ca792188bf7133f
F20101113_AACWQJ silva_h_Page_071.tif
b9a5cc62e0b0eda3348b4dddb40d64cc
121ad559a3d9266576be47b80ab5d980c2b987b9
F20101113_AACWPV silva_h_Page_054.tif
ede687d25caa9d438a862a9173f49d62
f1f256a812df808304881e47f635871181b33561
F20101113_AACWQK silva_h_Page_073.tif
a22783288c0dde196e815dbf1b114796
aa4447d991b6f21120b354184ede76c7b3651036
F20101113_AACWPW silva_h_Page_055.tif
6c006e367d5bff33b1819cc31cde1edb
0c0d396ec4553340b0429829d9079817a8e494aa
F20101113_AACWRA silva_h_Page_096.tif
7d8a7a700636ec0f1dab4b3a6ef0e4c4
5fcc802c66d7b0cb4b1054ca9e142ab9913e290e
F20101113_AACWQL silva_h_Page_074.tif
b073e7b5c2c0d76ce757340aa89be25e
dc2cfc4ad483816b74222b08de1f1fb1f37c0813
F20101113_AACWPX silva_h_Page_056.tif
56388b8e3529fad32a9936a7196f903e
13d1c3880f5922c0fc427f6cb6a008d062529c8a
F20101113_AACWRB silva_h_Page_097.tif
5d5f733aa18c8430d3a399e63efb54b2
31ba192e4b7fa392b7dfc9514d3794d3f5e8ac3c
F20101113_AACWQM silva_h_Page_075.tif
45699db68035b8032b3e6f8186c0827a
b83bb4fd2951d9a7565826a340a6d4fabea6b95e
F20101113_AACWPY silva_h_Page_057.tif
b2532128788f29949b1eb3da3012864b
cdc09f3f19da37e191662ce692c7d4d2c1d14e81
F20101113_AACWRC silva_h_Page_098.tif
f5794b77603559de739b4efa30561db8
944e2a7325e7822e37bc7bd29867a7c60948985d
F20101113_AACWQN silva_h_Page_076.tif
7b7f39ed309b96034027469be8ba17a8
9d4c76f569a61483acd7ecd6ddba1db1bc6f1129
F20101113_AACWPZ silva_h_Page_058.tif
75436afce368299da68581a16177e4f0
06272305d4a627d9169f12d750f9ab346d8a313f
F20101113_AACWQO silva_h_Page_077.tif
51dd8fd63a6d5f407bdecb6e7894ee5c
c1a3807367b278e7481e72fe2bdfcddbbb9799b9
F20101113_AACWRD silva_h_Page_099.tif
966ba9d6df2ebb463db4f107cb180d0f
59e22def4dfd0fd32f5b6191b8d6b197e0b5a7a1
F20101113_AACWQP silva_h_Page_078.tif
c7d4ff886edd0000567ab7479159bec3
d26f067bbde214adf331e7414f0304cf3fd44d49
F20101113_AACWRE silva_h_Page_100.tif
78f8a757391e67414df3771ce8df8b9d
77ee0be84138a17ea566dfbe03ce42bc0db38624
F20101113_AACWQQ silva_h_Page_079.tif
11f3fcbf4810fe6f4fc1c473ac49ad8b
913a5362de9852112fc421cc9cb40567afcbffe0
F20101113_AACWRF silva_h_Page_101.tif
b7f9d43efa2202925deaff57d309cea5
caddab2cba27bcfeda8d25b2dd95c73bd4abf920
F20101113_AACWQR silva_h_Page_085.tif
1763cb2bd9028d666e45d7578ff659a6
92db3d0200afa6740be53b70c685838070794270
F20101113_AACWRG silva_h_Page_102.tif
75e23b5872e53bcd7e5403644c3e44f7
1c35120fa75854fa0201971a9f58fef9572b59be
F20101113_AACWQS silva_h_Page_086.tif
3a656d98f4f07f0a0eee9644ad0afd4a
0e0b70b8e5182a4c0fd3a9d3a06b7bc6639e982f
F20101113_AACWRH silva_h_Page_103.tif
b60d7cefdfd90583f8330d28eef7c227
e888455b358b7b9226d85050dfd1d1047c7bc3ce
F20101113_AACWQT silva_h_Page_087.tif
5aa62a49c54579fb35f759130d051bdf
9bcdc3caacf4b963662f32a4307f87d724d4a346
F20101113_AACWRI silva_h_Page_104.tif
d5511f4ca453913aeff367ecedde3be4
2719156544c2a5089a3a27f4dea793b075fd9f6d
F20101113_AACWQU silva_h_Page_088.tif
e4da197e531fa24f9a08e6dad6f00d51
d082582640cccea7e6c1fe7d62824b7e851cb339
F20101113_AACWRJ silva_h_Page_106.tif
75596706f6a38b62a542093ec1efa299
a399462464c0d5857c64e860cebd7e27f39e5127
F20101113_AACWQV silva_h_Page_089.tif
5b601368056940def2eb98ab2bd54e50
ad7ec79195a9184b29a29f9729a211e2ed51fee2
F20101113_AACWRK silva_h_Page_107.tif
24f9ab170ef55a7a615bd6c2b5304886
b1472928e66038e6833d0949b5d7f7ece199ac2c
F20101113_AACWQW silva_h_Page_090.tif
560c6f3bed0e7ee35fe9387d31718128
fd0e1cca513ad4cc4b94f158ca99a7e85228491b
F20101113_AACWRL silva_h_Page_108.tif
da8e75ce2ecec18fed97c63f572973c8
3e1635d03649ffb53f0aea4b1d7cfbe59284a584
F20101113_AACWQX silva_h_Page_093.tif
689c7ecfeca999929259ef023b3cf539
7f5b6407774384dd0dd814288965a5b4a36005aa
F20101113_AACWSA silva_h_Page_128.tif
13bc3f7c1ed892c2f4e7f1e95b5ce69b
eacf8f7ba36b24d9097fece9b9fc1b3de33d785e
F20101113_AACWRM silva_h_Page_109.tif
9cbe1fdd41b12b60905ac5fa7fbbdab9
60029d7ddd8389c9877d0d9505f09e1d5dde31a3
F20101113_AACWQY silva_h_Page_094.tif
1f85dadcb8371bb0a8a68c5823cf2c06
324bb6d69db2dda2fdf1723ba2780643711b8b2d
F20101113_AACWSB silva_h_Page_129.tif
8ed8016673a699d1eabccc0282d36f93
cefd222ed6383d0721730a3a266af6afe221a70d
F20101113_AACWRN silva_h_Page_110.tif
04507aaa601d324ad065fd76d1d2dae5
9f15c610e406c98b638ffb4150b1102ee002bef9
F20101113_AACWQZ silva_h_Page_095.tif
e0a2daf46af6fe2795500a85c3603878
46595a9de027ccf50616c2390baffa2d2ddafe3f
F20101113_AACWSC silva_h_Page_130.tif
86f5f86c7e35db2efbcae408dca8e5b3
20ae3d9de1a8414470de0cc708f6f3fd396c9d27
F20101113_AACWRO silva_h_Page_112.tif
0a37fd65929f4d5d54947c40430922a3
f5ecffe70db76a8ce5fdae4923653f3fa3b956d9
F20101113_AACWSD silva_h_Page_132.tif
1670a4a7e24f29ee15429a8fbbafde4e
01f08ce1c2e2bfb1f11775ae9a9c8cfa400ca902
F20101113_AACWRP silva_h_Page_113.tif
3e1ad8a9ce533bdf2e001ca8d8705902
52c285ceee0e5281484da6be3bc650cd88c8ab57
F20101113_AACWRQ silva_h_Page_114.tif
0e36028e2e4f521a0ff7cc9d34a6f048
654c35484d9c2e86c91dd157764bb64105345501
F20101113_AACWSE silva_h_Page_133.tif
eb04148e431e8f33f6afdca9dd3d84b6
8079f65eb34fcaca0ebe7a5f128a93f6df485082
F20101113_AACWRR silva_h_Page_117.tif
db7a73d3a54c0aa63a9bc0b4ea082f86
56377dc2b3ce5b01d0aaf4fab6ead56fcfc452b7
F20101113_AACWSF silva_h_Page_134.tif
e076f748e646666484770c3c79961a94
6174fcaa64ee32b1a117f0badcb346a60fa475b8
F20101113_AACWRS silva_h_Page_118.tif
ac878809dbe37ec04a2656a3ff25131a
3e40662b7fb1cb4e137f556c6dd5ee413db7f54f
F20101113_AACWSG silva_h_Page_135.tif
b2876598d1101352cf79956f945d7a32
405911baa8e5f9fd8e39c9526770677936978850
F20101113_AACWRT silva_h_Page_119.tif
45234f36743f5e3f460fbd17b2089625
e0ff8ca33e0e87cda906b037d7ea9e7bfd82a781
F20101113_AACWSH silva_h_Page_136.tif
12c7171d2782e0e5675559cc9ac76ac5
05694b4a73d58cdafa65c2dc6d0b34f6cd585dcc
F20101113_AACWRU silva_h_Page_120.tif
5c86449faecdab4c428bb4528d90a118
53458562fda60fc090d6afc8495fd46775a8c0a8
F20101113_AACWSI silva_h_Page_137.tif
1c6baa6e0848763b63b0eaed1f16a401
f52951f34428a6e8504380338617a039a7eca84d
F20101113_AACWRV silva_h_Page_121.tif
9902c5e835e868ffb33d5e258972bf1f
a5fca3b43e84e6f825da420dbdd156a404e2c94c
F20101113_AACWSJ silva_h_Page_138.tif
f61f73b5aea35fc3d9413d6ab615ec5d
ccc9609760ee48c3a16015de3b2a58656be1734c
F20101113_AACWRW silva_h_Page_122.tif
421b8578eb5a033e66f555ea3ffeeb45
e52f6143351c6cdb90620a2149c7e26c330382f5
F20101113_AACWSK silva_h_Page_139.tif
48fc865633e37ec4e3a2984602e00635
9dff3aa1b7cfaeb93a6c43ea86f30e0639955240
F20101113_AACWRX silva_h_Page_124.tif
72cdc46c924ac8bc5d56b31a7f2d702a
db7a42708d399f5e49c37c507d10e8e37573bd23
F20101113_AACWTA silva_h_Page_157.tif
6a5f5e8e8ac3679a1dc8ef6f5c4a2e1b
79139238e3b3a11b3006ffe1bbeb39fbebf6f829
F20101113_AACWSL silva_h_Page_140.tif
b4f5cb09b8c8264d7f752295c9655145
c49fd6e55290e525fbe5e512c93497fe9e61ff58
F20101113_AACWRY silva_h_Page_125.tif
b3eb2bc274ba10a5fa8d93c7adda2eb5
e6c44a468fce76f29e85843a7cd801b6b0598da6
F20101113_AACWTB silva_h_Page_158.tif
651884650dd4f35a60e3dd61914946b0
6f2ad90b0302ad5cb55563301096545dd8e45515
F20101113_AACWSM silva_h_Page_141.tif
d91cf07214fc44b5eb947080586ece59
0fdf2ca71c02313b34e1aa1da470fd5ff7149d99
F20101113_AACWRZ silva_h_Page_126.tif
8c9f433de57fb4db0669fa28fb5a76f1
001f03bd00c77fd92fd07da5ed8a82172e06f9b1
F20101113_AACWTC silva_h_Page_159.tif
6eb78ae4543b969c1ef685e44d4fa334
e771fb4ef61e9704650ffac42355630fe3fe34fc
F20101113_AACWSN silva_h_Page_142.tif
aac5b9d748a92a40bbb0074d77a0a76d
7f16a14100c57b3fd2f2c41603be4444863b03cd
F20101113_AACWTD silva_h_Page_160.tif
0bfca6bf4b7f4554ec0b2f6232e6721d
d8f6b3aa385a3a28c124cce411d24bb1cae0405c
F20101113_AACWSO silva_h_Page_143.tif
29508d344d48d020b75009aa9d603a6a
71fcfc525fba7b0b306d4aff8707b52f216eedce
F20101113_AACWTE silva_h_Page_162.tif
9b5549b983b4183e2d4272fdbc7492c6
b27822c110c1744c2aa0e2dbd178084a39e860ea
F20101113_AACWSP silva_h_Page_144.tif
89121d468b2b966174580ac0a485e019
b554a9e476666b62bd506cfed8116e0ae219fe4d
F20101113_AACWSQ silva_h_Page_145.tif
a1cbb283c2b87c89637b6cd90b00a70d
d8d15bfff76eae3496e6f366db45dc7bcf2ecdec
F20101113_AACWTF silva_h_Page_164.tif
393aac9216c6ddef61cf7fca3f78b4bf
a77fc2310a9c275daca22a4069a2cf95cf5152f5
F20101113_AACWSR silva_h_Page_146.tif
f82b38b55c550e05bf4ef5984dc856cc
c518b8504e08b40e24da4b3223c328997abd1de8
F20101113_AACWTG silva_h_Page_166.tif
963d3cb1acbd93e74f56480698fb0e6f
8b023cf4db6593d87d5e0a1b3a2852558ce33b5b
F20101113_AACWSS silva_h_Page_147.tif
6a3f0c5c811957aa62e6cfa3cadec9bb
b9d3503b85b04fda48afcfaa6ee5e5abb8568913
F20101113_AACWTH silva_h_Page_168.tif
d8dfb0daf9fd6e27b3c9777b1f6878df
8487b54c46ec9ea7dc4d9bee8669e7059fdffca3
F20101113_AACWST silva_h_Page_149.tif
62b2a79585628b897e311053bb19a0bf
82942b8761973571c198e3a75eda780f9f06d056
F20101113_AACWTI silva_h_Page_170.tif
5f98b17f4c9311fcdf13d87e8b5c29e1
dab168bb4b322b30474f7b9c6f4adfec3194e55f
F20101113_AACWSU silva_h_Page_150.tif
d7456cf7862423fa23d8102ddff9f167
1c4180511ed1a3f50eb69d9c17e50107ad95781c
F20101113_AACWTJ silva_h_Page_171.tif
1f4ccca98b5424c58c8a514195db308e
6c2f5d061d0be90ea805a29a82e1d8364da373ee
F20101113_AACWSV silva_h_Page_151.tif
b8815c96f61ffefe752f34683540eea0
e00577180d2038fb17e83a7d16b167823c6b0547
F20101113_AACWTK silva_h_Page_173.tif
7dff56fc7d0724b7ee1735e307485498
b2777455494ab732944088eb83f18b95dedc19eb
F20101113_AACWSW silva_h_Page_152.tif
6e1eccac207491cc0bb78dcc9706df39
8841671f96ce7134b1106473828665cef7ab5d24
52462 F20101113_AACWUA silva_h_Page_027.pro
4eac7bc075706b4f81cfd7ead0c1862d
560cbfc160813dcface21552c592c684d679bd6c
7718 F20101113_AACWTL silva_h_Page_001.pro
0d617cdf31eca5c131425e7ce41abae1
30ef7e581f9c153cd2534a2c482a561327db8491
F20101113_AACWSX silva_h_Page_153.tif
9f206153e551a5cb78364729929716c0
d4ad07d11a5bdcb74c2572345e1c918e5a7d6edc
53113 F20101113_AACWUB silva_h_Page_028.pro
ec2a4411753b6f9d6aa48af7ae38d102
df1bd55a5569ae85e3a90556a58c6ee526058cc2
1223 F20101113_AACWTM silva_h_Page_002.pro
83913f68c93717ee1229ae432332a431
1b66a77061c3cce47420e8b5e5abbd2aff08ed5c
F20101113_AACWSY silva_h_Page_154.tif
9f70bbcbfea47d567fb7bd128d96fc73
b712633e80f4c8a8b04cf5ba5bd435a64a0f1af7
64814 F20101113_AACWUC silva_h_Page_029.pro
eaf568668bb9e0948bfde66b2700e022
be943b73e53a99342a4d66fd63e086fa60113a0f
38587 F20101113_AACWTN silva_h_Page_006.pro
e1ec6dc25f209a08590e9d4aa4dc67cb
9236e42cfac0fae1ae2318bf5214ded25533d58e
F20101113_AACWSZ silva_h_Page_155.tif
ee1882ec647486269566f4822e6c06ab
b09d653dbc097a164a84104c66288723a61282fe
60766 F20101113_AACWUD silva_h_Page_030.pro
69c052bdbd08d02a73efa1c5073e430c
232ad58567ba66c3921bab7ec042eccd8324c020
68456 F20101113_AACWTO silva_h_Page_007.pro
32ec409598836aba8b9b8809e173c8b7
2ad6b9fe14fcbb727c726435da6c1238c4935424
62481 F20101113_AACWUE silva_h_Page_033.pro
a791f492c3e478afef47e5fc330ad5df
0478f9fa015d710b10050b4f20bade2f85813e7a
44807 F20101113_AACWTP silva_h_Page_009.pro
1b8473928484e0601fe0acacde8921a2
12e5ce9a065abaad9d5382ee21d3931fd43b8874
79955 F20101113_AACWUF silva_h_Page_034.pro
fdf839e329c7ac81caaafad2e546b8f3
02a0dd6e62243f7dfa57963981f4681929b4a8b6
14712 F20101113_AACWTQ silva_h_Page_011.pro
3668d3460bed054e08bf229a3796532c
c1096f8862f79ae215c64ce361bed19727132729
4451 F20101113_AACXAA silva_h_Page_056.txt
105ff797b9072db62361d84442e52f21
ee87dccf312b9a5568fe45c099eed8c28e8b2766
51037 F20101113_AACWTR silva_h_Page_012.pro
2d108e4091f88cee4e462e97693977ff
44ac1b2396f915a16dd28aca08725c7705afb5ca
4392 F20101113_AACXAB silva_h_Page_057.txt
d6c73f58454b5a71688b1a5b089740ec
f87c1cc73af92c8ab2561335f15d2d15c7d174df
82474 F20101113_AACWUG silva_h_Page_035.pro
4edf87456483468b69e9b85e9dbb8d0c
bdc025cf1f9c3384ff10d2eee15c8dfb82fe3245
53650 F20101113_AACWTS silva_h_Page_014.pro
2d17d5b85d4c335c56895ffe84192bba
85d8b9e6ee68eadb5e6e46a3ce577f1a4eb38e55
7272 F20101113_AACXAC silva_h_Page_058.txt
6e207a10ac2c77c37877c4329f7b7d5e
8b14c87de613a189767b94b5543aa4dc5618e766
69439 F20101113_AACWUH silva_h_Page_036.pro
bc8bf7c45123dc52c92892ea15d03aef
5b3470ef0acf04221be28a72acdfddb22b17de76
37264 F20101113_AACWTT silva_h_Page_015.pro
2ea849301ddb295ed7d22a3ddd1f2a44
fa34a3e56963f2f9563b3bb3cad4438c07882ad3
2481 F20101113_AACXAD silva_h_Page_063.txt
6b277fde77546eef8c6e4052812417fd
9b10052dd2055856fb012bc95f59bef3df47603f
55112 F20101113_AACWTU silva_h_Page_017.pro
521af7432249db312297bfc368e16354
3d6e7d843cfe6ac05892e50fb10b499a56924821
62699 F20101113_AACWUI silva_h_Page_038.pro
b6691ab0eedd9439f33bb3abe1a32363
c7783bf6d4bdd6d7fd0df0cd35b72b15296dcaa9
1837 F20101113_AACXAE silva_h_Page_064.txt
74d3f73ab6504819d0c8df7833cdd444
ad8930553f2b92a9d1ad694a1881b83530bf72bb
78887 F20101113_AACWTV silva_h_Page_019.pro
7b4666d3f04d82ac742535257376b7e7
71d17c499be0c0ae1b8616e17714966a69873be8
65021 F20101113_AACWUJ silva_h_Page_039.pro
001aaab77e8eb9299510221f1486b7e9
4aed6cda12da3cd429e4c9081c99466c036f18be
2915 F20101113_AACXAF silva_h_Page_065.txt
d8e057d10886c7e98effc227bf6de67b
d9470ccd7877c76895efa61d271d75219fb9f2a4
76730 F20101113_AACWTW silva_h_Page_022.pro
3e4d362c8d2d831c148d57c498425dc6
04a86edfccc37b9d062eca3660b492aab53dd7d7
66864 F20101113_AACWUK silva_h_Page_040.pro
f8ce0835177b0ec8f0c13c36506185e3
4cdbd5c96e37fcc6c854c697b37ddb90de32383b
2082 F20101113_AACXAG silva_h_Page_067.txt
e34dd47a987981f02bf48e55aee503f6
9e7d5b0da80e9ed7f02aac03495f540200691e6e
52584 F20101113_AACWTX silva_h_Page_023.pro
34e2a9819b26f95e18808518fb68a439
b498da806eb7d744aedae09b882308e0754f3255
75656 F20101113_AACWVA silva_h_Page_065.pro
6b8644a6917bb8b8223cc53fc8545f6d
04d8aed271b632ada097c4dd6a4a123a9c58247e
59450 F20101113_AACWUL silva_h_Page_043.pro
b234bd95c8d8152a5043027d9c3fb016
18ae57eb219dd6463a22938b0d8c094a5d930f04
2071 F20101113_AACXAH silva_h_Page_068.txt
86f9c425896d8671a836522d538dc29a
62317b719f5dfb47f2079207b7a9b6e261897bed
55813 F20101113_AACWTY silva_h_Page_025.pro
c17c83a114fa4adb311d853ea5da793d
70cf4dd7df83aa1beefbc510c07d6df4ff0fed92
53182 F20101113_AACWVB silva_h_Page_066.pro
910f6aeb5cedc201b8d2e8d8a1b37e53
ec1c2a85267591327fc580578d012f0ba3c08606
59608 F20101113_AACWUM silva_h_Page_044.pro
cdb8b00a4373d1acfeb3bab354267c4b
a9f7af1923833c8bf08c835d16ef22164e90ec1c
2094 F20101113_AACXAI silva_h_Page_069.txt
46eb7a0be342274047c13f3d54acda5c
039bafb7557b7061868c2c284bf759a3ca98c83f
72292 F20101113_AACWTZ silva_h_Page_026.pro
cff975eb8dbce59adc5a8acf77a790ba
61ed55892f3ded2e7f4fc47e456111b8c46f82a8
58017 F20101113_AACWVC silva_h_Page_070.pro
04725473a94b3c1f0638de0fd5814182
b8e46249b0952d2689eab2279193eb44ac09b5c2
91994 F20101113_AACWUN silva_h_Page_046.pro
a8e8051dfdcd7dbf8eb4c5c9aebab42d
bbbfb070e8a117de31736ae3ad50440d733641b1
2269 F20101113_AACXAJ silva_h_Page_070.txt
c3f114d9491dda98261cb123e2e29782
08556d17ae94baf873c433e1e9b302eb88df70ef
62716 F20101113_AACWVD silva_h_Page_071.pro
56cde156eb33bb35d1da6c63f0847fa2
e68fcb12f6674cb70edcddfe6cb95f715b1d1b25
53094 F20101113_AACWUO silva_h_Page_047.pro
fb063adeeb091da1e694e57cae6ac8ac
974b22889004e3c488453d9f729facdfc4544619
1744 F20101113_AACXAK silva_h_Page_072.txt
a5cbaf366a82364148f1d3f786b0906b
ab5db4b2e528bf9ffc3e9190ab768e7e3dee6156
40767 F20101113_AACWVE silva_h_Page_072.pro
66bbe3f19c58b1f382c696419b45f94f
8cbc516e96d8c9713ea31a7300d23bd307c97df4
108998 F20101113_AACWUP silva_h_Page_048.pro
f1724b00459bc3d387a7e02a6150edff
1e4d8d85cd805ed5e351b09c5b737354e923ec07
1994 F20101113_AACXAL silva_h_Page_074.txt
5b9d2d8e7d934cfc2dcd4f130cfa2906
2f6b3bd08c2e0b39c46293e7faf4236ac6f5698c
45805 F20101113_AACWVF silva_h_Page_073.pro
39c28c5eecf7b4de8239677bd780ae8e
68239ad8ee0978b3d02fcfeb1f31c58626450c56
50330 F20101113_AACWUQ silva_h_Page_050.pro
c430e3970b0626c6f5bc8cf42a10ce8a
f72d99cb2c83ffcf86a5ec5b54dc84fb71d02762
2295 F20101113_AACXBA silva_h_Page_093.txt
d9842b5ace0421d1945c24378b859c01
652a7b717e606fb1d1ccff31338fb18cdb47e4ed
2399 F20101113_AACXAM silva_h_Page_075.txt
b0aafa418aea8033db0df09a666bd2a7
cf4f6e0f0d55b8b1f18a7ec74cc862490d5938cc
47121 F20101113_AACWVG silva_h_Page_074.pro
3816739787d2e8e848445f220e4471d6
9f010e6f8f0954c997cfd26d1daa26b02cabf3ed
36624 F20101113_AACWUR silva_h_Page_051.pro
21626e7765c815cb95563bf63745e5ac
2816f75e6efbba2d360f6dc98f0b558103002c4b
3267 F20101113_AACXBB silva_h_Page_094.txt
7e737a59f9daaa27831b2aee516cd2f7
36f062f8899bad24f637e418b2074a7962a91b46
2182 F20101113_AACXAN silva_h_Page_076.txt
2ef76ac60573e72e1a97e35c9188998a
8b370ea3da0613f7dab3f8fc8145a2d32eaaffba
66965 F20101113_AACWUS silva_h_Page_053.pro
56f25f161fe19a17578b52e683c3eadf
1d970e3f3630a5997db53cb654ca6a2efe451c8b
3111 F20101113_AACXBC silva_h_Page_095.txt
bfc28398db589f5ead3cd6ab5b2479aa
9b4bb68768cb0e490f723244c7b26e6f2df2186c
2680 F20101113_AACXAO silva_h_Page_077.txt
74d353c0c26093ec25797111ec721a1f
203029c056a298c454b11135193635049d527489
54495 F20101113_AACWVH silva_h_Page_076.pro
c1fe0f1571e24a86db14b26126ab4517
d3461634c47a9f49fe142e1318042e6e42e81364
67338 F20101113_AACWUT silva_h_Page_054.pro
9b00aad7d76fd06eb05944f46c6e91f9
c073aa8ff1951b4fb43aebd1839aec7463143c2f
2639 F20101113_AACXBD silva_h_Page_096.txt
48d31979663ee80cb31d86bb8e998eb4
d5f5fc17453d9f24c02186f788ad922a166fdd8c
2325 F20101113_AACXAP silva_h_Page_079.txt
f8820f4605c3d3480360c6dd6ee3c89d
ab5c8d0676c2f323be3211b292181195cd98cb97
59851 F20101113_AACWVI silva_h_Page_078.pro
68ada88cfc08079957444dc20d76d1a1
bd0cb4c64507e5d46a25e246fb47b930eb2b787e
84780 F20101113_AACWUU silva_h_Page_056.pro
80a864227e2c391e28cb2f2d45b9b5ad
d5030edb879ed7875cff1a53388525b77f1cc1bf
2387 F20101113_AACXBE silva_h_Page_097.txt
0bd5b54701e2ad6883e7aaa78b153513
1bf5364213ff75f5f277e835912d37b4dddfadf0
2514 F20101113_AACXAQ silva_h_Page_081.txt
4e2b246e73cdf8d955a0e636664ef053
9da96b7e279ea0e27bbc83431ab4ee05a8e3fbae
52933 F20101113_AACWVJ silva_h_Page_079.pro
a4884273b18a0101508bfd7d3f857833
62cd54e4e7878ea22ea901811d6612addd6085ba
92665 F20101113_AACWUV silva_h_Page_057.pro
04a7c4f60e28e4c6d2430262867cd0a8
f263a2ba336e6ccd918ae53a8568ee5868663338
2623 F20101113_AACXBF silva_h_Page_098.txt
d5ca7857e963f1d7a49908e78e30344c
e91034e3f2ab0bb4bf311128e9eec35203acfd9c
56137 F20101113_AACWVK silva_h_Page_080.pro
191e602def11f174231726d064e8f72f
afe1febcd79701fde00167ea78401b332d3dae3d
59142 F20101113_AACWUW silva_h_Page_059.pro
0d576c876859af3bf3f1dc21b98b954b
daf398bdf8bd430b7b6237735350c6997e07e23b
2230 F20101113_AACXBG silva_h_Page_099.txt
446749043a823a3b1e1d0f0d0632e4c8
20e6ac7309930ef84d1f1a714803829dd8d892a3
2168 F20101113_AACXAR silva_h_Page_082.txt
80d36863290c7199692ccbafa40db164
c728617b8e0c3f1f8770a048e7aa4fb4de5c13c8
61384 F20101113_AACWWA silva_h_Page_097.pro
b90828ecdafbfaf597648c2280e1d8d0
72593b95b561c3c84ff8f6fe311125c9a8753124
64005 F20101113_AACWVL silva_h_Page_081.pro
c0722041ce1c8dc798713406398e9e02
fcb5e9fb89e6c8fbb2bce9815922690fd0f827e7
48454 F20101113_AACWUX silva_h_Page_061.pro
0f1d390959328891c7d8cd51a5073857
77c97f6b5303172a1783848f0dcc031e1cd47caa
2183 F20101113_AACXBH silva_h_Page_100.txt
702fb19c030795220ce12ac73a331784
d2dff7cad2df28f32f008e36283ab35c9a43e967
2583 F20101113_AACXAS silva_h_Page_083.txt
6db2cdc8f7cec8c56323c3ab8d915e7b
69b57849608eada2955e5df81ce237dd2d5e49fd
67819 F20101113_AACWWB silva_h_Page_098.pro
a90e5ee54efe155991af4c531bda1c2a
1682c5d138e730fc2bceb3a966d0fdff65c1ea1a
55283 F20101113_AACWVM silva_h_Page_082.pro
1aaab6813dd2e3c9de8bb689d0893faf
18e356a7b0a420b625e4134ff3d9236ac6f644cd
70423 F20101113_AACWUY silva_h_Page_062.pro
ef5dedddaaba70bc1e7f5db8da89d531
d5f522079a121d9b391baab5463506afb29df4aa
1207 F20101113_AACXBI silva_h_Page_101.txt
6ed54379ac457edf187a75201f957357
9d5a88f7f128959cc0651731faf270e5c3d241aa
2590 F20101113_AACXAT silva_h_Page_084.txt
b63ec0b61895e08b6fdf248372b208ac
f28fc155e81d0b845e63eb8cfd5fb8de3d62da43
26856 F20101113_AACVTA silva_h_Page_136.QC.jpg
c84118eaafc11bd96f59c14b5ae431cf
c42eec50e7840e8dc4e34b23244c5fcb31b11902
55748 F20101113_AACWWC silva_h_Page_099.pro
212adcd4ce8536898a7b58f731b7b29d
af543881bfde88e06a5fb1b9c2468b2e5bda0f68
66908 F20101113_AACWVN silva_h_Page_083.pro
7006684612d11de5161f302051ee13ff
847a1a44eda8669b0d73e198f8428771c1575a95
45471 F20101113_AACWUZ silva_h_Page_064.pro
11ad868079042ce934b109444116308c
94559af7b1234de4e5e59451191afa9ffb3f6f32
2998 F20101113_AACXBJ silva_h_Page_102.txt
29344309638a43630c4a2f974dcdadd1
e1b58b9d3fb741bce39ee0e09f87e840ce2b88fc
2003 F20101113_AACXAU silva_h_Page_085.txt
af98a57fd0fe62c2f489494d1ce920eb
99becad6bc85311fbf6da07af34cb231cba3e879
82994 F20101113_AACVTB silva_h_Page_060.pro
d4f9280234b7c35795cb9ea0b1abc96f
fad4fd4d6a8bc45fd1ae027e6bbe4f2ec11e24b5
55304 F20101113_AACWWD silva_h_Page_100.pro
0b4ead6501bd0664a7f3d67ddbf627b9
733f6eac59602caeb296d5234b04c9acdfd286d4
65781 F20101113_AACWVO silva_h_Page_084.pro
769a9303cd1cf26b9615e20e6888dc5a
c875a319ee9d8bc9b2107e84018a464413566267
4355 F20101113_AACXBK silva_h_Page_103.txt
1cf0b2015d97f71eb077c40faea69275
ef1f7dfbc860e9da64026d8aff2e26b156254626
2147 F20101113_AACXAV silva_h_Page_086.txt
83046c128d44b40d70892a8929371782
30e038a695086273f323d8d5ff61acecacab3b71
26289 F20101113_AACVTC silva_h_Page_031.QC.jpg
f7a34d7ec42f2663e0cac0ce5df634e2
96bab45c0f0abd8f2f287e6f5b0214c3a638e9b4
30314 F20101113_AACWWE silva_h_Page_101.pro
138e8bb8b5e0de3c5447794c59cb07e7
5ba72bf2b42ba0dd0ff1281a2ad65196114beced
50610 F20101113_AACWVP silva_h_Page_085.pro
6287963687f3ea300c74830bb23e9ddf
b43c3923335fca5f3c22f0e2df333f270d4976b0
1166 F20101113_AACXBL silva_h_Page_104.txt
690c459e8897bae949e1b87fa445b3a5
9e18a11caceefd1424e23967fd9ac6f7dfe7025b
2700 F20101113_AACXAW silva_h_Page_087.txt
dc2e5773ecdc814d09fdfd979dba7d3d
0fa1bfb9b11cac4c709852adcaea92f2fbc9a31c
1641 F20101113_AACVTD silva_h_Page_003.pro
f4f8edf8eb360e34f58aac1550cf8f03
936465dee4a3a50812a7f3c7af2107102a16e43a
64070 F20101113_AACWWF silva_h_Page_102.pro
ecb804f8ff243b74dff583f407a29202
ccb014afeeab16f97f726446beae580688849156
69793 F20101113_AACWVQ silva_h_Page_087.pro
565ee77383b2c5feb69ffd1937b16c0a
ea3c8ee04c76be6d9c1bfe677fcb1f61ace5e1d7
2035 F20101113_AACXCA silva_h_Page_123.txt
42d98cb8249f525178f0a6c531362af4
7dc61468619e72c3fef33e617fdaa5be83768cb3
3513 F20101113_AACXBM silva_h_Page_105.txt
7b1e7b4bba780619d312c053d3e98900
8f7004cfda6d6fa11724f239f2def9264dc7f73d
2167 F20101113_AACXAX silva_h_Page_088.txt
5e98720f37f4a7a04e3e31f6913e58a5
0dafac38895517b36d9833dfb057b3594de0bcd0
F20101113_AACVTE silva_h_Page_123.tif
ddb77a46812b8272d6694029552fb152
d99fc6e4a4978a7252d1cf84c9870e23f9d0d5e0
65627 F20101113_AACWWG silva_h_Page_107.pro
d93802a1403efc8e3ce189661c959191
e8445e3eac9682207259b7bbaeec235d9670fa82
55138 F20101113_AACWVR silva_h_Page_088.pro
464bcd760102169352499d8c05d3d05a
03f2d948c4bd82b12b4eadef659472b738d98267
2735 F20101113_AACXCB silva_h_Page_124.txt
bae18f4a1db1fb4485e4b71e46edb112
62f784a35475108968e881348d04ee6011886ca1
3936 F20101113_AACXBN silva_h_Page_106.txt
37978cf64fde2270b215fd2e0fef74be
c9116ab367bf929e8d90be6c3c3b50458e9f55e4
2451 F20101113_AACXAY silva_h_Page_089.txt
7e2f969871b79dba826b95af817efcdd
34c8018a9b319baa06bb2a4449209b17e19afa22
66182 F20101113_AACVTF silva_h_Page_041.pro
a006b04387f4783aaab5cc5634f02442
77b2b59eea19ef5cf9e10ba689dd49633e88f4bb
73153 F20101113_AACWWH silva_h_Page_108.pro
0ff54598ee1db144600a43521551a31d
f60f6b7ade9128f632cc99894c75ea36d0c61426
62759 F20101113_AACWVS silva_h_Page_089.pro
260a25a38b5585aeb625ef228611ff66
c13d2f4e27fda378066abae3bbbd0952992549ff
2212 F20101113_AACXCC silva_h_Page_125.txt
e83a390202dcfd5072bbd8ad0733a357
17682ae963e4358a87ae6a4c990cfb3b24aac4a1
4285 F20101113_AACXBO silva_h_Page_107.txt
091afc37cbd9e1d49673ee17de15d44a
d358216645b07aa6dde15b2f84eb5949bc455968
2439 F20101113_AACXAZ silva_h_Page_092.txt
739c8647210fff627ab452a635de337e
2080bfe2a880de3520b1c1c6cf6b176a805a417b
57979 F20101113_AACWVT silva_h_Page_090.pro
dc47d9ad20d59bad1672ebcc64610cdb
a85c16544ddc4708019ca1e108f408a317518b6a
2790 F20101113_AACXCD silva_h_Page_126.txt
795d8703856639dfe51d7124cd2b0a60
731348e9f84714f43b3d45e4666e8588060d0189
3212 F20101113_AACXBP silva_h_Page_108.txt
4df4802ed9886bb3d9034e058a33e064
467909de567a720238a791ead3e7957cc89814e3
2354 F20101113_AACVTG silva_h_Page_043.txt
fd303266cb1a9eaf9c4d11003fbccada
6ea3dd9acb84d89e596198ae64bf6bb092698fce
57597 F20101113_AACWWI silva_h_Page_110.pro
21d9e4cd2d22981f5a3b06acf2e85084
f5f89a40880e8fd464e7074772977d87bd5ef421
60466 F20101113_AACWVU silva_h_Page_091.pro
358f29f76f84967331fd50d0e0b784a9
50361c41c84b0458cf225589e9e89f873675beac
2263 F20101113_AACXCE silva_h_Page_127.txt
ef0fd789a564ea635362a9e61809a369
35447ce1df72e3866dbecabba569ac3426cb70a9
2443 F20101113_AACXBQ silva_h_Page_109.txt
7cdcc6d59cf6c07cda7566e5518816d1
cbbaada9504bb85acd01e4a0bb4d9f502f15e172
82125 F20101113_AACVTH silva_h_Page_170.jpg
845edbfad27eacb391f4ae05411821de
40c6252e51d8447485a66c0a94ae9c19a74db5cd
52348 F20101113_AACWWJ silva_h_Page_111.pro
697e4935382e001c73ebe44cf8a6091b
515e80c98bb6ee3fc6379dfa4c04f557140f8a9d
63067 F20101113_AACWVV silva_h_Page_092.pro
eeb290d662a589c502d5e70d63825e2c
8de9cb4b90663de82b85e207d408f2cb1dd9085e
2946 F20101113_AACXCF silva_h_Page_128.txt
dfb3065284cac693d91c96d1581ff180
751ce4dbf724eb0b0a868a232e4eb45ded5f9016
2061 F20101113_AACXBR silva_h_Page_111.txt
21472753fd543bd0837c05b96ed592e7
541a5b2a7ea5cab9c7505eb8274259a2eb114344
24964 F20101113_AACVTI silva_h_Page_092.QC.jpg
895a8465ff0943e8de013da75731434d
8c32eba5be38c43e391c1ebcbf42868a0ce514ec
54579 F20101113_AACWWK silva_h_Page_113.pro
91cddae6d2b0f6072731a266fcf3d23f
697e4da76e580ca0e25b5ce190d96b0f99cb2c1e
58902 F20101113_AACWVW silva_h_Page_093.pro
9353cf920b07c6e0b2f791cfc531beb8
d8c0b1fd7a9d4b52d590c9d2318d9219e616993f
2818 F20101113_AACXCG silva_h_Page_129.txt
d9e3ef3a52cd060d283f26b4bc855a56
8ca6e4c12c08a747899f5a24247cd2ce88f0c92a
104350 F20101113_AACVTJ silva_h_Page_121.jp2
120ace4b215eb86395480a484f6ee1f7
50338c6138a65760adf73f29400f3fee5bec5e19
59481 F20101113_AACWWL silva_h_Page_114.pro
86d95bfa84d0c73a2fce41b1b8c70f4f
b55675b0095a50ff22588fd6cfe050f0b62dc516
84644 F20101113_AACWVX silva_h_Page_094.pro
c577d90e84624d0546880e8c96a46eae
0e79d80ad075c7af2e9a71c8ee22b00df17172c7
62577 F20101113_AACWXA silva_h_Page_132.pro
86b67f5c8cbffbc7470ef74ed0c43932
3cb7f25494536b8048b08ce009044ceb7bfdf1be
1996 F20101113_AACXCH silva_h_Page_130.txt
ee6a6add5cd88b4f87c68bdd7656a999
85bcacc23790329ac3accd8fe530cf3fe2e975a7
2159 F20101113_AACXBS silva_h_Page_113.txt
9ac780b3e978af00aabda53a26c61006
dbba3af6f36f08efc37bc44e316ab4ceb52c4c00
25167 F20101113_AACVTK silva_h_Page_098.QC.jpg
60428d6880f259095127ffd781bb09e7
2711963846a6b6cdf8a93e447498daf6eeb75225
51328 F20101113_AACWWM silva_h_Page_115.pro
75a7a1d1bf14a97da3afa72208ed6163
5bb7050d80fe32b3ce2d34b643ab65d6063a97e1
24786 F20101113_AACVSW silva_h_Page_032.QC.jpg
78b173ac70c46d2ddfb51204c9a1e738
fb5f5769a576d4d693e9fe08bebe2b31e1cc2e99
80679 F20101113_AACWVY silva_h_Page_095.pro
1e278cb3190225aa8318ad4b9aa7705f
89398c1e22acd3e24109c6eb2e241996080cb9cb
69815 F20101113_AACWXB silva_h_Page_133.pro
8a3c91c995d14e5afc3243c06ef8aa12
6d8986333613aedc6519fe3d765b872b3c859edf
2090 F20101113_AACXCI silva_h_Page_131.txt
736d53510dc5c5076e3d78492ea804da
ebeeffb19fed8684e7cc7468870a33f377af94fc
2316 F20101113_AACXBT silva_h_Page_114.txt
b1b91a09a5fc2771c35ecf9f57681e56
2d67fe9ae921e0132e832bb6971cdb44e36c5f8d
70522 F20101113_AACWXC silva_h_Page_136.pro
ec48f567804e392fee8554e50050ff9d
e834ffdc4a9110ff8e11cfd70292522069298cad
1590 F20101113_AACVTL silva_h_Page_172.txt
f5548f3fc1406cc7e118868e280d16c8
cc34ac0f9c2455deb1b1aeb7de733746c3db1c99
56003 F20101113_AACWWN silva_h_Page_116.pro
07689c7cb3b85d82070538802ec4199f
fbcec5636c2ccee473f48fced7337aa6d4eb6da8
25740 F20101113_AACVSX silva_h_Page_030.QC.jpg
d7df1a5cdbf5cc68ab4701f44703772c
0643e3cb4aec9e1a811fd3d0269bbc2739f422c4
68568 F20101113_AACWVZ silva_h_Page_096.pro
70e300b3f42286815b063b3cf059d3f4
afca5c456317d5f6f0f09f90bee0fd57dd1c920d
119672 F20101113_AACVUA silva_h_Page_112.jp2
c59edb6f6921ba1a8f677084d20cd27d
2286e6441a1ff5f4f625cba6e8bff8a58e199eaf
2429 F20101113_AACXCJ silva_h_Page_132.txt
c65ca3faf89c4de909eea387c83d05b9
5e7c9ed688305ff019e3c9025a67d811e29b02f2
2196 F20101113_AACXBU silva_h_Page_116.txt
cabc544ad32f9ea44b2e6801d6b23038
76cf13f65dfbd736086e00790fc6cd87a30ed857
55366 F20101113_AACWXD silva_h_Page_137.pro
a9aa84600cfb187e9aeba017064f2a0b
ba23ef60ef508c17457650bf94a5e13ad831e7b8
2678 F20101113_AACVTM silva_h_Page_142.txt
46b3c8e39e5feae4324763b159c9d97b
c095630e3af47b7f6f8e34a2f8dc297b589021e4
56459 F20101113_AACWWO silva_h_Page_117.pro
31b47cd42561623d314e27ccdd825e9b
be393de5a3e0a711755387bff6e0847ecfbfde4e
26072 F20101113_AACVSY silva_h_Page_046.QC.jpg
97e83fea214516dc32f29abb1f3ae602
f417312cc85337ce2a8dd092c304b90ad6e21e14
69857 F20101113_AACVUB silva_h_Page_103.pro
6f8468d55af91de9459f81d6f1c467cc
6b1c3f8fdea5d21d1b3461b550586ddc21725551
2697 F20101113_AACXCK silva_h_Page_133.txt
457fd080746d784188e014692031b137
fe4b9704953b54e62cd707b9bfd58fb79ab0b518
2215 F20101113_AACXBV silva_h_Page_117.txt
24e1128dc4ab5b3f060d3ec47288eb8e
612d2eeafc1de8e8aa92b83a4f7ceec2f62b3869
58027 F20101113_AACWXE silva_h_Page_138.pro
d7b4a10859bf88e4a2e1cbfac7c7a834
7e7c5b2a73573224b6907f0431cc1bed2a9b525b
26979 F20101113_AACVTN silva_h_Page_011.jpg
56587df22795857ac9aa121617d38837
7109aa5f3fa5d839c9c4a29af27bccacee2ffc68
64765 F20101113_AACWWP silva_h_Page_118.pro
2d509f8d39e1f8b835cbb63277b212c8
85515067ca7af0b4e13c26c84cfa0a8b6d89d564
F20101113_AACVSZ silva_h_Page_167.tif
da3c8291c0d003cf17f4e74e993cd96f
7e1d9d5926ec2d583b8dea66933b50c97139bd21
7173 F20101113_AACVUC silva_h_Page_065thm.jpg
d4ba7a028343468a932fd5f8badb1faf
72501ff608038fd114ef02560ad3560796b6bf4a
3037 F20101113_AACXCL silva_h_Page_134.txt
7bff234b3d383f1a647f4ff8d7ac40a8
13855d49e49014df75ba5493b1d9b61e14bb95d8
2537 F20101113_AACXBW silva_h_Page_118.txt
d2ec4a6107c693e6dd5f44ac3451fedd
a9c067f5a6092d21cc1a8f53df9d27dd92776f88
52949 F20101113_AACWXF silva_h_Page_139.pro
3b5d808903c2ebd1c005d6d32bc8c2b0
7ed9d5f78e1376190d8b0a01cc5440874559b038
126545 F20101113_AACVTO silva_h_Page_063.jp2
c43d3c54b6806bc7003939a4d08296a1
ea87510ea598a7727fe118f9053c2f945962d58a
58968 F20101113_AACWWQ silva_h_Page_119.pro
2c1ef012d9e380b85a3044b3251d814a
bf7f60ee102b6ae10bb34845be5b5a7f4b1a6509
46667 F20101113_AACVUD silva_h_Page_141.pro
f9b05b8e8bb0d91b61e40d88822dd864
0e09c40d5e6c7ca190c4cac444f8bf34a9d61b25
2091 F20101113_AACXDA silva_h_Page_155.txt
f117d9837124ab4d8fc506f43b034847
f13c5bb980f1f34e1818bb31771ccbf0ffc59178
F20101113_AACXCM silva_h_Page_135.txt
e6449009a4509d15c20519fc6ab1216f
2009371cd55354ca5433f043bee19d68baa41890
2309 F20101113_AACXBX silva_h_Page_119.txt
ce7fc1a1c6195fff7a09e551c0a9840e
f11a50a0a020cc553f1c040cbd26d36d96522627
56682 F20101113_AACWXG silva_h_Page_140.pro
5df8fa94c0ee78f2be265decd34689ac
0894d9e6c5036a9b8b5a40328d6d8a07098b92ca
36313 F20101113_AACVTP silva_h_Page_045.pro
8a16a04855db859be17d2049eb3642ea
a6b995bf7e32103655ca399ca94160a1e6afd96c
73972 F20101113_AACWWR silva_h_Page_122.pro
29558682a50da7e83eae74674d9c922e
a8835d8f56c98c912f6ff700f024d051440065ed
2101 F20101113_AACVUE silva_h_Page_066.txt
2fcf123f53cf176a3b2c024e439864b8
38b8e0d52adfaf3d3f173b40ea90da42913a70fa
2163 F20101113_AACXDB silva_h_Page_156.txt
576ba84c496631317ed3a6140ada8f2b
9a2221d8a83e3c016d6b1245c61c1c2edc2c8ec6
2272 F20101113_AACXCN silva_h_Page_138.txt
2a75bf8edaf1cd2e4e7933c1fd2d15cd
32243b9af8287b44436c614cc1606dcadb71f405
2143 F20101113_AACXBY silva_h_Page_121.txt
4142ccb4bbaacce9f6d046a799a43a52
2434d6e0195ef630b03e7988c3bbfd9a0569a69d
68668 F20101113_AACWXH silva_h_Page_142.pro
0f72923cb4e290892f03ebb8a1488673
3f7b23c88c55b48983a20c0004c3b685d18df221
2121 F20101113_AACVTQ silva_h_Page_014.txt
21e00000b2c35b4fa55196836480d458
ee7e40642c70acc2163cc29ab972c76b40d77de5
51347 F20101113_AACWWS silva_h_Page_123.pro
3fd725aa6c89af68589978dc2dae54ca
ec7b522839d2f5fc5940f66635dd31c42e1a5015
F20101113_AACVUF silva_h_Page_127.tif
3f70e40f4ca91bbd2d1573151f062ae9
fbb40cd71d8d97139dc6ee149311924c1f59a5e3
2057 F20101113_AACXDC silva_h_Page_157.txt
26b4653eeb223c9b83092a52f57ccd32
43d183b791c481486b602053dd52be2f5e596f6d
2085 F20101113_AACXCO silva_h_Page_139.txt
51c159a6c43b458441fb0e44092c0134
d0a1fbdfc580c729111a88f6a9640007a1bc7bae
2896 F20101113_AACXBZ silva_h_Page_122.txt
8f4eb8102d524fdecdd7b1d66c46fe2d
de4c76b70e3cd635756b75ef3c6d165cd2dc29ed
49177 F20101113_AACWXI silva_h_Page_143.pro
44966812329ae05fc05bd45fb79629f0
4bdff292d98f6f87a4e7ae170f8fba0e71467917
26269 F20101113_AACVTR silva_h_Page_169.QC.jpg
7c2ef7d59c0d93a79fedcddea62802db
cafee7289fb2f63cd29360ee1a97f274d91a1d43
69960 F20101113_AACWWT silva_h_Page_124.pro
08e3f0f215a825397dbad6c0a6b7e8b9
dccc639f081889db763f85f784e99d828d1a6fe6
52367 F20101113_AACWAA silva_h_Page_068.pro
68d621763dcbbf954abdbe323c5d664f
0ee31cbf35c441a0a2c16023c8a61d5f3f4d8d7a
59808 F20101113_AACVUG silva_h_Page_042.pro
a060f434b72bd7b5a9de8da7e91d42d2
0b50e9a697e8298fe92922abbe98afa5d670858b
2012 F20101113_AACXDD silva_h_Page_158.txt
dda054f61c5dece67a56c95491f269ba
b9284eadd2256dac66defd3eac8ff9797ca5f843
1905 F20101113_AACXCP silva_h_Page_141.txt
b377c402b7a01fd0c2aa3f91033aa37e
3a15b3fb2025211415fdba9a797491b446b6629e
77170 F20101113_AACVTS silva_h_Page_156.jpg
10cd006e4d8f977cdadd283a1f8cb9d1
6a19e7159d5c570cf67cc0e9802213ff3c389244
56367 F20101113_AACWWU silva_h_Page_125.pro
13d1c05ade739081b96c16886ed7df3e
fc1be6c818f91a377729a49b0deafa02a61b6693
6887 F20101113_AACWAB silva_h_Page_154thm.jpg
bb13fb073386b8cd30d55cff79889a2a
8454b8314ec18072fce84aec02999267f96efae3
2148 F20101113_AACXDE silva_h_Page_159.txt
19a1859898446c0dfd2b571f3532a654
a4c66fe26ba0f8bd35fde27ecc4ab070310ea880
1944 F20101113_AACXCQ silva_h_Page_143.txt
0ef500eea16dd204b6fa1a1cee7118bb
3548e86d7d5361b4ea1e909300fe40a827f6ad0b
56191 F20101113_AACWXJ silva_h_Page_144.pro
18cab3bb4bec8a72a62313dfef6e223f
ee0c25260dbfc590dfcf2ab3c211d4fb43d8e2e2
25585 F20101113_AACVTT silva_h_Page_146.QC.jpg
5974227dcabd918599313d47cba93a80
cb2ddf74767de15835421b3cd8534aed6dc3881b
72460 F20101113_AACWWV silva_h_Page_126.pro
823b3dcdbd3b7588b643a76382fdd764
289a00c06a17193e0c4c8182ef7c076c96ee4c2f
27142 F20101113_AACWAC silva_h_Page_160.jp2
b4b2b13d269778de07a8c254d74fa5ef
7e0ea8d746ddb74ad02eb87257b5a182e6b3df31
90563 F20101113_AACVUH silva_h_Page_005.pro
942716cef797c7c334dfd0f8db43562a
7b24413e64c42ff6936cf8c7a614ee461352b7e4
449 F20101113_AACXDF silva_h_Page_160.txt
12dffbda9e08f515c70677859c1b1714
f993f108f13ce649253df8de4df50630d7bd7391
2246 F20101113_AACXCR silva_h_Page_144.txt
84dc5fedd15d12fad3e527a43e997e78
045ec4730c3c8ee30b20ae535cf7bb159c290dfc
53046 F20101113_AACWXK silva_h_Page_145.pro
a4608de995b14494c7b44ace5205ab89
294a85111b34242ac8faeb9fb8b389d4ab5b9180
F20101113_AACVTU silva_h_Page_052.tif
9b715c2f227ae30838b51493409048bd
cc19e3d37c460e6ef6574d84153ee28648024cef
56900 F20101113_AACWWW silva_h_Page_127.pro
ced07f28c5947c801b5a04022ba69074
1131714ec1afd6afecfc3ad5a396a6a2aece62fd
77840 F20101113_AACWAD silva_h_Page_090.jpg
5e1700e2cfbf12a01beed94bd74b9a4f
3a541a3534fada5d5b92051c32e1900daa8345f1
2253 F20101113_AACVUI silva_h_Page_140.txt
e3736fe7284ef881e65011c2066c7631
3a0274ff7d2d184c51a8ab2d41fd00dfdb1a41be
83 F20101113_AACXDG silva_h_Page_161.txt
df531de11b922a0fac272e4be1dfa03c
3b7dced5ecbf7ff4f8fb97fa3e8f229126034e33
F20101113_AACXCS silva_h_Page_146.txt
0a4a889217aba03d5064184a038374cb
d9eb2791844e40cad9695bdd1a0de7cf782ecac6
45412 F20101113_AACWYA silva_h_Page_163.pro
d3a5064ace958330243cff56012802e6
90cfb4a232bb252a03c9ec39770e1c38e2277bf9
55431 F20101113_AACWXL silva_h_Page_146.pro
4744bcf277c82529407f33bd94d66e56
3810ea55a9189d2fccc82770f3b419c9e0674b13
25379 F20101113_AACVTV silva_h_Page_150.jpg
d0a693197294e59ebf47cd5a71efa350
9ad5fabde8b976941c2796d61132f1956b1d96b5
76656 F20101113_AACWWX silva_h_Page_128.pro
94b1069d56072044e9539677eee1b01e
efa7c7ead799cffa9c2294b0584200aaed2141f3
75153 F20101113_AACWAE silva_h_Page_165.jpg
3a87eaac196261ab75945ee72e8d64b3
f0bf8c4caa42494201b5fe43869ff2ec0c6a4be6
20053 F20101113_AACVUJ silva_h_Page_141.QC.jpg
894976d14436e2c1cf7fdfe20288f656
3dba7e55d23b29d112e8670fb4edb08ac4be6cec
1702 F20101113_AACXDH silva_h_Page_162.txt
6e4aa68b4d34480f40db02b715c23db2
a2629175ec3427b3d86fed755765a30596952bfa
136404 F20101113_AACWYB silva_h_Page_164.pro
4efb84bde04ee67695deed484cb2e441
f1843cd6d96751a4c63b76af80e780d2996203cf
112540 F20101113_AACWXM silva_h_Page_148.pro
8e005b496ccc1438e987bf3da419f968
0c8b1f87a1183148a6c3c5ef03178b36f7bb949d
73493 F20101113_AACWWY silva_h_Page_129.pro
0ee53ed6d7aa0d7b909d0be1038850f1
caad4686f8ec3761433fe37137712720a623d2dc
75938 F20101113_AACWAF silva_h_Page_018.pro
95ef940a23a8404b553e85f2f08adf70
f12d195a38e38aa3cf54a850389ab69e4792d77d
F20101113_AACVUK silva_h_Page_111.tif
643aca3a9dce5f5607e5efc239944369
5b7fbf1b984663214f551001c578835ffe3d0a3b
80718 F20101113_AACVTW silva_h_Page_075.jpg
b9b7edb2b01b212c948092e4f1438ac4
b669d8b17cc8effa6b8dd86a4219f8a6519525e6
2355 F20101113_AACXDI silva_h_Page_163.txt
952526eb387f48d8c6cbd26a9660e569
8a7376838e015fdb0f5bfbfff304a03997d928cf
1616 F20101113_AACXCT silva_h_Page_147.txt
9a950d0b0585c43e9bcfe85f582c359d
8e7db29d40d31e00e9a2ea20416584b24780ab62
74594 F20101113_AACWYC silva_h_Page_165.pro
b90b9940cc9c6d03fa96225e45927650
725b9e1699d7775634e97277fdd06674d85e8212
63641 F20101113_AACWXN silva_h_Page_149.pro
fd8cb9569df2988f3756fa5a22c9df7a
8df48474c7c1b0a615d61b0d4eb20d03f088deaa
50301 F20101113_AACWWZ silva_h_Page_130.pro
2fd19ffb0c09ff4f49a7cc682ff50683
6b56450b0855ce723884860281f4c119d142b0bc
57067 F20101113_AACWAG silva_h_Page_162.jp2
9e97daad33efb9c86058c4edbe81d03c
548720bc29d691f4b47566859c8c5e898f9dff73
F20101113_AACVVA silva_h_Page_165.tif
da7d9cab5931a803f55facbfa88d08b0
dcf65fdf6c753829f3261f17357bd3f73590f8a0
F20101113_AACVUL silva_h_Page_003.tif
cb81ab544a8b4921746a3edebe654825
9faf646f755b23b7ab23095d29c041e6d08c56e1
F20101113_AACVTX silva_h_Page_033.tif
af284010a6182bf43c758aef6dc4c9ea
798bf7546f0aba80c6b01575daf4d9bc15c12bec
5716 F20101113_AACXDJ silva_h_Page_164.txt
d2a7c0dbff134981b0c2ecd07a9bbf9b
857c880f4471e3dbf25a20e659d178f2f58cfd70
5462 F20101113_AACXCU silva_h_Page_148.txt
9006c11a83c54bfce93f3344eff51337
323038280cd4d02041c491ec9982f43c1cb82a86
58821 F20101113_AACWYD silva_h_Page_166.pro
3b4a0a605c3732422335b600d395714b
d929f305fff89e71da6b598d897429731f0b7558
21257 F20101113_AACWXO silva_h_Page_150.pro
e70e25100e784a21ffa537965e845049
2eab2af719cec223ba8d6589b3fd9bba029e6384
54716 F20101113_AACWAH silva_h_Page_013.pro
c9b42628a529363a92d1fba0bf55a183
ee818d692e74a2dd9ef96b9d447d15fc1e1eecaf
F20101113_AACVVB silva_h_Page_066.tif
83f412d9f46457d6d148bbf5cb6ae810
d461399bef6652b1c576f720bf73095de1525a27
86691 F20101113_AACVUM silva_h_Page_005.jpg
facf461185880b326747370ce9a4c1c0
9493407af596d0b7b22475745e610f5bef44b247
F20101113_AACVTY silva_h_Page_016.tif
a98599888db2fc4e9133839d387b2757
e864dca00fc0030321d5e87854eaf731b9c093df
3221 F20101113_AACXDK silva_h_Page_165.txt
793b42adb24a324d30dc0df0cf275d4f
51abdbb95ea45ebad74ec6bd104e83d4e781afae
2979 F20101113_AACXCV silva_h_Page_149.txt
a3eee21484d4ea6708f71eebe39c5c41
9cebe5d3bfcaa7f6c71659b8ea7fc90ba5e6e98c
63005 F20101113_AACWYE silva_h_Page_168.pro
9c232fcdb3d7812dfd05afa258892ba3
b6c37b59d2148010d91fee81b978b95bbff6641c
70957 F20101113_AACWXP silva_h_Page_151.pro
1804c7c92130fa0c84c80c4f72b9a058
d6304887fbc7f6b0b2879c914b0ddc670f0ae7f8
61713 F20101113_AACWAI silva_h_Page_105.pro
67cf2ea9cdba1891aae9e9c74048a885
a0ca5643092ec49b7d3c1138263ce143cb37157a
54683 F20101113_AACVVC silva_h_Page_107.jpg
c645e41137abec754d25a245d76bc49a
7841d79dec4ec1d5be60bdfa042a84559d7a36f8
42115 F20101113_AACVUN silva_h_Page_103.jpg
7dc0a227a426e14d0f53043eb97b566d
bd4d572105d63294058c9998aee200db27b19194
94598 F20101113_AACVTZ silva_h_Page_035.jpg
bb43525f6a2ad381eb355c4eb25a73ef
2b5d4fe3d7f3de32879c3efad4659385fc5d9eb9
2480 F20101113_AACXDL silva_h_Page_167.txt
81c96d4b2f92a972192335fe77aa0849
c1c06e805a0a1664462ba0cc07cc39099c88a129
1038 F20101113_AACXCW silva_h_Page_150.txt
ed2bf7c8614e3bfb30548d08ddb637a6
1a27bd88d97611dd5bf98f599b5b6ca59bac48d1
62255 F20101113_AACWYF silva_h_Page_169.pro
0310c6add3cf0973cd88716367d0b983
d6a02b3fdd482e8e815965576d62663aa51acbd9
70227 F20101113_AACWXQ silva_h_Page_152.pro
759e78c582ff5907083abfc098f2f8a7
b3cb0f71e9e368fca727cdb9eeb58f99ecee5996
103 F20101113_AACWAJ silva_h_Page_002.txt
351bbacd095b3bc436f2068e47c8b623
a323c00b3886ce195734aebd7a972f203b8907d3
6692 F20101113_AACVVD silva_h_Page_081thm.jpg
c435c8e0ebd0b342d398c08e7042d945
b937ab1773ae1dc9e691eefb2a7266c25132c451
6596 F20101113_AACVUO silva_h_Page_131thm.jpg
a4964e2f390c5fd12cd34c58e629992d
48bd18ffa43e89761151b8e0ec577a4538911e00
12848 F20101113_AACXEA silva_h_Page_008.QC.jpg
072537ef256772280826eb8c3b324939
2a2524bb805d75c167bd04a42cea1fdcac74635b
2544 F20101113_AACXDM silva_h_Page_168.txt
98cc54db2fd8bf55376a3529ce1b0d90
bed4f2ce68c9ff949493c1cdff6dcdd64aa5efbf
3299 F20101113_AACXCX silva_h_Page_152.txt
67afc350960f213af2db485837d8cc38
ef24cd5cb7532ce4b25e4e1e92f290b543d6855f
62234 F20101113_AACWYG silva_h_Page_170.pro
b604fdad8c81ebb2c7fdbdc9cbcd8f49
26d566dd73c485d35882745e8aeb054b6c5a3009
30007 F20101113_AACWXR silva_h_Page_153.pro
273f103ae065f37d1fdaf3387e0860f3
16e69fef0bebec02cd7a1cb3cf2c52cb11b62b5d
6949 F20101113_AACWAK silva_h_Page_033thm.jpg
2f7a9ca2edb90ee8381e3c35edc801d0
63d897f26f03d8d7c0d7012839c42d5a82fd2a0b
F20101113_AACVVE silva_h_Page_084.tif
44aeb604f3ea3f97612e364c90ca9ad2
266291057d3ab9d99fecc7085c951680931d50b0
F20101113_AACVUP silva_h_Page_156.tif
8edd3426e06106fd8211d603d7719556
0052780d81541aa3abaa885df301a3cbeb9a311e
2862 F20101113_AACXEB silva_h_Page_049thm.jpg
f686e17cfa029b12b8e33a1ae09ae94f
3248dad0753526980a68ed0759d570f6709bd497
2515 F20101113_AACXDN silva_h_Page_170.txt
a5ccee4982b0060ead78f2313a0ec5c8
65078f461c4e57ad21d803e5428a2a11afeb776d
1419 F20101113_AACXCY silva_h_Page_153.txt
ef5e5187271c3ed775be05620246ca9e
2b178d7716db8a20fdd0126535d27af4b829022b
63185 F20101113_AACWYH silva_h_Page_171.pro
411b0628ea86ebfa763ddc28b21ddcde
145190a42c3842f5a088896525b8488dd3059b45
54620 F20101113_AACWXS silva_h_Page_154.pro
d9d6481306cbbe4a240b29d51249c1e4
6fe1cf649360cffd8148bd1ff79d266af6b48cbd
63929 F20101113_AACWAL silva_h_Page_063.pro
474322af72a4798f22de8fc9f663eb94
d7df6fef94700da7f2cc0414de00d6a508318318
100374 F20101113_AACVVF silva_h_Page_079.jp2
4d6a5bb66a95aed802b51bbadf6cde28
51851c4da01d184c7cf7c3be9ced24cebdebf818
126465 F20101113_AACVUQ silva_h_Page_118.jp2
f07f1f905b6a4aa0020e2a4787f3a784
1d0c826be3f5644bf1d5a5f7cc81ebc2bbe6b0a1
21885 F20101113_AACXEC silva_h_Page_007.QC.jpg
3ea42e44b6cc7820bc54ecc5c498226e
370c2c614b82bb28ec47174ca4d613d7f54ad5a6
2542 F20101113_AACXDO silva_h_Page_171.txt
99973a90c5d82b194276c4b342a60308
c5291e303a460b611d48a050d5375a7dafef251d
2218 F20101113_AACXCZ silva_h_Page_154.txt
3e2873bf42fce0d81f5ce42215cb59ea
432c7344f6342652c17c706f7f8d952a05fe9a4f
39118 F20101113_AACWYI silva_h_Page_172.pro
97524157f24d19a11e9288f6e205fe1e
4b3fae2a8d0e8ff986809bcc72de8f3d6815aa5c
52742 F20101113_AACWXT silva_h_Page_155.pro
952aabcd25b2bb57a330b4c8bf55b7d6
f1deaf32a10a23692dcba89604f5e23eacd665e3
7204 F20101113_AACWAM silva_h_Page_098thm.jpg
c66a29affefa48be2543ead87b5befed
ade0845f3df32c22fbce47370fab6fec8e32a4aa
123149 F20101113_AACVVG silva_h_Page_033.jp2
c263bba99baa397977f86cba57c4b025
8627cfc7381c3ecc919fed1fda28ce73b87a2408
6837 F20101113_AACVUR silva_h_Page_016thm.jpg
bc5895f011e9c9e5de740ca173c800b1
87346d3143e958c861cc4a620d16a99447096bb0
2172 F20101113_AACWBA silva_h_Page_025.txt
18f3cd15b0158dc9f5a3da2f9b4b5fac
085f0f5989304a5ee4d01e663b86b4c58c6eb1df
6810 F20101113_AACXED silva_h_Page_166thm.jpg
4ecc04fe1b0525a6ccf96dee7988b2d1
a74f23aedf6e8544ee7e3f4160f4de09668bc7c3
993 F20101113_AACXDP silva_h_Page_173.txt
48f2c034b583fd9426af931576fc425a
54570204bf3603f1671a6d5babaa785609b96c50
23799 F20101113_AACWYJ silva_h_Page_173.pro
6858b6b965a779c16cd7b355cc0d20c3
716b51788dcf284496339a82966660c44c068e89
54885 F20101113_AACWXU silva_h_Page_156.pro
f15110ce0e16c83d24b4a7164e1a0898
95688fff3de187195f6f43c9c78537049b587cf9
4816 F20101113_AACWAN silva_h_Page_149thm.jpg
c27231a691143ad18f9310a04e383dd9
baa49965c448c2eea75329914be719ea20c88eb7
980381 F20101113_AACVVH silva_h_Page_006.jp2
22e81c4ee5428bfd8a4573d143c35811
03ea4c8e10b3f7015c47c55baea9e33fff841d97
34637 F20101113_AACVUS silva_h_Page_008.pro
a21c77aab036df5888f25553756cca65
199c3a4c9ad90a8dc4433fb2d446a997fd3269ba
79786 F20101113_AACWBB silva_h_Page_140.jpg
4a7e8f4f82bfac0064e149346e7b816a
a5cc5163a89f1c82bcd6bd835a3115001aacc81f
6817 F20101113_AACXEE silva_h_Page_068thm.jpg
0baa0dc67a75c7d675668c6e0dec6f04
086a0d265e445feef239ec369e3fbfb5068c968b
735545 F20101113_AACXDQ silva_h.pdf
c70036cda4bf65f34d73addeafdcd9b0
730cc65e964d1175cb1caa0fb4a9db2bb8d6d446
52040 F20101113_AACWXV silva_h_Page_157.pro
614dd59e3e3ca98acddc5f67ec5331b4
6e5a40a8b18d6659fcc1150f395e20920336846e
115482 F20101113_AACWAO silva_h_Page_014.jp2
24dd66ba1a898da5568b24b19351614b
dbe9ef93bd8c1e27ee063b43c37724707b0b74a6
24570 F20101113_AACVUT silva_h_Page_145.QC.jpg
01bc31bd8fbfaaec9b31e1b793da400a
3d3e53f78e5f1ef0b468e30b18567f95c69f0e15
7312 F20101113_AACWBC silva_h_Page_128thm.jpg
a25013dad5411c3df7d6782ccb6cb88c
476edc51234a107380e219c31981e8cde87a8538
15937 F20101113_AACXEF silva_h_Page_147.QC.jpg
4a113c8fc793ee00a820ba7edaf75ef7
a62d181451b3baf796f5b7aafc395a25920376ad
23719 F20101113_AACXDR silva_h_Page_012.QC.jpg
c976a32a4f5efc0902901002edf69491
0187241cd3d1dedc2db91ea0676514c67b3600af
416 F20101113_AACWYK silva_h_Page_001.txt
36be9ed4be7f248e8f2a2e2b29eb33da
49b53214b78ccb9aca702e2ad828221e9ff410c9
54406 F20101113_AACWXW silva_h_Page_159.pro
3bd660eba211a4d8bbef131e05dca218
ea1785ff5c0e70f310f560f31163a8322e4b5ae2
96801 F20101113_AACWAP silva_h_Page_054.jp2
606076c9423f3beac770101417763627
b3dc6720474ac7871138a8620c1f5bf7a4b0de27
37445 F20101113_AACVVI silva_h_Page_054.jpg
16186289a8d1585a6c72c4d19955caab
afcc99417af0148955bb8e698d79e59c3e908dc3
71024 F20101113_AACVUU silva_h_Page_031.pro
1446593bd41d3d01075065792fb3080e
3fb66ad3d42ccbe0b2f083f705aeb64d8a9b042b
46623 F20101113_AACWBD silva_h_Page_101.jpg
affa22003f5eaa13a96ca023ade548ef
26a5d18786d6bd606cf642e077cc70346d5d6d4e
4112 F20101113_AACXEG silva_h_Page_163thm.jpg
4f85de659d11ae5d8892b0413c186162
b688854b003eda78881ff9d41e4ccc90dda7ae15
6720 F20101113_AACXDS silva_h_Page_125thm.jpg
b62e419985c8c6d6ff40c00fcd0f23a9
a33165942ea5b6901f0a51640a6eeb58c2cc2621
2107 F20101113_AACWZA silva_h_Page_021.txt
e8d7fa3a5ac18e72035b18a7d804d22b
13b293eb40e3fbb401c89bc8d5efd598200f61c8
126 F20101113_AACWYL silva_h_Page_003.txt
f661e9e3d77087f918385bafc7e5d759
02ab20390a0e45eba681445b0df2735d10754751
11109 F20101113_AACWXX silva_h_Page_160.pro
bb9bb9a63b5bc2e4ee4ae020939990ab
d31367809ccd82c424407e8bfabee2c297b85d8d
52872 F20101113_AACWAQ silva_h_Page_067.pro
2eaf17394a9bd1d816005037d714d693
788271b63d5d3066e079588082b733e73b7a9b4f
34498 F20101113_AACVVJ silva_h_Page_059.jpg
38958b9939cbbfe88d34238ddf08dd58
70101d2b8c86aa57cfdac7d6f3a0b8b1a8f81379
108560 F20101113_AACVUV silva_h_Page_058.pro
d852a53a92c72a3fe341c8e1eb1624c9
2f7c85dc8ed3cfbbb7a2aa316517f029bff11232
7121 F20101113_AACWBE silva_h_Page_062thm.jpg
714007548820ec7424831b9f1b4929c7
378f80a3a71045260d37e31263cb7c5ab720b647
26186 F20101113_AACXEH silva_h_Page_119.QC.jpg
9dcb95969d8cdedc54bc104b2743f721
e1f7b42b9ace3c9d9b135bd091530c34d1149120
24457 F20101113_AACXDT silva_h_Page_068.QC.jpg
7682d8f693fdb6ee26b803af390e396f
562e51a2dde7186200bde25b254026f1e4a17099
2958 F20101113_AACWZB silva_h_Page_022.txt
340fecf759dcafc0c02dce12d214b8c3
0ad65d87a8dfcc79c872511c2bed6cae593a7bd5
1678 F20101113_AACWYM silva_h_Page_004.txt
a7acf3720ac656dd95eee58744f6def1
50aaf979245a0e68ef3aef978f5e1750b285fa5e
1626 F20101113_AACWXY silva_h_Page_161.pro
8cfa67af03b66deeaaa8e39fc0fd5802
3363d0050c6671e45ac5c64f73cf5cc32ad7e308
81164 F20101113_AACWAR silva_h_Page_033.jpg
9147b68159fe203093f845c34a23bac6
8ca8fd6abe99a17992fbbe2da192b85934771c81
4329 F20101113_AACVVK silva_h_Page_101thm.jpg
7cbdb3fcc7357510c3b5fb13609c71d6
ccb01a2b9eec7059aacf97508b491db7476a2bc9
6065 F20101113_AACVUW silva_h_Page_141thm.jpg
baf92ada6f9076e9aaad641f2a5b7478
bbc032a166aa9a32205d12a4070de4ff07f1d770
F20101113_AACWBF silva_h_Page_091.tif
455ea32075493d23b174fdc71c108fb8
55996c3806cd6cd177fe3628eeeaf98278663b4b
7099 F20101113_AACXEI silva_h_Page_082thm.jpg
d0730072ca032b4343f2a1ee723f55a2
f4799fa7a2ce8f66051b3193fe0ee1309185f68c
2108 F20101113_AACWZC silva_h_Page_023.txt
46dc4ea146badd5fb79b79456985d886
9b466638a36a5b86aac7f76b307219a37739e9ad
1531 F20101113_AACWYN silva_h_Page_006.txt
9b0ec8f044d536880300643b5f21819e
dab1a8e51adb030a14f2af88112c1af4dd91e68f
34068 F20101113_AACWXZ silva_h_Page_162.pro
d06d2604251240d051bbe6a988f2f0ae
c40c5b1da4e79b412ad51ef6837a2ae0888c1b37
F20101113_AACVVL silva_h_Page_041.tif
0e24ccad7594236a96b12e891743f6d9
0dee10115e774c5ea7c317b7ecce511d8cb199a9
1453 F20101113_AACVUX silva_h_Page_045.txt
76cb39db945684a612becf1b8df09fa7
4c234e90d8ad228990a09efcb5fca579b83a32e4
2895 F20101113_AACWBG silva_h_Page_162thm.jpg
2e42d85243a85cb79f26715104a29122
5394dd0577d707e402e0984ca35cb86de2a44655
7001 F20101113_AACVWA silva_h_Page_080thm.jpg
84ac06c1cd462efb236f71bb6093d3be
c03bbc539c0cea09379b32749bfb4d83c6bf3aef
7127 F20101113_AACXEJ silva_h_Page_075thm.jpg
ef4d244692d6091823e1cec1e2d3c72c
add52f5b7a1299d6b7f59abd73afd40b368e6f72
23513 F20101113_AACXDU silva_h_Page_023.QC.jpg
82c81d67761a467ceaa5f2099b47f57a
5a31901a4952abdb0d5d358ea1c74ff9214c5a62
F20101113_AACWZD silva_h_Page_024.txt
654f3b6872d6e85860ef498837911b3d
d898dab7cbe2723b764cb130e498af941da040aa
2719 F20101113_AACWYO silva_h_Page_007.txt
45893b0a85c6b7e0f52e4e985e1f6a36
44185679dbc5e8edd69f2732563f462ae679329b
55747 F20101113_AACWAS silva_h_Page_075.pro
6380b1b2a7adcce1d5761589ce6f53fe
86dc718d8585f4c33f8eca9e5b62b69365e893e2
25956 F20101113_AACVVM silva_h_Page_167.QC.jpg
6d3c6f37a6e9ea6ebde60f378f330c93
44921d84597ba5585a10b1b33e842e2be77f76a3
55544 F20101113_AACVUY silva_h_Page_112.pro
f6cc0130618119942d7e662eb431535c
ced97838d643ce3261478f60bb71ca9dc83d2a7f
53433 F20101113_AACWBH silva_h_Page_045.jpg
3fe469f662b0092fb156fae9fb1d4fe9
2236357fbf108939ce30990c1c81a15725431b9e
2502 F20101113_AACVWB silva_h_Page_169.txt
2a6595bb4dea26fdf758746689868579
50ba227336843b02086739aede9bba3d71389017
11530 F20101113_AACXEK silva_h_Page_053.QC.jpg
21e7ac6fb4e78b3640a15b9d2163429b
e4000b1444179710bc90f95c43bdc7a8565b0dc7
23642 F20101113_AACXDV silva_h_Page_088.QC.jpg
e900be0978ad1d7424026d889178988c
781a9307c909d69b1608123ea08d3b2380307e74
2084 F20101113_AACWZE silva_h_Page_027.txt
4c95fd949a179b20330545c64b09a26a
729004031ca8590a92dc2ebfe6d9f16038b8e704
1376 F20101113_AACWYP silva_h_Page_008.txt
02d2f201e3b129e9d611f1cd310e4c69
61faf1b298c86c4a24e3c756c52a3da015845d6b
F20101113_AACVVN silva_h_Page_013.tif
d1af3a5d0cdee4d3ea59fc42fabf224e
c2d97e0a7e1ce6b9e3e316a6f0269fc46d761c15
54379 F20101113_AACVUZ silva_h_Page_086.pro
1d5584242b767a0be41e208361271fbd
7d440c72b7f437dabba75d9ad929177cee435f72
28681 F20101113_AACWBI silva_h_Page_035.QC.jpg
03bb23e754b12f55b55911e2ae31895a
8e81b8182b2f39c0c90e787620ffe323f18728b3
F20101113_AACVWC silva_h_Page_034.tif
7bd1f3ba0037672d432289f9795a7d58
09158bbd8dc8dc7a41423c2a4740ee50d1a7e48f
109234 F20101113_AACWAT silva_h_Page_131.jp2
afa4cd35663e216075bc79caf626962c
ae14e8369577b0f3bd2f76a5a38883713b96d2c0
6648 F20101113_AACXEL silva_h_Page_021thm.jpg
18c3f75f0581038da17c9c47affdf2cf
cb0209d6e0adf648d50d3eed0bb9e55ab657d57d
2743 F20101113_AACXDW silva_h_Page_011thm.jpg
5972e3011a090429374df479843d1a9e
0ff0d9aec4ce2bbed64d2999b32cae3d6a26f91b
2126 F20101113_AACWZF silva_h_Page_028.txt
41e6cb6176d460b6c894b09d66728831
fdc78b0bc9d2296b5143a184b7bcc39e612988c9
1947 F20101113_AACWYQ silva_h_Page_009.txt
2377f022c630c6413f905b1dd316ad52
8af004e87076d0a8cd59d61e77496d0227d48c19
53138 F20101113_AACVVO silva_h_Page_021.pro
530831b5d9460d59bfaad48c0e2bdcda
1087f836eb3404a026130cbc5c00ff2dfe47847c
55274 F20101113_AACWBJ silva_h_Page_010.pro
8deaad31a08329b10bb3283d29ca9aba
cf8fb5cfe991026ff5712a55541edbb400f50cd1
145981 F20101113_AACVWD silva_h_Page_165.jp2
fdf984f564b9e88ca7ce93080c9d6093
634687cf6a1bf13ca1c337fb64b3cd3651653612
79416 F20101113_AACWAU silva_h_Page_097.jpg
30c11451dd19ab41f9e0f8a645cd2248
a65d0aad34e965675e7b29bf05494f79f024e61a
12486 F20101113_AACXFA silva_h_Page_054.QC.jpg
09dc15269330367267b09ac6b6e95e99
5171da2c7ef7824045cb8f1801cfbea5a413d956
6685 F20101113_AACXEM silva_h_Page_023thm.jpg
86f4a767cc46c8a6980532f8c5f0e2f1
fad2e206a72d24beee6084b542bea513720ae183
3888 F20101113_AACXDX silva_h_Page_055thm.jpg
b36af6d53068d18f574c88229c26ffa9
4cc1c1a48456a6c8794c16a9fa182c35359bf3a3
2508 F20101113_AACWZG silva_h_Page_029.txt
cd18903cdc53498537d8c6e842195769
d54d82ca62e3c6401994a07e0d03b9b9c56a885b
2171 F20101113_AACWYR silva_h_Page_010.txt
0b92c2b71c5c5d04a4a81868f33567af
0cc8c4f1bc2efa82e7900018223b2c24dd1c222e
7178 F20101113_AACVVP silva_h_Page_087thm.jpg
d1d072302e9a876bdde325b101b6239d
75072a19a2042c4a3c870a647f184d19ed76b438
67951 F20101113_AACWBK silva_h_Page_049.jp2
6e5bac06b430b702abc71ad6cd872a5f
aeaf10c4f3baf5604175f83cb70da89aedd0d6b9
F20101113_AACVWE silva_h_Page_068.tif
2722514c8e526f1e5c4ec4b733e52fa1
0144736ca16b71faba32553b6dd0b89d325442a3
F20101113_AACWAV silva_h_Page_115.tif
7243ba2a794a4ea5fb17d7fc9bb40118
bc82887bb596d15d8f22acdac06f3de880c0e7a2
24775 F20101113_AACXFB silva_h_Page_111.QC.jpg
d34b0671ec39af08a13791e91f530e05
85f48a3f8d77d760be077142c87fdfc05179d59b
25753 F20101113_AACXEN silva_h_Page_038.QC.jpg
d5cbf0b9d8618d9b42cc88594f3c61c4
b1fb87ece665c9ef269a0f3841d115938f4a6542
6473 F20101113_AACXDY silva_h_Page_088thm.jpg
47222dd343a5db071c586d62c524654e
e45504ad794d0c38e0128f33658dc6dc6b76a6de
2366 F20101113_AACWZH silva_h_Page_030.txt
1b384a302c1c43fd1e1338f0c4505af2
fa1550e8b86a25913acb51527157145e022b5c99
584 F20101113_AACWYS silva_h_Page_011.txt
f0d7472791e85408fd9b2940eaf0145c
ffedeabc44171d2f26176ccddb7ff15f458b95d3
F20101113_AACVVQ silva_h_Page_105.tif
58bb5ab61cd1fe3849bf41fe33f5b57a
127c7aec988758bba461824ef8d2c14fa99ee048
21118 F20101113_AACWBL silva_h_Page_104.pro
2cf49a3c76045c5ddcec29a8aea37b57
fc4bee3600a52f78bc8b7022135354308a25bb03
2447 F20101113_AACVWF silva_h_Page_032.txt
076a5284b7672de9dd60aaf64e8bf4be
4d411c2ea76097eebd6a006eca28d4f780e17c3a
F20101113_AACWAW silva_h_Page_074.jpg
3187af3971e9125fc15b95996cf6e509
41aabe1b61cc59d34eba31e42fbfbfde5b23ef41
26342 F20101113_AACXFC silva_h_Page_084.QC.jpg
92b05087f3cd418efa2de8fe9883a410
9a31473587f6534ee25fe6e48b8fb8d6c3aa2a10
20503 F20101113_AACXEO silva_h_Page_004.QC.jpg
d91ee3163546ebff43f34a02b74aa1e8
ca66a9c16bc13e1ac17282cda4c24555e8656594
24477 F20101113_AACXDZ silva_h_Page_110.QC.jpg
140101c60aa87f9e0276bebc8ce4a795
8fe551eaaf1687ad4ca92755d522a342b2895cb4
2733 F20101113_AACWZI silva_h_Page_031.txt
87805bf6e96ee7a91689643cd9451d6a
8f772fffe6adfd26e8178f6181e93a37f63a0c1c
2093 F20101113_AACWYT silva_h_Page_012.txt
0116d27bbfea1ccb3634a52b54ee4b4a
2f78d34cc744b37495576bb8401afb57d354d297
122307 F20101113_AACVVR silva_h_Page_048.jp2
04df8794789b3db80840513693a2f5a1
1185da83d6a6c80c7b95ac0f4c33b3eedeb19224
40561 F20101113_AACWCA silva_h_Page_004.pro
8333fd6b7c13391bcaf234752fc82d51
4f4f2fc37fb7ffb0247ff4f9f9b90d6515109173
63237 F20101113_AACWBM silva_h_Page_141.jpg
befe55b4b045bed31f4311cbd7ea5257
bc3349c89505a759d319751b86bb6d1b224bfaaa
4827 F20101113_AACVWG silva_h_Page_102thm.jpg
186a6e7ec7f2f85d5535a2ce59e8798a
9f4db6b5b55c25c9f8967dc8ebb6b7c8c253e684
121409 F20101113_AACWAX silva_h_Page_032.jp2
0d2280ee829f2bb2938fe54927e05ec1
3b943509605bb11ebe635e00f3207c8c0025541c
9108 F20101113_AACXFD silva_h_Page_049.QC.jpg
cb84802680d0728317caf3c670fbbf37
1e07d4b20e6e0196fd30856c81018b2619b723d7
F20101113_AACXEP silva_h_Page_009thm.jpg
4140d1085ee53ca038463b75a6bf8e17
7035505c26d9314b55d2bbb8ea4804d83b941757
2423 F20101113_AACWZJ silva_h_Page_033.txt
514570af7736cd4ef4aa0a0c6da37527
ccf4222253be2739e7a001fc1b4948a4ae3c6bb6
2153 F20101113_AACWYU silva_h_Page_013.txt
9d0e0b9857c2c7db2eb399030ce8b7a1
b044ae5e8a7d88959284309d17f5b16281e209c0
91746 F20101113_AACVVS silva_h_Page_059.jp2
46fa60d550ce7fa7cc830f19d00c9f79
a45a1584aac743844b230063e13a6ff1253cd69f
115350 F20101113_AACWCB silva_h_Page_070.jp2
25b761b4bc9f1e0e67a536dc432feeef
9a2de3c70c065826c005ef5194987a9251d80da2
150080 F20101113_AACWBN silva_h_Page_094.jp2
269834e1885fd0423fea5c19df0aad05
ba93007b339e5e94082ff2b3c5182ec43ff46714
3073 F20101113_AACVWH silva_h_Page_034.txt
8d4cca773b3a34a2e2e19ee875acdd19
f92086dc5f4f13361cfb456042bdb7db7ce5c737
18683 F20101113_AACWAY silva_h_Page_172.QC.jpg
af8b912dcfa724acb1bdc4d92617ad55
2419ca97a1e0d843178ce5bc81bd62679849ad32
27242 F20101113_AACXFE silva_h_Page_124.QC.jpg
fcd90bc4e97a216502a6d9da9cfaebfa
e78aa89848f6cae762cc8a64dc0bfb5067eba72b
25682 F20101113_AACXEQ silva_h_Page_100.QC.jpg
536174af12e62f82e1ace36f9c96e4bc
4f903f941c7755a60a61f1dfe4b72f29605d28f4
3154 F20101113_AACWZK silva_h_Page_035.txt
c620c98e160601a9c3c64d268c7dd4a0
2b53f625e940f9a4c76ace83254b3882d73369a9
1490 F20101113_AACWYV silva_h_Page_015.txt
598237a515f2bb2323748066a9c1c7fb
21f8f9d0aba2cf5c3b30b8bded43186d37bda5c3
6801 F20101113_AACVVT silva_h_Page_003.jp2
249fdeda21ee2c9b56d0514ae4827a54
08b8202e4c87085d438d2a568b3272deceb58670
25464 F20101113_AACWCC silva_h_Page_099.QC.jpg
dbbdb8236f233de8e5b2136aeec0ac8f
dbd04f59ab89081c51e884e3769603975b4ce4f1
F20101113_AACWBO silva_h_Page_082.tif
6a863eed7857302045bcd0564b4dcb95
4089df27471c8025156fade967d27113cc3f16c1
14023 F20101113_AACVWI silva_h_Page_051.QC.jpg
7d777db91ef7ec2f5c0ff00baac05b69
4ffd3bfb586baf3258cf66391337e4ed36dbfb70
13873 F20101113_AACWAZ silva_h_Page_055.QC.jpg
418d9ba01504d0394e7aaae86fbfefe1
e29183f5e598915d4858ddc394917995251ae509
21697 F20101113_AACXFF silva_h_Page_005.QC.jpg
d22ab285d33db63ba743965b8723d026
2b249cdc1aead9881d80a772926ac819c2275c1a
23905 F20101113_AACXER silva_h_Page_070.QC.jpg
1a9edc265bf7722c514bae960572871a
a940d852c1de556dfa1b97707db68664df14411d
2929 F20101113_AACWYW silva_h_Page_016.txt
8cd8e2531b6773c847b21da08065bed2
7068e5b42adf7aba0b79b48c4484273d040735b5
60340 F20101113_AACVVU silva_h_Page_109.pro
3d61ce36f3f4b543f5bbb2587e15308e
6e7aaed2129327cde7a7ed04e1b60423b9fe1c6b
5221 F20101113_AACWCD silva_h_Page_015thm.jpg
ad750145b01eff40db5279485db59a03
61e7899f91db7333cd23a3776ea42523f2b1d738
71752 F20101113_AACWBP silva_h_Page_066.jpg
fa36c7fdf3d69cd059cd1b88f05e3201
e5728b369b1215525a892823bce57d5fcdb186a2
6920 F20101113_AACXFG silva_h_Page_063thm.jpg
27735f3715334b48fdc0025b4fb83347
77e87e2254a55ab535801e61bdb0e99e94ec9e12
25668 F20101113_AACXES silva_h_Page_135.QC.jpg
bf9e9d43ea33ccdaa18cf13ede113ca2
f725d5f2d3e9a0212cc8777511eb0cf754e32a6d
2683 F20101113_AACWZL silva_h_Page_036.txt
8f1f7e335168ca898cccf806115f99a9
067e3cf8956e43973996f158dd8901bd8d6788b9
2165 F20101113_AACWYX silva_h_Page_017.txt
0b67d94359045fe3270323592246993b
5a777b3cb550710ee2fd15d00dc873dcbb0d2e78
24026 F20101113_AACVVV silva_h_Page_067.QC.jpg
2d736310323948a4a89897fd1d8e49d4
5ce433bcb14801fcd9088bbc439c5decb145ebf2
22384 F20101113_AACWCE silva_h_Page_114.QC.jpg
7a45b784f4a713d4913094b755c3dd66
59970d4ae262865bb15c8fe0694a4fc6d22e666a
4064 F20101113_AACWBQ silva_h_Page_052thm.jpg
ae46770e68a7e75cdee8cc990c8f5892
833ef624d0bea085972b951518907d305f03c94b
72728 F20101113_AACVWJ silva_h_Page_023.jpg
992f1ea0f5a11fd618c81f493cc4550f
a0fd97c68e77eb856af750f0ca636b5a3813c244
6367 F20101113_AACXFH silva_h_Page_061thm.jpg
d513be8d33bc35a717f398528074365f
fa36106c89b57196e8d8b412a61f056fbdf26835
6800 F20101113_AACXET silva_h_Page_092thm.jpg
3bdbc8903d3b05ccfd8fa2c9c630e1c2
3724287b41caa1f8f8c3a0189e7406c8d08633a3
2617 F20101113_AACWZM silva_h_Page_037.txt
e61ab15b70fa1aa6675bb349f74e43b5
15588298e8c837b90bdf67aed2ea0c55d974b5f5
2912 F20101113_AACWYY silva_h_Page_018.txt
3c14a87ffcbea3725854597241ce552c
eccb5607539fce14ca3b85d48e14945bcca506e8
6675 F20101113_AACVVW silva_h_Page_076thm.jpg
181044be7432e04e6743dbbfee29f9f0
08116f4f68b31df8ba36a2d0013c9476f09b812f
25265604 F20101113_AACWCF silva_h_Page_163.tif
f13b57e638a1572870ddbd02f4f4bf88
85339a2fa5a93003f84f63050c993fbab5c84fee
F20101113_AACWBR silva_h_Page_148.tif
ec90912c9c4394b88f4618f38a965e01
e799021ba5b363b5e5e5254d3542a8aa5bd71be6
7086 F20101113_AACVWK silva_h_Page_129thm.jpg
d3cb50aa1b5e2e096c3118d23ff57641
43e2468426e22b9400ce7cf88250cd1c491a6cad
23949 F20101113_AACXFI silva_h_Page_115.QC.jpg
748ab6ce4f2fdda7c4d400bd4f082016
678fc1853b5f89dad82fa25d280666422e2da2dd
6775 F20101113_AACXEU silva_h_Page_097thm.jpg
5d6dd79eeafcd65348312c651fac0817
3e931c4dd2ebc8c70fb148eb0e7eb8afa2c1e3c4
2513 F20101113_AACWZN silva_h_Page_039.txt
bd85daea4267af1e33209f90f9a87073
7e05d6eb8b4d15ed4fc8f79eacb3ddba9aa4cc80
2287 F20101113_AACWYZ silva_h_Page_020.txt
640abfb813f7ffd81bde66758c0223b0
638a666be15547ff90031a30b44c705028f85a6f
5080 F20101113_AACVVX silva_h_Page_172thm.jpg
dd816c1d8210c78e25b9ce3315a834f7
61df33733d6cff523309a872ad30bd804a44b919
6234 F20101113_AACWCG silva_h_Page_152thm.jpg
37e5962d48e6a8f4bcaf923cb6e1b68e
473fbe8b491f1b73982c9c353186267d31327785
132236 F20101113_AACVXA silva_h_Page_167.jp2
c6d5d51136697fce51387647e53cde7f
a28b5b13cb5ed1d726231c8ea74cea6dda715f04
3714 F20101113_AACWBS silva_h_Page_005.txt
96eb5cd70c585eabb6ebdb3e263593a1
0e6c292ff14278e1467342b174c2cf67b2e3fb60
2383 F20101113_AACVWL silva_h_Page_166.txt
d8e54e96787c649bc5194c2a76d0057e
b559132eec6c5dd150e632e76f61852af262e070
6691 F20101113_AACXFJ silva_h_Page_070thm.jpg
ed9af11c8c553006c42f791beedf4478
890655196e68494d38d900e1693edf68792f9247
2576 F20101113_AACWZO silva_h_Page_040.txt
50316c2dad297a618c1bb711964e1529
2867d1262e1b8ce5c684a98d0d3916480d3a404f
84040 F20101113_AACVVY silva_h_Page_084.jpg
f166ff7516aca7c2dadc1bf607dd0e13
30594647f6dbb569cf4d20bfdc210e3c0c81634b
95092 F20101113_AACWCH silva_h_Page_120.jp2
ed9fe75c1f6f5e4a1276ca424d1865d4
ee3505ecb4dd465409ad5d599f8087a2ffab1c86
86908 F20101113_AACVXB silva_h_Page_172.jp2
e92a9b80325baaa630e39a0ae8fe9efa
fea80702a9a53df18d227da193fe2ec065d590a7
74404 F20101113_AACVWM silva_h_Page_021.jpg
e8187642a54dbd0390c373a5dbf85012
53e9c033db11fdd1f065212ecb24bc20cfb81b19
25823 F20101113_AACXFK silva_h_Page_118.QC.jpg
8434865e89fe876fb12776640c5f33e4
9bcf23aa349f5d135e6b513446b8501924cde677
25567 F20101113_AACXEV silva_h_Page_013.QC.jpg
cf39b9894ba1144c6463840b4868b86b
70e2038ebf76bbce81841194ec789a8f3d0f94af
2547 F20101113_AACWZP silva_h_Page_041.txt
93f12963337696de71f6ad9a9d75328b
89e218f3bcd10e190b14ee51cf01d83a2406b397
7624 F20101113_AACVVZ silva_h_Page_026thm.jpg
23093a31afab7b914c963de6dad6b2c7
d50d22f881341ec7fc6b4d54f724d8dc472ba107
F20101113_AACWCI silva_h_Page_069.pro
5ba2b6724283a6fd30943dfc507dd816
00bd5ecc914918a0c6f0ea81381b6556f6d66dbe
F20101113_AACVXC silva_h_Page_115.txt
625fa7888d278a493a23eb5ab9b9dde3
a19ceab382157f53a5189c5abd5828efa4f61979
42762 F20101113_AACWBT silva_h_Page_060.jpg
c694ecbac6a2b83a28acaf0bbb54fc8c
2b2a1338c540a6ef8f063f54dd4113c6d345deb0
F20101113_AACVWN silva_h_Page_014.tif
0a6b726aa20be0f514750bb0e3cfe88e
9609905c4ab89bd68c20d75d64bb251c119c4dc6
13028 F20101113_AACXFL silva_h_Page_060.QC.jpg
dbae1bfced4d6ba0d0adcd217379f3a4
e5382b664727d6c79446c2b3473d7a9e9cf6cf4c
7284 F20101113_AACXEW silva_h_Page_160.QC.jpg
84cd1a9430c1c0764dd77e5b85ca9dc6
70eecc91418db97ce70d3910d9cd41ad9dc2cb4e
2342 F20101113_AACWZQ silva_h_Page_042.txt
6e1adde633665d40882c6731cde45161
73c57bc90657c870bacd4145a8fd8f7991cd8eb2
2715 F20101113_AACWCJ silva_h_Page_062.txt
b244cfd8dc21a3c2deefc8d9e9a44757
077d1024d032a448edeaf9ef7be8e226c7fc2653
136550 F20101113_AACVXD silva_h_Page_168.jp2
3d61141282289efe291ee48babd921f0
3354fd1757f2248c5026d9c928f55e11189ef596
F20101113_AACWBU silva_h_Page_083.tif
3951fb39526190688368f76acf0cef4b
8df824db7bb040a352fafb541812839a6fd893af
136153 F20101113_AACVWO silva_h_Page_136.jp2
d178fd1bcc7015ac2c2faff15ca762e8
4784ee8e9116d9cd16ffcdfdb86dd270a1aa1d71
26234 F20101113_AACXGA silva_h_Page_071.QC.jpg
1b51bf1ddabdddce466ed6fdedb6e550
b3c10a775740efe3710836bdacabc1265ab04310
7200 F20101113_AACXFM silva_h_Page_171thm.jpg
371112356022ce02d8a5675d83395f2d
7edef182594bf3ffb1a61e2c5147006e74330878
24838 F20101113_AACXEX silva_h_Page_155.QC.jpg
68a31838642256c61421268c74cd22a7
68bc44b529281ede5f8f72ed12c4de180273eb28
2377 F20101113_AACWZR silva_h_Page_044.txt
3b9438a0a5dfc33d06951b3ea4440c08
8c3c4dcdb012fb38de499ecf0b6d2a5832710871
2428 F20101113_AACWCK silva_h_Page_038.txt
88a359d7c2e1f1782a41dbe3590c66fc
e1fcb2490205db57ff13f3d103ea90a67c9fae47
F20101113_AACVXE silva_h_Page_131.tif
dc5db2b5c88488284c08761b5b434008
b6fe8143751352a2327f5ecd7735c22f45f0d1a9
6848 F20101113_AACWBV silva_h_Page_112thm.jpg
70180f32945140126449c1b1d2482cf8
06f9017b57a268d29792a5097308bca4ae25f7d4
6427 F20101113_AACVWP silva_h_Page_066thm.jpg
146de0a10680e74824d1b9519d1c3af5
70b1075d366a9644576b2b66e389496428d3dcd9
3555 F20101113_AACXGB silva_h_Page_060thm.jpg
67a83056d6c5c505d34db7087fce13b5
03e27eed0ca1ead2d1d7a1f501b11ec05f33bf48
23482 F20101113_AACXFN silva_h_Page_131.QC.jpg
2d3528bf51d3cf62ba935b4c0971323e
e570cddc67d333a602443f456814ea8b77a2e42d
22242 F20101113_AACXEY silva_h_Page_064.QC.jpg
d7d2336b39e3752878ad08016ce1afae
4c16eaa24f3db7b73dc544c6cd22357af491a4d6
2175 F20101113_AACWZS silva_h_Page_047.txt
b96de8801404bbd12abd80f9e778e121
565446003a2277983c73e6790d6992302d1868cd
F20101113_AACWCL silva_h_Page_045.tif
d08114374624fc9706b634b0a2b84f10
c18c14cef01cc8cea48834ea146725d915083e6c
20466 F20101113_AACVXF silva_h_Page_120.QC.jpg
426e765622d45404e6a229618fae87d9
18d41196618ca456fa370a41645f5afa45531a14
2985 F20101113_AACWBW silva_h_Page_059.txt
6930b83425a9007e897fd0ddb224cb46
496cfb93465bd8a9fb6df4432c6d75f0ace0123e
F20101113_AACVWQ silva_h_Page_116.tif
a3b4010c1a4c629f7f393a7ebec6f860
451fa9a032218fbe53b2cb0743ce0e606cb4fab3
2993 F20101113_AACXGC silva_h_Page_006thm.jpg
d1912a78aee5782dade3c2a19d41abfa
9185d036317cb18e716e977c05378faaf004ef6a
6742 F20101113_AACXFO silva_h_Page_142thm.jpg
1d2c97a70be701794600e9cff2165000
59d8b68c62e3565c175f59af3befa7310d93e0bd
7198 F20101113_AACXEZ silva_h_Page_169thm.jpg
afa6c9f5a9f6a2d21e34546828b63250
84812ca31f9887dfec269bd3e48f08c534070c10
5832 F20101113_AACWZT silva_h_Page_048.txt
f2c325d78d3b830755d8409f6694e0fc
243679892c00b4c6c7a975cf045e411978dc6a71
F20101113_AACWCM silva_h_Page_053.tif
23c5feb14d99bedd5981f462abce2e63
d42acab76c2183aab23c5a067b6ba75549ecc5e8
107599 F20101113_AACVXG silva_h_Page_028.jp2
a52d99b07fa7e40bfc52ee497db0084d
9427cf2d996985ea126ff9bbd59fc37d15282298
F20101113_AACWBX silva_h_Page_137.txt
18a1e984aadce44a2437455fa3be2e78
65f5b62140dd6e8fa011917b015273eaef766a04
23291 F20101113_AACVWR silva_h_Page_130.QC.jpg
62876aedcb61952e51e26fbc8bb19f82
75a63aa27ca75a72064a99d25628eb6f208d1a0a
26935 F20101113_AACWDA silva_h_Page_126.QC.jpg
7b82231a739efa3ddae4a08995fefe62
a9ddb4f0050f834635177d7a831671a1150b1aac
24138 F20101113_AACXGD silva_h_Page_021.QC.jpg
46998f13f811be1018e8f6da26fa5b11
7201192453aac06ddb0a853ade053ea53dced76d
25579 F20101113_AACXFP silva_h_Page_010.QC.jpg
c2ceb7f626fe9143f68b0200a557b16a
ff531979be482e2760592c6a2036831d3dc35ebb
2859 F20101113_AACWZU silva_h_Page_049.txt
8876beb34c881824a11e29fef92c4adc
605f8dad99bc81b4f29e6ba7cf3e36943875bc43
83128 F20101113_AACWCN silva_h_Page_041.jpg
624c50d7b5fdc6ca961b851bc146996c
12fec9a3de1ab8d628ab73c9ca4f4a017b04389f
2027 F20101113_AACVXH silva_h_Page_120.txt
7096339867e903c6b19b91373c806389
2eb0ae3a700f47be6db731ec3c6ebb8c1a5354c8
6346 F20101113_AACWBY silva_h_Page_028thm.jpg
1b8ed30196d8fab343a96c2345fb959f
579572ebb530a54f4f86689475187de0d4418488
21288 F20101113_AACVWS silva_h_Page_009.QC.jpg
e8285b7421d5d083adfd00bc58aa6b2c
139479e377622822cedb07bebb1ace8d864ea29d
16062 F20101113_AACWDB silva_h_Page_151.QC.jpg
5cae75cffc2bd0ff9c988a97cad04061
18fdb1f4f23809c6ec21d82c247272d5fd7c525c
7053 F20101113_AACXGE silva_h_Page_127thm.jpg
818c8ffadbeac0986da9df8f3e5322d7
b6639da252dd858cafd75d0e3132c9e42a45fea4
6657 F20101113_AACXFQ silva_h_Page_069thm.jpg
283f3fcd99eea7c3496aa7ca6b55c235
cd42a05bb38817bd0797d77c7fec180b0665002f
3338 F20101113_AACWZV silva_h_Page_050.txt
28edd1adecf85e5575a0285e8e4d7696
5155056181eede0545bc2e091312cf68599775c3
F20101113_AACWCO silva_h_Page_031.tif
53e719db1ca3c4aa0ef99174b953f142
c6c503e613a41f11f8c247210c9cad95cafb6cec
80382 F20101113_AACVXI silva_h_Page_038.jpg
6c5262f738528feb0cba9a218a2b2bff
087d559231d581f2522d6bda26841619978d1113
12006 F20101113_AACWBZ silva_h_Page_104.QC.jpg
7d86f7712574721f35d950681bf585cd
1124ec3b705b17af5685a32accc1a2917a1f07a1
7302 F20101113_AACVWT silva_h_Page_135thm.jpg
f2a58f168b8d8696bb1781a27231f88c
016972ebc0af21476c0678329347496931e3e9bb
58067 F20101113_AACWDC silva_h_Page_020.pro
ca0815f0b52f50f54a492d24a4857023
abec4ebf1241dcbb57a21a11d929a71a645738f2
22919 F20101113_AACXGF silva_h_Page_066.QC.jpg
228484f38a2cc260f0e0e7343c386cdf
226fa929dd26aa0220a236dcdb0132a2072aba96
13709 F20101113_AACXFR silva_h_Page_056.QC.jpg
244a4d685abe1ff434509dd1215ef8b9
95a038519a4c45d57193b7943182a5fec836195a
1821 F20101113_AACWZW silva_h_Page_051.txt
73a80eef7a478f94c6b371a624ecc2f2
de2857ac86ca349018e8c5f114551d16d64a711f
25190 F20101113_AACWCP silva_h_Page_159.QC.jpg
cf763a4e92b160a7051ffe2051bb3314
b0c7bea36c7bd4f49df35e43b6a54209aead3df7
87798 F20101113_AACVXJ silva_h_Page_167.jpg
fb5b87b803abdd395ffa687a31413d94
2ab845b718dea0cb9537064f449712b0726c899c
F20101113_AACVWU silva_h_Page_092.tif
0d96fdb485eb9ff613b04fc02c6c3514
cdadd2250c94e129195c976b5a515e44d165b183
80725 F20101113_AACWDD silva_h_Page_071.jpg
d56b215055d68114fa0eaf3831c77d7d
41738811a1b1f6817cdbab23a4339a9378cd6579
15393 F20101113_AACXGG silva_h_Page_101.QC.jpg
a489346605a051276871a192af5cfbc1
3d926fc20c8488b2ebdddee12b814d5a458c4aa1
3886 F20101113_AACXFS silva_h_Page_056thm.jpg
e78b4577c5ea14cd0600437a5c8fe449
c9b43b3d362443cbb2720316f688b85946080b95
3628 F20101113_AACWZX silva_h_Page_052.txt
c2f7b733f84f3e8b43f37004b6ca6014
6cf4819d279a632dd7cc354ba3cc7d797006b84d
82778 F20101113_AACWCQ silva_h_Page_106.jp2
1f3ed157e324af34ad3fbf7eb49adb9f
9cc0b2bdb0f6f83c414ecd8b72a18332d2d90763
F20101113_AACVWV silva_h_Page_050.tif
506a2d7d2e9b19f62152a5e0c1936e8c
ba1d6b06f103588e65ce0eaf93b6592346a0c0db
F20101113_AACWDE silva_h_Page_037.tif
83600b4592435ecb5977beac1520ab72
314a59a5cf244d92020f70a4c276bea0232cc6ff
24665 F20101113_AACXGH silva_h_Page_117.QC.jpg
be63657ea324ae2f4998f5ff2ea16b2b
748e0feecae9b301dbe800b45defc65b79095007
6790 F20101113_AACXFT silva_h_Page_090thm.jpg
a068882101231caf180d65f00a9bf14d
4787b02880a28ce7f2093b6dd555b7068e171c55
3078 F20101113_AACWZY silva_h_Page_053.txt
81b12055761dc6c68fda6bf266e3ad5f
0db0952921a5fbcdf68adf8ea65adda99fc10643
4480 F20101113_AACWCR silva_h_Page_047thm.jpg
330102033e26369ee91428a46eef2368
92d5d0ded2d64286489a379f8dc3332b8f161e81
1423 F20101113_AACVXK silva_h_Page_161thm.jpg
887da84f6752fb99a22c4bbe6f1ddd88
3074c8c89233455847b953da21445f7d4ee217f5
121570 F20101113_AACVWW silva_h_Page_097.jp2
3526b9869e5d91e174745a2656c0f5c2
606b9ac9b23516b8bedb1781038666683970ff74
2713 F20101113_AACWDF silva_h_Page_136.txt
e6c46c9ddb68edf47af5129e679e0302
fbe830d18488a2d078346f5cd054902536f27c4e
F20101113_AACXGI silva_h_Page_048thm.jpg
b74b1be8d1fb6951cfdaa5c5bfa3b3eb
e0b74df43f322630db5c061f4017786691effa55
24007 F20101113_AACXFU silva_h_Page_158.QC.jpg
5e4476dfe6e16b68d740e00441eb731b
933983f0e054b939c9de79a3a0e580c6a92ed8ad
4568 F20101113_AACWZZ silva_h_Page_055.txt
b3ca1f793408c41cce6a0702da83d6cf
7182cd347ba3b5eb88b1d9cdc26a5486054b946a
2357 F20101113_AACVYA silva_h_Page_091.txt
834c6260eb317a2ab1b1b6232551a7ef
c6093e7c3d4571421eb72748a57e1b1367b9a419
7010 F20101113_AACWCS silva_h_Page_113thm.jpg
91eb9222ce9db46a3634a866c97f9304
bb177e5ba88c155a817eb49d761072c0277cfb92
27937 F20101113_AACVXL silva_h_Page_022.QC.jpg
43f3aafb2ec333ba2b276af282c4c3e0
aa380b3325edc952c5a6dd1ea763f24da9a113c5
106542 F20101113_AACVWX silva_h_Page_148.jpg
87e6ee8e80961799aa4c7a7330cbd37f
0cb733661d401069992a6034a6e5d450792da930
6905 F20101113_AACWDG silva_h_Page_099thm.jpg
28783ef8bd2b8ba8678ec24617496c94
2441b85d097623c6b2a7bd2da92328c611e9ec89
7218 F20101113_AACXGJ silva_h_Page_042thm.jpg
49368917c970faaaa9e84dba5f4e0970
3502776984c13098d7cc8b2d7df1b67c9bc484fe
25997 F20101113_AACXFV silva_h_Page_062.QC.jpg
a92a18f4e485f87e1cd0e2ce5cbd3f5f
cec46e560416b6deaeac0cb41d0f79d52a35b09b
68831 F20101113_AACVYB silva_h_Page_077.pro
479e6db79d9cb76b958b562341f56654
4adec5e4eb1c15c5ce56d3e4b79d6c1ea57b6d36
27588 F20101113_AACWCT silva_h_Page_128.QC.jpg
4ba66650d8711d9f406ec361d0972105
babd968df76953a782068b885e79a68878eca04f
148397 F20101113_AACVXM silva_h_Page_035.jp2
696ce7af908f0947ca3c1a792533b692
8b0c74ea31470872c5a8386591d2693a5fd3aad9
1051952 F20101113_AACVWY silva_h_Page_046.jp2
4c28d98f67faa39f68380dd582a20c9a
9cb99f84113a45077c1013910c24b304f72e1191
2781 F20101113_AACWDH silva_h_Page_026.txt
53a1d39a5a71c7496aed6e5a86ac4579
0f5a3a4a6a1a3aee160a8c621938f30d4e00f62a
4718 F20101113_AACXGK silva_h_Page_058thm.jpg
9a3d08c547314318c06e5ea3847798aa
aa85db0a00e46fc378ee43aac74f1799e7e5ebd7
59286 F20101113_AACVYC silva_h_Page_153.jp2
cd98ceba4c17fb1baa2631daa3338fca
9d47df436e170ce028488492546c8ea3d5330e60
112453 F20101113_AACVXN silva_h_Page_123.jp2
abcd3a420013db056e945ead7d198089
dcae77f407b69aaa9e0e6f2ea941ee646e17fe4e
7059 F20101113_AACVWZ silva_h_Page_144thm.jpg
75477992d9ca9c269ed204ee66f0bf72
3b409da163c5f7031a51cfe4544087cbfe60c893
76947 F20101113_AACWDI silva_h_Page_013.jpg
7f09fb7236ab4a64c7815cf7bfa77bd2
cd3c7e6acdea6fa6480b1a8ec134df6640a5db3c
7326 F20101113_AACXGL silva_h_Page_095thm.jpg
18728e3d9915af3e1988ae729c018e0a
121999d63de05ca6b41c9b921c9e991141a82dc4
6989 F20101113_AACXFW silva_h_Page_031thm.jpg
5bdf556db659ac32bc6715159448e3d3
ead4e182f3f10a8d8fc40a847c24899881698506
F20101113_AACVYD silva_h_Page_172.tif
486aec2e9db1df8751e6c3f8a7527849
1205b0e0a888548a89133b6e96f8ccd8cd37ceee
52673 F20101113_AACWCU silva_h_Page_121.pro
c560b17db4fa149a98c51a0e2988618f
af748aede021055d69207ecb2cf153125dd1c2af
3042 F20101113_AACVXO silva_h_Page_019.txt
2b87fc4fc8a53d5f20331a69a7f58009
b1502eee8cd88b79f4002126df270703efa56526
53863 F20101113_AACWDJ silva_h_Page_049.pro
e069bc76be0a15054ff9fe679fb5e656
f2659bb02c3448d34b057cec730f1482c78f8179
22415 F20101113_AACXHA silva_h_Page_165.QC.jpg
3eceddc97681038f85e4e7f48c6ad6c4
62d9c07a535d6d30bfa5f61aa0ede56351abc90e
28558 F20101113_AACXGM silva_h_Page_018.QC.jpg
37e80de2fa568aa2c6277e4b4864abff
5891400f21dc85957573734d3101291c4894dec6
3866 F20101113_AACXFX silva_h_Page_161.QC.jpg
4ad93e9744688ef51512dd3f56d49e96
f883209bd633334b38fed5a3baea5d03ad81c92b
53768 F20101113_AACVYE silva_h_Page_024.pro
2be4e143fa0f517de4c682d3a62d9a6b
88196f129e26d95d90efb0f0405021375449938f
73523 F20101113_AACWCV silva_h_Page_123.jpg
ba6070d29c6004be6f23a8873a5591a5
b36c1ace9d628b13a57aa88dc0e7354d2e0c1794
110525 F20101113_AACVXP silva_h_Page_025.jp2
0fee41c996d66fc6d4abc25f12a6778d
ee44a9e4b7c8db97ccf6c07aa584ec67b28a09b3
F20101113_AACWDK silva_h_Page_072.tif
0387af336f407dd2d08c05a2b90536d0
1a4e1a610a0c3a7aa09ef042150936c7cfe58738
7038 F20101113_AACXHB silva_h_Page_167thm.jpg
b4e98f02af32f5406a0dcb19ee6818c3
55f9a3a09f37d5a1b7bec5b95c85cf08720cdb15
2222 F20101113_AACXGN silva_h_Page_001thm.jpg
19416d2705239fbcd59088df03440831
e62f927401b3ab87ec720619336bd6ac5484f7fc
26001 F20101113_AACXFY silva_h_Page_042.QC.jpg
be48771161541a548abc882e65dc4a7b
e934891acd82813ef6dde14133f4072161f99d10
3560 F20101113_AACVYF silva_h_Page_054.txt
74ac41d7d4f15fa0939604abc834855b
2064d6f4bd22881ed5f7b4d27a6a437c3ff8d799
85763 F20101113_AACWCW silva_h_Page_037.jpg
676711e54abfa1899594539875ec0250
56a8f2f4f81ade159f57f3bdbdb334517b84ac38
62181 F20101113_AACVXQ silva_h_Page_032.pro
688913879882d571d8590821d0beba4a
f72bba46ffe362c0a45fb23fd26504971ff31067
2210 F20101113_AACWDL silva_h_Page_080.txt
e1a9947c8e27e81d9d51b21bff58ff62
54b663dc3b9eafa00447e577124cc2e837b63306
27085 F20101113_AACXHC silva_h_Page_168.QC.jpg
367f15da2b53cdc814cf4b1d7ff6aa63
0739a46cf3d6d38533926df555c33c3ce0b46045
2364 F20101113_AACXGO silva_h_Page_160thm.jpg
e1fcfa56434483b34ef60554ae24da71
6dc53026c021ac87f5509e837dc2d09f6be2e582
26132 F20101113_AACXFZ silva_h_Page_039.QC.jpg
79b705e2f47875bda582a3f0e26da833
391d34c73ef8b7155d02146eff898127388575f5
6602 F20101113_AACVYG silva_h_Page_157thm.jpg
83924fa3256e38bf9c89deea691d5fbc
1c202ae7f4c5f56d2685d887516557ed8ef02d37
20863 F20101113_AACWCX silva_h_Page_105.QC.jpg
21dad0252d25517113e603f76e2f53d7
6ea3a40912ab70cf4bfd4d67755fc26548c72956
19826 F20101113_AACVXR silva_h_Page_108.QC.jpg
3966e9fe8d82378e0f1a7042116e0d63
370ad286a1b4ab997f5f52a51a6b6976ad14b482
77497 F20101113_AACWEA silva_h_Page_159.jpg
5f6a5b8cf3802a7a5904d451430aaeb7
a965ce8b5488145157d8d3b794d39c5c5f66a1ed
108671 F20101113_AACWDM silva_h_Page_027.jp2
c2e8759643c5e81bb86f7112c76df4b9
52e60e665c61dd0b8bfb67f2d6d583cd00fa691f
25732 F20101113_AACXHD silva_h_Page_075.QC.jpg
5f2716d06182f3266c60ccc2b4339c23
3f940fe893102c6507afa738704ca7242b77f87d
6951 F20101113_AACXGP silva_h_Page_013thm.jpg
88a429762c587fe7dc3ed5136e8a738e
133f8f695b17055714ac10536d05d79a4910f574
F20101113_AACVYH silva_h_Page_080.tif
960136a8fcb265c68ace1a7628c04110
1e339b9570e7cb255fff588164722ac38f21a30b
73704 F20101113_AACWCY silva_h_Page_016.pro
c86248caebdb7b7849e3ba5f13d285d1
e84e26c1b261f2caa84ffc10b6ebfa43d5786c4e
51130 F20101113_AACVXS silva_h_Page_158.pro
93dd771310efd488e940424fb1e7f417
519708f439730e9b28bd07540540a5714b5d6791
74186 F20101113_AACWEB silva_h_Page_052.pro
cf1a4576619c0e8b186779a07d4799bd
42cc630974d4b84b546004906b1f0a1fba648353
6791 F20101113_AACWDN silva_h_Page_110thm.jpg
ab461a3f7dc700594f4665a097b737b3
8c40e651701a733be7e4d787450b601255f570dd
24899 F20101113_AACXHE silva_h_Page_139.QC.jpg
2b492bb74e35476c6f16e0de1c998820
dc01c985f7f4ec180059fab28fc34eaff9b65d8f
26701 F20101113_AACXGQ silva_h_Page_037.QC.jpg
f42accff3331144a70bd65b59b06d628
3972ddd32b31b737996ad35eecdaad4d00bf4439
3201 F20101113_AACVYI silva_h_Page_002.QC.jpg
585800b471d7df58cc3ede158a661832
2522b7cbdc785245992571e143dff1fb46163019
F20101113_AACWCZ silva_h_Page_030.tif
ab3dc135f2ebfc787854d903abe0f875
da03f41a716c8d8bc0d77198534616ec6f90438a
2180 F20101113_AACVXT silva_h_Page_112.txt
97837538049ef5f4d216b573ba58b6c5
ad665d773bc153201848b02fac2df42738d7a6b1
119391 F20101113_AACWEC silva_h_Page_146.jp2
8c78ddcf6077ccf039be64eb71e84656
bae49fb2578af255a49245ec17f538c7076b69f9
5732 F20101113_AACWDO silva_h_Page_105thm.jpg
ed75447e53be844eb35fd6255cd72ff6
08f99841658f00fef37a7440241a95ab5b1a868a
23549 F20101113_AACXHF silva_h_Page_069.QC.jpg
12da6acad8a6d5dc38a0e9fb4b77c781
a6892c1020b12b548ec0f00866d1ec4136ad67be
7184 F20101113_AACXGR silva_h_Page_132thm.jpg
d900a4e016d0d42dac5efe041c7da73a
69bdf0abe443292ded5076a9556582f32dd97c54
5765 F20101113_AACVYJ silva_h_Page_004thm.jpg
29acb15fd842657358294fc00e91c3f3
63f9bfe0d6693afdc6cd9de45894c0a89ace3c64
1051902 F20101113_AACVXU silva_h_Page_026.jp2
ef9b758ec3f08e5a41a1c51052ad1e0b
fd24222788205a66a673d6f5479114fb976610ac
55548 F20101113_AACWED silva_h_Page_015.jpg
ac4ffe9e831b0c7c851996dee3fdcf28
1dd69dab1e8f0041748dbac9284882d77cf068b2
23099 F20101113_AACWDP silva_h_Page_025.QC.jpg
9afc6c2ad2c95f0b80c7d693d2b54d46
6ab20b108f9d2d7b7a32023a0e6c14bef1ecfd8c
6038 F20101113_AACXHG silva_h_Page_120thm.jpg
1662c83fbfc3247ad77066958cf65110
e3854bdc05b1394dc5955c794cf6603bb4226932
8368 F20101113_AACXGS silva_h_Page_150.QC.jpg
eeec89c83a1bd1d5f60dc705fb33535e
c5af0f0e98facf7ca7401664786635175194eaad
3971 F20101113_AACVYK silva_h_Page_046.txt
b68e8d883db6c875d7925dff1a808ed9
ce124ef466dfbd6ae95d2d68f6f3010cd4642318
86418 F20101113_AACVXV silva_h_Page_169.jpg
bd6ea67b9d01645b85f1bb88650c24d6
e4fd8bfda96079a9e261ddcf2ba7a71be9a9c4eb
161491 F20101113_AACWEE silva_h_Page_164.jp2
3f74be16d234fb786f651939a73567ac
570b654c1011bf013ac887a24d13d2f3ca848066
80948 F20101113_AACWDQ silva_h_Page_044.jpg
52615b28848cb111417c54e13f5faf8f
2ab370fa27dd0aa73118cd8f138f84563558ef9a
25658 F20101113_AACXHH silva_h_Page_089.QC.jpg
a6cb7e32803a27fc241d40d5a2632578
a6e943203bcaa277dc83d04a6c5de80a6573b2c2
25558 F20101113_AACXGT silva_h_Page_063.QC.jpg
f8259380777da1a3a3c5725937f6bd1e
13f6bfdde658cb64bd4869b38ffc3b49293b30ba
6885 F20101113_AACVXW silva_h_Page_010thm.jpg
5e24e8a4709d8270f4e4f2192b79cb43
3b211841346a953ec15b7076478261a52f741c21
6869 F20101113_AACWEF silva_h_Page_146thm.jpg
83e422e6c2a4f3027503f0f8e416710e
178464fc39eed4e8032ea6c67266595418f82fbd
F20101113_AACWDR silva_h_Page_081.tif
968efa48cb677943fa6c166b668a38f2
8672522da3e6a2faa38a7e0855aadf180b7e25d3
24505 F20101113_AACXHI silva_h_Page_020.QC.jpg
764eb89a75cdcd027233f733b1e166a4
c5f8ba6ba7172a37cbbfbd3ce762700a66fd616f
24960 F20101113_AACXGU silva_h_Page_097.QC.jpg
61f770d47aafacf93f4cd3db8598e4c7
8d52dd9dee7c020dc9203cfc0f1d1909ca2c304f
2009 F20101113_AACVYL silva_h_Page_061.txt
b60fb62a4eff3031bbc8f6c944c92d55
f9865fb84b8c5baa9f00718588084fc013946e1c
23037 F20101113_AACVXX silva_h_Page_027.QC.jpg
c3bb5544cdd7df06b057260dba4f9212
86ae12b6147960321e0e26aba438440317b78e78
3756 F20101113_AACWEG silva_h_Page_054thm.jpg
79cabbd8ef839b8808bf500ac883c8c4
0c6d05f5eecf698a4e1144fa4b988adba986fd13
41158 F20101113_AACVZA silva_h_Page_147.pro
919509bad43b429eb318bd88619e4a51
4297bc3e30970adcde9dfd07bdf60f5cc4e0b19d
21842 F20101113_AACWDS silva_h_Page_001.jpg
242b94e3db5eb6c50dce8f1bcbd44727
90896bcab6bf79894516765df3098f44f350acae
24941 F20101113_AACXHJ silva_h_Page_014.QC.jpg
e58727a0a8c87aa157d200a425bd9b39
da662298400a7f62783e534daa1f4f753e563523
5942 F20101113_AACXGV silva_h_Page_005thm.jpg
66e709703c5b0b62001be75b9e347df8
6eae53f7c73282587f5b2cf0c22bbac4ba08ddf7
4919 F20101113_AACVYM silva_h_Page_060.txt
bca07bedeecfcdee256a1879c9110fa3
1d8a7b3eef153f6b3b5f7b07fb17aa54c0df1498
61035 F20101113_AACVXY silva_h_Page_106.pro
5cea4bf4a9a3743882b9bfb2895a89a2
d0e1b67b1a7cff17f62f6c0fa7f57e8ad9b5af50
891818 F20101113_AACWEH silva_h_Page_163.jp2
03fbd76c5ac9b9bf857a8dc4b146b8a5
03119d6ff9053393948b29b152db518a28ecb885
142417 F20101113_AACVZB silva_h_Page_019.jp2
29253a7390b532009e6ee9d154685fcf
d32a086e249ce8f267baccfce1c46ee79b291d0c
18754 F20101113_AACWDT silva_h_Page_015.QC.jpg
8fe0c77cc96add4684a161c655953e50
71e1fdc8cb9f01b58d1744e474860015d28656b1
5877 F20101113_AACXHK silva_h_Page_121thm.jpg
e792fd0db44ed3a1d9cb358ae47b7ef1
f984415fd6f14d1dfbcafe74340fff05bc7b3f54
6918 F20101113_AACXGW silva_h_Page_118thm.jpg
7204c5e142bda8efe4af151d3d77d600
1895d1f294e16e25bae6e3ab89345dd0dbf78ed8
81679 F20101113_AACVYN silva_h_Page_118.jpg
027f93919396d46f361cef0292b25e6f
92bfe50b7ca603412b9219d5fd2144622ae02658
25494 F20101113_AACVXZ silva_h_Page_033.QC.jpg
a9a8e02c8b5d6a023c8dd286a1fec81c
2ab71535bb49cbbda84a31d787b500c7af1e55f0
7413 F20101113_AACWEI silva_h_Page_022thm.jpg
fc898ab82999f9233b0e9d905389cd27
0946b24c34faddc46589f92a58b61d53569b3f63
6911 F20101113_AACVZC silva_h_Page_089thm.jpg
f37834fefd4ea332ccb732d7ada75e85
1bf254d1d61803d4f8777c7f0b950ecef8b1d25c
112319 F20101113_AACWDU silva_h_Page_157.jp2
f983758f14361cf76e36a67f334d8e29
7e8a982ad08ac99dc633287142a2dc7e5b851a7b
26209 F20101113_AACXHL silva_h_Page_116.QC.jpg
eb08ed3b92cb6e6e1a89f51a7b1c53b0
d4d0bbaac0384d64c026e1118c813e46355de15d
122846 F20101113_AACVYO silva_h_Page_044.jp2
c20f9481807857b00720de6335f81594
f9d7f3cb8324a14faf72addcc7e78411e1e1347b
87841 F20101113_AACWEJ silva_h_Page_124.jpg
7c81c4f2d240f5f5de1e71829d3ed1f7
fd353e5877216226c360b280b6648a760eac3a38
6960 F20101113_AACVZD silva_h_Page_036thm.jpg
1f74006cf8f8532d108ffadd78053319
e7e9b47f828c7ea51b701da3e5aa0308ebfe1ad9
7461 F20101113_AACXIA silva_h_Page_035thm.jpg
09be7a033931d44b20df0dc4250297ce
26dfa3a28ccd3bcb4eb3cbb65e4aa89e16bcd168
F20101113_AACXHM silva_h_Page_041thm.jpg
f7f42432303ce1a6fa3e82b7180b3825
59711de02d133c8da753a8e4381265ade58cf673
4793 F20101113_AACXGX silva_h_Page_151thm.jpg
f97ec1bc0c0dd041987243c7f0f39581
82b50fe2983909b74f1db15bcdcd573286c90f8e
44987 F20101113_AACVYP silva_h_Page_120.pro
e25c20bf473788decead2bc11e90e37f
dedf5e19558a7cb3fc5be1a3c0d0dd15d39be1e3
2349 F20101113_AACWEK silva_h_Page_078.txt
a7bb85e9e0ba42a17d156f5e907070fc
fa8495c702f88c4a2c8370c5bb575cd22b8599f6
121517 F20101113_AACVZE silva_h_Page_109.jp2
607ce2278efe17550f79e99a1d69d957
20a53c719c48125a220b901f993f446096f34179
83176 F20101113_AACWDV silva_h_Page_132.jpg
048891d522eb281db9ac8c15500e44f2
dd61c716356e0101ab002faf8cf52322aaba9024
7040 F20101113_AACXIB silva_h_Page_168thm.jpg
4c1fd40dcae7016fef40d34133c5043f
b4c1dfc5fc1af47a4bb1c20e61b20fbfc8f08058
1373 F20101113_AACXHN silva_h_Page_002thm.jpg
9253dc4a8077f4b96aef16ba46efab66
b27e69fb05f390e05ed5e4dfbf6207a617123a96
5291 F20101113_AACXGY silva_h_Page_045thm.jpg
03da2b2d11f65a2185eda4f93e5bb1c8
a9fb0bff3b25c2ad0ba5f1115522df2f2ca9304f
16412 F20101113_AACVYQ silva_h_Page_107.QC.jpg
f74e15b538ef519d2a56c4f0b36723b5
949efc0d937d4f7d2c44875b82a6405adb01a289
3655 F20101113_AACWEL silva_h_Page_151.txt
0e39530d669abb9f211b9767224d0600
21ea1371fd0a2846146501d4ab9d43a9747fc556
116682 F20101113_AACVZF silva_h_Page_117.jp2
410f64c44d283341f15ee6a8cba17d7b
37a498fd388b591076035c6788f95e5c98fcd8d0
78940 F20101113_AACWDW silva_h_Page_134.pro
d5c8ba6072bbaab214458d93043afa43
91405f8f6209acdc37261c2b661789e97456c72b
6562 F20101113_AACXIC silva_h_Page_145thm.jpg
653e29f5c21df0fd1eefa60a3bc4de8a
c562d4d980bd11fb2d4ac7b1182901e4a25465fd
6970 F20101113_AACXHO silva_h_Page_091thm.jpg
4e4c753210f3ee72832e11430d37df46
bffd03a9e2914cb748191e62386efe1ada079fa9
6593 F20101113_AACXGZ silva_h_Page_123thm.jpg
f057c64addc97ff5531a69532a1e253d
d34314b2bbedd90ddc08b703340ed41758e393a1
7102 F20101113_AACVYR silva_h_Page_140thm.jpg
3b799236541c02d31f61df19faa0dcdd
035ef26235243758fc772f2afcc4f69babed3d2f
89966 F20101113_AACWFA silva_h_Page_016.jpg
ce18982f7ea5baa84749c0e31a78305c
53caedb23c3d7835823d5e9565ff19d07ab45b1c
26443 F20101113_AACWEM silva_h_Page_132.QC.jpg
544e9a42ff4e26a19fcfcf582cf17347
f1c02622600332d3073d322173c4f38ff54ed71c
F20101113_AACVZG silva_h_Page_161.tif
1ceb563cee43e577c4a51ad6200093b9
1267a770efdf8124f31c3eb3c85f460297ca3d0a
40921 F20101113_AACWDX silva_h_Page_006.jpg
ddcba38ed0d24c1cdcb1857a28bacdd4
d31b3481001862c6d4fc2d0bba7ac224b0fa7e4e
25373 F20101113_AACXID silva_h_Page_086.QC.jpg
dfa93855a83d1455d81fd7d9e736a323
2b2f905c398bf6cf3a505b30c7226dc44b35a41e
18331 F20101113_AACXHP silva_h_Page_045.QC.jpg
02b3db7406a32df47557c0545ebd6ff3
b92294e160836c1153387648b360402561d44b6a
F20101113_AACVYS silva_h_Page_145.txt
9bcde539810959def6e8ab1fbb2e7fe5
f0a18b4f111fd713e36be87afd07784101755f01
77252 F20101113_AACWFB silva_h_Page_017.jpg
161e3bd5058f609040e68a4fe92c9cbf
ba42bbf5b3e36cb35da461a3539216663f7b5cdd
75271 F20101113_AACWEN silva_h_Page_020.jpg
7ece6a27cc8533b88275d0a59eff5af1
eaad886ce274cd10ae544f2a3e7db118a94a4b06
1051984 F20101113_AACVZH silva_h_Page_034.jp2
194cd824a28ff6eeb9fa397bad3c663a
dc2bb992198281273ac082375cb9529b074aa40a
F20101113_AACWDY silva_h_Page_169.tif
d074b41a4188d80e9b6e4525baa988de
14deaa7ddd73e407c61e6ac5c679860579ec21de
7885 F20101113_AACXIE silva_h_Page_034thm.jpg
105c84a8d1715b3d04120dd0aafe6aef
e64e88c2f1b74ff8a452bff4f02b7166e25d3c4b
24963 F20101113_AACXHQ silva_h_Page_017.QC.jpg
df58af7622636871c6247c7c2e321a38
c837924a846a9098a658174a502c906155709ba9
98540 F20101113_AACWFC silva_h_Page_018.jpg
05ef106adb85f5427189c23ad7699938
9672635f972225d8b96adb81deeb6160bb9b3832
198026 F20101113_AACWEO UFE0021070_00001.mets
8e1f83f6afc37c26883f68167617a833
684a4dc2b25dbe3004b969820711d251df6fe9b3
1938 F20101113_AACVZI silva_h_Page_073.txt
7d47bff8859efb9bfd0f64a5a608e594
03e7c4ed2906a76f142b4816511356c3f01d9ee5
F20101113_AACWDZ silva_h_Page_062.tif
8c3d5c3fef70113b7cebe29046ff57de
e20f84e0482251adb422147efb2008f81f2e581c
F20101113_AACVYT silva_h_Page_090.txt
ebe6be053b1ea398f699615913eaf967
1fb6c96574edb5d706ecc47f1c6b213bfef59fca
7171 F20101113_AACXIF silva_h_Page_138thm.jpg
3f04ef0a28b24e4ce10f67a037725c41
6627b25e328ef58eefc44331ac779bfe6304c594
5841 F20101113_AACXHR silva_h_Page_108thm.jpg
ec69eb5f5d86ae8c933ea377fdfbd38b
1902a30d0467a8d442a47a8d2cb4bb09e3dcd1a5
92575 F20101113_AACWFD silva_h_Page_019.jpg
bdc4dc191491e9900ef685c244811ab9
2bca5a1b94b3e2fd28108f6e151eece8a84a1e80
93001 F20101113_AACVZJ silva_h_Page_095.jpg
4284bb7f810d12ac152bc04007395e65
e346811f70f67b3fe7748c46402d915ea7f7c6bf
F20101113_AACVYU silva_h_Page_006.tif
87e6a09abc45f48a89426881cc686aea
a3d9f6bc46983cbd12c0b6917f8997b66ed01254
26583 F20101113_AACXIG silva_h_Page_127.QC.jpg
02aeb679fa11eda1d1a134895c8bd34f
b89e3a4ac7e89e53ecb2ad56c57080f8de4f9589
6876 F20101113_AACXHS silva_h_Page_017thm.jpg
12991d93a4d75b68eb17d20370067365
07c90f107978c55422bf4d62a317f29b0af970bc
91017 F20101113_AACWFE silva_h_Page_022.jpg
b9934611f3fbb8bcd4f80407e4488d33
83fd96aa56b4866e6aaddc1adb493283ea0a25c4
6500 F20101113_AACVZK silva_h_Page_143thm.jpg
995dbba2e8d5c0ee40c2ba0f42f2370a
c59602322c93c9217d36e950c6643b01cf8ce989
74921 F20101113_AACVYV silva_h_Page_145.jpg
604d24358a3d3b5b04e2afbce3cb88e5
ffb8e23c87f157b4edb0e236b860e2d856c96db3
3438 F20101113_AACXIH silva_h_Page_053thm.jpg
2ff3572793012a6e3c1b5a53ef72d34c
60bf578e17a485f209f6e2b5963be034d856ddfd
27999 F20101113_AACXHT silva_h_Page_148.QC.jpg
9b98f2bf1e0f60534673df63b0f44a2d
139ef1ccbb31a1ff21100dc63e2203ffb39a7067
71756 F20101113_AACWFF silva_h_Page_024.jpg
3ba70a248223e53e46c69ccb7cbcf844
9a4ed2ce6eec136572cae6e07a97908b9a87a755
10450 F20101113_AACWER silva_h_Page_002.jpg
c78476f98ec5ed073e179a57b8380b7b
a6b6a3eb9b5eea1cde76dc352474e6fe9cb180f1
67890 F20101113_AACVZL silva_h_Page_037.pro
16479ff63612659bb9cc615dfb9107b6
0120bb53257829fba603120b8b75dae0e5c51e50
61652 F20101113_AACVYW silva_h_Page_167.pro
a29a36129974620513d46bd71b7e3555
e33935e1bc362bb5b23db93414b49f0721c26a0f
F20101113_AACXII silva_h_Page_007thm.jpg
325e0cef65435eae2c01e7179983fc42
07d263ac8c1408d1d90c48c0cd274023e49cbb43
6220 F20101113_AACXHU silva_h_Page_079thm.jpg
7e5ace094ba7cf97b8cf2d5ce436307f
e26f6adec17bc8a56c2f780ff0abbb3dda3ce9fc
71861 F20101113_AACWFG silva_h_Page_025.jpg
54964d6e74c4b753195dfef061eb0ba9
cf193a235de99acb948167adbd18a11346a43415
10786 F20101113_AACWES silva_h_Page_003.jpg
50775e362272f153d8995a100885a0af
5111b2b0d0be05a32d55d9c76b547b829572f042
59280 F20101113_AACVYX silva_h_Page_135.pro
1d30a180fd59936dce609ef4501c8e3e
29a2dcb6c4ff3d00d841f49b2cdbfd205fc6a69c
23518 F20101113_AACXIJ silva_h_Page_143.QC.jpg
5abd6da9df974e73f894a446ae39dd4b
4adfa337b4846682d5d6be3eca807ecc55b38aea
25787 F20101113_AACXHV silva_h_Page_140.QC.jpg
8693cdadaa5c1e5cb6af5bb87e68920e
b874d57bdffa511686d810d2220a26e4bd2a508e
96870 F20101113_AACWFH silva_h_Page_026.jpg
6aadc34f6f90420873e2d4e5a7c94000
6a3e72ec68c503a4d3c280fbd7414d251261b7a8
61333 F20101113_AACWET silva_h_Page_004.jpg
4c415ff2e699b9bf01b9e2907ce4db92
8c48e339ca599d184d210a96aa4c3f0108d8bb79
25563 F20101113_AACVZM silva_h_Page_041.QC.jpg
43a6774852bec147d4de55415521f2fc
01e6ed3a35e0ea155a89e425bb22772b9606ea7a
115089 F20101113_AACVYY silva_h_Page_145.jp2
41e01c6a3d0a7cb50a8e4b7b526042c1
eb886d81eefc2850b13221e7fa354d45d19e601e
6756 F20101113_AACXIK silva_h_Page_155thm.jpg
04799f092c0cdb55ba682d962230631e
710b3821e95facfdfa62a718c1c5eb903e0aaa9f
6606 F20101113_AACXHW silva_h_Page_032thm.jpg
cb9cf83da7751b4132864978426bad5e
6d128c8d0dcde8a290ab022edcb88019d675dfcd
72608 F20101113_AACWFI silva_h_Page_027.jpg
81ae3b4db2a6854f9d359801a1bc5a09
3fbfeed946b1eeecfccea0b327235ad391ff078f
74715 F20101113_AACWEU silva_h_Page_007.jpg
943747d6f5878bef47491b7d1c7e6959
6ece6824cea7bf8b4e304d9430f292522a342fbe
78197 F20101113_AACVZN silva_h_Page_154.jpg
512a4cf4ad4e3ea16030cf780036762d
43cc6de5254b7ec3a32cb6c4fa0390ec6486221e
2247 F20101113_AACVYZ silva_h_Page_110.txt
31328b4be0b7c3e89f7d3a0863783d57
92f603e933be5b56720b28899bb0fbd9c7b478f9
26690 F20101113_AACXIL silva_h_Page_029.QC.jpg
c68b701bfc1daf46b6e86204f1b08c8e
4f145d7acba41a7b5b5761c63c63cb3743f67dfd
6223 F20101113_AACXHX silva_h_Page_025thm.jpg
85d5b828eb358c7d0b59e1a9c3f606c9
53b82b5f227d7ac58d30f5f86e2dfc3f3bc59e74
71435 F20101113_AACWFJ silva_h_Page_028.jpg
0649f80af53bf02b7268d1ccc2a566b7
243205eea450e3d75f435d5a2a57d4756b422fd8
44611 F20101113_AACWEV silva_h_Page_008.jpg
6a48177fa99627061991b2de2514ce56
52d00d37c6be9b7c604c8175c84599e2bb353b13
2431 F20101113_AACVZO silva_h_Page_071.txt
c630c825dd9b0863e83d20886ca02962
aadb829c3c08f6ca470043916e59b2cabae046f7
4834 F20101113_AACXJA silva_h_Page_147thm.jpg
19bc21405b5d6af86b634d5287c46179
f767f992232272ea3a1348d12be6c272ff6ef62c
25534 F20101113_AACXIM silva_h_Page_109.QC.jpg
f2149763a2239d5d89e07be5ce0d9713
1d226aa903091bc22415ef063be61ac43d60babe
83920 F20101113_AACWFK silva_h_Page_029.jpg
060addc60c1502d8a92de8e8b525115d
088560141da07ac6a6a37f08c4e8b04816a3f0b8
91976 F20101113_AACVZP silva_h_Page_055.pro
a3ea6c7eff107b757128b3963407ec29
3e3b3a8872dd71f320cfdde23667277e080c7888
6858 F20101113_AACXJB silva_h_Page_117thm.jpg
cd938424a91e314a02a4862997d87c32
f2521bdf11317c1caca7af39faaddb88771d62c7
6630 F20101113_AACXIN silva_h_Page_093thm.jpg
6e34588bc4f38d9668163a4feb414646
ba230fe1aa6b4aca2c2471f1595d4682b1ac6f3b
6969 F20101113_AACXHY silva_h_Page_037thm.jpg
13a694b7430e9649b28fcd040ef862b1
e705f3863578b98281b42d91812fb2612b39cce8
81607 F20101113_AACWFL silva_h_Page_030.jpg
f01f326634b5b12829b3efd81a10edc7
72c952b2ba9f1abb30acbc4156d54d9fbe839d9a
66318 F20101113_AACWEW silva_h_Page_009.jpg
b1f3ccb9248727bd44abd5ff49356686
d51389eea1e7bb13e63fd39ee162370b4b852677
4007 F20101113_AACVZQ silva_h_Page_051thm.jpg
53f3111d25d6705863f754eecac15cc1
5df20e5c0d6532e73a77026386981043f0ae37af
6766 F20101113_AACXJC silva_h_Page_100thm.jpg
97e28192ac756909c5fca85c37eb0ecf
319df73e09581143d0a4c7ef99d6091c1bd92d68
3943 F20101113_AACXIO silva_h_Page_057thm.jpg
77cd3e3cd93305252247d5ca03741b19
a77d0863087e53599218d34fe3abfbee636191c9
7111 F20101113_AACXHZ silva_h_Page_078thm.jpg
cc2d4fe3f914a75c999800f8cd15cfba
49cfe14ec64c2658f5f6f71b8583b8de3f9e0794
86377 F20101113_AACWFM silva_h_Page_031.jpg
b594f9cbb2264b14ac81f373c14172d6
3c3b3f059d352ba44232d76d5634f5c3a833e7b7
78095 F20101113_AACWEX silva_h_Page_010.jpg
122b485f40edfad1235ee537929d9e99
1a1a16bd67dbcf33ea5a5a5b17784fe576dbe78e
6991 F20101113_AACVZR silva_h_Page_137thm.jpg
830a31b3bb4169a7f23c7e505b2f6b9d
2ce01210a202274924595da8890b3106a272cdc6
46269 F20101113_AACWGA silva_h_Page_052.jpg
a1e2d283e359210904b563a5e3c6ad63
2834516f4c0fbc2c497608000bce645026b0a704
7242 F20101113_AACXJD silva_h_Page_019thm.jpg
6c74d2b3ef59c922bd7661a50a01eeca
1bfede78fb77cee41f5cb92829ba9cc98ea4edf9
12834 F20101113_AACXIP silva_h_Page_103.QC.jpg
50fb09393449637b5b7b3f0c95e70a23
cf1579dea2f3cf982ababf6178d043716e2ef2bd
79328 F20101113_AACWFN silva_h_Page_032.jpg
3a13fcc9a7d6f8e7a9767bfe455e4280
9ef192e6ec516af88fcb4d6fa3b53cedd920f9a4
F20101113_AACWEY silva_h_Page_012.jpg
a72426a69b8a4829fefa74633cc13d5a
ed4dea8247f56ec16d51312f73ec385047842cfc
86434 F20101113_AACVZS silva_h_Page_077.jpg
377651d92dc50b2cd8350e542d7c93e0
66d3624a17ed6c78af0f7e42fa7fb399377b33a5
36401 F20101113_AACWGB silva_h_Page_053.jpg
fd257f5ecabdc85238f178aecfa9877a
ee2a8d34e8d9e0dbb24bdbe1ca0e231e5b091308
6512 F20101113_AACXJE silva_h_Page_130thm.jpg
5b0aa898045b9a04be4c92bd62e8c3f5
21b753efc4cc8629239ec21c5f667a0ede8c40d9
6574 F20101113_AACXIQ silva_h_Page_115thm.jpg
e691564e3bfb0160506614f82541191b
2afc06b73f5c4f2e5a1af810431409b556375f33
105042 F20101113_AACWFO silva_h_Page_034.jpg
3bd00f1ef09da4318628b3eca2eb093c
997cc908ce96d7d9dc5974a75b5ef2374aeaeb6c
75766 F20101113_AACWEZ silva_h_Page_014.jpg
7dea5c9227cb7ad86b1ce44b6f013cc3
cc889018ae145816f92c2364cf50eef5c7afbf05
20637 F20101113_AACVZT silva_h_Page_152.QC.jpg
175b3a7993329d2ee1b8e821159e33fe
366d91da407c50e7e20d964ba11dc1bdf723ceb7
44736 F20101113_AACWGC silva_h_Page_055.jpg
a43dfe9d01552fd2d9e210e2bd441587
989f0d8b2bc91b739f108559445ef3f238d60e96
26038 F20101113_AACXJF silva_h_Page_078.QC.jpg
c4a4ada657d571227417f33a30bfce5c
a23ee13fb4cb9b04550baae4d1f03fe457860c7a
20852 F20101113_AACXIR silva_h_Page_073.QC.jpg
264a03908a172e63d9584d6c6681acfb
902323b83dd5201c57c701becdf1994e1dae3222
84614 F20101113_AACWFP silva_h_Page_036.jpg
48bcc36ae3c552a79df9f050ee63bd01
ed270225336177b786db5324039d873dedee47d8
75026 F20101113_AACVZU silva_h_Page_111.jpg
20218f5cce719ef8668a4522055ab8be
609c25d4b44a6399c55f9e52ad3fa2860bb42b33
46169 F20101113_AACWGD silva_h_Page_056.jpg
8c0f685babaf04445cbeb2d02fa5a96e
edba8018cd3420aa02920193faaa6e3daf920ad0
13550 F20101113_AACXJG silva_h_Page_057.QC.jpg
2a770f217a6f4e45a2ded30b1be944e6
4ebde8f14f2089804d232d0236df9d2b404273b0
7320 F20101113_AACXIS silva_h_Page_050.QC.jpg
0b458a750926177d3d56385bc17ddb9d
526989a7f126cab125b86065da003cd73c95c5a2
82196 F20101113_AACWFQ silva_h_Page_039.jpg
13f9ba6a7cb6b001ea67ceffc250e751
54fc3276412c2014d1ed691d1a1ccdcb09df6938
7211 F20101113_AACVZV silva_h_Page_170thm.jpg
568f5444603f90e5b186ff1e28268a2e
c0c41a02f3226140d7c0e56860fdae83caf742e8
46017 F20101113_AACWGE silva_h_Page_057.jpg
34ea837d84a80c84342cf6304adaaef5
9e9ff18b0ab4e21b7da7f2f8f3eaeb360bb3f99c
26444 F20101113_AACXJH silva_h_Page_096.QC.jpg
6a082b9c05c944b458e1a711631c117e
9e5b52e14919c84e3e885f50c63871ec676a94f8
6841 F20101113_AACXIT silva_h_Page_043thm.jpg
477730869d07840bd84f620493ee9c97
60e67eb7163cdafdc59d33bf1dab69c77e451ff6
6860 F20101113_AACVZW silva_h_Page_159thm.jpg
8b3cf8b73d7b20a42c703f1997381bd2
50772bc2a61513234b1247431080ce2356f7cd0b
54450 F20101113_AACWGF silva_h_Page_058.jpg
21755574c414756a3d9df49a96aa914f
f48efae2af88ede40e81aee36d48872a7e2ac0e5
83892 F20101113_AACWFR silva_h_Page_040.jpg
31ff51468ba59efc33768fca2cabc7b7
d6469513de46acc976badb21fdd4912f32bf1162
6216 F20101113_AACXJI silva_h_Page_073thm.jpg
f29a14756842b690a991415468c93c25
eb60a854664124c911e8e826aa661030657496ae
6748 F20101113_AACXIU silva_h_Page_085thm.jpg
bbaeaadfaa777eb28c5d2c8208a38528
6f63220c6398f7c786cb476120a57a3acdb589cd
75603 F20101113_AACVZX silva_h_Page_070.jpg
564b903488938f1a92b8bf78ca71b6fd
49d5be31cc58d31b0155c69dcad9a037016a4f50
70126 F20101113_AACWGG silva_h_Page_061.jpg
7b4ee76e74e5d0410e4d77f01da09fc6
74452099efc87e4e315631ff67c445ad74353801
80713 F20101113_AACWFS silva_h_Page_042.jpg
accd39859c1d3628d7fc4787dcc35053
7695e6a18da63b12d7693cf5977309e0626a6667
6940 F20101113_AACXJJ silva_h_Page_083thm.jpg
1ebaee99fdc06a483c1c7756265e5d8a
4f21c0c846f3e0a62fce20b622aede41cde8b2b6
26853 F20101113_AACXIV silva_h_Page_065.QC.jpg
43ff4494a6567405c679e56498fb9036
5d1317e00843db1307bf54de9a37bdcbd7b1561a
7197 F20101113_AACVZY silva_h_Page_038thm.jpg
eef653626ba2ed71ad9e9805d1235128
d2a285a41752da737a24d439fc92ba383c2b962e
83763 F20101113_AACWGH silva_h_Page_062.jpg
d78b2b85e09a5bb3adae8bdf46ca61bc
4a38f86394eafb889f11d0727ab91868759592e0
79551 F20101113_AACWFT silva_h_Page_043.jpg
52524a341b8fa57ca2058456eb45d9a8
bdf2531cffdc08976b46afea5528d2b1b9ba6946
25171 F20101113_AACXJK silva_h_Page_090.QC.jpg
a30e05eb933d66546bc7457e5400781b
849db8e11e3e6bdf2d2041294590f387d511ddb2
F20101113_AACXIW silva_h_Page_109thm.jpg
26d0158f91949d0ad0ec4995fce2bda4
b8591614570d22179082048796e9fc4c9855afb7
49395 F20101113_AACVZZ silva_h_Page_131.pro
ed42337c72b02eccc62e746a1cc76ee4
c7489b932a4c1b0d00eb7dc71f69b0be24cb80ec
82038 F20101113_AACWGI silva_h_Page_063.jpg
e1ad5f9fd2c26cf878ca7a5d25482525
9567e15cffc671646a89a8e1890068cfc2505878
94402 F20101113_AACWFU silva_h_Page_046.jpg
30dcbd4123cc86cd0a3b0d3bddfb4830
3d4169b991da256163ba240b2b5cc7fdd72b5120
24463 F20101113_AACXJL silva_h_Page_043.QC.jpg
d0708d809f71ddf84ae7b2df56480e36
be1743ee89bb734e162606db3e4f931db876e864
17506 F20101113_AACXIX silva_h_Page_047.QC.jpg
9f7471e472fc2ead367555f8531a1b95
d7af745466c2f70fa1073bc79bd9acace7412bd5
66041 F20101113_AACWGJ silva_h_Page_064.jpg
af9cb9fe3a3485a55a850d6588b8dbc4
339edd7dbee4b9426afff73fc19a83dffbaa02e7
61513 F20101113_AACWFV silva_h_Page_047.jpg
c6c871d728cd41173d1f317ed9d13312
f4ce263095f5751e910d97000f76ba84409f86d1
2497 F20101113_AACXKA silva_h_Page_050thm.jpg
707e73ac4f97a02a5e5f036509f74919
ba9bdb8d49b29aeef1709a013f118004f7bed59a
F20101113_AACXJM silva_h_Page_081.QC.jpg
11ac642232c427ecd21a2141175539de
7e122fb3176dd1fed504153db611eb30ca7942ab
16850 F20101113_AACXIY silva_h_Page_149.QC.jpg
7ac58941937e0c39fdf1e92c8ce0ad5f
3cb3fbd404a4b1f58e0923d15a565e19e5268e90
88071 F20101113_AACWGK silva_h_Page_065.jpg
cbf4660de1cac5c7dddc9c2fcf43cfde
1efc18be06ae3919bf85f75ad1cf9aba37f27eba
51520 F20101113_AACWFW silva_h_Page_048.jpg
552ed3920d11588e408ba5331595132e
12984f51ce99c7f8f06bb1deca52fd54ee9fe8de
3753 F20101113_AACXKB silva_h_Page_008thm.jpg
f074796598d413d4062c118f09e0c7cb
6657309f302c7b968c93816cad5c68bb016525f8
7205 F20101113_AACXJN silva_h_Page_044thm.jpg
0ea14cb2693d98a4dcdf0b6640bc9966
3c83fa95d8ebe2d0212d9c3c136276c18bc7698d
74883 F20101113_AACWGL silva_h_Page_067.jpg
7a50f0afd6831040cedccabbe164aa29
3ab8d35d740cf700e327da294387659758b9ab15
2704 F20101113_AACXKC silva_h_Page_150thm.jpg
b634d1664819425e7ec24e3aa77d4540
b86522c9a496e2decce7869fb09182622a9721e2
7028 F20101113_AACXJO silva_h_Page_040thm.jpg
5a2ce07f7964524faecc38ea8c7ab3a7
007154a035a2b6277070e15e04db6651d94b080f
5997 F20101113_AACXIZ silva_h_Page_165thm.jpg
b599bcb312f0e3b50acccd1e3f24b494
ee98d6630ad6d53a50bfda0796d75da5bc1144c3
73552 F20101113_AACWHA silva_h_Page_088.jpg
3b53d9b956c0ba81b2e0bfe76e10380a
5b9218184252cb6b9cb98ccaca1771ee438b0215
75637 F20101113_AACWGM silva_h_Page_068.jpg
892fb5dae8eb402459504bbee60ebc10
5c951e32de9580a9c07fa70dce4942b560be6789
29682 F20101113_AACWFX silva_h_Page_049.jpg
6778a0027aea1f292768deb8d8392401
c58fe73b47a5bb76096522427329d405d0c955fc
4074 F20101113_AACXKD silva_h_Page_103thm.jpg
93e4b21b54d6085267e8f5666e4908a9
88f5dfab0dabaf39d2c59b895c494d21d9356eb0
6468 F20101113_AACXJP silva_h_Page_158thm.jpg
ee80149c80afb43bff0e15d2875863d6
d5955d92fdb01584be0358074d8df6985aba1f23
79823 F20101113_AACWHB silva_h_Page_089.jpg
30f79cb9d7afc5d16e12e607c7a3d5f5
8fa55b49dcddb84d612b7bb3db84261da5c59881
72791 F20101113_AACWGN silva_h_Page_069.jpg
212814ba02229b47356fb0091e2e0982
4e8e630683348ffa5aba79f9e5359c7a527bbb2d
23029 F20101113_AACWFY silva_h_Page_050.jpg
7c6d1b5f556c0c471d1c6dc8665935ce
2572028acfe66e77db29ab3a31e439a32f7fc9be
3812 F20101113_AACXKE silva_h_Page_173thm.jpg
7d3675cbac2c988f9690a6e240c87ec9
23b9a3eb42efa8739e67674fccc058f5ead77ada
24642 F20101113_AACXJQ silva_h_Page_113.QC.jpg
08ac1e1472f4cb994834637d68e96c6b
3c816c1ac35845de7ed537706eabec8927d9f9dc
80370 F20101113_AACWHC silva_h_Page_091.jpg
32604e52355017842a5a5b7fcb4b2ae0
7ef86c7cce306a0fa209c696825dbcf029c297f5
55998 F20101113_AACWGO silva_h_Page_072.jpg
28d2a83a8e3107817ffe8196601c2b3c
f69479e977efcb359399c5c1f1713d7ebebbabab
41020 F20101113_AACWFZ silva_h_Page_051.jpg
51b49124753e739f6b348f31cb491ce1
3cb3df26a81b7489675bac52cf05d6b0182e365b
27477 F20101113_AACXKF silva_h_Page_095.QC.jpg
c4edf17595e843aec9ddbf8626029738
9f90397f8c4c8a3b09655aec233d03a311ed54a0
29588 F20101113_AACXJR silva_h_Page_034.QC.jpg
ccb44c11a2f269ec7f0821c33803d192
9df1be0fc727bc28b6ce005f4fc8510584319c0f
80762 F20101113_AACWHD silva_h_Page_092.jpg
fa3553f429b1df153b2fe35f2f721b16
c011af92c10ab1e9a02e06dc344d16324f8c077e
62725 F20101113_AACWGP silva_h_Page_073.jpg
d0e55f711711e4f0cb6826fc9b4f14c4
66427e02370602e3780eec6aa36dd7c82610f5d8
7231 F20101113_AACXKG silva_h_Page_030thm.jpg
112f0d2ae8812d836c0ba866a37614f2
494b548332c811818ce4f1f164fd76a8b61434a8
7258 F20101113_AACXJS silva_h_Page_077thm.jpg
dc70d9d4dcfa5a541b3b5580ef7995d2
96daafc489b258b1e955ff388ea49f78fe6e334f
76995 F20101113_AACWHE silva_h_Page_093.jpg
fe59643d7377624c1085aa930b11aa50
40ebb99b340a6479d117e5e54c3380a3dc8d3611
73955 F20101113_AACWGQ silva_h_Page_076.jpg
792cceaea49df1ddd01a4d62b365183b
3cbc50d785d852dafa5f75b275cbf789880e75c2
26386 F20101113_AACXKH silva_h_Page_138.QC.jpg
27f7284f2d2bb30e9b046fb54cfc2b5e
de3e100389e48e13ce527035790d665423828197
7060 F20101113_AACXJT silva_h_Page_039thm.jpg
0c3fea8af50e4dc8e1ae2d636b421798
867b1c3cb37e941efb1ac76355db472836810516
96556 F20101113_AACWHF silva_h_Page_094.jpg
d880bd74f22c5beb0748b02d1a6ea3e4
a19eb34828d54e8bd7ad4a32412e40d3754ffa82
79889 F20101113_AACWGR silva_h_Page_078.jpg
6034463dbcfefe7a4be976b5da26537d
5770a209bf2f03fd5bc628d87e4f4653c884454a
25910 F20101113_AACXKI silva_h_Page_112.QC.jpg
df5c2385ebb523573aa045af0dc8a3f1
04935bea426f1ad44b73c81baa16dbf78abd72f8
F20101113_AACXJU silva_h_Page_136thm.jpg
5e233c453e8a2280b4b24dd3e4fd59eb
6870ef9eb06059764e12ee1fae6732cc9138a971
84401 F20101113_AACWHG silva_h_Page_096.jpg
dc3a2f6e62f7ceaa36e512fbd4fa9205
ca0961d4c523cae54b7426b87482cf5ab1ca0259
65179 F20101113_AACWGS silva_h_Page_079.jpg
3a8c5ac80667538552a7ec3f770e7f3a
c9d8e4fda8ab95dae94586b3a0e91fc2d58a8752
27197 F20101113_AACXKJ silva_h_Page_016.QC.jpg
63b865b08b7e40309bf44227eb463112
5af2e15c9d5e2e32b4d1ddd82bfa4e1aee3960d7
24632 F20101113_AACXJV silva_h_Page_137.QC.jpg
9d93e8aaaafa80d2a90da97d7ac9e85d
9bbae327e6be07c74f0571c0784ecc3c95889c9a
80699 F20101113_AACWHH silva_h_Page_098.jpg
95f65cc0b0c5ea0900070175fe25bffc
0b2b62746184cc132dc17838d48557e498e7680c
79926 F20101113_AACWGT silva_h_Page_080.jpg
44ed539bccc4c9a77ec82c16a4f74609
36104120a890a7f68bd8c0ce737937e45ed9b88b
9174 F20101113_AACXKK silva_h_Page_162.QC.jpg
effbde6ce3245c8b773c7f603cd131f0
de3cc5bfdac6f8c4b38af7a4b04af47a8e58c763
26176 F20101113_AACXJW silva_h_Page_170.QC.jpg
c884facfd402b815189398979b5adc09
3ba6808402028a7e255743a71a920e8ef43319a1
77114 F20101113_AACWHI silva_h_Page_099.jpg
309b37b6d830fd4e30c6e5021d46a8cb
6dc9784e512f119c0c518c10a78184e5a3d5e44b
78975 F20101113_AACWGU silva_h_Page_081.jpg
0e86ffcebf6f3f0e7c5eda3ab1951620
4e2921a9df4186fb49a91977930418b81e415d6d
11210 F20101113_AACXKL silva_h_Page_006.QC.jpg
a1d49371793381ed62ed04dc4c22a049
fb68919dc458aa0fd44d09bb3fa03ce475429e32
7206 F20101113_AACXJX silva_h_Page_126thm.jpg
4f7f1dea83b6e31edb8f180dfd954bb7
d8cb0905942bbac35719409c3c983a5446854db8
77765 F20101113_AACWHJ silva_h_Page_100.jpg
4b1e9d68057d2c18521c28e2b26b7a21
e34f6f4f8955bf18f84220a52ad0f6db53af6cb6
76880 F20101113_AACWGV silva_h_Page_082.jpg
b8637c71bb99ac2b51346bc8b8a9bb6f
10363c41cf2bc346da6d7a42d1e1d8cb665992ec
23181 F20101113_AACXLA silva_h_Page_024.QC.jpg
f04e13c3b220453b42acd22be2daf341
a3ed751a9ed31e8c33b685d232cd688eb93337db
6912 F20101113_AACXKM silva_h_Page_156thm.jpg
f5d5a45c9890ddd30c6962e17009cfe4
a0a4e74977cc3c532f1e47d8b5652839039b7e00
25826 F20101113_AACXJY silva_h_Page_087.QC.jpg
d0b104059b1d84d7c3d9cbd738492a87
fcea55fda06baa304b299ac4d932138d2005065e
53124 F20101113_AACWHK silva_h_Page_102.jpg
ab7d50317ca480c346ca2136669357c9
37de5e2f4e24b124e36678f8844c9ded2b117c38
81874 F20101113_AACWGW silva_h_Page_083.jpg
9e619f71d1f3aa299482f6fc639af715
69b9d430d706593424baf2431818a6b1e136b89c
6382 F20101113_AACXLB silva_h_Page_046thm.jpg
a0bef6135e8b323afbaafbbea9a19157
9e752ae0948dfb2d755bf9b408c2da8069df1803
22460 F20101113_AACXKN silva_h_Page_028.QC.jpg
726692f06e02afbd63bd3bea2491af55
ccc3fdd7e212b3dd77369aecce6c54682031b53c
7180 F20101113_AACXJZ silva_h_Page_124thm.jpg
c00853da8df461864d8bd42bfe6a191c
3521d4215b7b3e7032d2e3ecc942e98c14867b1a
33889 F20101113_AACWHL silva_h_Page_104.jpg
89e94cf52bd16301af778a9ca4a737a2
2474971150afb322e66245c042fb3df4dc291a95
72571 F20101113_AACWGX silva_h_Page_085.jpg
e0e007c8944683cfe7f410c154825209
a010a0cd87e30c97c34cc03447d4273e2421e397
3270 F20101113_AACXLC silva_h_Page_059thm.jpg
2eb8849e3e793ea02448035790866f6f
7dbdb4898ddb287b838061e479b909645b8485bc
13369 F20101113_AACXKO silva_h_Page_173.QC.jpg
db217eae40ecfc447ccc941d76d9b033
4709aeb6e5efc6da360b024c5aea6db1e07ef24b
68634 F20101113_AACWHM silva_h_Page_105.jpg
09f5023b97a7ea83a7bef2cef86abf3d
2de6271b57fdff60941aa7b466e6a69acbaa856f
88962 F20101113_AACWIA silva_h_Page_122.jpg
76cd2d2ae888802eb1ce461eccd052ef
856587ef625d9ec17b3f948c186fef8221fb7f16
27359 F20101113_AACXLD silva_h_Page_133.QC.jpg
43dde294feea839f057cd3db8023f750
494ac42282fbed3e3cf46b124a787a8833679718
6762 F20101113_AACXKP silva_h_Page_111thm.jpg
821f0f7b7e26b4d6f3644ce05c737c66
cc9b91aefa672ae3892912ac8e01e56db61b9e48
53004 F20101113_AACWHN silva_h_Page_106.jpg
9f23002a204e8a89305de33f35473354
30bbd62da0559658a0d7a1a6a15c42e1a75ca206
77436 F20101113_AACWGY silva_h_Page_086.jpg
af37a5a51e2af3ba668e9a851a20c0b1
6e4cf33ea6d9796ba2075e61351fded00d557b29
78768 F20101113_AACWIB silva_h_Page_125.jpg
f4397c60fbbfe59b766f1e6c68e83de4
4bc50f4043e86ab277a57dea413c0b3895e3e5dd
6609 F20101113_AACXLE silva_h_Page_164thm.jpg
106ebe76787b69a8456c420153026d85
22f9b09a955302866e303fbc348485c9746151d2
13558 F20101113_AACXKQ silva_h_Page_163.QC.jpg
a305a0a5da025bcf3a77a7ed4e931c2d
849859a33d2c31856a508b8d2cf44fd1aae13ad5
67866 F20101113_AACWHO silva_h_Page_108.jpg
5c389b7f2725998b498c566cae299a56
51ab97b48fcad66dc180395b8984a54b3f07b942
82662 F20101113_AACWGZ silva_h_Page_087.jpg
84afb33a6915f2235461755394fee3b5
7779b4de6918e09e171649d6802d4fe2b43a6aad
87589 F20101113_AACWIC silva_h_Page_126.jpg
b7268bc3aad465995fcefee15507cc4b
d29f00a07d1dfd085fae971e2d32c749b37762b8
6483 F20101113_AACXLF silva_h_Page_012thm.jpg
ec39011d3e9180433b1bd1f31e4a9ca0
39dc973853ea098e2f16ed25470f8bb5182b441c
26398 F20101113_AACXKR silva_h_Page_142.QC.jpg
a63ba8ce1863db6659a73202d5cdd083
cb6dbdcdf1a63978d3b5a428e5861bde00dda44d
81387 F20101113_AACWHP silva_h_Page_109.jpg
cd55f716a82d7172c820f6596503fc47
e621d7ba7c59a03e630a7fd6c157a7e7097f8f45
80894 F20101113_AACWID silva_h_Page_127.jpg
2e674dc4a704949e801719129d72699b
12f30f8a4315971c91e345370221fe6f44f358a8
6852 F20101113_AACXLG silva_h_Page_014thm.jpg
a5eacb3643dfd438044f724a3aede3fb
e90e52ba1686b046783629b130a7627b6f2b3264
6998 F20101113_AACXKS silva_h_Page_116thm.jpg
1108b98dba12564d4e65f97af84d3b30
f0148bee1fd8909fe5a848b1ee304949a884b798
75989 F20101113_AACWHQ silva_h_Page_110.jpg
4362858aa50dabb9b22a390515f7fcae
4f939bf7b57673859b71593e139d24f8953bfcf6
91623 F20101113_AACWIE silva_h_Page_128.jpg
28eb8ed86fcea0c51523aab45e9c2cb5
8c3bb4ad795b53fe20ce9852357f18bfb4ed14a2







EMPIRICAL ESSAYS IN THE ECONOMICS OF REGULATION


By

HAMILTON CAPUTO DELFINO SILVA















A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2007

































2007 Hamilton Caputo Delfino Silva





























To
Ethevaldo and Arlete
Rita, Leonardo, and Feranda









ACKNOWLEDGMENTS

The enduring process of obtaining a doctorate degree involves sacrifices, encouragement,

and support from several sources.

I am deeply grateful to my parents. Their unconditional love, encouragement, and example

have always been the major inspiration to achieve my goals, and to them I owe my education.

I am extremely indebted to my wife Rita. The decision to pursue the doctorate, in a foreign

country and without any help from my organization in Brazil, imposed an extraordinary burden

on her. I owe this accomplishment to her strength and courage, as well as to her resilience in the

not so rare difficult moments. I also thank my children, Leonardo and Fernanda, for being very

accommodating and understanding the importance of this project for their dad.

The support provided by my brothers Emilson, Edilene, and Anderson was decisive as

well. Their words of encouragement were often remembered when the strength and motivation

became to fade. In addition, Emilson's help was critical for the starting of my doctorate.

I would like to express my gratitude to my advisor Sanford Berg for his support since the

first day of doctoral studies. His back-up and dedication to this project were decisive. A special

thanks also goes to Professor Larry Kenny, for his commitment and very helpful comments and

suggestions.

Finally, I am grateful for the financial support of the Public Utility Research Center and

the Florida/Brazil Institute. This project would not have even started without the help of these

organizations.









TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ..............................................................................................................4

L IST O F T A B L E S ...................................................................................................... . 7

ABSTRAC T ...........................................................................................

CHAPTER

1 INTRODUCTION ............... .......................................................... 12

2 AN EMPIRICAL ASSESSMENT OF THE REGULATOR'S PERFORMANCE IN
THE BRAZILIAN ELECTRICITY SECTOR.............................. ..................... 16

Introduction ............... ....... .................................. .......... ............................ 16
In stitutional B background ............................................................................. .....................20
M e th o d o lo g y ..................................................................................................................... 2 3
D a ta ............................................................................2 8
Findings .........................................32
R o b u stn e ss C h e ck s ........................................................................................................... 4 3
C onclu sions.......... ..........................................................44

3 PRIVATIZATION, INCENTIVE REGULATION, AND EFFICIENCY
IMPROVEMENTS IN THE BRAZILIAN ELECTRICITY DISTRIBUTION
IN D U S T R Y ................... ...................6...................1..........

Intro du action ................... ...................6...................1..........
Institutional B background ................................................................64
M eth o d o lo g y ..................................................................................................................... 6 6
The Electricity D distribution Technology .................................................................. 66
Comparative Efficiency Studies .................... ....... .............................. .... ...... 69
Stochastic Cost Frontier and Treatment of Environmental Variables .............................72
Specification and D ata ..................................................................................... .............. 77
F in d in g s ................................................. .... .....................................8 4
Service Quality and Economies of Vertical Integration ................................................. 94
C o n clu sio n ................... ...................9...................9..........

4 THE ASSESSMENT OF FIRM'S EFFICIENCY IN PERIODIC TARIFF REVIEWS:
AN EVALUATION OF THE REFERENCE UTILITY APPROACH ........................... 109

Introduction ........................................................ ............ ..... ...............109
Institutional Background and the ANEEL Model Company Method ............ .................113
The Tariff Review Methodology ... ...........................................................114
M odel Company Estimates .................. .............. ............ .. .. .. ................. 117
C om parative E efficiency A analysis ........................................ ................... ..... ..............118









SFA M odel and D ata .................. ................................ .. .......... .............. 120
SFA R results and Com parison................................................ ............................ 123
E conom etric M modeling ........................... .......................... .. .... ........ .. .... .. 127
Specification and D ata .................. ............................... .. ...... .. ........ .... 131
R results A naly sis ....................................................... 137
R obu stness C heck : D E A .................................................................................. ............ 140
C including O b servations............................................................................ ....................144

5 SUMMARY AND CONCLUSIONS........................................................ ............. 154

APPENDIX

A DETAILS ON EVENT STUDY' S SAMPLE AND DATA ...............................................161

B SFA AND DEA EFFICIENCY INDEXES ......... .........................................................164

C M ALM QUIST TFP INDEXES .................................................. .............................. 165

REFERENCE LIST .................................. .. .... .... .................. 166

B IO G R A PH IC A L SK E T C H ........................................... ......... ................... .......................... 173









LIST OF TABLES


Table page

2 -1 E v en ts L ist ...............................................................................4 6

2-2 Descriptive Statistics .................. ................................... ....... .......... 48

2-3 Correlation M atrix ................................... .. .. .... ....... .. ............50

2-4 Securities Returns, Stock Market Returns and Exchange Rate Variation, by Year. .........51

2-5 Individual R egressions R results ............................................. .................. ............... 52

2-6 Hypothesis Tests Results for Events in Year 1999.................... ................ ..............53

2-7 Hypothesis Tests Results for Events in Year 2000... ............... ........... 54

2-8 Hypothesis Tests Results for Events in Year 2001 ......................... .............. ..............55

2-9 Hypothesis Tests Results for Events in Year 2002... ............... ........... 56

2-10 Hypothesis Tests Results for Events in Year 2003...................................................57

2-11 Hypothesis Tests Results for CARs Before Significant Events............... ..............58

2-12 Hypothesis Tests' Results for Events' Overall Effect ............. ..... ..................58

2-13 Significant Announcements' Categorization. Direction and Estimated Magnitude of
Regulatory Announcements' Effect on Security Returns..........................................59

2-14 R andom E vents' R esults........................................................................... ....................60

3-1 D descriptive Statistics ................................................... .......... .. ............ 102

3-2 Stochastic Cost Frontier R results ....................................................................... 103

3 -3 E la stic itie s .................................................................................................................. 1 0 4

3-4 Technological Change by Year and Ownership ................................... ............... 104

3-5 E fficien cy E v solution .............................................. ............................. .................... 10 5

3-6 Decomposition of Privatized Firms' Efficiency Evolution..................................105

3-7 Productivity Growth Rate and Decomposititon................. ............. ...............106

3-8 Productivity Growth Rate and Decomposition by Ownership Type ............................106









3-9 Productivity Growth Rate and Decomposition by Ownership Type Firms with Q >
400,000 M W h/year .................................................................... ......... 107

3-10 M ean Service Quality Indexes ..................................................................... 107

3-11 Vertical and Quality as additional regressors or as mean inefficiency parameters .........108

4-1 Initially Estimated OPEX, Final OPEX, and Firm's Reported OPEX ............................148

4-2 SF A D escriptiv e Statistics .................................................................... ..................... 149

4-3 GLM and OLS D descriptive Statistics ........................................ ......................... 150

4-4 Stochastic Cost Frontier R results ........... ......... ....... .......................................... 151

4-5 Efficiency R ankings and Indexes.......................................................... ............... 152

4-6 Regression Results ................. .............................. ......... ........ .. 153

A-1 Missing Observation Problem (Number of computed stock returns missing)...............162

A -2 C om panies in the Sam ple........................................................................ ...................163









Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

EMPIRICAL ESSAYS IN THE ECONOMICS OF REGULATION

By

Hamilton Caputo Delfino Silva

August 2007

Chair: Sanford Berg
Major: Economics

Reform processes undertaken on network industries have encompassed unbundling,

privatization, introduction of market-oriented regimes for their competitive segments, and

implementation of a new regulatory framework for the remaining segments with natural

monopoly characteristics. The new regulatory framework has involved the creation of

independent regulatory bodies and the incorporation of theoretical advances from the economics

literature on incentive regulation, under the ultimate objective of providing the conditions and

incentives for efficiency improvement and for the possible achievement of second best prices.

This dissertation adds to the literature on the impacts of the restructuring measures implemented

and contains three empirical essays on the reforms accomplished in the Brazilian electricity

sector.

In Chapter 2, information on stock market reactions to regulatory announcements are

employed to investigate the level of consistency and predictability of decisions taken by the

Brazilian electricity sector regulator and to identify the extent the aforementioned regulatory

body has been able to act independently and to balance multiple interests in regulation. The

findings indicate that the regulator's decisions have not favored a single interest group. The

evidenced unpredictability of the regulatory agency's decisions suggests the need for









improvements in the regulatory discussion process, with the adoption of measures to increase the

transparency and to promote more substantive public hearings. Additionally, the study shows

that the need to provide incentives for new investments has played a significant role in the

regulatory process. The estimates indicate that regulatory decisions accounted for roughly 12%

of the difference in performance of the sample securities with respect to the market index.

The Chapter 3 examines the impact of privatization and incentive regulation on firms'

performance in the period of 1998 to 2003. Evidence indicates performance improvement after

the implementation of sector reforms, with both privatized and public companies reducing the

efficiency gap with respect to companies that were privately owned before the reforms. The

results show that privatized firms responded more aggressively than public firms to the new

incentives brought by price-cap regulation, and suggest that the high performance improvement

experienced by privatized firms did not come from mere reduction in costs brought by

deterioration in the quality of service. The findings also indicate a possible strategic behavior

associated with the periodic aspect of price-cap regulation, as well as to cost shifting

implemented by companies that operate in the electricity generation segment.

The Chapter 4 is motivated by the increasing use of the model company approach to

determine electricity distribution tariffs in Latin America, despite the criticisms made to the

method's subjectivity and obscurity. The study examines whether the use of the engineering

approach in the Brazilian electricity distribution sector periodic tariff review enabled the

attainment of the welfare maximizer regulator's rate setting objectives, by comparing the

method's implied performance scores to efficiency measures provided by economic

benchmarking approaches. Results show that some firms, mainly the ones serving more affluent

consumers, operating in more densely populated areas and having a lower proportion of









electricity delivered to industrial customers, received substantially lower repositioning indexes

than the economic benchmarking methods would recommend, pointing to a possible violation of

firms' break-even constraints. The findings also reveal that significantly higher repositioning

indexes might have been given to companies with the opposite characteristics, negatively

affecting the incentives for further productivity improvements, as some of the possibly benefited

companies do not appear in the top ten of the benchmarking efficiency rankings.









CHAPTER 1
INTRODUCTION

Network industry reforms recently implemented have involved unbundling, privatization,

introduction of market-oriented regimes for their competitive segments, and implementation of a

new regulatory framework for the remaining segments with natural monopoly characteristics.

One of the features of the new regulatory framework has been the emergence of

autonomous regulatory bodies. The associated increase in regulatory discretion, however, has

raised concerns over the possible influence of interest groups on regulatory outcomes,

strengthening the debate over the pattern of government intervention in business, represented by

two main opposing theories: the "public interest" approach, which asserts that the government

acts to lessen or eliminate the inefficiencies engendered by market failure; and the "capture" or

"interest group" approach, which argues that the government acts to transfer wealth between

interest groups in an effort to maximize political support. Examining the actual behavior of

regulatory agencies, in this context, constitutes an empirical question, addressed by Sawkins

(1996), Dnes, Kodwani, Seaton, and Wood (1998), and Dnes and Seaton (1999) on the

perspective of the energy and water industries in United Kingdom.

The new regulatory design has also comprised the incorporation of theoretical advances

from the economics literature on incentive regulation, notably the implementation of incentive

mechanisms, such as the price-cap method. In case, the firm and its managers are the residual

claimants on production cost reductions, and bear the disutility of increased managerial effort

(Joskow, 2005). As a result, the conditions and incentives for efficiency improvement and for the

possible achievement of second best prices are settled. However, whether price-cap regulation

effectively leads to efficiency improvement also constitutes an empirical question.









This dissertation examines these questions in the context of the reforms accomplished in

the Brazilian electricity sector. The reforms began in 1995. While constitutional amendments

abolished the public monopoly over infrastructure industries and allowed foreign companies to

bid for public concessions, the Law 8,987/95 (General Law of Concessions) set the stage for the

beginning of the privatization process, represented by the auctions of Escelsa in 1995 and Light

in 1996. By the end of 2000, a total of 20 distribution companies had been privatized.

In addition, the implementation of a new regulatory framework involved the establishment

of an independent regulatory agency (ANEEL) in late 1996 and the institution of a new model

for the electricity sector in 1998. The model focused on privatization and unbundling of

generation, transmission and distribution assets, gradual transition to a competitive generation

environment in nine years, creation of a wholesale power market, operation of the transmission

network by an independent operator, and use of the price-cap regime to regulate distribution

tariffs, replacing the previous cost of service system. Price-cap regulation was implemented

through the signature of new concession contracts, which took place from 1998 to 2000, and

scheduled the first tariff review for after five (for contracts signed in 1998) or four years. As a

result, 61 companies were submitted to a tariff review process from April/2003 to

February/2006.

The analysis is performed in Chapters 2 to 4. The Chapter 2 ("An Empirical Assessment of

the Regulator's Performance in the Brazilian Electricity Sector") uses information on stock

market reactions to regulatory announcements to investigate the level of consistency and

predictability of decisions taken by the Brazilian electricity sector regulator, and to identify the

extent the aforementioned regulatory body has been able to act independently and to balance

multiple interests in regulation. The investigation improves upon previous studies by focusing in









the context of a developing country, where the interests are presumably more pronounced, and

by employing an identification strategy that explicitly accounts for the possibility of event

anticipation. In addition, the study sheds light on the debate over the possible capture of

Brazilian regulatory agencies.

The Chapter 3 ("Privatization, Incentive Regulation, and Efficiency Improvements in the

Brazilian Electricity Distribution Industry") concentrates on the effectiveness of the price-cap

incentive mechanism issue. The study examines the impact of privatization and incentive

regulation on performance of Brazilian electricity distribution companies in the period of 1998 to

2003, employing a Stochastic Frontier Approach (SFA) that controls for heterogeneity in

operating conditions, influence of macroeconomic factors, and random shocks.

The investigation evaluates the efficiency evolution and the productivity gains that

occurred in the period, checks for difference in performance between public and private firms,

and looks at the possibility of efficiency catch-up. It also examines whether vertically integrated

firms might be behaving strategically, shifting costs from unregulated to regulated activities, and

whether efficiency changes are associated with variations in service quality.

The Chapter 4, entitled "The Assessment of Firms' Efficiency in Periodic Tariff Reviews:

An Evaluation of the Reference Utility Approach," is an extension of the previous one. The

obtained SFA efficiency estimates and measures of firms' productivity improvements, along

with efficiency measures provided by an alternative benchmarking technique (Data Envelopment

Analysis), are employed to examine the results derived from the use of the Reference Utility

approach at the distribution companies' first periodic tariff review.

The study is motivated by the increasing use of the reference utility model to determine

electricity distribution tariffs in Latin America, despite the criticisms made to the method's









subjectivity and obscurity. The analysis checks whether the methodology has enabled the

attainment of the welfare maximizer regulator's rate setting objectives, and is based upon the

reasoning that substantial (and consistent) divergences in the Model Company's implied

performance scores, relative to efficiency measures provided by economic benchmarking

approaches, reflect deficiencies in the engineering method's application.

In sequence, the investigation checks for the possible causes of the divergences found, not

only exploring the predictions of the same interest group theory of regulation employed in the

second chapter, but also accounting for the fact that the regulator's decisions were taken in an

incomplete and imperfect information context. In case, it is examined the possibility of flaws in

the engineering cost parameters employed to estimate the efficient costs and the potential use of

some of the available data as signals for firms' profitability and cash flow availability, as a

subsidy for the regulator's decisions regarding the distribution of productivity gains among

stakeholders.

The three chapters mentioned above are self-contained, presenting the results obtained and

the corresponding conclusions. Nonetheless, the main results and conclusions evidenced in the

three studies that compound this dissertation are outlined in Chapter 5, along with directions for

future research.










CHAPTER 2
AN EMPIRICAL ASSESSMENT OF THE REGULATOR'S PERFORMANCE IN THE
BRAZILIAN ELECTRICITY SECTOR

Introduction

The analysis of regulatory impacts provides insights into the level of predictability and

consistency of regulatory actions, and is essential if we are to understand how regulatory

processes and regulatory decisions affect infrastructure performance. The subject's importance

has increased with the emergence of autonomous regulatory bodies, in the context of network

industry reforms recently implemented, as the associated boost in regulatory discretion has raised

concerns over the possible influence of interest groups on regulatory outcomes, although the role

of special interests within predecessor government ministries was not insignificant, and probably

less transparent.

The issues raised by regulation continue a debate over the pattern of government

intervention in business, which has long been carried out in the fields of economics and political

science and resulted in two main opposing theories: the "public interest" approach, which asserts

that the government acts to lessen or eliminate the inefficiencies engendered by market failure,

serving as an impartial referee that aims to maximize social welfare; and the "capture" or

"interest group" approach, which argues that the government acts to transfer wealth between

interest groups in an effort to maximize political support.1 Recent agency theoretical models



1 According to Stigler (1971), regulators are self-interest maximizers and stakeholders face costs of organization and
information. Consumers, being dispersed and having less at stake, face higher costs of organization than producers
and usually do not have the required incentives to spend the necessary resources to become informed. Consequently,
the prediction was that the producer interest would win the bidding for the services of a regulatory agency. Stigler's
formulation brought theoretical foundation to the "producer protection" view that characterizes the capture theory of
regulation, based on the accumulating evidence of empirical research done before the 1970s. The Stigler's argument,
however, was further developed by Peltzman (1976), who posited that regulatory agencies would not exclusively
serve a single economic interest. Utility-maximizing regulators would allocate benefits across interest groups
optimally, attempting to equate political support and opposition at the margin. Peltzman's contribution helps explain
the location of policy in the competitive price to the monopoly price spectrum. Consumers who spend a larger share
of their income on a good have a higher incentive to participate in the regulatory process and should drop more









further developed the interest group theory by explicitly recognizing both the existence of

informational asymmetries and the principal-agent relationship which exists between the

Congress (or the Government) and their delegates in regulatory agencies.2

The possibility that the regulator may favor specific interest groups underscores the need to

devise regulatory frameworks incorporating a system of checks and balances. However, to

determine how regulatory processes and decisions affect sector performance, it is necessary to

assess the actual behavior of the regulatory agencies. The present study uses information on

stock market reactions to regulatory announcements to investigate the level of consistency and

predictability of decisions taken by the Brazilian electricity sector regulator, and to identify the

extent the aforementioned regulatory body has been able to act independently and to balance

multiple (and often conflicting) interests in regulation.

With respect to network industries, the empirical research conducted to date focused in the

United Kingdom's context. While Sawkins (1996) analyzed the performance of the water sector

regulator (Ofwat), both Dnes, Kodwani, Seaton, and Wood (1998) and Dnes and Seaton (1999)

concentrated on the behavior of the electricity industry's regulatory agency (Offer). In all these

studies, the similar findings that no interest group had been systematically favored led to the

conclusion that the regulatory bodies were performing their duties reasonably well.

The conclusion of "balanced" decision-making, however, was drawn on the basis of

securities' abnormal returns evidenced at the moment of the regulatory announcements. Studies

neglected the possibility that the findings were actually reflecting a change in market


votes for the politician in response to a price rise. Therefore, goods with a high share in the consumers' budget are
more likely to have prices close to the competitive price.
2 See Laffont and Tirole (1993), and Armstrong and Sappington (2003), for contributions on the subject.










expectations.3 The present research attempts to correct for this flaw by also looking at the

significance of the cumulative abnormal returns in the period before the event's disclosure. Thus,

when there is evidence of an early incorporation of the event's effect, the regulatory

announcement's impact is based on a computed measure of the event's overall effect.

The present investigation also distinguishes from the previous ones by focusing in the

context of a developing country, where the different interests in regulatory outcomes seem to be

more pronounced. In addition to both Brazilian firms and consumers having a higher incentive to

participate in the regulatory process than their counterparts in Great Britain,4 it may be argued

that the Brazilian Government exerts a higher pressure on the regulator to keep tariffs at low

levels than a Government in a developed country, by virtue of the memories of the recent

hyperinflation period and the consequent still strong concern with the effect of tariff increases on

inflation. Moreover, Brazil faces an acute need to expand network access (provide universal

service) and to increase the amount of energy distributed, to avoid energy shortages in the near

future and promote economic growth. The resulting greater necessity to provide incentives for

(private) investments in the sector should be reflected in a higher pressure on the regulator along

these lines in Brazil than in Great Britain.





3 A finding of negative and significant abnormal returns at the exact moment a regulatory announcement is made,
for example, may not indicate that the decision affected firms' value negatively, but, on the contrary, that the
decision was not as good to firms as the market was expecting. Thus, when a significant abnormal return is found,
one has to check for the possibility of event anticipation. If this possibility is confirmed, the conclusion regarding
the event's impact has to be taken on the basis of the event's overall effect.
4 Brazilian firms face higher regulatory and country risks than firms in Great Britain, which impacts the necessity to
mitigate the possibility of not getting a rate of return that covers their correspondent higher cost of capital. The
affirmative regarding Brazilian consumers, on the other hand, comes from the fact that they are poorer. Since the
income elasticity for electricity is less than one, electricity is a greater share of the budget of poor people than of rich
people. Thus, in case of a tariff increase, poor consumers experience a higher percentage change in their cost of
living than those who are more affluent. Therefore, poor consumers (or consumers in poor countries) have a greater
incentive than rich consumers (or consumers in rich countries) to become informed and participate in the regulatory
process.










The Brazilian electricity sector's regulatory framework has been critically explored and

analyzed relative to international experience,5 resulting in some policy changes. Nevertheless,

there remains a debate over the regulator's possible capture by the industry,6 indicating a need

for further investigations of the regulatory agency's performance.7 The present study attempts to

fill this gap.

The paper's contributions to the literature are thus the following: (a) it extends the event-

study methodology employed in previous studies, by explicitly recognizing the need to account

for the possibility of event anticipation; (b) it is, to our knowledge, the first empirical assessment

of a regulatory agency performance in the context of a developing country, where the interests

are presumably more pronounced; (c) it examines the level of consistency and predictability of

decisions taken by the Brazilian electricity sector regulator; and (d) it sheds light on the debate

over the possible capture of Brazilian regulatory agencies.

The following section presents the characteristics of the Brazilian electricity industry and

describes the reforms implemented in recent years. Section 3 explains the methodology

employed and Section 4 outlines the data set. Section 5 presents and interprets the results




5 The main initiative on this respect was represented by Brown and de Paula's reports (2002, 2004) prepared under
the World Bank's PPIAF Project for Brazil Power Sector, Task 4: "Strengthening of the Institutional and Regulatory
Structure of the Brazilian Power Sector". The first report, issued in December 2002, analyzed the state of
institutional arrangements in electricity regulation right after the energy crisis and proposed specific
recommendations for strengthening the regulatory structure. The second report was prepared in July 2004 and
reviewed the previous recommendations in light of the regulatory changes occurred in the period, detailed in the
following section. The vast majority of the recommendations made were effectively implemented by the
Government (Ministry of Mines and Energy) and the regulatory agency (ANEEL).
6 The debate was reinforced when the Government submitted a proposal to the Congress to better define the role of
the regulatory agencies, in September of 2003. The proposal, which is still pending in the Congress, contains
elements that limit regulatory agencies' independence, such as establishing performance contracts and defining
ombudsman duties. The justification that accompanies the proposal refers to the need of social control over the job
done by all infrastructure regulatory agencies.
7 Based on a literature review, Pires and Goldstein (2001) appears to be a unique study on the subject. The authors
evaluate the performance of the regulatory agencies responsible for the regulation of the telecommunications,
electricity and oil sectors. It is not an empirical study, however.









obtained, while Section 6 portrays the robustness checks performed. The final section provides

concluding observations.

Institutional Background

In event studies, the interpretation of results requires a reasonable understanding of the

context in which the regulatory initiatives were undertaken.8 In Brazil, fluvial basins with an

enormous hydroelectric potential and a large territory account for the configuration of the

electricity sector, where hydropower generation9 is linked to a large transmission network.

Transmission lines deliver electricity at high voltage to, for the most part, regional distribution

monopolies.

Most of the investments on hydropower facilities were made in the period of 1965 to 1990.

These investments were primarily carried out by Eletrobras, a state-owned enterprise which was

a holding company comprised of four concessionaries engaged in generation and transmission

(Furnas, Chesf, Eletronorte, and Eletronuclear) and two regional distribution utilities (Light and

Escelsa).10 Nevertheless, there occurred a progressive deterioration of the public sector's

capacity to invest, caused by the deepening in macroeconomic instability in the 1980's, as well

as a widespread inefficiency emerged from a version of rate-of-return regulation. The system

utilized a single nationwide uniform tariff accompanied by a compensation scheme to equalize

price and cost differentials among firms. The resulting weak sector performance brought about

the need of structural reform.11


8 The following is partly based on Goldstein (1999) and Mota (2003).
9 At the end of 2000, hydroelectric power plants accounted for 88.4% of the total capacity of 67,713 MW.
10 By the early 1990's, Eletrobras accounted for 57% of total generating capacity, with the remaining 43% being
mostly provided by Itaipu, the binational enterprise between Brazil and Paraguai, and vertically integrated utilities
owned by the states of Sao Paulo (CESP), Minas Gerais (CEMIG), Parana (COPEL), and Rio Grande do Sul
(CEEE).
1 For details regarding the determinants of structural reform in the electricity sector, see Oliveira and Pires (1994),
and Ferreira (2000).









The power sector reforms began in 1995. While constitutional amendments abolished the

public monopoly over infrastructure industries and allowed foreign companies to bid for public

concessions, the Law 8,987/95 (General Law of Concessions) set the stage for the beginning of

the privatization process, represented by the auctions of Escelsa in 1995 and Light in 1996.

In addition, part of the implementation of a new regulatory framework involved the

establishment of an independent regulatory agency (ANEEL) in late 1996 and, in the same year,

the commission of an international consultancy to study and propose a new model for the

electricity sector. The consultant's report was released in 1997 and its proposals were

incorporated into Law 9,648, issued on May of 1998.12 In essence, the approved model focused

on privatization and unbundling of generation, transmission and distribution assets, gradual

transition to a competitive generation environment in nine years, creation of a wholesale power

market, operation of the transmission network by an independent operator, and use of the price-

cap regime to regulate distribution tariffs, replacing the previous cost of service system.

Two main points emerge from this reform process. First, the market was opened to private

investors while the restructuring was still under study, and before crucial points such as pricing

regulation and tariff review procedures were defined. Second, as a consequence of this uncertain

context in which it was born, the regulatory agency, in its first years of existence, devoted its

resources to refining elements incompletely specified in the law. Staff were engaged in the

process of discussing and resolving issues, including developing specifics for the new sector

model still to be implemented. 13





12 See Ferreira (2000), Mota (2003), and de Oliveira (2003), for detailed descriptions of the new model's
characteristics.
13 The events identified for years 1999 and 2000 are symptomatic of that (Table 2-3).










By the end of 2000, the privatization process had concentrated on distribution companies,14

reflecting a deliberate intent of privatizing generation only once some key elements of the new

model had been established, such as the wholesale power market, and a significant delay in the

implementation process itself. Nonetheless, the prospect of an imminent privatization resulted in

generation companies controlled by Eletrobras dramatically reducing new investments.

The slow pace of the new model's implementation, the failure in designing an arrangement

that could make investments in thermo plants economically viable and a severe drought in the

end of 2000 were the main factors behind the Brazilian energy crisis, culminated by the rationing

measures announced in May of 2001. The rationing imposed severe losses on both generation

and distribution companies, ultimately recognized and authorized to be compensated for by the

Government in the so-called Acordo Geral do Setor (Sector General Agreement).15

The energy crisis also led to a re-evaluation of some of the foundations of the model that

had been implemented. In January of 2002, the Government decided to postpone the flotation of

the generation companies Furnas, Chesf and Eletronorte, and announced that it would continue

establishing the prices of energy supplied by state-owned generators.16, 17 Then the new

Government introduced extensive modifications in sector arrangements, which were officially


14 A total of 23 companies had been privatized, with 19 being distributors and 4 generators. In 2000, the private
sector participation in the distribution and generation markets was 60% and 20%, respectively.
15 The Agreement established that the losses would be covered by special loans that distribution and generation
companies would contract with BNDES, the Brazilian Development Bank. Moreover, it determined that the loans
would be paid back with the proceeds from the extraordinary tariff increase (2.9% for residential and rural
consumers, and 7.9% for the other consumers).
16 The Government decided to abolish the disposition (approved by Law 9,648/98) which stipulated that the energy
supplied by the state-owned generators would be progressively sold in the free market, at the proportion of one
fourth per year, starting in January of 2003. The argument employed was the need to prevent a boom in electricity
prices. However, given the negative reaction to its initiative, later on 01/31/02 the Government partly revised its
position, defining that the energy would be sold in public auctions, with the minimum price being the one that the
state-owned generator had been charging on its contracts.
17 As a result of these initiatives, the Governments of the States of Sao Paulo and Parana reconsidered the
privatization of their electricity companies as well (Cesp and Copel).









announced in July 2003: the formation of a generation pool, discontinuation of the wholesale

power market, prohibition of self-dealing, and the commercialization of energy through long

term contracts (20 years), to be signed between all generators and distributors. These

developments provided the back drop for rate review announcements affecting individual

distribution utilities.

Methodology

The event-study technique consists in testing for the existence of significant changes in

prices (abnormal returns) of firms' securities at the moment that the event was initially released

or announced. The methodology relies on the assumption that stock markets operate efficiently,

so that any unanticipated event that has an impact on firm's value will be immediately reflected

in security prices, with the price change being an unbiased estimate of the change in firm's future

cash flows. As a result of this feature, the methodology has been used to assess the impact of

regulatory initiatives in several industries.18

In the identification strategy employed in the present study, the standard event-study

technique is initially used to check for the existence of abnormal returns at the moment of the

regulatory events' announcement. The investigation, at this point, is accomplished by performing

both a joint hypothesis test and a constrained regression procedure, where the event's

coefficients are forced to be the same across securities, in order to gather evidence on the

direction of the event's impacts. The sequence explicitly recognizes that any conclusion

regarding a regulatory event's impact has to account for the possibility of event anticipation.

Thus, identified cases of significant abnormal returns are checked for the existence of abnormal

returns in the 5-day period before the event's announcement. If the anticipation is confirmed, the



18 Some examples are Rose (1985), Beneish (1991), Prager (1992), Carrol and Landim (1993), and Landim (1999).









event's impact is given by a computed measure of the event's total (or net) abnormal returns.

These procedures are detailed below.

Event studies in general usually adopt the statistical method where abnormal returns are

modeled as the prediction errors of the regression of security returns on market returns, and

hypothesis tests are conducted under the assumption that the residuals are independent and

identically distributed (the "market model"). In the context of a regulatory event-study, though,

some problems arise. Here, it is quite common to have all or most of the firms in the sample

belonging to the same industry, or the regulatory events occurring during the same calendar time

period for all firms, violating the independence assumption. Moreover, there is evidence that

market model residual variances differ across firms, as a result of the positive relationship

between market model residual variance and systematic risk (as well as between the variance of

returns and systematic risk).19

To account for these potential problems, this study employs the approach developed by

Binder (1985b) and Schipper and Thompson (1983), which assumes the following return-

generating process for a firm i:

t
Rit = ai + PiRmt + AiDOLLARt + yiLIQUIDt + I ijyDjt +it t = 1, ...., T (2-1)
j=1

where Rit is the return on the security of firm i in period t, Rmt is the equally weighted return on

the market portfolio in that period (IBOVESPA),20 DOLLAR is the percentage variation in the

exchange rate in the period t,21 LIQUID stands for a liquidity measure, given by the ratio of



19 See Fama (1976, pp. 121-124).
20 Regressions were performed using either the equally weighted (IBOVESPA) or the value-weighted (IBX) indexes
of market returns. Better specifications were obtained when the equally weighted index was used.
21 DOLLAR is given by ((ER,IER, )- 1)* 100 The exchange rate expresses the amount of reais needed to buy one
dollar, with its increase reflecting an appreciation of the dollar.









negotiated volume at time t to company's market value, Djt is a dummy variable equal to 1 if

event j occurred at time t and 0 otherwise, and Cit is a random disturbance. Note, for instance,

that a, and /, are the market model parameters, and ,j is a measure of the abnormal return

associated with event j for firm i.22

The DOLLAR's incorporation in the model was motivated by the belief that the oscillation

of the exchange rate might have caused differential impacts on companies in the electricity

sector. In the period investigated, electricity firms were more exposed to exchange rate

fluctuations than the average firm in stock market, since most of them either bought energy from

Itaipu (whose tariff is quoted in dollars) or were more indebted in foreign currency than firms

from other sectors. Given that changes in the exchange rate and stock market returns are

negatively correlated, a negative sign is predicted for DOLLAR. The expectation is that an

increase in the exchange rate produces a higher (negative) impact on electricity companies,23

compared to the impact on the average firm in the market (represented by the stock market

index).

The LIQUID's inclusion, on its turn, is due to the existing evidence that liquidity helps

explain time-series variations in stock returns.24 In case, the prediction is that LIQUID has a

positive sign, since it is anticipated that an increase in liquidity over time leads to an increase in







22 It was tested an alternative specification, where abnormal returns were measured as CAPM prediction errors. In
case, the risk-free asset return (given by the SELIC daily variation) was subtracted from both Rit and Rmt. However,
lower R2 was observed for all individual regressions performed.
23 Except for the cases of Eletrobris (which holds a participation of 50% in the Itaipu Enterprise), and the
transmission company Transp.
24 Beneish and Whaley (1996), Lynch and Mendenhall (1997), and Elyasiani, Hauser, and Hauterbach (2000) find
evidence of permanent excess returns associated to improvements in the stock's liquidity.









stock returns, as a result of adjustments in stock prices to incorporate the corresponding change

(reduction) in transaction costs.25

An equation of the above form is specified for each firm in the sample, resulting in a

system of N equations for N firms, estimated with a seemingly unrelated regressions procedure.

Given that the individual return equations are estimated jointly with generalized least squares,

the procedure has the advantage that heteroskedasticity across equations and contemporaneous

dependence of the disturbances are explicitly incorporated into the hypothesis tests. Moreover,

the approach makes it possible to test the joint hypothesis that all dummy variable coefficients

(i.e. the abnormal returns of all firms) for a given event equal zero, which has particular

relevance in the context of regulatory events, where it is quite common to have differential

impacts (including in sign) among firms in the industry.26

The analysis is complemented by another set of hypothesis tests, where the system of

equations is estimated with the event coefficients being constrained to be the same across

equations (securities). The estimates obtained from this procedure are equal to the estimates

which would be obtained from a single regression run on a weighted portfolio of the original

securities, where the weights are proportional to the inverse of the estimated covariance matrix

of residuals used in the joint GLS estimation (Schipper and Thompson, 1985).27 The constrained

regression procedure provides estimates of the regulatory event parameters and, as such, enables

25 Investors value liquidity, the quick execution of their orders at the lowest cost. Liquidity, therefore, represents the
ease of buying or selling the stock, which can be measured by the bid-ask spread. The more illiquid is the stock, the
higher is the spread, or the transaction costs to the investor, and the lower should be the stock's price, which adjusts
to incorporate the higher transaction costs. This return-spread relation is taken as a rational response by an efficient
market to the existence of the spread (trading friction and transaction costs), rather than an indication of market
inefficiency (Amihud and Mendelson, 1986).
26 When the abnormal returns differ in sign across firms this will frequently be a more powerful test of the
hypothesis that the event affects security holder wealth than the test that the average abnormal return equals zero
(Binder, 1998).
27 The estimates represent variance-minimizing weighted averages in which greater weights are given to
observations with low variance and low or negative covariance with the other observations (Salinger, 1992).









to get evidence on the direction of the events' impacts and to draw inference in terms of

economic significance.

In addition, the sample is composed of companies that operate in the generation,

transmission, and distribution sectors, but some of the events are essentially "distribution

events," in the sense that they are expected to have an impact on distribution companies only.

Since the incorporation of generation and transmission companies' abnormal returns in these

cases might end up obscuring the event's effects on distribution firms, hypothesis tests were

performed considering two groups of firms, one consisting of all companies in the sample, and

another consisting only of the distribution companies.

Hence, the investigation on the existence of abnormal returns at the moment of the events'

announcement is carried out by testing the following hypotheses:

Hi: The regulatory event parameters for each firm in the sample are all equal zero;

H2: The regulatory event parameters for each distribution firm in the sample are all equal

zero;

H3: Under the assumption that the event parameters are the same for each firm in the

sample, the abnormal return during the event window equals zero;

H4: Under the assumption that the event parameters are the same for each distribution

firm in the sample, the abnormal return during the event window equals zero.

For the specific events where the null hypotheses H1 or H2 are rejected, the possibility of

event anticipation is checked. The same joint hypothesis and constrained estimation tests are

applied to the cumulative abnormal returns (CAR) earned by each security in the period of five

trading days before the events' announcement.28 If there is evidence of event anticipation, a


28The test for cumulative abnormal returns followed the procedure suggested by Salinger (1992, p.42), which
consists in the inclusion of "dummies" for each day t of the event-window as explanatory variables. In the









conclusion regarding the event impact on security returns is taken on the basis of the "event

overall effect", given by the results of a new set of hypothesis tests where the event period

includes the announcement and the (five) pre-announcement days.

Both stock returns and stock market returns are measured on a daily basis. Consequently,

the base specification employs a 1-day event window, which means that it is being tested

whether or not the regulatory event had an impact on firms' value in the day it was initially

released or announced.29 Nonetheless, in order to account for the possibility of information

leakage in the day before, as well as for the chance that an announcement was made after the

closing of the stock market, the study also makes use of 2-day and 3-day event windows.30

Due to the missing observation problem (detailed in Section 4 below), the present study

investigates the significance of regulatory events of each year, from 1999 to 2003, using the

observations from the year under examination and the previous year. This procedure prevented a

further reduction in sample size, by enabling the inclusion of stock returns from firms that have a

missing observation problem in some years, but not in others.

Data

The sample was limited by availability of data and by the fact that only 34 electricity

companies had stocks traded at the Sao Paulo Stock Exchange-BOVESPA, the main Stock

Exchange in Brazil.






observation for period t (t>l), the "dummy" for period t takes on the value 1 and the "dummy" for period t-1 takes
on the value -1.
29 Brown and Warner (1985) provide evidence that shorter event windows increase the power of the tests.
30 In the 2-day window, the event dummy was set equal to 1 in the event day and in the day immediately after, while
in the 3-day case the same was done for the three-day window centered on the event day. On the few occasions
where events were close together, the dummies were truncated to prevent overlap.









The information obtained from BOVESPA included daily stock price data related to 31

companies,31 for the period of 03/16/1998 to 09/30/2003. The data, however, presented a severe

missing observation problem. Seventeen companies had more than 40% of observations

missing,32 and were consequently dropped from the sample. Daily stock returns were computed

for the remaining 14 firms, resulting in a total of 1372 observations for a company with no

missing data.

The computation of stock returns incorporated the necessary adjustments for share splits

and rights issues occurred, stock dividends given, and all forms of cash payments made to

stockholders.33 Nonetheless, the transformation of share prices into stock returns increased the

number of missing observations. The final sample was then defined considering the number of

missing returns in the period of two years of data yeart and yeart-l) employed to run the

individual regressions for years 1999 to 2003 (Appendix A). Ultimately, stock returns from a

total of 12 companies in the electricity sector were used to perform the statistical tests, with the

exact number of companies' stock returns employed in each year varying from 9 to 11.

Detailed information regarding these 12 companies is provided in Appendix A.

Importantly, the numbers for participation in the distribution and generation markets indicate that

the sample is composed of some of the large companies in the electricity sector.34 These numbers

also show that, except for a few cases, it is not possible to come up with a pure classification of

the sample companies in terms of belonging to distribution or generation sectors. The

31 The data received from BOVESPA did not include daily stock price of firms CPFL Piratininga, Energipe and
VBC Energia.
32 From the mentioned 17 firms, 09 had more than 80% of missing data and 15 were in the above 60% range.
33 In order to ascertain the correctness of the adjustments performed, the computed daily returns were aggregated
into monthly returns and compared to the information on monthly stock performance provided by BOVESPA.
34 As of December/2000, Eletropaulo, Cemig, Light, Copel, and Celesc were among the eight biggest distribution
companies in the country. On the other hand, Eletrobras, Cemig, Cesp, Tractebel, and Copel were, in this order, the
five biggest generation companies (Setor E16trico Ranking 2001, Vol. 1 BNDES).









classification reported on the third column, which will be utilized throughout the paper, comes

from the comparative analysis of the data regarding electricity delivered to final customers and

electricity generated.

The sample implications to the present study are twofold. First, the paper assesses the

regulatory events' impact on the large players, which implies that the results cannot be taken to

draw conclusions with respect to effects in the electricity sector as a whole. Second, given the

existing evidence of an inverse relationship between firm size and systematic risk, as well as the

already mentioned relation between systematic risk and variance of returns, the use of larger

companies should strengthen the power of the statistical tests.

The identification of the regulatory events was made through the careful analysis of the

following material: (a) resolutions, resolution proposals submitted to public hearings, and press

releases issued by ANEEL during the period of March 1999 to September 2003; (b) reports in

Gazeta Mercantil, the main financial newspaper in Brazil, related to economic regulation of the

electricity sector, released in the period of March 1999 to June 2002; and (c) daily reports in the

electronic newsletter IFE, jointly provided by Eletrobras and the Economics Institute of the

Federal University of Rio de Janeiro (UFRJ), over the period of November 2000 to September

2003.35 The 65 events listed in Table 2-1 were selected on the basis of their relevance in terms of

the new regulatory framework that was being implemented, the expected magnitude of the

impact to the stakeholders involved and the repercussion in the media. Note that the events'

selection has the date 03/16/1999 as its starting point, given the decision to use at least one year

of observations (250 trading days) as the estimation period.


35 The electronic newsletter IFE collects reports related to electricity regulation released in the following
newspapers/magazines: Correio Braziliense, Diario Catarinense, Diario do Grande ABC, Diario do Nordeste, Folha
de Pemambuco, Folha de S~o Paulo, Gazeta Mercantil, InvestNews/GM, Jomal do Brasil, Jomal do Commercio, O
Estado do Parand, O Estado de Sio Paulo, O Globo, Zero Hora, Valor, and Canal Energia.










The descriptive statistics provided in Table 2-2 depicts a higher variability of stock returns

in 1998 and 1999, compared to the other years under examination. Both the range and the

computed standard deviation of stock market and securities' daily returns decrease sharply from

1998 to 2003. The higher volatility in stock returns in the first two years examined should make

it more difficult to reject the null hypotheses associated with the 1999's events. Note also that the

exchange rate varies more in years 1999 and 2002, implying an expected higher impact of this

variable on security returns in these years.

The mean liquidity measures indicate that some securities (Fcata, Tract, and Cerj) are not

actively traded. Here, the main concern resides in the fact that stocks with irregular transactions

may not have incorporated in their prices the change in firms' value brought by an unanticipated

event. The correlation matrix (Table 2-3) indicates that this seems to occur with respect to

macroeconomic events that affect the market as a whole, since the returns from the less traded

securities are the ones which have the smallest correlation with stock market returns. Table 2-3

also reveals, as expected, a significant direct association between volume negotiated and stock

market returns (with the logical exception of the least traded stocks)36 and a negative and

significant correlation between market returns and variations in the exchange rate.37 Notice that

some of the sample firms' stock returns display a quite high negative correlation with the

exchange rate, which corroborates the need to include this variable in the model specification.



36 For some of the not actively traded securities, there is at least an indication that they were more negotiated in days
of announcements specific to these firms (their liquidity measure is significantly correlated with their own returns,
but not with the market returns). This, however, does not occur with Fcata (the least traded stock), which is also the
only security whose daily returns are not significantly correlated with the variation in the exchange rate. Thus, the
fact of not being regularly negotiated can be taken as the main reason for the bad specification of Fcata individual
regressions, mentioned in Section 5 below.
3 Due to the evidenced correlation between these variables, we have also performed statistical tests with the
DOLLAR variable being defined by the residuals of the regression of exchange rate variation on stock market
returns. Practically no changes were observed in relation to the event parameter estimates shown in the present
study.









The annual exchange rate variation and the annual performance of each stock are displayed

in Table 2-4. The exchange rate depreciated highly in years 1999 and 2002, which is in line with

the picture shown by Table 2-2. On the other hand, the comparison among sample firms' average

stock returns and market index returns reveals that the stocks from the electricity companies

included in the sample outperformed the market in the period examined,38 in spite of the rather

poorer performance in year 1999. The present study investigates to what extent the difference in

performance can be explained by regulatory decisions.

Findings

The regressions incorporated the necessary adjustments for correlation in security returns

over time. The individual regressions' estimated parameters are presented in Table 2-5.39 The

findings indicate that the model is well specified (R2 in the range of .30 to .80), with a few

exceptions in the cases of the three least traded securities already mentioned.40 As expected, the

IBOVESPA's coefficients are positive and strongly significant. The average betas estimated for

each firm are in line with the sector of activity examined,41 whereas the sample average beta

shows a U-shaped evolution in the period. The decrease in firms' systematic risk from 1998 to

2000 reflects the reduction in uncertainty brought by the progressive definition of some crucial

points of the model being implemented. On the other hand, the beta's increase after 2001 might







38 The same holds when the Electricity Sector Index (IEE) reported by BOVESPA is compared to the market index.
As expected, the sample firms' average stock return and the IEE were highly correlated (p = 0.6956).
39 For reasons of space, we do not show the event dummies' estimated coefficients of each individual regression.
40 The main concern was with respect to Fcata, included in the sample of years 2001 to 2003, whose regressions
showed an R2 in the range of .06 to .11. The results obtained when this security was excluded from the sample were
similar to the ones reported in this paper, though. The three cases in which the hypothesis tests provided stronger
results are mentioned later on in the text.
41 For some firms, the low negotiated volume might have resulted in an underestimation of their beta values.









be indicating that the model revision carried out after the energy crisis amplified the regulatory

risk.42

The DOLLAR coefficient is negative and significant in about half of the individual stock

returns' regressions, denoting that variations in the exchange rate did provoke a differentiated

impact on some firms in the electricity sector, compared to other firms in the stock market. The

impact occurs in practically all years examined, and is not more pronounced in years 1999 and

2002, as it was expected. Additionally, the results indicate that liquidity contributes to explaining

variations in firms' stock returns, and support the prediction regarding the variable's sign, since

the majority of the LIQUID's estimated coefficients is positive and significant. Here, however, as

it is possible to anticipate an increase in trading activities in days of important announcements, it

should be noted that, in case of events that impact firm's value positively, the LIQUID's

estimated coefficient might be capturing part of the estimated announcement effect, leading to an

over rejection of the null and to an under estimation of the abnormal returns associated to these

events.43

Tables 2-6 to 2-10 report the results from the hypotheses tests performed on the sixty-five

regulatory events. Before proceeding to their analysis, though, the peculiarity of the regulatory

process must be emphasized. In a regulatory context, where the main issues follow the ritual

"initial proposal public hearing final decision," the events' information content is often

leaked or flagged before the regulator's final decision, with the corresponding impact on firms'

value being progressively incorporated into stock prices before the official announcement is


42 It is beyond the scope of this paper the exam of the regulatory events' impact on firms' systematic risk. For
studies that address the issue, see Antoniou and Pescetto (1997), Buckland and Fraser (2001), and Morana and
Sawkins (2002).
43 This problem is possibly reduced by the fact that the statistically significant LIQUID's coefficients were, in their
majority, the ones related to larger (and with higher trading volume) firms, given that Chordia, Shivakumar, and
Subrahmanyam (i 4) have found that these firms' liquidity are the least affected by information shocks.










made. As a result, a finding of no significant effect may occur for events that impacted firms'

value. It follows that it is harder to reject the null hypotheses H1 and H2,44 when the decision is

taken after the implementation of the mentioned process.45 On the other hand, the occurrence (or

not) of significant abnormal returns on rulings set forth after the ritual constitutes evidence

regarding the level of consistency and predictability of regulatory decisions.

The peculiarities of the regulatory process, associated to the high volatility that

characterized the Brazilian stock market in years 1998 and 1999, may explain the non-rejection

of the null hypotheses in any of the 1999's events. In contrast, in years 2000 to 2003 there is

evidence of significant abnormal returns for a total of 23 regulatory announcements. Here, the

surprise verified in more than one third of the investigated events might be attributed to two main

factors: a) the ruling deviated from what the market had anticipated; or b) there were still too

much uncertainty regarding the final decision, due to a possibly wide spectrum in terms of

regulatory choices and methodologies or simply to the fact that no (or very little) information

was provided to the market previously, particularly in cases of rulings not submitted to the

mentioned regulatory process. We will return to these points later.

For all of these 23 events, the analysis that follows focused on the outcomes of the tests of

hypothesis H2 and H4 in case of a "distribution event" and incorporated the results from the

44 It also contributed to tougher joint hypothesis tests the incorporation of the LIQUID variable in the model, due to
the already mentioned expected higher traded volume in days of important announcements, and the fact that most of
the firms in the sample were included in the portfolio employed to compute the market index. Consequently, the
IBOVESPA captured part of the variation in the sample firms' daily stock prices caused by events specific to the
electricity sector (as of December/2001, electricity companies accounted for 12% of the market index, in terms of
market capitalization value).
45 The ritual initial proposal public hearing final decision is a practice adopted by the regulatory agency.
Therefore, all rulings in which the decision took place at the Government (Ministerial) level did not follow this
process. The rulings of this nature amount to 23 out of the 65 events listed in Table 2-1. In addition, some of the
regulatory agency's announcements did not follow this process either, mainly because they were just the initial
proposal regarding the issue. The ritual occurred in 22 of the (42) ANEEL's announcements listed in Table 2-1. The
decisions which do not follow the ritual should convey more information to the market at the moment of their
announcement. It is thus expected that they lead to a higher proportion of stock price reactions than the decisions
that follow.










constrained regression procedure (columns H3 and H4 in Tables 2-6 to 2-10), the cumulative

abnormal returns' hypothesis tests (Table 2-11), and the "event overall effect" (Table 2-12),

when applicable. Moreover, the examination of the results took into account the findings in terms

of each security's abnormal returns at each of the 23 significant events, especially when the

evidence indicated the occurrence of differential impact among the sample securities. The same

findings, on the other hand, were employed to examine the possibility of the results being driven

by a factor other than the regulatory announcement under investigation.46 In particular, the

analysis revealed that in two opportunities (E012700 and E010901) the event's abnormal returns

cannot be attributed to the regulatory announcements made on those days, but to some specific

factors that affected Light and Fcata, respectively.47

It is worth noting that the evidence confirmed the importance of checking for event

anticipation, given the finding of significant abnormal returns in the 5-day period before the

announcement of 7 out of the 23 significant events (Table 2-11). The need to perform this

additional analysis is also illustrated by the results found for E021700 and E040803, which

denote that the abnormal returns evidenced in the announcement day are actually an adjustment

in market expectations, since the total effect is not significantly different from zero.48



46 The focus on the results from the tests of hypothesis H2 and H4 in case of a distribution event is already a way to
control for that. Note, for instance, the results for the distribution event E080503. The rejection of H1 only (in the 3-
day window specification) indicates the occurrence of significant abnormal returns on generation companies'
securities, and consequently cannot be credited to the regulatory announcement under exam.
47 In the case of EO 12700, the null hypotheses H1 and H2 are not rejected when Light is dropped from the sample (X2
= 6.00 (p=. 539), and 2 = 5.14 (p=. 274), respectively). Moreover, the event's coefficient on Light's individual
regression is highly positive, in contrast with the event's expected effect. The same non-rejection of the null
hypotheses H1 and H2 occurs in E010901 when Fcata is excluded from the sample (x2 = 9.32 (p=. 502), and 2 = 5.85
(p=. 322), respectively).
48 Take, for example, the announcement of the "Emergency Plan for the Short-Run" (E040803). In case, while the
positive CAR indicates an early incorporation into stock prices of the adopted measures (early on March 31st some
newspapers had disclosed the information that the Government would provide resources to help capitalize firms in
the sector), the negative abnormal returns on the announcement day reflects the market's surprise with the decision
to postpone the incorporation of increases in non-controllable costs (higher price of energy bought from Itaipu,
caused by the dollar appreciation) into tariffs, taken to alleviate the impact of tariff adjustments on inflation. In sum,









Information regarding the remaining 19 significant announcements is summarized in Table

2-13. The majority of the significant events caused a positive-as opposed to negative-effect

on firms' value, which is in line with the already mentioned better performance of the sample

firms' securities over the period, compared to the market portfolio. On the other hand, the results

show that a higher proportion of decisions taken at the Ministerial level turn out significant,

when contrasted to regulatory agency's decisions, confirming the prediction that Government

announcements should convey more information to the market.49

As previously stated, finding abnormal returns at Government announcements should be

credited more to uncertainty regarding the final decision, than to deviation in market

expectations. The distinction is relevant because in the last case the event's estimated impact

cannot be taken as a direct measure of the decision effect on firms' value, since the market prior

is unknown. o In this context, it is interesting to note that the estimated impacts on security

returns portrayed in Table 2-13 are, as a general rule, related to the nature of the Government

announcements. While a negative impact is evidenced at the disclosure of the rationing plan

measures (E051801)51, reflecting their deleterious consequences on companies' future cash

flows, positive effects are found for the subsequent compensatory measures undertaken by the





the non-rejection of the null hypothesis of no "event overall effect" suggests that the unforeseen tariff decision led to
a reevaluation of market's previous expectations that firms would be benefited by the Government measures.
49 43% (10/23) of Government's rulings turn out significant, against 21% (9/42) of ANEEL's rulings. As
anticipated, the difference might be attributed to the fact that Ministerial decisions are less susceptible to information
leakage or anticipation, since they do not follow the regulatory process. This result, however, may also reflect the
difference in the nature of the decisions taken at the Ministerial and the regulatory agency levels. Since Ministerial
rulings focus on the broad definition of the model for the sector, it is expected that they have a higher impact in
firms' value than agency's rulings, which in some cases only detail what has been previously defined by the
Ministry.
50 This point is detailed later on in the text.
51 For both E051801 and E080701, the results are stronger when Fcata is excluded from the sample (2= 18.61
(p=.046), and 2= 10.85 (p= .054), respectively).









Government (E090601, E102501, E09080352, E091603). Additionally, the finding of differential

impact across firms is consistent with the decision to control the price of energy supplied by

state-owned generation companies (E010902), a typical within sector redistribution policy,

whereas the positive abnormal returns found for E013102 are in line with the Government's

reconsideration of its previous position, materialized by the determination that the state-provided

energy would be sold in public auctions.5

The negative reaction to the Government's proposed revision in the regulatory agencies'

job (E092303)54 is according to expectations as well, since the initiative to limit the agencies'

independence constitutes a step back in the electricity sector regulation and induces the fear of a

political use of the regulator, increasing the regulatory risk. Note, however, that in two occasions

this alignment between the nature of the Government announcement and the estimated effect on

security returns is not observed. First, negative abnormal returns are evidenced in case of a

compensatory measure (E081602),55 reflecting the decision's implications for Eletrobras and

indicating that the market was expecting a better compensation system for the distribution

companies. Second, differential impact across firms in the sample is shown for a change in

ANEEL's board of directors (E050201), as if the new composition would be beneficial to some

firms, but not to others, when no abnormal returns were expected for this event. The new

directors were indicated by the same Government that had designated the previous ones, what


52 The settlement between BNDES and AES in 09/08/03 solved a dispute that had lasted for more that one year, and
was interpreted as a signal that the Government was concerned with electricity companies' solvency.
53 The estimates of this announcement's impact on each of the sample's securities indicated that one distribution
company (Eletropaulo) was negatively affected by the Government decision.
54 See footnote 6.
55 The Government decided that resources from the 'Reserva Global de Reversao' would be used to cover the
reduction in concessionaries' revenues brought by the new criteria for low-income customer (see footnote 57). The
negative impact on Eletrobras' security should come from the fact that the mentioned resources were being used by
this company to capitalize the seven highly indebted former state-owned distribution companies that were
transferred to the Federal Government.









should not signal a change in the way the regulator would rule on regulatory issues in the future.

Here, though, the result is not conclusive. The individual securities' coefficient estimates are not

consistent by firm type and the CAR's findings suggest the possible early incorporation of the

rationing plan measures.

Among the nine regulatory agency's initiatives displayed in Table 2-13, some consist of

decisions taken after the implementation of the discussion process previously mentioned. In

these cases, the observed effect on security returns denotes that the final decision deviated from

market expectations formed along the discussion process, what poses difficulties to an

interpretation of the estimated effects on firms' value in the absence of information regarding the

market prior.56 The result found for the ruling defining the basic transmission network

(E111000), for example, should be revealing that the action taken towards the effective

implementation of the model for the electricity sector reduced regulatory uncertainty, but it

cannot be ignored the possibility that the positive impact is just a review of (negative)

expectations formed on the basis of all the information released or flagged along the regulatory

procedure.

One possibility to overcome the problem is to consider the observed reaction to the

agency's proposal concerning the issue as a proxy for the market prior.5 ANEEL's initial

proposition on the ruling imposing restrictions on agents' participation in the market (E041700)

had an impact on stock returns. In case, the estimated event parameter is not significant,

indicating the occurrence of differential effects among firms, with some benefiting and others


56 Regulatory discussion processes usually take two to three months. Thus, eventual abnormal returns occurred along
this period, as a result of a progressive incorporation into stock prices of expectations concerning the ruling's impact
on firms' value, will not be captured by the CAR check employed in the present study, limited to the 5-day period
before the final announcement.
57 This possibility relies on the assumption that the expectations initially formed on the basis of the regulator's initial
proposition do not change in the period that goes up to the announcement of the final decision concerning the issue.









harmed by the proposal. By taking these effects as the market expectations concerning the

ruling's impact on firms' value, one may conclude that the positive abnormal returns associated

to the final decision (E072100) result from adjustments made by the regulator on its original

position, making it more aligned with companies' interests.

Similar analysis might be applied to the events related to Escelsa's periodic tariff review.

The announcements were embedded with a high signaling power, given the difficulties faced by

concessionaries in the rationing period and the fact that it preceded the rate reviews of other

companies. Under the generalized uncertainty regarding methodological choices and methods to

be used, the positive impact evidenced for the regulator's initial proposal (E062101) provides

indication of its robustness and soundness. And the finding of another positive impact when the

regulator takes its final decision (E080701)58 denotes that ANEEL made further adjustments in

the methodology, interpreted as favoring companies in the sector, or revised upward the

repositioning index initially proposed, as if the regulator indeed wanted to signal that was aware

of the problems imposed on distribution companies by the rationing plan and would compensate

them in the future for that. 59

Conversely, the results found for the norm issued to define low-income customer

(E050202) suggest that ANEEL's decision on this topic impacted firms' value negatively.60 In

this case, however, there is no proxy for the market prior, since the previous Congress decision

regarding the issue (E041002) caused no reaction on stock prices. Nonetheless, the fact that a


58 See footnote 51.
59 When asked about compensatory measures to the electricity companies, the President of BNDES affirmed that the
Government had been signaling that it would solve the questions pending in the sector, giving as an example the
recent tariff increase given to Escelsa (Folha de Sio Paulo, 10/19/2001).
60 The prediction was that the new criteria for low-income customer would increase the number of households under
this classification from 10.3 million to 18 million, or 36% of the total market (Gazeta Mercantil, May 16,th 2002).
The negative impact on firms' returns should come from the fact that low-income customers were exempt from
paying the extraordinary tariff increase set forth by the "Sector General Agreement".









negative surprise is found for E050202 shows that the regulatory agency had some discretion in

defining the concept, and used the flexibility to set up a ruling which was not as favorable to

firms as the market was expecting.61

The findings related to ANEEL's decisions regarding the asset base valuation62 were

unexpected. The voluminous press coverage related to the issue, which invariably reported the

distribution companies' disappointment with the proposed reposition cost methodology, led to an

expectation of a strong negative impact of ANEEL's initial proposition (E062102), not

confirmed by the hypothesis tests' results. The findings provide evidence that abnormal returns

did occur,63 but suggest a differential effect across firms, as if the market had been able to realize

that the methodology would harm recently privatized companies, by not considering the price

paid for the corresponding assets, but benefit the concessionaries that had invested in facilities

promoting universal service and built a large asset base.64 Additionally, the evidence concerning

ANEEL's final decision (E090402), which kept the same methodology, does not provide support

to the alleged surprise that this decision produced on concessionaries.65 Hence, if the distribution

companies expected that the regulator would reconsider its initial proposal, the findings indicate

that the market did not. This result, it must be stressed, is consistent with a market belief

regarding the regulator's impartiality.



61 ANEEL's decision may be interpreted as favoring consumers if it is assumed that the market did not expect a
ruling which would favor electricity companies.
62 Foster and Antmann (2'1 14) provide a detailed overview of the discussion process that resulted in the adoption of
the repositioning cost methodology by ANEEL.
63 The event is significant at the 5% level when Fcata is excluded from the sample (X2=11.63, p-value = .040).
64 Positive coefficient estimates were found for Cemig, Light and Celesc's securities. For Eletropaulo, on the other
hand, the results suggest a negative impact.
65 The Brazilian Electricity Distributors Association (ABRADEE) issued a note, released on 09/04/2002, expressing
its surprise with the decision taken by the regulatory agency. According to the Association, there was an ongoing
negotiation process with Aneel and other sectors of the Government in order to revise the reposition cost
methodology initially proposed (Canal Energia, 09/04/02).









The results related to the announced repositioning indexes, on their turn, provide insights

into the level of objectivity and transparency of the methodology employed in the periodic tariff

review.66 Here, the evidence is mixed. The no impact on stock returns found for the repositioning

indexes proposed initially (E021703, E030703, E031103) denotes that the numbers released

were close to market expectations. A significant impact is observed for E041703, though, due to

the positive abnormal returns caused by the regulatory announcement on Eletropaulo and Light

securities.67 The final indexes announced on this day must have incorporated adjustments made

by the regulator, and consequently led to a reevaluation of the expectations concerning these two

firms' future tariff reviews. However, the market expectations were not confirmed for

Eletropaulo (E052603), since its stock prices fell considerably when its own repositioning index

was disclosed.68 This negative surprise suggests the need of improvements in the periodic tariff

review methodology, moving to greater predictability.

The results obtained by the present study, when taken together, show that the regulatory

decisions which impacted positively electricity firms' value were essentially represented by

initiatives taken to effectively implement the model approved by Law 9,648/98 and by measures

adopted to compensate firms for the losses imposed by the energy crisis of 2001. These results

denote that the need to provide the proper environment and the suitable incentives to the





66 The rationale is that, in case of a transparent and objective methodology, the regulator's disclosure of the first
repositioning indexes, along with a detailed description of how they were computed, is sufficient for the accurate
prediction of the other firms' indexes. The two main points of the methodology employed in the periodic tariff
review were the asset base valuation by the repositioning cost and the estimation of 'efficient operational costs'
using the reference company approach previously adopted in Spain and Chile.
67 The estimates for the abnormal returns caused by the announcement on Eletropaulo and Light securities were 5.70
(p=.069) and 8.70 (p=.003), respectively. The event is significant at the 5% level when Fcata is excluded from the
sample (2= 11.15, p-value = .049).
68 The estimated abnormal returns caused by E052603 on Eletropaulo's security was -5.96 (p=. 057).









realization of the so desired investments in electricity generation, transmission and distribution

sectors had a high weight in the regulator's utility function69 over the period examined.

Interestingly, the evidence reveals that most of these "favorable regulatory decisions" were

discussed and deliberated at the Ministerial or the Congressional levels, preserving the regulatory

agency's impartiality, which is indeed supported by the results. This investigation suggests that

ANEEL is seen as a neutral institution and acts as a relatively independent organization. While

some of the ANEEL's decisions which significantly affected security returns led to an increase in

electricity firms' value, some provoked a differential effect among firms, and some were

effectively contrary to companies' interests.

Some caution on this interpretation must be exercised, though, given the fact that every

event-study restricts the analysis to decisions which could not be correctly anticipated by the

market, a problem that strongly affects regulatory agency's rulings which follow the ritual

"initial proposal-public hearing-final decision," compared to Ministerial rulings. On this

respect, it must be stressed that finding 5 stock price reactions in 22 cases where the information

was flagged or leaked in the regulatory process exceeds what would be expected. Actually, this

result constitutes a sign of unpredictability of regulatory agency's decisions.

This finding also provides insights into the applicability to the Brazilian electricity sector

regulation of the two opposing theories concerning the pattern of government intervention in

business. As a higher predictability should be expected if the regulatory agency consistently

serves as an impartial referee that aims to maximize social welfare or consistently favors a

specific interest group (Government, industry, or customers), the evidenced unpredictability

indicates that the regulatory agency has favored different interest groups at different times,


69 The word regulator, here, is used in a broad sense. It encompasses not only the regulatory agency, but also other
actors with the power to adopt regulatory initiatives, such as the Government (Ministry) and the Congress.









confirming Peltzman's (1976) prediction that the utility maximizing regulator will not

exclusively serve a single economic interest.

In terms of economic significance, the results found in the present study indicate that the

regulatory decisions examined were responsible for an increase of around 8% in sample firms'

market value (Table 2-13). As previously stated, the equally-weighted sample firms' portfolio

outperformed the market in the period investigated. A variety of factors might have contributed

for these "extra-returns", including efficiency improvements resulting from privatization and

implementation of incentive regulation.70 This research, however, suggests that regulatory

decisions account for roughly 12% of the difference in performance of the sample securities with

respect to the market index.

Robustness Checks

Two procedures were adopted to examine the robustness of the results found in the present

study. First, it was verified the possibility of a small sample bias. In light of the existing evidence

that the use of a statistic-such as the chi-squared-whose distribution is only asymptotically

known may lead to an over rejection of the null,71 new hypotheses tests were conducted using

Rao's F exact statistic. The findings were practically identical to the ones obtained previously.

Consequently, the likelihood of a small sample bias was disregarded.

Secondly, a counterfactual analysis was implemented to check the extent to which the

significant abnormal returns evidenced in this study may effectively be attributed to the

regulatory announcements investigated or simply result from the Brazilian stock market's

volatility. The analysis consisted in the application of the same hypotheses tests to a random set

70 We are currently addressing the issue in another paper. The results obtained do indicate that privatization and
incentive regulation led to efficiency improvements in the Brazilian electricity distribution sector.
1 Binder (1985a) presents evidence that joint hypothesis tests in the multivariate regression model, using the chi-
squared statistics, are biased against the null hypothesis when there are 60, or in some cases even 250 observations,
per equation.









of 69 dates.72 The results displayed in Table 2-14 show the rejection of the null of no abnormal

returns in 8 random events. In three of them, however, the same analysis of individual securities'

coefficient estimates applied before revealed that the abnormal returns are due to specific factors

that affected CERJ (E090199) and Eletrobras (E112601 andE031303).73

Given that one would expect 5% of the random events to be significant, finding significant

abnormal returns in 5 of the 69 random events is not out of line. In addition, the result may be

due to information leakage, very common in a regulatory context, and thus reflect an early

incorporation of regulatory announcements made later on. Nonetheless, it should be emphasized

the high disparity between the numbers of significant random events and significant regulatory

events, which gives credence to the results provided by this study.

Conclusions

The paper examines the regulator's performance in the Brazilian electricity sector using an

event-study methodology specially designed to a regulatory context, which explicitly accounts

for the possibility of event anticipation. Despite the more pronounced interests that characterize a

developing country's regulatory environment, the results indicate that the regulator has acted

relatively independently, with its decisions not favoring a single interest group. The findings are

similar to the ones obtained in previous studies that focused in the United Kingdom's context,

and do not support the claim that Brazilian regulatory agencies are captured by the industry.

On the contrary, the evidenced unpredictability of regulatory agency's decisions suggests

that the regulatory agency has favored different interest groups at different times, supporting the

claim that the utility maximizing regulator will not exclusively serve a single economic interest.


72 75 dates were randomly generated in the period of 03/16/1999 to 09/30/2003, but six of them had to be dropped
because they were already included in the Events' list.
73 The test statistics were the following: X2=4.03 (p=.854), for E090199 without Cerj; 2=8.66 (p=.565), for El 12601
without Eletrobras; and X=3.76 (p=.927), for E031303 without Eletrobras.









The observed unpredictability, on the other hand, reinforces the need of improvements in the

regulatory discussion process, with the adoption of measures to increase the transparency and to

promote more substantive public hearings, as recommended by Brown and de Paula (2002,

2004).

The study suggests that the need to provide incentives for new investments has had a

significant role in the regulatory process. In addition, electricity companies have been

compensated for the regulatory risk they face. The estimates indicate that regulatory decisions

led to an increase in firms' market value over the period examined and account for part of the

difference of the sample securities' performance with respect to the market.

Some specific findings are also worth noting. The adoption of the asset base repositioning

cost methodology was not that harmful to distribution companies, as one would anticipate in

light of the press coverage related to the issue, and the evidence raises the concern over the

objectivity and transparency of the methodology employed in the distribution companies'

periodic tariff review, suggesting the need of improvements. Moreover, the results indicate that

the Government's proposal to review the regulatory agencies' responsibilities and performance

was seen as a step back in the electricity sector regulation and increased the regulatory risk.











Table 2-1. Events List
Event Initiative Description
E032399 Aneel Resolution proposal: access to the transmission system
E051099 Aneel Resolution proposal: definition of the Normative Values
E061099* Aneel Extraordinary tariff review
E070199 Aneel Resolution: regulates the commercialization of energy not previously contracted
E072999 Aneel Resolution: defines the Normative Values
E080399* Aneel Resolution: Escelsa's periodic price review final numbers
E092199* Aneel Resolution proposal: quality of service
E092399* Aneel Resolution proposal: defines rights and duties of consumers and utilities
E100199 Aneel Resolution: rules for access to the transmission system
E121499* Aneel The regulator denies the requests for an extraordinary tariff review presented by 08 distribution companies
E012700* Aneel Resolution: quality of service
E021700 Gov CELPE's privatization
E041700 Aneel Resolution proposal: limits for participation in the market (market concentration)
E042400 Aneel Resolution proposal: rules for the wholesale power market
E050300* Aneel Resolution: defines the quality parameter for the X factor in Escelsa's tariff review
E061500 Gov CEMAR's privatization
E072100 Aneel Resolution: limits for participation in the market / market concentration
E080400 Aneel Resolution: rules for the wholesale power market
E111000 Aneel Resolution: defines the basic transmission network
El 12900 Aneel/Gov Resolution: rights and duties of consumers and utilities / SAELPA's privatization
E120700* Aneel Resolution: new quality standards for distribution companies
E010901* Aneel Resolution proposal: defines procedures for ordinary, extraordinary and periodic tariff reviews
E042001 Aneel Intervention in the wholesale power market
E050201 Gov Change in Aneel's board of directors
E051801 Gov Announcement of the rationing plan measures
E062101* Aneel Regulator's proposal for Escelsa's periodic tariff review
E080701* Aneel Resolution: final numbers for Escelsa's periodic tariff review
E090601* Gov Compensation for increases in non-controllable costs of distribution companies (MP 2,227/01)
E102501* Gov Authorizes that variations in distribution companies' non-controllable costs be adjusted by the Selic interest
rate, until the annual tariff review date (Port. 296/01)
E111301* Gov Agreement with distribution companies regarding the compensation of rationing losses
E112101* Aneel Resolution: details measures implemented by MP 2,227/01 and Port. 296/01
El 12301 Gov Agreement with generation companies regarding the compensation of rationing losses
E121101 Aneel Decision favoring distribution companies in their dispute against Eletrobras concerning the energy from
Itaipu
E121201 Gov Change in Aneel's board of directors
E121701 Gov General Sector Agreement
E010902 Gov Revitalization Plan for the Electricity Sector
E013102 Gov Announcement that energy supplied by state-owned generators would be sold in public auctions
E020102 Gov Details of the Revitalization Plan for the Electricity Sector
E021902 Gov President announces the end of rationing period
E041002 Congress General Sector Agreement is approved, incorporating a new criterion for low-income consumer.
E042502 Aneel New criteria makes it more attractive to invest in electricity transmission networks
E050202* Aneel Resolution: new criteria for low-income consumer











E060402 Gov

E062102* Aneel
E081602 Gov

E083002 Aneel


E090402*
E101502
E103002*
E110702

E011403*
E021703*
E030703*
E031103*
E040803
E041703*
E043003*
E052603*
E070303*
E072103
E080503*
E090803
E091603*
E092303
E092503*


Aneel
Aneel
Aneel
Aneel

Aneel
Aneel
Aneel
Aneel
Gov
Aneel
Aneel
Aneel
Aneel
Gov
Gov
Gov
Gov
Gov
Aneel


Revitalization Committee announces additional measures of the Revitalization Plan for the Electricity
Sector
Resolution proposal: methodology for asset base valuation
Decision to use resources from the RGR to cover revenue reductions brought by low-income consumers'
criteria (Decree 4,336/02)
Resolutions: detail Decree 4,336/02 measures; define normative values for contracts with thermo-electric
plants; allow distribution companies to contract repairs in the transmission network below 230KV with
transmission companies
Resolution: methodology for asset base valuation
Resolution: credits from pending operations in the wholesale power market must be adjusted for inflation
Resolution proposal: X-factor methodology
Resolution: reviews the previous definition that credits in the wholesale power market would be adjusted
for inflation
Adoption of the model firm approach in the distributor's periodic tariff review
Repositioning indexes proposed for the periodic tariff review of Cemig, Cemat, Enersul and CPFL
Repositioning indexes proposed for RGE and AES SUL
Repositioning indexes proposed for Coelce, Coser, Energipe and Coelba
Emergency Plan for the Short-Run
Final repositioning indexes for RGE, AES SUL, Coelce, Coser, Energipe, Coelba
Resolution: Electricity Universal Plan
Repositioning index proposed for Eletropaulo
Final repositioning index for Eletropaulo
General guidelines of the New Model for the Electricity Sector
Emergency Program to help the distribution companies (MP 127/03 general guidelines)
BNDES and AES settle an agreement
Emergency Program to help the distribution companies announcement of its conditions
Law project: revision in regulatory agencies' job
Repositioning index proposed for Light












Table 2-2. Descriptive Statistics
1998 1999 2000 2001 2002 2003
Mean Range Mean Range Mean Range Mean Range Mean Range Mean Range

Retlbov -0.196 [-15.8, 18.7] 0.421 [-9.9, 33.4] -0.024 [-6.4, 5.0] -0.025 [-9.1, 7.6] -0.053 [-6.5, 6.3] 0.200 [-3.8, 3.6]
3.960 3.123 2.074 2.140 2.070 1.576
RetCopel -0.196 [-15.3, 31.4] 0.382 [-16, 40.9] 0.017 [-6.3, 9.4] 0.092 [-10.3, 9.5] -0.148 [-8.4, 14] -0.008 [-9.7, 6.9]
5.735 4.171 2.858 3.222 3.200 2.970
RetEletr -0.492 [-31.9,29.1] 0.376 [-16.5, 12.1] -0.003 [-10.9, 11.4] 0.004 [-9.7, 11.4] -0.310 [-12.1,21] 0.271 [-8.5, 13.2]
5.822 3.890 3.168 3.299 4.598 4.046
RetEletb -0.293 [-14.1,25.1] 0.356 [-13.4, 38.5] -0.005 [-9.6, 19.8] 0.034 [-12.1, 11.6] -0.024 [-8.8, 11.5] 0.176 [-8.1, 7.7]
5.767 4.287 2.947 3.466 3.227 3.013
RetCemig -0.209 [-24.3, 28.4] 0.335 [-14.6, 30.2] -0.091 [-7.3, 9.6] 0.111 [-9.0, 8.8] -0.005 [-10.1, 10] 0.166 [-6.5, 7]
5.616 4.072 2.764 2.819 3.171 2.305
RetCesp -0.177 [-15.6, 17.6] 0.294 [-17, 11.1] 0.362 [-10.3,25.5] 0.041 [-20.1,20.3] -0.236 [-10.1, 12.1] 0.214 [-7.8, 12.7]
5.062 4.020 4.649 4.738 3.528 3.371
RetLight -0.418 [-23.1, 16.7] 0.197 [-18.1, 15.6] 0.099 [-12.5, 14.5] -0.169 [-18.0, 16.6] -0.281 [-10.1, 8.7] 0.058 [-11, 15.9]
4.836 3.845 3.212 3.874 3.193 4.117
RetCeles -0.296 [-18.9, 19.5] 0.267 [-14.3, 16.7] -0.108 [-8.1, 10.2] -0.033 [-7.6,20] 0.100 [-7.2, 10.4] 0.070 [-5, 10.4]
4 5.212 3.897 3.173 3.524 2.697 2.541
0 RetEmae -0.326 [-31.7,20.5] 0.739 [-12.6, 34.7] 0.185 [-12.7, 15.9] 0.294 [-10,20.5] -0.122 [-11.5, 16.5] 0.032 [-5.8, 13.4]
8.415 5.859 4.177 3.621 3.251 2.765
RetFcata -0.138 [-10, 11.1] 0.307 [-17.2, 20.8] 0.178 [-8.4, 20] 0.097 [-11.5, 22.9] -0.052 [-12.3, 13.6] 0.105 [-12.1, 9.3]
2.750 4.639 3.639 3.623 3.426 3.163
RetCoelc 0.115 [-15.8, 16.8] 0.811 [-11.5, 37.5] 0.201 [-10.4, 9.1] -0.104 [-7.1, 4.1] -0.076 [-6.8, 6.5] 0.316 [-5.2, 7.5]
4.698 4.383 2.639 1.669 2.050 2.060
RetTranp 0.714 [-15.5, 19.5] 0.201 [-13.4, 35.9] 0.079 [-13.4, 14.4] 0.066 [-10.3, 8] 0.373 [-9.5, 6.6]
5.720 5.119 3.845 3.105 2.388
RetCerj -0.083 [-14.3, 34.6] 0.261 [-33.3, 20] 0.228 [-19.1, 15.2] 0.710 [-14.6, 20] -0.202 [-13.4, 16] 0.218 [-37, 30.4]
5.346 5.050 4.210 5.480 4.026 7.144
RetTract -0.028 [-18.7, 17.9] 0.117 [-13.9,21.7] 0.235 [-7.5, 14.4] 0.209 [-7.7, 12.1] -0.012 [-12, 8.5] 0.455 [-11.6,23.9]
4.899 3.405 2.989 4.057 2.849 4.087
LiqCopel 0.272 [.006, 6.59] 0.233 [.001, 1.80] 0.201 [.009, 1.53] 0.269 [.004, 1.46] 0.370 [.066, 2.31] 0.656 [.053, 3.63]
0.609 0.213 0.163 0.182 0.256 0.555
LiqEletr 0.150 [.001, 1.37] 0.193 [.001,1.30] 0.110 [.008, .712] 0.157 [.004, .963] 0.207 [.014, .855] 0.329 [.075, 1.35]
0.185 0.184 0.096 0.142 0.160 0.198
LiqEletb 0.476 [.074, 1.77] 0.652 [.001, 2.12] 0.478 [.088, 2.05] 0.515 [.081, 2.48] 0.653 [.124, 3.61] 0.683 [.169, 2.21]
0.261 0.338 0.265 0.281 0.371 0.323
LiqCemig 0.462 [.053, 1.65] 0.372 [.001,2.01] 0.541 [.120, 4.27] 0.459 [.034, 1.65] 0.487 [.083, 1.79] 0.629 [.166, 2.14]
0.262 0.244 0.472 0.215 0.246 0.306











LiqCesp 0.591 [.037, 2.43] 0.572 [.002, 2.42] 0.354 [.030, 2.13] 0.335 [.045, 1.57] 0.210 [.017, 1.15] 0.314 [.035, 1.51]
0.424 0.487 0.289 0.203 0.139 0.279
LiqLight 0.201 [.008, 3.17] 0.239 [0, 1.74] 0.184 [.008, 3.14] 0.069 [.003, .404] 0.036 [.001, .515] 0.027 [.001, .221]
0.282 0.228 0.350 0.060 0.051 0.028
LiqCeles 0.643 [.008, 3.83] 0.893 [.090, 5.28] 0.736 [.078, 5.43] 0.534 [.052, 2.89] 0.612 [.032, 3.16] 0.786 [.071, 6.80]
0.610 0.703 0.634 0.408 0.509 0.855
LiqEmae 0.202 [.001, 2.33] 0.162 [.001, 1.83] 0.124 [.001, 1.63] 0.182 [.001, 2.73] 0.049 [.001, .702] 0.104 [.001, .836]
0.296 0.267 0.216 0.306 0.080 0.120
LiqFcata 0.058 [.001, 2.81] 0.026 [.001, .227] 0.123 [.001, 4.16] 0.017 [.001, .427] 0.012 [.001, .362] 0.004 [.001, .040]
0.225 0.039 0.418 0.037 0.035 0.007
LiqTranp 0.343 [.028, 1.88] 0.386 [.040,2.90] 0.433 [.063, 17.15] 0.161 [.008, .645] 0.153 [.019, .448]
0.302 0.405 1.118 0.102 0.091
LiqCerj 0.041 [.001, .668] 0.113 [.001, 4.57] 0.187 [.001, 3.42] 0.004 [.001, .060] 0.002 [.001, .030] 0.001 [.001, .013]
0.066 0.303 0.390 0.007 0.005 0.002
LiqTract 0.123 [.002, .740] 0.080 [.001, 1.65] 0.053 [.001, .542] 0.047 [.001, 2.83] 0.036 [.002, .239] 0.039 [.001, .488]
0.123 0.143 0.072 0.183 0.036 0.067
Dolar 0.033 [-.413, .254] 0.174 [-8.44, 11.10] 0.037 [-1.36, 1.81] 0.075 [-3.15, 3.44] 0.181 [-8.93, 4.87] -0.097 [-2.98, 2.93]
0.078 1.742 0.511 1.022 1.553 0.967
Security Returns' Mean Standard Deviation, by Year:
5.346 4.403 3.503 3.633 3.255 3.382
Standard deviations are informed below the mean values.











Table 2-3. Correlation Matrix
Retlbov RetCopel RetEletr RetEletb RetCemig RetCesp RetLight RetCeles RetEmae RetFcata RetTranp RetCerj RetTract
Retlbov 1.000 0.7484* 0.5621* 0.8192* 0.7755* 0.6090* 0.5143* 0.6006* 0.3797* 0.1480* 0.5363* 0.2697* 0.3987*
(.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000)

LiqCopel 0.0670* 0.0889*
(.013) (.001)

LiqEletr 0.1040* 0.1780*
(.000) (.000)

LiqEletb 0.1117* 0.1552*
(.000) (.000)

LiqCemig 0.0874* 0.0780*
(.001) (.004)

LiqCesp 0.0917* 0.1474*
(.001) (.000)

LiqLight 0.0278 0.0990*
(.304) (.000)

LiqCeles 0.1009* 0.2016*
O (.000) (.000)

LiqEmae 0.0677* 0.3490*
(.013) (.000)

LiqFcata -0.0425 0.0377
(.126) (.233)
LiqTranp 0.0147 0.1233*
(.638) (.000)

LiqCerj 0.0861* 0.0937*
(.004) (.003)

LiqTract 0.0211 0.1737*
(.444) (.000)

Dolar -0.0578* -0.0910* -0.1897* -0.0723* -0.1190* -0.1577* -0.1660* -0.0876* -0.1077* -0.0503 -0.0942* -0.0827* -0.0672*
(.032) (.001) (.000) (.007) (.000) (.000) (.000) (.001) (.000) (.075) (.002) (.009) (.015)
The "*" indicates significance at the 5% level. P-value in parenthesis.











Table 2-4. Securities Returns, Stock Market Returns and Exchange Rate Variation, by Year.
1998 1999 2000 2001 2002 2003 1998-2003 1999-2003

COPEL -50.29 109.56 -5.64 10.50 -39.12 -9.30 -40.02 20.66

ELETR -73.08 108.93 -12.37 -11.69 -64.46 42.73 -77.92 -17.98

ELETB -58.94 94.09 -11.06 -6.29 -17.22 27.73 -29.77 71.03

CEMIG -51.16 87.00 -27.40 19.32 -12.83 29.87 -10.43 83.39

CESP -45.06 -34.68 88.17 -16.20 -52.40 34.45 -63.78 -34.08

LIGHT -64.85 35.37 12.56 -45.31 -6.57 -4.57 -73.88 -25.70

CELES -56.85 59.76 -32.50 -20.47 17.39 7.41 -53.34 8.14

EMA4E -71.59 303.85 26.71 75.38 -32.11 -1.01 71.35 503.17

FCATA -24.23 43.04 31.40 8.41 -24.05 10.64 29.74 71.22

TRANP 81.82 21.26 1.37 4.49 90.46 344.79 344.79

CERJ -34.23 37.50 30.92 122.13 -22.96 -5.23 92.02 191.95

TRACT -19.17 16.03 60.77 37.11 -12.28 101.43 265.28 351.92

AVGRET -48.49 97.08 19.54 20.80 -18.33 32.24 58.32 207.37

IBOVESPA -41.52 151.93 -10.72 -11.02 -17.01 42.08 38.01 135.99

DOLAR 6.60 48.06 9.30 18.67 52.27 -17.26 157.91 141.94
"AVGRET" displays the returns of the equally-weighted portfolio composed by the securities included in the
sample.












Table 2-5. Individual Regressions Results
Year' Variable RetCopel RetEletr RetEletb RetCemig RetCesp RetLight RetCeles RetEmae RetCerj RetTract RetFcata RetTranp Avg Beta2


1999 Ibovespa 1.024*** 0.770*** 1.291*** 1.087*** 0.977*** 0.730*** 0.828*** 0.814*** 0.559***
Dolar 0.070 -0.453*** -0.036 -0.225** -0.632*** -0.425*** -0.207 -0.521** -0.454**
Liquid 0.124 3.974*** 0.601* -0.471 1.253*** 4.184*** 0.461* 8.264*** -0.155
R-sq 0.6695 0.4730 0.8003 0.7010 0.5246 0.4788 0.4261 0.2985 0.1529
N 411 411 411 411 411 411 411 411 411

2000 Ibovespa 0.854*** 0.786*** 1.096*** 0.919*** 1.015*** 0.659*** 0.958*** 0.684***
Dolar 0.027 -0.490*** -0.023 -0.200** -0.566*** -0.437*** -0.175* -0.547***
Liquid 1.781*** 3.148*** 0.805*** 0.194 0.853** 2.709*** 0.472*** 7.745***
R-sq 0.6127 0.3560 0.6783 0.5524 0.3671 0.3277 0.4553 0.2976
N 475 475 475 475 475 475 475 475

2001 Ibovespa 0.871*** 0.750*** 1.054*** 0.893*** 1.102*** 0.827*** 1.015*** 0.699***
Dolar -0.200 -0.451*** 0.134 0.036 -0.095 -0.156 0.180 -0.419**
Liquid 2.231*** 3.307*** 0.736* 0.425* 0.177 1.862*** 0.402** 4.415***
R-sq 0.4847 0.3416 0.5060 0.5312 0.3604 0.3967 0.4684 0.3133
N 468 468 468 468 468 468 468 468

2002 Ibovespa 0.997*** 1.026*** 1.189*** 1.026*** 1.164*** 0.952*** 0.720***
Dolar -0.250*** -0.398*** -0.060 -0.191*** -0.066 0.075 -0.005
Liquid 1.297*** 1.316 0.432 0.237 -0.004 -5.055** 1.379***
R-sq 0.5635 0.4621 0.6182 0.5654 0.4716 0.4342 0.3749
N 487 487 487 487 487 487 487

2003 Ibovespa 1.082*** 1.179*** 1.265*** 1.107*** 1.186*** 0.887*** 0.487***
Dolar -0.288*** -0.449*** -0.154** -0.224*** -0.081 0.040 -0.085
Liquid 0.204 2.848*** 0.505* 0.092 2.933*** 0.407 1.262***
R-sq 0.5920 0.4940 0.6733 0.6184 0.5587 0.3604 0.3544
N 430 430 430 430 430 430 430


0.898


0.489***
-0.042
8.385***
0.3193
475


0.829


0.661*** 0.335*** 1.156*** 0.903
0.226 -0.192 0.152
1.168 0.340 0.386**
0.2211 0.1054 0.3392
468 468 468


0.804*** 0.154** 1.036***
0.229** -0.228** 0.042
0.107 -2.231 -0.037
0.2666 0.0905 0.4705
487 487 487


0.990


0.605*** 0.134 0.917*** 0.968
0.095 -0.070 -0.011
6.122* -3.616 1.012
0.1748 0.0611 0.4577
430 430 430


Avg Beta 0.966


0.902 1.179


1.006 1.089 0.811


0.802 0.732 0.559 0.640 0.208


1.036 0.918


Legend: p < 0.10; ** p < 0.05; *** p < 0.01
1. For each Year t, the regressions include data from periods t and t-1.
2. The computed annual sample average betas do not incorporate Fcata's measures.












Table 2-6. Hypothesis Tests Results for Events in Year 1999
EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW
H1 H2 H3 H4 H1 H2 H3 H4 H1 H2 H3 H4
Z2 Z2 Coef. Coef. Z2 Z2 Coef. Coef. Z2 Z2 Coef. Coef.


E032399



E051099



E061099*



E070199


E072999



E080399*


E092199*



E092399*


E100199


E121499*



# Securities
# Observ.


4.02 1.75 0.22
(.910) (.941) (.855)


9.35 7.82 -1.94
(.405) (.251) (.114)

5.89 4.73 1.63
(.750) (.579) (.183)

5.72 4.27 -0.23
(.767) (.640) (.848)

4.30 3.68 0.38
(.890) (.719) (.757)


3.07 1.98 -0.72
(.961) (.921) (.555)

2.08 0.26 -0.75
(.990) (.999) (.539)


8.50 4.09 -0.85
(.485) (.663) (.487)

2.14 1.60 -0.14
(.989) (.952) (.908)

3.07 1.84 0.26
(.961) (.933) (.832)


Q Q


411 411


-0.46 3.80 1.60 -0.30
(.760) (.923) (.952) (.732)


-1.87 4.16 2.13 -1.18
(.213) (.900) (.907) (.176)

1.47 7.00 4.24 1.88**
(.326) (.636) (.643) (.030)

0.39 5.22 3.43 -0.31
(.792) (.815) (.753) (.721)

0.76 2.98 2.30 0.00
(.614) (.965) (.890) (.999)


-0.02 6.09 3.10 -0.39
(.989) (.730) (.796) (.655)

0.17 3.79 3.31 0.65
(.911) (.924) (.768) (.452)


-0.73 5.57 1.84 0.26
(.624) (.782) (.933) (.766)

0.27 1.63 0.85 0.10
(.857) (.996) (.990) (.906)

0.40 3.14 2.16 -0.60
(.788) (.958) (.904) (.485)


411 411 411 411 411 411 411 411 411 411


P-value in parenthesis. The "*" denotes a "distribution event".
Legend: p < 0.10; ** p < 0.05; *** p < 0.01.


-0.89 2.81 2.42 -0.83
(.384) (.971) (.877) (.237)


-0.81 5.81 3.51 -0.93
(.433) (.758) (.742) (.191)

1.66 7.28 4.67 1.32*
(.105) (.608) (.587) (.060)

-0.27 3.67 3.05 -0.30
(.790) (.931) (.802) (.667)

-0.45 6.05 5.11 -0.11
(.664) (.734) (.529) (.877)


0.24 3.97 1.89 -0.92
(.814) (.913) (.929) (.193)

1.51 3.49 2.36 0.43
(.142) (.941) (.884) (.537)


-0.37 5.22 1.73 0.27
(.721) (.814) (.942) (.756)

-0.06 4.71 4.16 0.12
(.953) (.859) (.655) (.867)

-0.30 2.99 1.91 -0.67
(.768) (.964) (.928) (.344)

Q Q Q Q


-0.92
(.272)


-0.41
(.627)

1.15
(.170)

-0.22
(.797)

-0.63
(.453)


-0.74
(.377)

0.96
(.252)


-0.43
(.675)

0.00
(.999)

-0.43
(.611)












Table 2-7. Hypothesis Tests Results for Events in Year 2000
EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW
H1 H2 H3 H4 H1 H2 H3 H4 H1 H2 H3 H4
Z2 Z2 Coef. Coef. Z2 Z2 Coef. Coef. Z2 Z2 Coef. Coef.
E012700* 23.66*** 20.35*** -0.08 1.31 11.87 9.03 -0.76 -0.05 6.46 4.37 -1.09* -1.13


(.005)


E021700 19.39**
(.022)


E041700 21.33**
(.011)

E042400 2.54
(.979)

E050300* 9.49
(.393)

E061500 13.41
(.145)


E072100 25.78***
(.002)


E080400 5.90
(.750)


E111000 9.22
(.417)

E112900 1.04
(.999)

E120700* 6.57
(.682)


# Securities
# Observ.


(.001)

7.66
(.176)

13.90**
(.016)

0.88
(.971)

5.50
(.358)

5.74
(.332)

9.41*
(.093)

2.38
(.794)

7.92
(.161)

0.15
(.999)

5.33
(.376)


9 9


(.938)

0.58
(.595)

-1.26
(.246)

0.11
(.921)

0.66
(.546)

2.38**
(.029)

1.98*
(.069)

-1.77
(.104)

0.66
(.545)

0.01
(.994)

1.31
(.228)

9


(.330)

1.37
(.309)

0.43
(.751)

-0.01
(.991)

0.28
(.835)

1.61
(.230)

0.94
(.484)

-1.80
(.181)

1.20
(.374)

-0.39
(.771)

1.12
(.406)

9


(.220)

18.26**
(.032)

13.86
(.127)

3.92
(.916)

8.80
(.456)

12.49
(.187)

26.77***
(.001)

8.11
(.523)

7.14
(.622)

1.32
(.998)

9.40
(.401)

9


(.107)

8.89
(.113)

10.30*
(.067)

2.65
(.753)

4.44
(.487)

7.58
(.180)

9.11
(.104)

3.20
(.669)

4.58
(.469)

0.61
(.987)

9.08
(.105)

9


(.322)

0.13
(.862)

-1.27*
(.099)

0.71
(.355)

0.59
(.585)

1.87**
(.015)

1.61**
(.037)

-1.54**
(.046)

1.15
(.136)

0.35
(.652)

-0.35
(.652)

9


(.959)

0.91
(.338)

-1.08
(.257)

0.46
(.629)

1.47
(.121)

1.37
(.147)

1.00
(.290)

-1.03
(.279)

1.53
(.107)

0.17
(.854)

-0.60
(.531)

9


(.693)

20.61**
(.014)

11.82
(.223)

3.86
(.920)

8.73
(.462)

16.51*
(.056)

34.26***
(.000)

2.98
(.965)

16.69*
(.053)

6.03
(.736)

11.75
(.227)

9


(.497)

11.57**
(.041)

9.34*
(.096)

2.68
(.748)

4.46
(.485)

5.13
(.400)

12.87**
(.024)

0.49
(.992)

14.96**
(.011)

5.27
(.384)

7.99
(.157)

9


(.083)

0.51
(.413)

-1.27**
(.044)

0.70
(.360)

0.54
(.618)

1.25**
(.046)

1.55**
(.014)

-0.16
(.796)

1.21*
(.053)

1.01
(.108)

-0.40
(.520)

9


(.130)

1.49**
(.048)

-1.34*
(.075)

-0.20
(.791)

0.77
(.304)

0.43
(.569)

1.07
(.155)

-0.06
(.936)

1.87**
(.012)

0.98
(.190)

-0.46
(.539)

9


P-value in parenthesis. The "*" denotes a "distribution event".
Legend: *p<0.10; **p<0.05; ***p<0.01.


475 475 475 475 475 475 475 475 475 475 475 475












Table 2-8. Hypothesis Tests Results for Events in Year 2001


EVENT 1-DAY


# Securities
# Observ


11 11


WINDOW 2-DAY WINDOW 3-DAY WINDOW
H3 H4 H1 H2 H3 H4 H1 H2 H3 H4
Coef. Coef. Z2 Z2 Coef. Coef. Z2 Z2 Coef. Coef.


H1 H2
X2 X2
E010901* 18.19* 13.97**
(.077) (.030)

E042001 5.59 4.17
(.899) (.653)

E050201 13.94 9.70
(.236) (.137)

E051801 16.90 11.52*
(.111) (.073)

E062101* 19.79** 12.92**
(.048) (.044)

E080701* 5.47 2.81
(.906) (.832)

E090601* 23.10** 13.77**
(.017) (.032)

E102501* 27.01*** 19.62***
(.004) (.003)

E111301* 5.06 1.92
(.928) (.927)

E112101* 16.16 9.23
(.135) (.161)

E112301 4.33 2.45
(.959) (.873)

E121101 6.87 3.38
(.809) (.759)

E121201 17.12 8.70
(.104) (.191)

E121701 8.99 7.48
(.623) (.278)


P-value in parenthesis. The "*" denotes a "distribution event".
Legend: p< 0.10; ** p< 0.05; *** p< 0.01.


0.12
(.918)

-0.25
(.837)


11.05
(.439)

16.51
(.123)


9.39 0.28
(.152) (.693)

9.89 -0.99
(.129) (.163)


0.59
(.565)

-0.22
(.832)

-0.41
(.686)

-0.94
(.360)

2.08**
(.044)

0.74
(.471)

0.32
(.757)

-0.20
(.846)

0.68
(.507)

1.06
(0.300)

0.61
(.548)

0.46
(.653)

-0.61
(.552)

0.05
(.961)
11


0.10
(.906)

-0.96
(.245)


5.52
(.903)

15.64
(.154)


4.81 1.13**
(.568) (.048)

10.50 -1.17**
(.105) (.042)


0.24 12.12 4.93 -0.19
(.775) (.354) (.552) (.740)

-1.41* 12.78 5.21 -0.86
(.086) (.308) (.516) (.133)

1.08 46.36*** 12.73** 1.53***
(.204) (.000) (.047) (.009)

2.01** 8.33 6.90 0.70
(.014) (.683) (.330) (.222)

0.56 27.34*** 13.71** 1.42**
(.499) (.004) (.033) (.014)

-0.31 25.73*** 17.71*** -0.54
(.708) (.007) (.007) (.345)

0.55 5.20 2.79 0.33
(.499) (.921) (.834) (.560)

1.38* 11.30 8.17 0.72
(.093) (.418) (.226) (.208)

1.15 16.61 7.40 0.38
(.162) (.120) (.285) (.586)

0.47 17.73* 4.03 0.80
(.683) (.088) (.673) (.259)

-1.26 16.13 11.33* -0.68
(.124) (.136) (.078) (.332)

0.34 10.98 5.99 -0.12
(.675) (.445) (.424) (.836)


1.20*
(.071)

-1.16*
(.084)

-0.40
(.545)

-0.71
(.287)

1.03
(.128)

1.16*
(.083)

1.59**
(.017)

-0.52
(.436)

0.07
(.912)

0.76
(.252)

1.18
(.150)

0.60
(.465)

-1.26
(.124)

0.31
(.640)
11


46X 46X 46X 46X 46X 46X 46X 46X 46X 46X 46X 46X


0.11 21.53** 10.00 -0.10
(.925) (.028) (.124) (.885)

-1.04 18.63* 10.20 -1.77**
(.377) (.068) (.116) (.012)

2.42** 30.05*** 9.62 1.59**
(.042) (.001) (.141) (.030)

1.29 11.76 11.03* 1.33*
(.274) (.381) (.087) (.060)

0.72 30.08*** 16.20** 0.53
(.543) (.001) (.012) (.451)

-0.14 33.58*** 28.72*** -0.18
(.908) (.000) (.000) (.804)

0.54 4.72 1.60 0.91
(.647) (.944) (.952) (.196)

0.84 14.14 9.72 1.28*
(.473) (.225) (.137) (.069)

0.71 15.67 7.20 0.36
(.545) (.153) (.302) (.610)

0.46 7.28 3.43 0.46
(.695) (.776) (.753) (.646)

-2.03* 16.24 11.69* -0.68
(.084) (.132) (.069) (.334)

0.45 14.28 8.64 -0.07
(.700) (.217) (.194) (.924)












Table 2-9. Hypothesis Tests Results for Events in Year 2002


EVENT 1-DAY
H1 H2


WINDOW 2-DAY WINDOW 3-DAY WINDOW
H3 H4 H1 H2 H3 H4 H1 H2 H3 H4
Coef. Coef. Z2 Z2 Coef. Coef. Z2 Z2 Coef. Coef.


Z2 72
E010902 9.29 7.45
(.504) (.281)
E013102 33.22*** 19.27***
(.000) (.004)
E020102 6.53 4.53
(.769) (.604)
E021902 5.44 3.06
(.859) (.801)
E041002 6.48 6.14
(.773) (.407)
E042502 10.95 9.70
(.361) (.138)
E050202* 16.45* 14.14**
(.087) (.028)
E060402 5.00 2.62
(.891) (.854)
E062102* 13.43 12.16*
(.200) (.058)
E081602 22.16** 10.38
(.014) (.109)
E083002 30.91*** 21.10***
(.000) (.002)
E090402* 4.56 3.86
(.918) (.695)
E101502 9.95 7.11
(.444) (.310)
E103002* 5.12 3.35
(.883) (.763)
E110702 10.37 5.01
(.408) (.542)
# Securities 10 10
# Observ 487 487


P-value in parenthesis. The "*" denotes a "distribution event".
Legend: p< 0.10; ** p< 0.05; *** p< 0.01.


0.33
(.744)
1.97*
(.056)
0.82
(.422)
-0.64
(.528)
1.65
(.107)
1.07
(.296)
-2.05**
(.046)
0.59
(.561)
0.78
(.447)
0.82
(.426)
-1.02
(.323)
-1.27
(.213)
-0.90
(.380)
-1.18
(.252)
-1.65
(.106)
10


0.96
(.406)
0.97
(.402)
0.82
(.478)
-0.71
(.537)
1.83
(.114)
1.66
(.152)
-2.34**
(.043)
1.55
(.178)
0.88
(.451)
1.35
(.241)
-1.51
(.194)
-1.52
(.188)
-0.95
(.408)
-1.42
(.220)
-1.07
(.354)
10


27.77***
(.002)
30.18***
(.001)
8.25
(.604)
10.87
(.367)
6.05
(.810)
11.35
(.331)
7.72
(.656)
2.42
(.992)
5.52
(.853)
16.67*
(.082)
10.78
(.374)
8.23
(.606)
10.32
(.412)
10.93
(.362)
6.46
(.775)
10


17.00***
(.009)
18.22***
(.006)
6.16
(.405)
9.40
(.152)
2.95
(.815)
10.09
(.120)
4.63
(.592)
0.89
(.989)
4.15
(.656)
7.01
(.319)
9.06
(.170)
6.65
(.354)
8.74
(.188)
8.42
(.208)
3.17
(.786)
10


-0.92
(.202)
2.03**
(.046)
0.73
(.303)
-0.08
(.915)
0.90
(.206)
1.05
(.298)
-1.05
(.141)
-0.28
(.690)
-0.68
(.340)
0.63
(.380)
-0.71
(.320)
-0.65
(.359)
-1.46**
(0.040)
-1.05
(.144)
-1.03
(.148)
10


-0.15
(.851)
0.95
(.406)
0.83
(.302)
-0.35
(.666)
0.74
(.361)
1.69
(.140)
-1.04
(.197)
-0.21
(.798)
-0.58
(.477)
1.39*
(.084)
-1.14
(.157)
-0.90
(.263)
-1.75**
(.030)
-1.23
(.130)
-0.68
(.396)
10


18.66**
(.045)
28.67***
(.001)
8.01
(.627)
12.43
(.257)
2.37
(.992)
9.08
(.524)
6.18
(.800)
1.44
(.999)
7.45
(.682)
13.35
(.204)
4.00
(.947)
9.21
(.512)
6.34
(.786)
7.80
(.648)
6.44
(.776)
10


11.59*
(.071)
23.41***
(.001)
6.07
(.415)
10.64
(.100)
1.99
(.920)
5.54
(.476)
4.75
(.576)
0.85
(.990)
6.52
(.367)
4.47
(.612)
2.20
(.900)
6.19
(.402)
4.37
(.626)
6.98
(.323)
3.01
(.807)
10


-0.49
(.396)
2.14**
(.003)
0.73
(.298)
-0.03
(.962)
0.41
(.474)
0.97
(.169)
-0.78
(.179)
-0.35
(.543)
-0.21
(.714)
0.01
(.981)
0.12
(.841)
-0.94
(.103)
-1.20**
(.038)
-0.50
(.393)
-1.04
(.142)
10


487 487 487 487 487 487 487 487 487 487


0.12
(.852)
1.89**
(.019)
0.83
(.303)
-0.24
(.717)
0.39
(.549)
1.20
(.134)
-0.75
(.257)
-0.34
(.607)
-0.08
(.905)
0.77
(.243)
0.19
(.775)
-1.12*
(.088)
-1.30**
(.048)
-0.43
(.516)
-0.66
(.411)
10












Table 2-10. Hypothesis Tests Results for Events in Year 2003
EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW


P-value in parenthesis. The "*" denotes a "distribution event".
Legend: p < 0.10; ** p < 0.05; *** p < 0.01.


H2 H3 H4 H1 H2 H3 H4


E011403*


E021703*


E030703*


E031103*


E040803 2


E041703*


E043003* 1


E052603*


E070303*


E072103


E080503*


E090803


E091603* 2(


E092303


E092503*

# Securities
# Observ.


HI H2 H3 H4 HI
Z2 2 Coef. Coef. Z2
9.93 6.28 -0.04 -0.36 6.63
.446) (.392) (.968) (.744) (.760)
7.85 6.00 -0.18 -0.05 7.37
.643) (.423) (.851) (.962) (.690)
11.08 7.21 1.40 2.34** 9.51
.351) (.301) (.141) (.036) (.484)
9.00 3.97 -0.54 -1.23 11.56
.532) (.681) (.565) (.271) (.315)
2.65** 17.46*** -1.81* -3.54*** 46.45***
.012) (.008) (.059) (.002) (.000)
14.71 12.27* 0.78 1.09 14.76
.143) (.056) (.413) (.330) (.141)
8.25* 6.67 1.31 1.61 11.86
.051) (.352) (.169) (.151) (.294)
7.73 6.08 -1.40 -2.17* 9.39
.654) (.414) (.140) (.053) (.495)
8.28 5.80 -0.09 -0.44 13.04
.601) (.445) (.925) (.697) (.221)
2.43 1.37 -0.22 -0.49 4.89
.991) (.967) (.815) (.664) (.898)
7.71 2.88 -0.10 0.81 7.46
.657) (.824) (.919) (.469) (.681)
11.92 9.44 0.16 0.64 47.52***
.290) (.150) (.866) (.567) (.000)
0.55** 7.23 1.22 1.56 11.99
.024) (.300) (.199) (.165) (.285)
9.90 9.32 -0.33 -0.33 20.44**
.449) (.156) (.731) (.772) (.025)
13.60 7.69 -0.47 -1.73 8.03
.192) (.261) (.619) (.123) (.626)
10 10 10 10 10
430 430 430 430 430


X2

5.14
(.525)
3.54
(.738)
4.39
(.623)
6.64
(.355)
25.85***
(.000)
3.03
(.804)
8.10
(.231)
8.82
(.183)
2.71
(.843)
2.20
(.900)
3.58
(.732)
44.84***
(.000)
4.11
(.661)
18.74***
(.004)
3.53
(.740)
10
430


Coef.
-0.23
(.738)
-0.64
(.344)
-0.24
(.723)
-0.12
(.854)
-1.13*
(.098)
0.00
(.998)
0.32
(.635)
-0.50
(.464)
0.68
(.315)
0.07
(.922)
-0.94
(.166)
0.94
(.167)
0.57
(.399)
-0.40
(.557)
0.07
(.922)
10
430


Coef.
-0.53
(.515)
-0.39
(.626)
0.64
(.426)
-1.03
(.201)
-2.65***
(.001)
-0.22
(.782)
0.39
(.631)
-0.58
(.474)
-0.03
(.974)
-0.10
(.903)
-0.68
(.397)
1.97**
(.015)
0.60
(.458)
-0.42
(.600)
-0.86
(.285)
10
430


72
1.39
(.999)
5.42
(.861)
3.90
(.951)
10.79
(.373)
44.90***
(.000)
23.74***
(.008)
10.91
(.364)
6.81
(.743)
13.05
(.220)
3.24
(.975)
20.69**
(.023)
39.44***
(.000)
18.72**
(.044)
14.07
(.170)
8.13
(.616)
10
430


X2

1.19
(.977)
2.01
(.918)
2.67
(.849)
6.09
(.413)
24.81***
(.000)
6.19
(.402)
6.64
(.355)
6.54
(.365)
1.87
(.931)
1.35
(.968)
4.08
(.666)
36.50***
(.000)
5.89
(.435)
11.58*
(.072)
3.58
(.733)
10
430


Coef.
-0.19
(.726)
-0.37
(.501)
0.05
(.929)
-0.09
(.895)
-1.18*
(.079)
-0.24
(.665)
-0.03
(.956)
-0.24
(.659)
0.98*
(.073)
-0.29
(.602)
0.12
(.823)
1.08*
(.051)
0.75
(.176)
-0.62
(.258)
0.07
(.922)
10
430


Coef.
-0.27
(.681)
-0.37
(.576)
0.49
(.458)
-0.98
(.221)
-2.66***
(.001)
-0.62
(.341)
-0.33
(.614)
-0.54
(.413)
0.40
(.539)
-0.35
(.597)
-0.44
(.497)
2.00***
(.003)
0.43
(.511)
-0.44
(.505)
-0.83
(.297)
10
430













Table 2-11. Hypothesis Tests Results for CARs Before Significant Events
EVENT Hi H, H, Ha
2 2 Coef. Coef.
E012700 5.29 3.84 -0.15 -0.61
(808) (573) (951) (830)
E021700 27.03*** 12.69** -1.62 -1.57
(001) (026) (499) (581)
E041700 4.82 1.71 1.39 1.05
(777) (887) (588) (724)
E072100 3.33 1.56 -0.42 -2.05
(950) (906) (862) (466)
E111000 6.13 5.45 3.01 3.66
(727) (364) (207) (195)
E010901 21.69** 18.44*** 2.41 4.25*
(027) (005) (265) (080)
E050201 15.89 10.79* -3.54 -6.03**
(145) (095) (103) (013)
E051801 12.23 7.31 -1.12 -2.50
(346) (293) (604) (302)
E062101 2.79 2.30 1.63 1.54
(993) (889) (452) (524)
E080701 4.67 1.74 1.24 0.94
(946) (942) (567) (696)
E090601 7.99 6.56 3.50 3.55
(714) (363) (104) (142)
E102501* 32.46*** 26.88*** 4.95** 6.05**
(001) (001) (022) (013)
E010902 4.56 1.60 -0.82 -0.36
(918) (952) (709) (883)
E013102 17.24* 6.93 -0.38 -3.24
(069) (327) (863) (190)
E050202 3.22 1.12 0.27 1.53
(975) (980) (904) (537)
E062102 7.86 7.61 4.45** 5.29**
(642) (268) (044) (034)
E081602 22.71** 21.55*** -6.37*** -8.82***
(012) (002) (004) (000)
E083002 8.62 1.86 -0.89 0.65
(568) (932) (686) (792)
E040803 33.71*** 18.53*** 2.74 5.65**
(001) (005) (196) (025)
E041703* 38.71*** 4.22 2.01 0.74
(000) (647) (322) (757)
E090803 14.88 12.51* 1.81 4.43*
(136) (052) (375) (068)
E091603* 46.22*** 29.26*** 5.08*** 3.59
(000) (001) (007) (113)
E092303 25.99*** 18.60*** -4.37** -3.71*
(004) (005) (016) (086)
P-value in parenthesis. The "*" denotes a "distribution event".
Legend: p<0.10; ** p <0.05; *** p <0.01.


Table 2-12. Hypothesis Tests' Results for Events' Overall Effect
EVENT Hi H2 H3 H4
72 ,2 Coef. Coef.
E021700 11.48 6.47 -0.69 2.69
(244) (263) (797) (399)
E102501* 29.85*** 24.19*** 4.52* 5.55*
(002) (001) (067) (065)
E081602 17.71* 16.39** -5.60** -7.52***
(060) (012) (026) (008)
E040803 11.98 6.82 0.48 2.14
(286) (338) (847) (472)
E091603* 42.16*** 25.80*** 6.09*** 4.54*
(000) (000) (005) (081)
E092303 16.09* 11.14* -5.25** -4.94**
(097) (084) (012) (049)
P-value in parenthesis. The "*" denotes a "distribution event". The event overall effect was computed only for significant events
whose hypothesis tests rejected the null that the CARs in the 5-day period before the announcement was not different from zero (in
case of E010901, the analysis showed that the observed abnormal returns could not be attributed to the announcement examined).
Legend: p <0.10; ** p <0.05; *** p <0.01.













Table 2-13. Significant Announcements' Categorization. Direction and Estimated Magnitude of Regulatory Announcements' Effect
on Security Returns

Positive Impact Negative Impact Differential Impact Across Firms in the Sample

Type ofAnnouncement Event Initiative Estimated Type ofAnnouncement Event Initiative Estimated Type ofAnnouncement Event Initiative


Abnormal
Returns


Abnormal
Returns


Initiative to implement the
model for the electricity sector
rulng on agents'
participation in the market)

Initiative to implement the
model for the electricity sector
(definition of the basic
transmission network)

Rate decision (initial proposal)


Rate decision 'decision)


E072100





E111000





E062101


Aneel





Aneel





Aneel


E080701 Aneel


1.98 Rationing Plan





1.21 Initiative to implement a
Congress' decision (definition
of low-income customer's
concept)

2.08 Compensatory measure


2.01 Revision in
agencies 'ob


E051801 Gov -1.77 Initiative to implement the
model for the electricity
sector (proposed rulng on
agents' participation in the
market)
E050202 Aneel -2.34 Change in ANEEL's board of
directors



E081602 Gov -5.60 Within sector redistribution
policy


regulatory E092303


-5.25 Initiative to implement the
model for the electricity
sector (proposed rulng on
asset base valuation
methodology)


Compensatory Measure


Compensatory Measure

Partial revision of a within
sector redistribution policy
previously adopted

Compensatory Measure


E090601 Gov 1.42


E102501 Gov 4.52


E013102


Gov


2.14


E090803 Gov 1.08


Compensatory Measure

Rate decision .


E091603 Gov 6.09

S E041703 Aneel 1.09


E041700





E050201





E010902


E062102


Aneel





Gov





Gov


Aneel


Several rulngs issued in the E083002 Aneel
same day


Includes only the events for which the null hypotheses H1 or H2 were rejected. Events E012700, E010901, E021700, and E040803 were dropped, however, because a deeper
analysis revealed that either the abnormal returns could not be attributed to the regulatory announcements investigated or the impact observed in the announcement day was
actually a review in market expectations concerning the event's effect.












Table 2-14. Random Events' Results
1999 2000 2001 2002 2003
Event H1 H2 Event H1 H2 Event H1 H2 Event H1 H2 Event H1 H2
2 2 2 2 2 2 2 2 2 2


E041499 12.87
(.168)
E042399 16.38*
(.059)
E051899 10.94
(.280)
E060199 4.99
(.835)
E090199 21.03**
(.013)
E092099 3.85
(.921)
E110399 4.49
(.876)
E110899 3.63
(.934)
E120999 7.76
o\ (.559)
E121099 8.57
(.478)


10.82 E012400 8.54
(.094) (.481)
14.22 E021000 13.74
(.027) (.132)
7.71 E022500 4.14
(.260) (.902)
2.95 E032000 10.48
(.815) (.313)
20.27*** E032300 6.49
(.003) (.690)
1.44 E041400 11.08
(.963) (.270)
1.76 E041900 5.03
(.940) (.754)
2.99 E042000 22.02***
(.811) (.005)
7.51 E052200 2.38
(.276) (.983)
4.21 E062600 5.57
(.649) (.781)
E072000 13.21
(.153)
E090600 5.55
(.784)
E091300 3.09
(.960)
E092100 4.89
(.844)
E101000 4.38
(.884)
E102300 5.61
(.778)
E103100 5.28
(.809)
E111600 6.86
(.651)
E120100 10.14
(.339)


4.96 E011001 14.29
(.420) (.217)
1.69 E012901 5.48
(.890) (.905)
3.09 E021501 6.11
(.686) (.865)
8.55 E042701 13.71
(.128) (.249)
5.74 E070201 17.14
(.332) (.103)
2.51 E070501 26.33***
(.775) (.006)
4.36 E080601 6.51
(.499) (.837)
19.26*** E083101 9.99
(.002) (.531)
1.31 E100401 11.57
(.933) (.396)
3.74 E102401 13.89
(.588) (.238)
5.32 E102901 18.78
(.377) (.065)
1.03 E103001 8.05
(.959) (.708)
2.58 E112201 6.03
(.764) (.871)
4.47 E112601 21.79**
(.484) (.026)
1.08 E113001 18.4*
(.955) (.072)
4.84
(.435)
3.55
(.615)
4.91
(.427)
6.49
(.261)


8.62 E012302 6.40
(.196) (.780)
3.27 E030102 5.95
(.774) (.745)
2.97 E032702 1.78
(.812) (.997)
8.80 E040102 18.33**
(.185) (.049)
12.46* E042302 5.32
(.052) (.868)
20.00*** E042902 8.35
(.003) (.594)
3.62 E050902 11.92
(.728) (.290)
6.24 E060502 12.16
(.397) (.274)
3.83 E070202 3.70
(.700) (.960)
5.94 E072502 11.32
(.429) (.333)
8.43 E080702 7.40
(.208) (.686)
7.09 E080802 27.70***
(.312) (.002)
5.14 E093002 19.67**
(.525) (.032)
6.32 E101102 2.84
(.388) (.984)
5.54 E101802 11.29
(.476) (.335)
E102202 13.65
(.189)
E111102 4.50
(.875)
E111402 5.61
(.847)


P-value in parenthesis. Legend: p <0.10; ** p <0.05; *** p <0.01.


4.35 E021003 9.54 6.19
(.629) (.481) (.402)
1.79 E030503 8.89 4.21
(.937) (.542) (.648)
1.00 E031303 19.24** 1.70
(.985) (.037) (.944)
16.79*** E032503 6.76 4.80
(.010) (.747) (.569)
3.64 E032703 6.48 6.16
(.725) (.773) (.405)
4.96 E052803 4.08 3.63
(.548) (.943) (.727)
9.33 E070403 13.15 3.62
(.155) (.215) (.727)
6.46
(.374)
1.03
(.984)
5.45
(.487)
4.74
(.577)
19.81***
(.003)
15.18**
(.019)
1.57
(.954)
4.42
(.619)
11.33*
(.078)
3.96
(.682)
1.86
(.931)










CHAPTER 3
PRIVATIZATION, INCENTIVE REGULATION, AND EFFICIENCY IMPROVEMENTS IN
THE BRAZILIAN ELECTRICITY DISTRIBUTION INDUSTRY

Introduction

Network industries have experienced a remarkable change in the last twenty years. Once

characterized by state-owned, vertically-integrated companies, electricity power,

telecommunications, natural gas, water & sewerage, railroads, and ports industries from several

countries have gone through a reform process which encompasses unbundling, privatization,

introduction of market-oriented regimes for their competitive segments, and implementation of a

new regulatory framework for the remaining segments with natural monopoly characteristics.

The new regulatory framework has involved the creation of independent regulatory bodies

and the incorporation of theoretical advances from the economics literature on incentive

regulation. Here, the regulator is seen as a social welfare maximizer operating in a context of

imperfect and asymmetric information regarding firms' demand, cost opportunities, and

managerial effort. The regulator seeks to limit the regulated firms' rents and to allocate some of

these rents to consumers, subject to a firm break-even constraint.1 In this context, some incentive

mechanisms emerge as an alternative to the customarily employed cost-of-service or rate of

return regulation. Under a price-cap regime2, the most popular regulation, prices are fixed. The

firm and its managers are the residual claimants on production cost reductions, and bear the







1 For a comprehensive review of the theoretical literature on incentive regulation, see Laffont and Tirole (1993), and
Armstrong and Sappington (2003).
2 Sappington (2002) provides details on the design and implementation of price-cap regulation, as well as of other
forms of incentive regulation.










disutility of increased managerial effort (Joskow, 2005). Thus, the conditions and incentives for

efficiency improvement3 and for the possible achievement of second best prices are settled.

Whether price-cap regulation effectively leads to efficiency improvements, however,

constitutes an empirical question, given the different ways in which it is implemented in

practice.4 Benchmarking, or comparative efficiency analysis, is a technique used to address the

issue of relative performance since it enables the computation of efficiency scores and the

analysis of their evolution over time. More than that, by providing information on each firm's

inherent cost opportunities, the benchmark exercise helps alleviate the potential adverse selection

problem faced by the regulator, consequently allowing the establishment of cost-effective prices

at the scheduled tariff review. From the researchers' perspective, though, it allows an ex-post

evaluation of the new prices the regulator has set.

The present study uses a benchmarking methodology to assess the impact of privatization

and incentive regulation on firms' performance. The related literature on the electricity

distribution sector is limited, possibly due to a reduced number of observations and data

availability constraints.5 Nonetheless, for the empirical studies that have addressed the topic, the

general finding is that privatization has been associated with improvements in efficiency,





3 The theory states that price-cap regulation provides incentives to improvements in performance in other
dimensions as well, such as innovation, efficient choice of operating technology, and even service quality. In this
paper, however, I focus on the efficiency improvement possible impact only.
4 The length of time between schedule reviews and the degree of association of prices to realized costs, for example,
may mitigate the efficiency improvements incentives brought by price-cap regulation. When a price-cap plan links
future prices directly to realized costs and the time between schedule reviews is relatively short, the incentives under
a price-cap regime are similar to the ones under rate of return regulation (Sappington, 2002).

5 Given the natural monopoly characteristic of electricity distribution, there is a small number of firms in most
countries which have undertaken sector reforms. In addition, there does not exist, up to this moment, a widespread
understanding among regulators of the need to collect and keep the longitudinal data necessary to perform studies of
this nature. This seems to occur even in Great Britain (see footnote 17).









although only when accompanied by incentive regulation mechanisms (Estache, Perelman, and

Trujillo, 2005, p. 8).6

Mota (2004), in contrast, compares the performance of 14 Brazilian privatized electricity

distribution companies to the average performance of 72 U.S. investor-owned electric utilities,

using data from 1994 and 2000, and finds that privatization had no statistically significant impact

on efficiency when operating costs are used as an input, but resulted in a strong drop in

efficiency for the models that used total costs. Mota's study finds that the Brazilian distribution

companies experienced annual average productivity gains of around 5% during the period.

The present study of 52 Brazilian electricity distribution companies evaluates the

efficiency evolution and the productivity gains that occurred in the sector from 1998 to 2003,

checks for difference in performance between public and private firms, and examines the

possibility of efficiency catch-up. It also investigates whether vertically integrated firms might

be behaving strategically, shifting costs from unregulated to regulated activities, and whether

efficiency changes are associated with variations in service quality.

The investigation provides evidence of performance improvement after the implementation

of sector reforms, and finds that both privatized and public companies have reduced the

efficiency gap with respect to companies that were privately owned before the reforms. The

results show that privatized firms responded more aggressively than public firms to the new

incentives brought by price-cap regulation. The findings also indicate a possible strategic

behavior associated with the periodic aspect of price cap regulation, as well as to cost shifting


6Berg, Lin and Tsaplin (2005) also find that privatized firms respond differently to incentives than public firms.
Their empirical analysis of 24 Ukraine electricity distribution companies from 1998 to 2002 indicates that privately-
owned firms not only respond to incentives that add to net cash flows, but also respond more aggressively than do
state-owned distribution utilities to cost-plus regulatory incentives that increase profits but decrease efficiency. The
authors point out that comparisons of public and private utility performance need to be explicit about the incentive
regimes facing both ownership types.









implemented by companies that operate in the electricity generation segment. Moreover, the

paper suggests that the high performance improvement experienced by privatized firms in the

period comes essentially from a more efficient operation of their units, in line with what was

expected under an incentive regulation scheme, and not from mere reductions in costs brought by

deterioration in the quality of service. Some of the results' implications for policy are

highlighted, and it is underlined the possible use of the paper's findings to evaluate the

regulator's decisions taken and methodology employed at the periodic tariff review.

The following section briefly describes the reforms undertaken in the Brazilian electricity

sector. Section 3 discusses the technology of the electricity distribution industry, presents an

overview of the different benchmarking techniques, provides a detailed picture of the stochastic

cost frontier approach, and shows how exogenous factors that influence producer's performance

may be incorporated into the analysis. The model specification and the data set are described in

Section 4. Section 5 presents and interprets the results obtained, while Section 6 explores the

possibility of strategic behavior and the relationship between observed efficiency changes and

variations in service quality. The final section provides concluding observations and directions

for future research.

Institutional Background

The power sector reforms in Brazil began in 1995. While constitutional amendments

abolished the public monopoly over infrastructure industries and allowed foreign companies to

bid for public concessions, the Law 8,987/95 (General Law of Concessions) set the stage for the










beginning of the privatization process, represented by the auctions of Escelsa in 1995 and Light

in 1996. By the end of 2000, a total of 20 distribution companies had been privatized.7


In addition, part of the implementation of a new regulatory framework involved the

establishment of an independent regulatory agency (ANEEL) in late 1996 and, in the same year,

the commission of an international consultancy to study and propose a new model for the

electricity sector. The consultant's report was released in 1997 and its proposals were

incorporated into Law 9,648, issued on May of 1998.8 In essence, the approved model focused

on privatization and unbundling of generation, transmission and distribution assets, gradual

transition to a competitive generation environment in nine years, creation of a wholesale power

market, operation of the transmission network by an independent operator, and use of the price-

cap regime to regulate distribution tariffs, replacing the previous cost of service system.9 This

paper focuses on the distribution segment only. For the corresponding concessionaries, therefore,

the sector reform changes which might have affected performance were privatization and the

implementation of incentive regulation. 10



7 Privatization concentrated on years 1997 and 1998, when nine and five firms where privatized, respectively. Only
three firms had been privatized up to the end of 1996.

8 See Ferreira (2000), Mota (2003), and de Oliveira (2003), for detailed descriptions of the new model's
characteristics.

9 With exception to companies Escelsa and Light, price cap regulation was implemented through the signature of
new concession contracts, which took place from 1998 to 2000, and had their first tariff review scheduled for after
five (for contracts signed in 1998) or four years. Light was the first to have price-cap regulation applied, by order of
the concession contract signed in November/1996, in which the first tariff review was scheduled to occur after seven
years. Escelsa was submitted to price-cap regulation in August of 1998, and had tariff reviews every three years
thereafter. Except for Escelsa, all companies had the X factor set equal to zero in the first period prior to the first full
review.

10 Competition in electricity supply to high-voltage customers began only in 2004. Therefore, its possible effects on
distribution firms' performance are not captured by the present study. On the other hand, distribution companies
were affected by the unforeseen electricity crisis in 2001, caused by severe drought conditions and under
investments in generation and transmission. The subsequent rationing measures proposed by the Government
reduced significantly electricity consumption and firms' revenues in that year, and need to be controlled for in
the analysis.









Methodology


The Electricity Distribution Technology

The modeling of electricity distribution technology is not straightforward. Many factors

influence electricity distribution costs and, with respect to some of them, a controversy exists

over their exogeneity.

Neuberg (1977) provided the theoretical foundations for the four factors that have been

considered the main distribution costs' drivers. The higher the amount of kilowatt-hours sold, the

greater the wear and tear on transformers, while increases in the number of customers induce

higher meter-reading and billing expenses. The probability of a wire-outage is assumed as a

monotonically increasing function of network length, and geographically dispersed areas,

encompassing several cities, imply higher repair costs because of the greater labor input required

by increased repair labor travel time, along with higher meter-reading and billing expenses.

Roberts (1986), on the other hand, emphasized the role of demand density (demand per

unit of area), as a factor affecting scale economies and, consequently, the average cost of

delivering a unit of power.11 After noting that demand density can change if either the demands

of existing customers change or new customers move into the service area, with the latter

requiring customer-specific investments for the delivery of the product, the author argued for the

inclusion of both factors (output density and customer density) in the model specification.

Roberts' work also pointed out the multiproduct nature of the electricity distribution activity,

with the consequent importance of treating low-voltage and high-voltage deliveries of power as





1 The study finds that average distribution cost falls as output per customer increases, a result consistent with
previous empirical studies' findings. According to Roberts, output density (output per customer) is useful for
explaining not only differences in efficiency across companies but also differences within firms across time.









separate products, as well as the need to consider the percentage of more costly underground

facilities on firm's total distribution equipment.

Some other distribution costs determinants are noted by Bums and Weyman-Jones (1996).

Maximum demand determines the overall capacity of the system and at individual nodes, and

together with energy delivered affects system load factor (reflecting the extent of peak use in the

system). In addition, transformer capacity affects network losses. The type of customer,

measured by the share of industrial KWh delivered, for example, determines the extent to which

power lines operate at different capacities at different times, or the effect of delivering energy at

different voltages. This same effect is sometimes captured by the load factor, used by Filippini

and Wild (2001) in place of maximum demand, and defined as the ratio of average over

maximum demand. The higher the load factor, the lower the average distribution costs, given the

smaller fluctuations of electricity demand over time.

The way these factors should be incorporated in the analysis depends, initially, on the

objective for the environment under study-output maximization or cost minimization. An

important characteristic of the electricity distribution industry is that, in general, the

concessionaries are required to provide service at specified tariffs. As a consequence, output is

demand driven, with firms maximizing their profits by minimizing the cost of producing a given

level of output. In this context, a cost function is the appropriate approach to deriving

performance comparisons. This approach also has the advantages of being able to accommodate

the multiproduct nature of the electricity distribution activity, and to treat variable and quasi-

fixed inputs differently. The knowledge that some inputs are not variable during the period under

study can be exploited by replacing a cost function with a variable cost function (Khumbakar and

Lovell, 2000).









However, in a cost function approach, a measure of cost is modeled as a function of

output(s) and input prices, leaving the question as to how all the above-mentioned factors that

affect distribution costs can be incorporated into the function. Neuberg (1977), taking the

separate marketability of components as a necessary property of a vector of outputs, argues that

only number of customers and electricity delivered might be considered as outputs. Nevertheless,

the other factors could be included in a cost function if they were taken as exogenous factors

reflecting differences in distribution system from firm to firm (environmental variables).

The exogenous characteristics of some of the cost determinants listed above are subject to

debate. Forsund and Kittelsen (1998) question the exogeneity of energy delivered and number of

customers in a context where firms can decide upon their prices, which fortunately is not the rule

on this regulated industry. The major concern involves network length and transformer capacity,

which Fillipini and Wild (2001), among others, do not use as explanatory variables, arguing that

they are capital inputs endogenous to the firm. These same factors are sometimes considered as

fixed capital inputs in the short run, and consequently included as explanatory variables in a

variable cost specification, such as those used by Salvanes and Tjotta (1994), Burns and

Weyman-Jones (1996) and Botasso and Conti (2003).

Moreover, Neuberg (1977) had already raised the point that the network length's

exogeneity might come from taking it as a proxy for a linear measure of the territory, assuming a

fixed geographic distribution of customers, whose importance is emphasized by Kumbhakar and

Hjalmarsson (1998) in the context of sparsely populated countries, where, according to the

authors, the amount of capital in the form of network reflects the geographical dispersion of

customers rather than differences in productive efficiency.









An additional insight is provided by the empirical literature on comparative efficiency

analysis. Jamasb and Pollitt (2001) survey this literature and report the frequency with which

different input and output variables are used to model electricity distribution. Mota (2004)

performs a similar task, but concentrates only on academic studies. Both surveys find that the

most frequently used inputs are network length, transformer capacity and number of employees,

and the most widely used outputs are units of energy delivered (with a high proportion of the

studies decomposing it into high-voltage and low-voltage sales), number of customers and

service area. Mota (2004) also reports that load factor, customer density, and output density are

used as environmental variables in the studies surveyed.

Comparative Efficiency Studies

The increased emphasis on efficiency analysis has its origin in the implementation of

incentive regulation. The adoption of incentive mechanisms aimed at improving the performance

of companies formerly subject to rate-of-return regulation brought the need to measure the

expected efficiency gains at the firm level, to be reflected in the X factor of a price-cap regime

with an RPI-X rule.12 In addition, the policy change stimulated research on whether the incentive

mechanisms have effectively attained their performance improvement objectives.

In comparative efficiency studies, the estimation of an efficient frontier is the shared goal

of the different methodologies. In all benchmarking methods, firm's efficiency is given by a

measure of the distance of the observed practice to the efficient frontier. What differs is the

technique for estimating the efficient frontier.





12 The use of a comparative efficiency study with this purpose is subject to criticisms, however. Shuttleworth (2003)
and Irastorza (2003), among others, argue that it provides misleading results, by confusing inefficiency with
heterogeneity, and therefore should not be used.









A first possibility is to perform a bottom-up efficiency study, where the theoretical

yardstick comes from the engineering knowledge of the industry process. This model (or

theoretical) firm approach has not been the rule, but it was used in countries like Spain, Chile,

Peru, and more recently by the electricity regulator in Brazil. Academic researchers have focused

on estimating the efficient frontier based on an empirical estimate using observed data, with its

estimation being implemented with either a parametric or a non-parametric technique. 13

Non-parametric methods, like Data Envelopment Analysis (DEA), use mathematical

programming techniques and do not require specification of production or cost functions nor the

imposition of behavioral assumptions. These methods are generally easy to implement, but carry

an implicit restriction in the number of variables that might be used, and do not allow for random

shocks.

Parametric methods, in turn, entail applying an a priori functional form to the frontier,

estimated with econometric tools. They allow for hypothesis testing,14 enabling the analyst to

investigate the validity of the model specification. Tests of significance can be performed for the

functional form and for the inclusion or exclusion of factors, which is of special relevance for the

electricity distribution industry, where the inclusion of several factors is theoretically justifiable.

Moreover, with a parametric method it is possible to allow for stochastic factors or measurement

errors, which avoids the assumption that all deviations from the best practice frontier involve

inefficiencies.




13 For a detailed description of the different methods to perform efficiency analysis, and an assessment of the
strengths and weaknesses of each, see Kumbhakar and Lovell (2000), Coelli, Prasada Rao, and Battese (1998),
Cubbin and Tzanidakis (1998), Sarafidis (2002), and Atkinson et al (2003).
14 In non-parametric models, a bootstrap technique may be used to produce confidence intervals around the
estimated individual efficiency and thereby assess statistical properties of the efficiency scores generated (Simar and
Wilson, 1998).









Thus, parametric methods can be deterministic or stochastic. A deterministic approach, like

the Corrected Ordinary Least Square colsS), does not allow for random shocks of elements

beyond management control, which might have also contributed (positively or negatively) to the

discrepancy between the individual firm performance and the frontier. 15 This problem can be

addressed by using the so-called stochastic frontiers (Stochastic Frontier Analysis [SFA]), which

use a mix of one-sided and two-sided error terms, with the former capturing the firm's

inefficiency and the latter capturing the effects of random variation in the operating

environment. 16

Jamasb and Pollitt (2001), Mota (2004), and Estache, Perelman, and Trujillo (2005)

perform comprehensive reviews of the comparative efficiency literature on the electricity

industry, providing several examples of the use of the methods mentioned above to examine

firms' performance.

In some of the existing studies, however, the choice of method was determined by ease of

use or limited by sample size or data restrictions. 17 Ideally, the decision regarding the appropriate

method depends on the purposes of the study and the context under examination. This study aims

to investigate efficiency evolution through the period of 1998 to 2003, the decomposing of

productivity growth for each firm into technical change and efficiency change, looking

separately at public and private firms, and the degree of convergence in efficiency scores. The

investigation, in turn, is conducted in an environment where random shocks were present and the

15 Another drawback of COLS is that the structure of "best practice" production technology is the same as the
structure of the "central tendency" production technology, since the estimated frontier is parallel to the OLS
regression. Thus, the frontier does not reflect the production technology of the most efficient producers, but the one
from producers down in the middle of the data (Khumbhakar and Lovell, 2000).
16 The stochastic frontier approach is described in Section 2.3.

17 For example, Pollitt (2005) reports that sample size and data limitations have restricted Ofgem's methodological
choices and prevented the successful implementation of SFA and the incorporation of stochastic factors into the
analysis of efficiency.









inclusion of several variables in the model specification, besides being theoretically justifiable, is

advisable due to the high heterogeneity in operating conditions. These considerations lead to the

use of a stochastic frontier approach, defined in terms of an input orientation, given the output

exogeneity that characterizes the electricity distribution industry.

Stochastic Cost Frontier and Treatment of Environmental Variables18

A cost frontier can be expressed as

E, > c(y,, w,; 0), i = 1, I, (3-1)

where E,= w, x, = ~,n wn x is the expenditure incurred by producer i,y, = (yi1, yM,) 0 is a

vector of outputs produced by producer i, w = (w,, .., WNs,) > 0 is a vector of input prices faced

by producer i, c(y,, w,; 3) is the deterministic cost frontier common to all producers, and 0 is a

vector of technology parameters to be estimated.

When the formulation above incorporates the fact that expenditure may be affected by

random shocks not under the control of producers, the following stochastic cost frontier is

obtained:

E, > c(y,, w,; ) f exp{vi} (3-2)

Thus, the stochastic cost frontier consists of two parts: a deterministic part c(y,, w,; 3)

common to all producers and a producer-specific random part exp{vi}, which captures the effects

of random shocks on each producer. In this context, a measure of cost efficiency of producer i is

given by

CE c(y,, w;) exp{v, (3
1 E8 section d u(3-3)
E





is This section draws upon Kumbhakar and Lovell (2000).









which defines cost efficiency as the ratio of minimum cost attainable in an environment

characterized by exp{vi} to observed expenditure. CEi < 1, with CEi = 1 if, and only if, E, = c(y,,

w,; 3) exp{vi}.

If it is assumed that the deterministic part c(y,, w,; 3) of a single-output cost frontier takes

the log-linear Cobb-Douglas functional form, the stochastic cost frontier can then be written as

InE, >/0 + ,f In y, + 1 In w,, + v, (3-4)
n
ln E, =/8o +/y lny, +nZ/ + lnwn +v, +u, (3-5)
n

where v, is the two-sided random-noise component, and u, is the nonnegative cost inefficiency

component of the composed error term e, = v, + u,. The noise component v, is assumed to be iid

and symmetric, distributed independently of u,. Thus, the error term ec = v, + u, is asymmetric,

being positively skewed since u, > 0.

Under the above representation, a measure of cost efficiency of each producer i is provided

by

CE, = exp {-u,}. (3-6)

Estimates of the production technology parameters, as well as of the cost efficiency of each

producer, can be obtained with the maximum likelihood method, which requires that some

distributional assumptions be made. While v, and u, must be assumed as distributed

independently of each other and of the regressors, estimation of CE, requires that separate

estimates of statistical noise v, and cost inefficiency u, be extracted from estimates of e, for each

producer, which, in turn, calls for distributional assumptions on the two error components.

Therefore, by assuming that the two-sided random-noise component is normally distributed, and

that the nonnegative cost inefficiency component follows a half-normal, an exponential, a









truncated normal, or a gamma distribution, CE, is obtained from the conditional distribution of u,

given 8,.19 More formally:

CE, = E(exp {-u,} ,) (3-7)

It is worth noting that additional information might be provided when repeated

observations on each producer are available. In a panel formulation, evidence of cost efficiency

change can be obtained by including time as a mean inefficiency parameter, when a truncated

normal distribution is assumed for u,. On the other hand, the effects of technical change can be

captured as well if time is included in the deterministic kernel of the stochastic cost frontier.

When a translog functional form is adopted, that would amount to the inclusion of a time-trend

(t) and its square (t2) as additional regressors, obtaining:

lnE +n l 2
Sk (3-8)

yn In Inln + 2tt + 8ttt2 +v, +u2
n

Technical progress will be evidenced by a negative partial derivative of observed

expenditure with respect to time. It should be stressed, however, that the inclusion of time in the

manner depicted in Equation 3-8 accounts for what is known as Hicks-neutral technical change.

This essentially implies that the cost frontiers (as production isoquants) are shifting each year but

their slopes (e.g. the MRTS) do not change (Coelli, Prasada Rao, and Battese, 1998). Non-

neutral technical change is obtained by also including terms involving the interactions of the

other regressors and time.20




19 See Kumbhakar and Lovell (2000, pp. 141-2) for the likelihood function and CE, point estimator expressions.
20 Likelihood Ratio tests might be employed to guide the decision upon which formulation should be used to account
for technical change (neutral or non-neutral).










The stochastic frontier approach provides different ways to incorporate environmental

variables, which exert an influence on producer performance in spite of not being inputs or

outputs of the production process. The point deserves special attention in an efficiency analysis

context, where it is essential to control for variation in producer performance due to variation in

exogenous variables characterizing the environment in which production takes place. The way to

proceed, however, depends on a previous judgment about how the interference of exogenous

factors occurs. The environmental variables may influence the structure of the technology by

which conventional inputs are converted to outputs, or they may influence the efficiency with

which inputs are converted to outputs. In the first case, environmental factors should be included

directly in the production or cost frontiers as regressors, producing efficiency scores which are

net of environmental influences. In the second, these factors should be modeled so that they

directly influence the inefficiency term.21

For this last case, one possibility is to assume a truncated normal distribution for the

inefficiency error term and relax its constant-mean property, by allowing the mean to be a

function of the exogenous variables (z,). More specifically, u, N+( ,t, C ), with ,t specified as

M
t = o + 2 ZJ,,t (3-9)
j=1

Another possibility is to relax not the constant-mean but the constant-variance property of

the truncated (or half) normal distribution for the inefficiency error term, by allowing the



21 This approach yields efficiency scores which incorporate the environmental effects. Coelli, Perelman and Romano
M
(1999) call them "gross efficiency scores". The study proposes the substitution of Y 6,5 z ,, in equation (3-9) for
J=1
M
min[ I Sz,,, ], in order to obtain net efficiency scores. It is argued that the modification enables the efficiency
J-1
measures to be estimated in a context where all firms are assumed to face identical conditions (i.e., the most
favorable).









variance to be a function of the exogenous variables. According to Kumbhakar and Lovell

(2000), this procedure makes it possible not only to incorporate exogenous influences on

efficiency but also to correct for one possible source of heteroskedasticity. More specifically, u, ~

N+(0, 2,,), with 2,, specified as


u = + jZ3j, t (3-10)
j=1

In spite of being sometimes neglected in practice, the possible violation of the

homoskedastic assumption requires special attention in parametric efficiency studies, since the

consequences of heteroskedasticity are potentially more severe in stochastic frontier models, than

in a classical linear regression model.22 Heteroskedasticity can appear in either error component,

as long as the sources of noise and/or inefficiency vary with companies' size, what is quite

possible. While unmodeled heteroskedasticity in the symmetric noise error component (v,) leads

to biased estimates of technical efficiency, unmodeled heteroskedasticity in the one-sided

inefficiency error component (u,) leads to bias in both estimates of the parameters of the cost

frontier and estimates of technical efficiency.23

The empirical literature on efficiency analysis provides examples of the use of these

different approaches to account for environmental variables. Burns and Weyman-Jones (1996)

and Estache, Rossi, and Ruzzier (2004) include these variables as additional regressors in the

functions employed, Hattori, Jamasb, and Pollitt (2003) and Mota (2004) utilize the

environmental factors to model the mean inefficiency, and Botasso and Conti (2003) employ

them to model the variance of the inefficiency error term. There are also some studies that apply

22 If the error term is heteroskedastic in a classical linear regression model, estimators are unbiased and consistent,
but not efficient.
23 See Kumbhakar and Lovell (2000, pp. 115-122), for details on the direction of bias on estimates of technical
efficiency caused by heteroskedasticity in either of the error components.









more than one of these approaches. Wang (2002) parameterizes both the mean and the variance

of the inefficiency error term, to accommodate non-monotonic efficiency effects of exogenous

variables. Coelli, Perelman and Romano (1999) and Hattori (2002), on the other hand, utilize

models where environmental variables are first included as arguments of the input distance

function and then as parameters of the mean inefficiency, comparing the specifications employed

on the basis of the likelihood-ratio test. Both studies find that the sizes of the estimated

efficiency measures differ significantly with the model selected, denoting that the

methodological decision is a key point to be addressed in an efficiency analysis.24' 25

Specification and Data

In the present study, the sample was limited by availability of data and by the decision to

exclude some very small concessionaries, which deliver less than 100,000 MWh per year. From

the total of 64 electricity distribution companies in the country, nine were dropped from the

sample due to small size and data for three others were unavailable. The unavailability of data

also prevented the incorporation of the period before 1998 into the study. Therefore, the sample

includes 52 companies, responsible for 99.47% of the total electricity delivered in the country in

year 2003, with the data being collected for the period of 1998 to 2003. The data were assembled

from the regulatory agency, the companies' websites, the financial statements provided to the

Sao Paulo Stock Exchange, the Brazilian Association of Electricity Distribution Companies

(ABRADEE), the Brazilian Institute of Statistics (IBGE), and the Caixa Economica Federal -

24 Coelli, Perelman and Romano (1999) argue that, in the absence of a strong preference for one approach over the
other, it is advisable to turn to the data for guidance.
25 No matter which approach is adopted to control for environmental variables, some unobserved heterogeneity
might still be embedded in the estimated efficiency measures. To address this point, Greene (2005a, 2005b)
proposes a "true fixed effects model", in which firm fixed effects are included either as regressors or as parameters
of the mean inefficiency, along with some environmental variables or not. This approach, though, might result in
biased results if the number of repeated observations on each producer is small. In addition, the method forces any
time-invariant inefficiency to be absorbed by the firm specific constant term, resulting in an underestimation of
inefficiency.









CEF, a public financial institution that is in charge of most of the social programs of the federal

government and provides financing to house construction projects.

One of the companies in the sample (Elektro) resulted from a split that occurred in January

of 1998. Consequently, its data for that year were not used. Another company (Bandeirante) was

split in 2001. The company's data for the year 2001 was disregarded, and the resulting two

companies were treated as new firms in the sample (after 2002). As a result, the sample has 50

firms in 1998, 51 in 1999 and 2000, 50 in 2001, and 52 firms in 2002 and 2003.

Among the sample companies, 23 were publicly owned in the beginning of 1998. Seven of

them were privatized during the period examined (four in 1998, one in 1999, and two in 2000).

For the purposes of this study, a privatized company was considered "Private" in the same year it

was sold only if the privatization occurred before June 30th. Therefore, there were 21 publicly

owned distribution companies in 1998, 19 in 1999, 17 in 2000, and 16 from 2001 to 2003.

We use a variable cost specification, given our belief that transformer capacity and

network length constitute capital inputs that are fixed in the short run, and provide important

information in terms of system configuration. In addition, environmental variables are included

as arguments of the variable cost function, instead of as mean inefficiency parameters, as we are

interested in having efficiency indexes net of factors exogenous to the firms.26

We turn to the stochastic frontier's hypothesis testing capabilities to guide our decision

regarding the functional form, the incorporation of technological change, and the distributional

assumption for the inefficiency error term. Likelihood ratio tests rejected not only the Cobb-

Douglas functional form in favor of the translog, but also the Hicks-neutral formulation of



26 It should also be stressed that the specification employed provided a better fit than the one where the
environmental variables entered as mean inefficiency parameters, according to the Likelihood-ratio tests that were
performed comparing these two alternative ways of treating environmental variables against a nested model.









technical change in favor of the non-neutral formulation. Moreover, the tests supported the half-

normal distribution for the inefficiency error term, compared to the truncated-normal

distribution.27

In light of the pronounced heterogeneity among the companies in the sample, in terms of

size and customer structure,28 we checked for the presence of heteroskedasticity on the two error

components. The null of homoskedasticity was supported for the two-sided noise component (v,),

but rejected for the one-sided inefficiency error term (u,). As a result, in our model the variance

of inefficiency error component is conditioned on a proxy of firm size, given by total electricity

delivered (Q).

Hence, the specification adopted is shown in equation 3-1129

1 1
lnE, = o + 8y lny, +1,/, Inw,,,, +-/P1 (nY,) +-Z ln Ink,, i r +1 ,, Iny, n,,,,
n 2 2n k n 311
(3-11)
+8i, InCap, + InLen, + S, InZ -,, +P,t+ft Iny,t+Zit Inwn',t+- tt ,, +t + -l,,
Sn 2

where E and y are the cost and output measures, respectively, Wis the vector of factor prices,

Cap stands for transformer capacity, Len represents network length, Z is the vector of

environmental variables, and it is assumed that vt ~ N(0, o ) and u,t ~ N (O, ,,), with cO

specified as

2.= (po+OQ Q




27 The null hypothesis that the mean of the inefficiency error term distribution is equal zero was not rejected, even at
the 10% significance level.
28 This point is detailed in the interpretation of the descriptive statistics, presented below.

29 The decision to not use a multiple output formulation, where outputs would be defined in terms of electricity
delivered to low-voltage customers and to high-voltage customers, was due to the inclusion of share of electricity
delivered to industrial customers as an environmental variable. The alternative specifications were compared, with
the one employed in the present study providing a better fit.









The modeling of technical change in the way shown in equation 3-11 is used to get

evidence of technological change over the period considered. For the computation of firms'

efficiency indexes and the consequent analysis of efficiency change, however, we turn to the use

of time fixed effects, to control for possible changes in macroeconomic factors that might have

affected firms' performance during the period under investigation. In the present study, it is

important to avoid attributing to efficiency change variations in cost caused by other phenomena,

such as industry-wide technological advances and changes in interest rates, exchange rates or

electricity sector policy. Note that the use of time fixed effects explicitly allows the computation

of efficiency indexes relative to yearly-specific frontiers.

The evolution of firms' performance (efficiency change), the possibility of differential

performance between private and publicly owned firms, and the existence of a privatization

effect on firms' efficiency are investigated with a conditional mean specification, where the

following variables are included as parameters of the mean inefficiency error term: (a) indicators

of time (time trend and its square); (b) a private dummy (PRIVATE); and (c) two other indicator

variables representing companies that were privatized (PRIVTZED) and the ones that were

already private before the beginning of the privatization process (ALWSPRIV). Additional

insights on efficiency evolution are supplied by the analysis of average inefficiency scores and

mean relative cost inefficiency measures, computed for each year in the period of 1998 to 2003,

which also enables the investigation of the possibility of efficiency catch-up over the period.

The observed technological change (ATC) and technical efficiency change (ATE) are then

combined to provide a more complete picture of the productivity improvements occurred in the

period under examination. This is done through the computation of Malmquist productivity

indices, following the methodology proposed by Coelli, Prasada Rao, and Battese (1998) for









stochastic frontier methods, adapted to a cost frontier context. For each firm, the Malmquist

index of productivity change between two consecutive periods is given by

MIJ = ATEJ ATC,

where

ATE= -1Eff.Inex +1 and ATC= {(1+TC,). (1+ TCJ,, }12
Eff.Index ,J

The resulting measures of productivity change, along with their decomposition into

productivity catch-up (ATE) and frontier shift (ATC) components, are employed to compare the

performance of firms by ownership type.

In our model (equation 3-11 above), the dependent variable is given by the operating costs

of distribution and retail service activities (Opex), computed as the sum of labor, materials and

third party service contracts expenses, as reported in the income statement.30' 31 Electricity

delivered, in MWh (Q), is the output measure32 and average wage, calculated as total labor

expenditure divided by the number of employees, is used as a proxy for the price of labor (LP).33

For the prices of materials (MP) and third party services (SP), the work uses two price indexes

provided by IBGE and CEF. The materials' price index reflects the observed change in the price

of a basket of items used in civil construction, by State, while the third party service's index

30 The computed labor expenses include firms' contributions to pension funds and to health insurance plans, profit
sharing payments, and management wages. Some firms already report these expenses under the classification of
labor expenses, but most of them do not. The necessary adjustments were made on these cases.
31 In case of vertically integrated companies, the computation of the operating costs of distribution and retail service
activities was made possible by the fact that those companies are required by law to report their expenses separated
by activity.
32 The use of two measures of output was prevented by the fact that electricity delivered and number of customers
showed up as highly collinear, with one of them being always dropped by the statistical software employed (Stata).
A better specification was provided by the former, when compared to the latter.
33 Total labor expenditure is employed to compute average wage because it was not possible to obtain information
related to number of employees segregated by sector activity, for the cases of firms that also operate on generation
and transmission.









portrays the observed change in the salaries paid to an electrician, also by State. The variables

Opex, LP, MP, and SP are expressed in 1998 values, being deflated by a general price index

(IGP-DI).

The regulatory agency does not keep track of changes in transformer capacity and network

length of each distribution company over time. However, information on these variables was

needed for the periodic tariff review, and gathered in year 2002. It is worth noting that by

employing these data on the present study we are implicitly assuming that both variables

remained constant during the whole period under examination, despite the observed increases in

output and in the number of connections.34 Notwithstanding the possible stringency of this

assumption, the variables were kept in the model due to their observed importance in explaining

variation in operating costs among firms, as is mentioned in the next section. Transformer

capacity is given in MVA, and network length corresponds to the sum of high-voltage and low-

voltage lines, in kilometers. Since these variables showed up as highly correlated with electricity

delivered, the variables Cap and Len in our specification (equation 3-11) actually correspond to

the residuals of the regression of transformer capacity on electricity delivered and network length

on electricity delivered, respectively.

Particular attention was given to incorporating exogenous factors that could control for

differences in firms' operating conditions, given the heterogeneity that characterizes the

Brazilian electricity distribution industry, as well as our interest in having efficiency measures

net of factors that impact firms' performance but are out of control of the concessionaries. After

checking the significance of their influence on firms' technology (as cost frontier shifters) and on



34 The companies in the sample experienced an increase of 23% in the number of customers, from 1998 to 2003.
Electricity delivered by these firms, on the other hand, increases 7.6% from 1998 to 2000, and 2.1% from 1998 to
2003.









firms' efficiency (as mean inefficiency parameters)35, the following environmental variables

were included in the modeling: customer density (CusDen), given by number of customers

divided by network length; share of electricity delivered to industrial customers (IndShare);

residential density (ResDen), computed as electricity delivered to residential customers divided

by the number of residential customers; service area36 (Area), in Km2; ratio of underground to

overhead lines (Undergrd); and income per capital, by State (Income), to control for variations in

socio-economic conditions among States.

Descriptive statistics are shown in Table 3-1. The difference between minimum and

maximum values of observations collected for almost all variables used indicate the considerable

heterogeneity among firms in the sample, in terms of companies' size, system configuration, and

customer structure. Electricity delivered, for example, varies from 103,191 to 37,540,051 MWh,

while the share of electricity delivered to industrial customers ranges from 3 to 64%, and

transformer capacity fluctuates in the interval of 120 to 22,728 MVA. The evidenced disparity in

firms' indicators corroborates the need to account for external factors in the comparative

efficiency analysis.

It is observed, from the evolution of the variables' mean values, that operating costs drop

around 26%, in real terms, from 1998 to 2003.37, 38 This reduction is partly due to the falling




35 The six environmental variables included in our specification showed up as statistically significant mean
inefficiency parameters. The results indicated that firms' efficiency increases with IndShare, ResDen, and Income,
and decreases with CustDen, Area, and Undergrd.
36 The Brazilian case justifies the inclusion of both network length and service area in the modeling and this is
reflected in the statistical significance of both variables as either cost shifters or mean inefficiency parameters.
While some companies have small service areas and relatively high network length (the ones that operate in the
more densely populated states), others have highArea but relatively low Len, because they operate in states which
are more sparsely populated and/or have a high share of the population not being served.

7 The indicated variation, as well as all others based on the numbers portrayed in Table 3-1 and mentioned in this
section, is adjusted for the change in the number of firms in the sample from year to year.









prices of materials and third party services, which fell 12.9% and 21.9%, respectively, in the

same period. It is also interesting to note the dramatic reduction in electricity delivered that

occurred in 2001 (-7.6%), as a consequence of the rationing imposed by the government. The

volume of electricity delivered increased in 2002 and 2003, but it still was, in 2003, smaller than

the volume delivered in 1998, and 5.1% less than the amount delivered in 2000.

The same "rationing effect" is observed in the average residential consumption, which

decreased 13.8% in year 2001. This indicator, however, does not recover in years 2002 and 2003

(it is even slightly smaller on these years). This fact, which may be due to higher electricity

prices in the post-rationing period, might also indicate a shift in the residential customers'

demand for electricity, as a result of a change of habits induced by the rationing measures.39

Findings

Prior to estimation, all variables were normalized by their sample median values.

Additionally, in order to ensure homogeneity of degree one in prices, the dependent variable and

the factor prices were normalized by the price of third party services. The models are estimated

by maximum likelihood, using the Stata 9 software.

Table 3-2 provides the results from the models estimated, while the elasticities of operating

costs with respect to output, factor prices, and time are reported in Table 3-3. It is observed that

the estimated cost function satisfies the monotonicity condition with respect to output and factor

prices at the mean. Moreover, as was expected, operating costs are negatively affected by


38 The higher drops occur from 1998 to 1999 (-14.1%), and from 2000 to 2001 (-9.5%), year in which there was the
energy rationing. Mean operating costs are also reduced by 5.7% from 2001 to 2002.

39 From 2001 to 2003, the number of residential customers of the sample firms increases from 38,160,276 to
42,237,897, with the amount of electricity delivered to this group of customers going from 66,399,207 MWh to
70,824,904 MWh. This volume of electricity was still smaller than the one observed in 2000 (75,283,460 MWh),
when there were 36,816,473 residential customers. Thus, the already existing residential customers in year 2000
consumed, in that year, more electricity than in year 2003. The numbers above are adjusted for the change in the
number of firms in the sample from year to year.









increases in the share of electricity delivered to industrial customers, and positively affected by

increases in electricity delivered, service area, factor prices, share of underground facilities, and

customer density. Note also that transformer capacity and network length show up as statistically

significant (at the 1% level) variables explaining variations in operating costs among firms. As

capital proxies, their positive coefficients met our expectation.

The income per capital coefficient estimate is negative and significant, which might be

reflecting higher maintenance costs incurred in low-income areas and, possibly, the impact of

work force qualifications on firms' productivity. Furthermore, the positive and significant

coefficient on Q as a variance parameter indicates that variations in firms' inefficiency levels

increase with firm size.

The time elasticity provides a measure of technological change. The results show that there

was technological progress during the sample period, with an annual rate of technological change

of around 6.55%, on average, which denotes that the efficient frontier has shifted considerably

from 1998 to 2003. Note also that the estimated coefficient on the squared term of time is

negative, indicating that the observed rate of technological progress has increased through the

period examined.

We investigate the possibilities of differences in technology and in the rate of

technological change between public and private firms through the use of a private dummy and

an interaction term private*time. The negative and significant private dummy coefficient (see

Model "B" in Table 3-2) indicates that private firms have had a better technology. This

conclusion, however, should be taken with caution, since the dummy captures the effect of any

systematic difference between the firms' groups, which might include a difference in firms'

performance.









The results also show a slightly higher rate of technological change for public firms, when

compared to their private counterparts. Nevertheless, when private firms are separated into

privatized and not privatized firms (called "always private"), it is observed that the above

mentioned relationship holds only for the comparison of public and "always private" firms, since

the privatized companies actually had a higher rate of technological change than the public

companies (Model "C" in Table 3-2). These results are corroborated by the information provided

in Table 3-4, where the technological change observed in the period is segregated by year and

firms' ownership type.

Firms' efficiency indexes are computed on the basis of the time fixed-effects specification.

An indication of efficiency evolution throughout the period and of comparative performance of

public and private companies is provided by the results of the conditional mean inefficiency

model (Models "E" and "F" in Table 3-2). The coefficient on the time variable is negative, but

not significant, suggesting that efficiency improves only slightly over the period. On the other

hand, the results indicate that private firms, specifically the privatized ones, are significantly

more efficient than public firms in the period examined.

A more detailed picture is seen in Table 3-5, which reports average inefficiency scores and

mean relative cost inefficiency measures, computed for each year in the period of 1998 to 2003

and discriminated by firms' ownership type. The relative cost inefficiency was obtained by

dividing each firm's inefficiency score by the minimum score observed in each year.

The efficiency improvement is reflected in the reduction of average inefficiency scores in

the period. On average, efficiency increases 1.51% from 1998 to 2003. The efficiency gains are

followed by a concomitant reduction in dispersion of efficiency scores, with the standard

deviation going from .0999 in 1998 to .0697 in 2003, constituting evidence of a catch-up effect.










This evidence is corroborated by the mean relative inefficiency measures. While in 1998 the

distribution companies were, on average, 8.7% more inefficient than the most efficient company

in that year, in 2003 they were only 6.3% more inefficient than the most efficient distribution

utility. This indicates that the observed efficiency improvement comes, essentially, from

companies that were relatively inefficient in the beginning of the period.

Additional insights are obtained when one looks at the evolution of these measures for

public and private firms separately. Here, to avoid possible distortions in the efficiency evolution

analysis resulting from the privatizations occurred in the period40, the average inefficiency and

the mean relative inefficiency measures computed for public firms reflect only the performance

of the 16 firms that remained public up to 2003.

Notice first that in the beginning of the period the group of private companies which did

not go through a privatization process is more efficient, on average, than both the public and the

privatized companies' groups.41 In addition, the public firms' mean inefficiency measure is

slightly inferior to the privatized firms' measure42, indicating that the privatization process did

not concentrate initially on the most efficient public firms.43




40 For example, if the public companies that turn out to be privatized in years 1998 to 2000 performed differently
than the average public companies in that period, an analysis that do not take these firms into account would
compare average public firms' scores for years 2001 to 2003 to distorted average public firms' scores for years 1998
to 2000.
41 According to one-sided mean comparison tests performed, ai\ \i private" firms' mean inefficiency in 1998 is
significantly lower than both public firms' mean inefficiency (p-value = .0291) and privatized firms' mean
inefficiency (p-value = .0192).
42 The null of no difference in mean inefficiency of public and privatized firms was not rejected, even at the 10%
level. The p-value of the alternative hypothesis that privatized firms' mean inefficiency is greater than public firms'
mean inefficiency was equal to .3517.

43 This is the same to say that the hypothesis that it is efficiency that leads to privatization, and not the reverse
(which Bagdadioglu, Waddams Price, and Weyman-Jones (1996) claim to have found some support for in the
Turkish case), is not supported in the present context. This fact, however, does not rule out the possibility of
endogeneity of privatization decisions, a point that we address later.









In line with what was mentioned before, the information in Table 3-5 shows that both the

efficiency improvement occurred in the sector and the identified catch-up effect come,

essentially, from the performance of privatized firms, which experience an increase in mean

efficiency of 5.58% from 1998 to 2003, associated with an expressive reduction in dispersion of

efficiency scores. Public firms' mean efficiency increases only 0.35% in the same period, while

the same measure for "always private" firms decreases almost 3%.

We have identified, among the privatized companies, the ones that were privatized after

June/1998, since their corresponding 1998 efficiency scores are free of any privatization effect. It

is interesting to observe that these firms experience an average efficiency increment of 7.8%

from 1998 to 200344 (Table 3-6), much higher than the obtained by the ones that remained

publicly owned. Furthermore, it is worth noting that the decrease in "always private" mean

efficiency in the period under exam comes, essentially, from the deterioration on these firms'

performance from 2002 to 2003, confirmed by a one-sided mean comparison test, which rejected

the null hypothesis of equality of mean inefficiency scores on the two periods, at the 1%

significance level.45 We will return to these points later.

The rates of technological change and of efficiency change, observed between two

consecutive years, are combined to provide a measure of productivity change, given by the

Malmquist indexes reported in Table 3-7. The results indicate that the Brazilian electricity







44 Some companies were privatized in 1999 and 2000, but their efficiency improvement in the period before the
privatization was not representative. It corresponded to only 8.4% of the above mentioned efficiency increment in
the period of 1998 to 2003.
45 The same null hypothesis of equality of mean inefficiency scores (of A.i\ .r private" firms) was also rejected for
years 2000 and 2001, at the 5% significance level, and supported for all other consecutive periods.









distribution industry's productivity increased 38.5%, on average, from 1998 to 2003.46 In

addition, productivity growth rates are increasing throughout the period and come, essentially,

from frontier shifts (37%), with a relatively small catch-up effect for the sector as a whole (1%).

The computed mean productivity growth rate of 6.73% stands out when it is compared to

the 0.9% mean total factor productivity growth rate of the economy found by Gomes, Pess6a,

and Veloso (2003) for the period of 1992 to 2000.47 Moreover, taking the results found by Mota

(2004) as a proxy for the distribution companies' average productivity gains in the period of

1994 to 1998 (around 5%)48, it might be concluded that the sector's rate of productivity growth

increased after 1998.

The observed increment in the sector's productivity comes from the performance of

privatized and public firms, which experience productivity growths of 57.2% and 39.2%,

respectively, from 1998 to 2003, against a productivity growth of only 16.1% of the "always

private" firms (Table 3-8). These results, taken together with the ones previously mentioned,

constitute evidence that the high productivity gains observed in the period of 1998 to 2003 are

associated to the closing of the efficiency gap present in 1998. Both the firms that were

privatized and the ones that remained publicly owned have decreased the difference in

performance with respect to the "always private" firms' group, with privatized firms actually

surpassing the "always private" firms' efficiency level.



46 The above mentioned measure of productivity change does not incorporate the scale effect. On the basis of the
computed output elasticity and the actual changes in output from year to year, we have estimated it to be equal to
3.56%, on average, from 1998 to 2003.
47 The authors report that their result is consistent with the findings of two other studies that examined the subject.
48 Mota obtains annual average productivity gains of about 5% in the period of 1994 to 2000 using data only from
these two years and from 14 privatized companies. Considering that the present study provides evidence of
significantly higher privatized firms' productivity gains for the period of 1998 to 2000, Mota's result can be taken as
an upper-bound measure for average productivity gains in the period of 1994 to 1998.









Note that public firms' efficiency has indeed improved significantly from 1998 to 2003.

The performance improvement, however, is not reflected in better efficiency indexes (public

firms' mean efficiency increase only 0.35% in the same period), or in a positive catch-up effect

(Table 3-8), due to the higher improvement of privatized firms' performance. Efficiency analysis

is comparative and dynamic. Public firms have essentially kept the same distance to a frontier

that has shifted considerably during the period, mainly in response to productivity gains of

privatized companies.

Looking at possible motivating factors for these findings, and in particular at privatization

and the implementation of incentive regulation, it must be stressed, at first, that in the present

study it is not possible to precisely assess the fraction of productivity improvement due to the

adoption of the price-cap regime, and neither if there is effectively a causality relationship on this

respect, given the lack of data from the period pre-reforms and the absence of a "control group".

Furthermore, it must be recognized that, despite the use of time fixed effects, other concurrent

factors may have impacted the results, by affecting some firms and not others. Distribution

companies with a high exposure to financing in dollar49, mainly some of the privatized ones,

might have had to cut costs and be more productive to compensate the higher expenses brought

by the real devaluation in 1999, whereas some public companies may have had to do the same to

put an end in the series of negative profits which became unsustainable in a context of increasing

State budget constraints.

However, no matter the specific reason that has induced each firm to expend some effort to

improve its performance, it might be conjectured that this action only took place because the



49 The real devaluation also impacted the amount distribution companies had to pay for the energy bought from
Itaipu. However, given that these expenses were considered non-controllable costs and, consequently, entirely
passed on to the tariffs, I do not consider them as an efficiency improvement inducing factor.









price-cap regime provided the conditions and incentives for it to happen, which would not be

present in a rate-of-return regulation context.50 In this sense, the observed increase in firms'

productivity growth, over the mean productivity growth rate found by Mota (2004) for the period

before 1998, could be associated with the implementation of incentive regulation.

This association is corroborated by the finding that productivity gains increase throughout

the period, which is consistent with the progressive implementation of incentive regulation in

Brazil.51 The increasing productivity gains may also result from the fact that the labor force

reductions implemented by many firms which began to operate under incentive regulation may

actually increase short term labor costs. Labor force reductions are typically achieved by paying

employees to terminate their employment voluntarily, and the associated expenditures are

recorded as short term costs (Kridel, Sappington, and Weisman, 1996).

It should also be noted that most of those firms regarded as more inefficient at the

beginning of the incentive regulation scheme have experienced very high productivity gains, in

line with the regulator's expectations. Given the similarity in the two firm groups' efficiency

levels in the beginning of the period, the performance of firms which remained publicly owned

can be taken as a proxy for the performance that privatized firms would have had if they had

remained public. Consequently, the difference of almost 18% in productivity growth from 1998

to 2003, which is equivalent to a (geometric) mean productivity growth rate of 3.36% per year,

constitutes a measure of the privatization effect. It also provides evidence that privatized firms

have responded more aggressively than public firms to the efficiency improvement incentives


50 This assertion comes from the fact that in rate-of-return regulation revenues are linked to realized costs. Hence, in
the usual situation of a one-year regulatory lag, it is expected that the regulated company will not have an incentive
to make a costly effort to reduce costs because possible higher profits in the current period have to be balanced
against lower revenues in all subsequent periods.
51 See footnote 9.









brought by the price-cap regulation, implying that incentives were higher for profit-oriented

managers operating under a shareholders' pressure to quickly recoup investments.

The analysis, however, has to account for the possibility of endogeneity of privatization

decisions. The mean comparison tests performed revealed that privatized firms, on average, have

a bigger size and a higher share of industrial customers than public firms52, which is consistent

with the hypothesis that companies with a higher potential for performance improvement were

the ones selected to be privatized. To address the topic, a new performance comparison was

implemented using a more homogeneous group of firms, given by the ones with Q higher than

400,000 KWh per year.53 With the new sample, although the computed Malmquist indexes

indicate a smaller productivity improvement for all firms over the period, compared to the one

obtained previously (Table 3-8), the difference in productivity growth between privatized and

public firms is even higher than the one found before (29% against 18%). This result

corroborates the conclusion that privatized firms responded more aggressively than public firms

to the price-cap incentives.

Conversely, the difference in productivity growth rate of these two firm groups provides an

indication of the additional improvement in performance that public firms might have obtained.

In light of the distributional consequences of these not-implemented-but-achievable productivity

gains, in the sense that they could have resulted in lower tariffs to customers, the issue is of




52 Privatized and Public firms were compared in terms of any variable used in the estimation procedure. The null of
equality of means was rejected for the variables Q, NumCust, Cap, and IndShare, with privatized's mean showing
up as greater than public's mean in all these cases.
53 Five out of the fifteen biggest distribution companies are publicly owned. Due to the new criteria, fifteen firms
were dropped from the initial sample (5 public, 1 privatized, and 9 always private). And when public and privatized
firms were compared using data from the new sample, the null of equality of means was rejected only for the
variables LP (Public > Privatized) and IndShare (Public < Privatized). The two groups of firms do not differ in size
anymore.









particular relevance for the regulator, who needs to target the appropriate efficiency

improvement incentives to this group of firms.

The electricity regulator has had the opportunity to act on this matter in the periodic tariff

review that started in April 2003. It is of interest, therefore, to evaluate the regulator's decisions

on this occasion, notably with respect to their consequences in terms of both distribution of

productivity gains among stakeholders and incentives for further efficiency improvements. Such

evaluation should not be restricted to the specific case of public firms, though, since its insights

are equally valuable in the case of the other distribution companies. It is important to note what

was done with respect to companies that have experienced lower productivity gains in the period,

such as the "always private" ones, and, by virtue of its signaling effect to all concessionaries,

whether privatized firms' productivity gains in the period were completely passed on to the new

tariffs. 54

"Always private" firms were more efficient, on average, than the other distribution

companies in 1998, and have experienced, from 1998 to 2003, a productivity growth higher than

the mean total factor productivity growth rate of the economy found by Gomes, Pess6a, and

Veloso (2003) for the period of 1992 to 2000 (3.04% and 0.9%, respectively). Hence, their

performance in the period might be taken as consistent with what was expected under a price-cap

regime in view of their initial efficiency lead over other distribution companies and of the fact

that the efficiency analysis performed in this study does not incorporate changes in capital

expenditures.55


54 I am currently working on these points in another paper.
55 Under rate-of-return regulation, firms have an incentive to overinvest in capital, moving away from the optimal
capital/labor ratio. Note that these firms will probably show up as relatively efficient if the performance analysis is
based on operating expenses only, as is the case in the present study. In this context, when price-cap regulation is
implemented, the prospect for efficiency improvements in operating and maintenance expenses is limited.










The results found for this group of firms might also be related to the fact that they operate

at a lower scale, with lower potential for efficiency improvements.56 Note, however, that the

observed decline of these firms' mean efficiency levels57 derives from their low productivity

growth in 2003, which is 23% less than the average of their productivity growth in the four years

before (Table 3-9). The deterioration in performance might have been caused by the need to

accomplish some regulatory requirements, such as the ones related to service quality (a point

addressed in the next section). Nonetheless, it cannot be ruled out a possible strategic behavior of

some of these firms, which could have inflated costs near the time of the periodic tariff review in

order to obtain higher repositioning indexes and, consequently, higher allowed revenues for the

subsequent tariff period.58, 59

Service Quality and Economies of Vertical Integration

Two additional points are investigated. First, in view of the well-known concern that price-

cap regulation may provide incentives to reductions in quality of service, it is examined if there

is an association between efficiency and service quality in the Brazilian electricity distribution



56 When always private firms were compared to other firms, the results from the mean comparison tests revealed that
Always Private's mean values were significantly higher than Public or Privatized's mean values for the variables SP
and Income. On the other hand, Public or Privatized's mean values showed up as significantly higher than Always
Private's for the variables Opex, Q, NumCust, LP, Area, Cap, Len, Undergrd, and CusDen. In essence, Always
Private firms are of a considerable smaller size than other firms, and operate in smaller service areas.
57 As mentioned before, hAl' .i s private" mean efficiency score decreases almost 3% in the period, with the most
significant change occurring from 2002 to 2003, according to the results of the mean comparison tests performed.
58 Although firm's own operating costs were not considered directly in the definition of the repositioning index,
under the reference company approach adopted by Aneel, a firm that artificially inflated its costs in the period near
the tariff review would be acting in order to increase its bargaining power in the tariff review process. The rationale
is that the probability of a firm request to increase its estimated efficient operational costs be accepted by the
regulatory agency increases with the difference between estimated and "real" (informed by the firms) costs.

59 This type of regulation game, associated with the periodic aspect of incentive-based regulation, is known to
regulators and was reported in the survey conducted by Jamasb, Nillesen, and Pollitt (2003). Evidence of strategic
behavior of the same sort was found by Di Tella and Dyck (2002), in a study of the Chilean electricity distribution
utilities. The study reports U-shaped cost reductions associated with the introduction of price cap regulation, with
strong initial cost reductions reversing every four years, coinciding with regulatory reviews.










sector. 60 Beyond that, it is checked whether the variations in efficiency levels identified in the

present study may result from changes in service quality. Secondly, given that some distribution

companies in the sample also operate in generation and transmission, it is investigated the

presence of economies of vertical integration and, as a corollary, the possible occurrence of

firms' strategic behavior.

Vertically-integrated companies generally experience economies of scope, which would

help them be more efficient than other concessionaries.61 They also have the opportunity to

behave strategically, classifying expenses incurred in their generation and transmission activities

as costs pertaining to the regulated distribution activity.62 In this context, a finding that vertically

integrated firms are significantly more inefficient than other distribution companies, despite

plausible economies of scope, would indicate that cost shifting may be occurring.

The 14 companies that generated at least 10% of the electricity they delivered to final

customers in 2003 are considered vertically integrated.63 Service quality is measured on the basis

of two continuity of service indexes implemented by the regulator, DEC and FEC, associated

with duration and frequency of service interruptions, respectively. While the DEC index


60 Actually, as it is stated by Ai, Martinez, and Sappington (i-" '4), economic theory does not provide unequivocal
predictions regarding the effects of incentive regulation on service quality. When authorized revenues are not tied
directly to realized costs, a regulated firm may be tempted to reduce service quality in order to reduce costs, and
thereby increase profit. On the other hand, because incentive regulation can allow a firm's revenue to rise
substantially above realized costs, incentive regulation may motivate the firm to improve service quality in order to
enhance the demand for its products and thereby increase revenue. Nonetheless, given that in the electricity
distribution sector the demand is essentially exogenous to the firm, especially when there is no competition, it is
possible to argue that the first effect should prevail over the second in this specific context.
61 See Lowry and Kaufmann (1999), for details about economies of vertical integration in the power distribution
sector.
62 The rationale for this behavior is provided by Joskow (2005). In light of the uncertainties the regulator faces about
the firm's inherent cost opportunities, the firm would like to convince the regulator that it is a higher cost firm than
it actually is, in the hope that the regulator will then set higher prices for the services it provides as it satisfies the
firm's long run participation constraint, increasing the regulated firm's profits, creating dead-weight losses from
prices that are too high, and allowing the firm to capture surplus from consumers.

63 Among the vertically-integrated firms, six are publicly owned and five are privatized companies.









measures the number of days of service interruption within a period, the FEC measures the total

number of service interruptions within a period. Note that service quality is inversely related to

DEC and FEC values. The service quality variable (Quality), therefore, is given by the inverse of

the equally weighted average of the continuity of service indexes.64, 65

A severe missing observation problem was identified with respect to DEC and FEC data

for 1998. Thus, Quality measures for each firm in the sample were computed for the period of

1999 to 2003. The mean values of DEC and FEC indexes, as well as of the resulting Quality

variable, are displayed in Table 3-10, which shows that the continuity of service indicators

improved substantially over the period, resulting in an increase of almost 50% in the computed

Quality variable from 1999 to 2003. The numbers portrayed provide an indication of the

effectiveness of quality regulation instruments implemented by the regulator.

The Quality decomposition by firm group reveals that the mentioned increase in service

quality was driven by privatized and always private firms, whose Quality measures raised 54.6%

and 51.6%, respectively. It also shows that the lower service quality improvement experienced

by public firms (35.7%) resulted in an increase in the distance of these firms to the others in

terms of quality of service provided to customers. It must be stressed, however, that in light of

the rather different operating conditions experienced by these groups of firms, the continuity of

service indicators incorporate factors out of firms' control and, consequently, cannot be taken



64 I recognize that some important dimensions of service quality are not incorporated in the Quality measure, such as
those related to customer satisfaction, the speed with which reported problems are resolved and the intrinsic quality
of the product offered to customers, an inverse function of variations in electricity tension levels. The Quality
measure, however, captures the adequacy of system maintenance, the amount of human and capital resources
allocated to network restoration and repair, and the existing facilities to recuperate the system after each interruption
(vehicles, communication, qualification of the work force, etc.).
65 In the Brazilian electricity distribution sector, quality regulation and enforcement have essentially been done
through the definition of target DEC and FEC indexes for each year, associated with the imposition of penalties for
non-accomplishment.









conclusively to compare the resources allocated and the effort spent by each firm group in

service quality provision.

To address the point, it was assumed that the different operating conditions are

incorporated in the target continuity of service indexes set forth by the regulator. Ratios target

over actual indexes were computed for each firm, and the resulting relative DEC and FEC

measures were used to derive an equally weighted adjusted quality measure.66 The

corresponding values found for each firm group in 2003 confirmed the service quality ranking

shown in Table 3-10.67 Public firms do provide lower service quality than other firms, a finding

that rules out a possible explanation for their lower efficiency level evidenced in the present

study. Always private firms, on the other hand, show an adjusted quality measure that increases

over the period and is well above unity in 2003, which denotes the provision of service quality at

levels well above what has been required by the regulator. It is investigated below whether this

overinvestmentt" in quality of service might be one of the reasons for the deterioration in

performance over the period experienced by these firms.

Two procedures are employed to examine the relationship between efficiency and service

quality, overall and for each firm group, as well as the possible existence of economies of

vertical integration. First, the Quality measure, interaction terms Qi1iiy'",
and a vertically-integrated dummy (Vertical) are included as additional arguments of the variable

cost function, with the resulting efficiency scores being compared to the ones obtained



66 The adjusted quality measure is given by: 1/2 DEC, .,, + /2 FEC, .,, where the relative measure equals Target
(DEC or FEC) over Actual (DEC or FEC). Since service quality is inversely related to DEC and FEC values, the
higher the relative measures, the higher the service quality. In addition, the higher is the computed adjusted quality
measure from one, the greater is the distance from the actual service quality provided to the target defined by the
regulator.
67 The computed adjusted quality measures for year 2003 were the following: 1.566, 1.731, and 2.183, for public,
privatized, and always private firms, respectively.










previously.68 The rationale here is to consider service quality as an additional output electricity

distribution companies must provide in order to accomplish the regulatory requirements.

Secondly, the same variables are included as parameters of the mean inefficiency error term, in a

conditional mean specification.69

The results obtained are reported in Table 3-11. Initially, the evidence tells that, after

controlling for other factors, service quality and the fact of being vertically integrated or not do

not contribute significantly to explain variations in operating costs among Brazilian electricity

distribution companies (Model "A"). Both Vertical and Quality coefficients have signs contrary

to what was expected and are not significant, while all other coefficients remain practically the

same. Likewise, the resulting efficiency indexes are quite similar to the ones obtained previously

(correlation of .9941). Note, however, that the service quality impact on operating costs differs

with firms' ownership type (Model "B"). While costs increase exponentially with service quality

levels for public firms, as it was expected, for privatized companies a negative relationship

holds.70 Among the privatized, the ones that provide better service quality have lower operating

costs.7

The conditional mean specification (Models "C" and "D" in Table 3-11) reveals an

unexpected positive, and marginally significant (p-value = .055), coefficient on the Vertical

68 Since service quality is costly, it is expected that firms that provide better service quality are penalized when this
variable is not accounted for in the analysis, by showing up as more inefficient than the ones that provide lower
service quality. Hence, the control for this factor should increase efficiency indexes of firms that provide higher
quality.
69 The ownership dummies themselves were also incorporated in the specifications that included the interaction
terms Quality *ownership dummies, to prevent attributing to service quality an effect due to the firm's ownership
type.

70 Although the coefficient on the interaction term lnqlt*alwspr is negative and significant, the computed elasticity
of always private firms' service quality was close to zero (-.0077).
1 In light of the longitudinal nature of the data, additional tests (random effects, between effects, and fixed effects
regressions) were performed in order to ascertain that the negative relationship found between Quality and Opex
comes from variations in Quality between firms.









dummy. Despite probable economies of scope, vertically integrated firms show up as more

inefficient than other distribution companies. The point deserves the regulator's attention, since

cost shifting is one of its possible explanatory factors.

The findings also show that quality of service provided-as it is measured in the present

study-does not help explain variations in firms' efficiency levels. The coefficients on Quality

and its squared term indicate that service quality increases inefficiency when Quality is higher

than 1.6, a value slightly higher than the 75% percentile (1.52). Here, again, the impact differs

with firms' ownership type (Model "D"). For public and always private firms the Quality

coefficients are positive, but not significant, a result that does not confirm the raised possibility

that the always private firms' overinvestmentt" in quality of service was one of the reasons for

their evidenced deterioration in performance. On the other hand, the findings do indicate that,

among privatized companies, the more efficient ones also provide better quality of service.

At this point, it is important to note that the evidence not only supports the effectiveness of

quality regulation instruments implemented by the regulator, but also suggests that the

expressive privatized companies' efficiency improvement identified in the present study comes

effectively from a more efficient operation of their units, in line with what was expected under

an incentive regulation scheme, and not from mere reductions in costs brought by deterioration

in the quality of service provided.

Conclusion

This paper confirms the theoretical predictions regarding the impact of incentive regulation

on firms' performance. Brazilian electricity distribution companies have experienced high

productivity growth rates after the sector reforms, above what was found in a previous study for

the period before the reforms. The productivity increment relates to the closing of the efficiency

gap present in 1998, and is driven by the performance of privatized and public companies.









Privatized firms responded more aggressively than public firms to the new incentives

brought by price-cap regulation, denoting that incentives were higher to profit-oriented managers

operating under a shareholders' pressure to quickly recoup the investments made. The study's

estimate of privatized firms' incremental annual productivity growth rate (3.36%), on the other

hand, brings about the need to tailor specific efficiency improvement incentives to public firms,

since it represents not implemented-but-achievable productivity gains, which could have resulted

in lower tariffs to customers.

The subset of firms privately owned before the reforms shows up as more efficient, on

average, than other firms in the beginning of the period examined. Its productivity growth rate

evidenced in the present study, therefore, is consistent with a limited space for efficiency

improvements-on operating and maintenance expenses-on the more efficient firms subject to

a rate of return regulation scheme. Given the sensibly higher productivity growth rates

experienced by other firms, "always private" firms face a decline in their efficiency levels over

the period. This research provides another possible explanation. It shows that the observed

decline in these firms' mean efficiency level derives, fundamentally, from their low productivity

growth in 2003, which might therefore indicate a possible strategic behavior of some of these

firms, associated to the periodic aspect of the price-cap incentive regulation scheme.

The results suggest a possible occurrence of strategic behavior of another sort as well. In

spite of plausible economies of scope, vertically integrated distribution companies show up as

more inefficient than other firms, raising the possibility of cost shifting. Stricter rules regarding

cost allocation and/or a closer look at these companies' accounting numbers may be appropriate.

Interestingly, the paper reveals that the high performance improvement experienced by

privatized firms in the period does not come from mere reductions in costs brought by









deterioration in the quality of service provided, a result that also indicates the effectiveness of the

quality regulation instruments implemented by the regulator.

All these findings ultimately provide a better understanding of the cost opportunities faced

by each firm, and consequently enable the establishment of prices conducive to a greater social

welfare. The regulator has had the opportunity to define new electricity distribution prices in the

periodic tariff review that started in 2003 and is still in place for some firms. On that opportunity,

the choice was for the use of the model company approach to estimate each firm's efficient

operational costs. This paper's findings provide the basis not only for evaluating the regulator's

decisions in those circumstances, notably with respect to their consequences in terms of both

distribution of productivity gains among stakeholders and incentives for further efficiency

improvements, but also for examining the model company approach itself. The approach's usage

is not pacific in the theory and its implementation in the Brazilian context has been disputed

among the parties involved.
















Table 3-1. Descriptive Statistics
Variable 1998 1999 2000 2001 2002 2003 1998-2003 Range


OPEX


Q


98,905 85,773 84,953 74,258 70,455 70,134
(132857) (111025) (113274) (100497) (97838) (97273)

5,074,129 5,260,394 5,520,603 4,790,657 5,063,016 5,110,973
(8442352) (8346455) (8719154) (7649132) (7569014) (7404106)

38.9052 32.4873 35.9164 34.0834 39.2159 41.9181
(189536) (147348) (18517) (147472) (223052) (21 2112)

78.6138 72.8605 70.946 70.9966 68.4405 68.595
(67104) (4419) (39627) (3 5699) (3 2352) (34071)

74.0168 66.802 64.4161 61.4267 53.5491 58.3822
(179825) (184658) (169462) (14435) (11 4705) (127854)

25.7095 26.6959 27.9056 28.7718 30.8484 32.0821
(18 6995) (19 1257) (20 0373) (204782) (21 8955) (224005)

0.2959 0.2980 0.3068 0.3132 0.3308 0.3257
(0 1461) (01434) (0 1432) (0 1413) (0 1498) (0 1568)

2.1026 2.0789 2.0028 1.7162 1.6803 1.6774
(06267) (06282) (05139) (04687) (04625) (04167)

129,178 129,210 129,203 131,495 126,671 126,725
(242029) (239567) (239564) (241463) (237902) (237882)

828,166 879,502 919,894 934,543 979,891 1,012,766
(1099440) (1134211) (1188028) (1228822) (1255942) (1287816)

5,769.74 5,086.45 5,160.16 4,996.71 4,386.60 4,642.68
(280422) (2351 86) (23793) (2272 11) (1880 11) (198973)

3,218.57 3,269.12 3,269.12 3,142.07 3,206.25 3,206.25
(4908) (4872 48) (4872 48) (4835 87) (4751 46) (4751 46)

41,998.10 42,957.20 42,957.20 42,959.70 42,131.10 42,131.10
(65700 6) (65399 9) (65399 9) (66063 9) (648945) (64894 5)

0.006592 0.006462 0.006462 0.005940 0.006338 0.006338
(0246) (0244) (0244) (0244) (0241) (0241)

50 51 51 50 52 52


80,640
(108952)

5,137,639
(7970465)

37.1144
(18 7994)

71.701
(5 5173)

63.022
(16 7229)

28.6965
(20 4544)

0.3119
(0 1463)

1.8749
(0 5537)

128,723
(237747)

926,545
(1193257)

5,001.43
(2317 13)

3,218.73
(4792 04)

42,520.10
(64850 3)

0.006356
(0241)

306


* Mean values reported for each year and for the period 1998-2003. Standard deviation in parentheses.


[2490,559072]


[103191,37540051]


[6.5398, 128.4681]


[60.008, 96.620]


[29.434, 98.120]


[6.747, 137.093]


[.0333, .6438]


[.663, 4.572]


[252, 1253165]


[19625,5744178]


[1060.012, 12747]


[.1, 22728.4]


[720.3, 379518.58]


[0, .1391]


CUSDEN


INDSHARE


RESDEN


AREA


NUMCUST


INCOME


CAP


LEN


UNDERGRD


OBSERVE .

















Table 3-2. Stochastic Cost Frontier Results
Variable Time-trend formulations Time Fixed-Effects formulations
A B C D E F


InOpex
InQ

InLP

InMP

Cap

Len

InlndShare

InResDen

InIncome

InArea

InCusDen

UnderQrd

T

lnO*t

lnLP*t

InMP*t

Tsq

Private

Private*t

Prvtzed

Alwspriv

Privtzed*t

Alwspriv*t

D1999

D2000

D2001

D2002

D2003

Cons

Mu
Private

T

Timesq

Privtzed

Alwspriv

Cons

lnsie2u
O

Cons

N
L1
chi2
Legend p


0 771***
(025)
0 442***
(062)
0 364***
(116)
0 096***
(027)
0561***
(064)
-0 007
(034)
0 169*
(089)
-0 179***
(035)
0 072***
(012)
0 500***
(061)
4 765***
(589)
-0 054**
(027)
-0 015***
(005)
-0 012
(017)
0001
(029)
-0004
(007)
























-0 150**
(070)















0 109*
(064)
-4 423***
(1 193)
306
128429
21833 984
<0 10, ** p<0 05, *** p<0 01


0781**
(023)
0 403**
(064)
0 374**"
(112)
0 108**
(027)
0561**
(063)
0 009
(033)
0 157*
(088)
-0 145**'
(037)
0 074**"
(011)
0 496**
(061)
4480**"
(582)
-0 052*
(0026)
-0 014**
(005)
-0010
(017)
0001
(027)
-0004
(007)
-0 110**
(047)
0006
(012)



















0001
(052)


0 803***
(035)
0 409***
(066)
0315**4
(114)
0 058*
(034)
0 522***
(064)
-0 013
(035)
0 143
(092)
-0 186**'
(039)
0 066***
(013)
0 466***
(064)
4486***
(563)
-0 051**
(025)
-0 007
(005)
-0014
(017)
0047
(030)
-0 005
(007)





-0 094*
(052)
-0081
(087)
-0 013
(0013)
0029
(0018)











-0 142
(089)


0 708**"
(018)
0 395**
(034)
0381**
(068)
0 103**
(028)
0 534**
(063)
-0 008
(033)
0 131
(089)
-0 168**
(037)
0 074**"
(012)
0 469**
(062)
4 830**"
(602)

























-0 039
(033)
-0 106**
(033)
-0 150**
(034)
-0 268**
(035)
-0 303**
(036)
-0 200**
(052)


0 739**)
(017)
0 348**)
(035)
0 401**)
(069)
0 102**)
(030)
0525**)
(066)
-0 003
(034)
0 099
(092)
-0 139**
(039)
0 073**)
(013)
0 462**)
(0 064)
4 465**)
(611)

























0049
(183)
0048
(294)
0029
(346)
-0 075
(355)
-0 137
(337)
-0 581
(367)

-0 100**
(027)
-0 140
(269)
0015
(032)




0649
(577)


0606 -0063 0131**
(761) (102) (051)
-15 104 -3 590*** -4 502***
(14 702) (1 029) (956)
306 306 306 306 306
13629 144157 120077 12498343 13431818
23964368 17857736 2153549 28332588 17974398
Standard deviation in parenthesis Coefficients on translog squared and interaction terms are omitted


0 756***
(017)
0 366***
(034)
0 480***
(069)
0 080**
(028)
0516***
(063)
0 009
(033)
0 145*
(087)
-0 185**'
(040)
0 073***
(012)
0 465***
(061)
4601***
(583)

























-0 044
(032)
-0 102**'
(033)
-0 147**'
(033)
-0 273**'
(036)
-0 308**'
(035)
-0 437)**
(105)







-0 155**'
(032)
0022
(036)
0359***
(102)












Table 3-3. Elasticities
At variables' median
Mean Std. Deviation Range values
Output 0.7077 0.0227 [.6394 .7530] 0.7092

Labor Price 0.3908 0.0602 [.2335, .5319] 0.3962

Materials'Price 0.3943 0.3246 [-.2229, 1.0662] 0.3911

Time -0.0655 0.0300 [-.1331, -.0058 ] -0.0694


Table 3-4. Technological Change by Year and Ownership

1998 1999 2000 2001 2002 2003 1998-2003 Geom. Mean

Overall 5.31% 5.75% 6.35% 6.58% 7.39% 7.83% 46.2% 6.54%

Public 5.62% 6.05% 6.66% 7.01% 7.75% 8.15% 49.0% 6.87%

Private 5.17% 5.62% 6.21% 6.37% 7.23% 7.69% 44.9% 6.38%

Privatized 7.37% 7.68% 8.36% 8.59% 9.37% 9.82% 63.4% 8.53%

Always Private 2.46% 2.89% 3.43% 3.81% 4.48% 4.96% 24.1% 3.67%











Table 3-5. Efficiency Evolution
1998 1999 2000 2001 2002 2003
Mean Inefficiency Score 1.1279 1.1206 1.1205 1.1133 1.1134 1.1109
(.0999) (.0917) (.0917) (.0722) (.0837) (.0697)
Public Firms'Mean Inefficiency' 1.1384 1.1513 1.1499 1.1342 1.1533 1.1344
(.1161) (.1422) (.1378) (.1145) (.1401) (.1125)
Private Firms'Mean Inefficiency 1.1229 1.1065 1.1071 1.1034 1.0956 1.1004
(.0928) (.0529) (.0583) (.0383) (.0272) (.0358)
PrivatizedFirms'MeanIneff2 1.1516 1.1258 1.1240 1.1095 1.0962 1.0873
(.1148) (.0616) (.0693) (.0415) (.0268) (.0261)
"Always Private" Firms'Mean Lin 1.0866 1.0809 1.0847 1.0958 1.0948 1.1189
(.0285) (.0201) (.0281) (.0335) (.0287) (.0400)
Mean Relative Inefficiency 1.0865 1.0738 1.0727 1.0645 1.0569 1.0629
(.0963) (.0879) (.0878) (.0690) (.0794) (.0666)
Public Firms'Mean Rel. Ineff 1.0967 1.1032 1.1008 1.0845 1.0948 1.0854
(.1119) (.1363) (.1319) (.1094) (.1330) (.1077)
Private Firms'Mean Rel. Ineff 1.0818 1.0603 1.0599 1.0551 1.0400 1.0529
(.0894) (.0507) (.0558) (.0366) (.0258) (.0342)
PrivatizedFirms'MeanRel. 1,,. n 1.1094 1.0788 1.0760 1.0609 1.0406 1.0403
(.1106) (.0591) (.0663) (.0397) (.0254) (.0249)
"Always Private" Firms' Mean Rel. 1,, n 1.0467 1.0357 1.0384 1.0478 1.0393 1.0705
(.0275) (.0193) (.0269) (.0320) (.0273) (.0383)
Standard deviation in parenthesis.
1. Includes only the 16 firms that remained publicly owned during the whole period
2. Firms privatized from 1995 to 2003
3. Firms already private in 1995


Table 3-6. Decomposition of Privatized Firms' Efficiency Evolution
1998 1999 2000 2001 2002 2003
Mean. Ineff Privatized Before1 1.1418 1.1175 1.1244 1.1105 1.0948 1.0872
(.1143) (.0649) (.0811) (.0410) (.0250) (.0287)
Mean Ineff PrivatizedAfter2 1.1790 1.1451 1.1229 1.1066 1.0988 1.0875
(.1249) (.0533) (.0338) (.0478) (.0320) (.0218)
Mean Rel. Ineff Privatized Before' 1.1000 1.0708 1.0764 1.0619 1.0393 1.0402
(.1101) (.0622) (.0776) (.0392) (.0237) (.0275)
Mean Rel. Ineff PrivatizedAfter2 1.1358 1.0973 1.0750 1.0582 1.0431 1.0405
(.1204) (.0511) (.0323) (.0457) (.0304) (.0209)
Standard deviation in parenthesis.
1. Firms privatized from 1995 to June/1998
2. Firms privatized from June/1998 to 2003











Table 3-7. Productivity Growth Rate and Decomposititon
Malmquist Index Technological Change
(Frontier Shift)


Technical Efficiency Change
(Catch-up)


1998 1999 6.04% 5.50% 0.51%

1999 2000 6.03% 6.05% -0.02%

2000 2001 6.94% 6.42% 0.46%

2001 2002 6.88% 6.92% -0.04%

2002 2003 7.77% 7.61% 0.11%

1998 2003 38.50% 37.00% 1.01%

Geometric Mean 6.73% 6.50% 0.20%








Table 3-8. Productivity Growth Rate and Decomposition by Ownership Type

Malmquist Index Technological Change Technical Efficiency Change
(Frontier Shift) (Catch-up)

1998 1999 Public 4.67% 5.83% -1.06%
Private 6.69% 5.35% 1.24%
Privatized 9.47% 7.45% 1.84%
Always Private 3.18% 2.68% 0.49%
1999 2000 Public 6.43% 6.35% 0.08%
Private 5.84% 5.91% -0.07%
-Privatized 8.13% 7.97% 0.14%
Always Private 2.80% 3.16% -0.35%
2000 2001 Public 8.11% 6.83% 1.16%
Private 6.40% 6.23% 0.12%
Privatized 9.43% 8.28% 1.03%
Always Private 2.56% 3.62% -1.03%
2001 2002 Public 5.62% 7.38% -1.60%
Private 7.47% 6.70% 0.70%
-Privatized 10.03% 8.73% 1.20%
Always Private 4.22% 4.14% 0.07%
2002 2003 Public 9.45% 7.95% 1.36%
Private 7.03% 7.46% -0.44%
-Privatized 10.29% 9.42% 0.80%
Always Private 2.45% 4.72% -2.18%
1998 2003 Public 39.24% 39.38% -0.10%
Private 38.19% 35.90% 1.55%
-Privatized 57.19% 49.46% 5.10%
-Always Private 16.14% 19.70% -2.99%
Geometric Mean Public 6.84% 6.87% -0.02%
Private 6.68% 6.33% 0.31%
Privatized 9.47% 8.37% 1.00%
Always Private 3.04% 3.66% -0.60%











Table 3-9. Productivity Growth Rate and Decomposition by Ownership Type Firms with Q >
400,000 MWh/year
Malmquist Index Technological Change Technical Efficiency Change
(Frontier Shift) (Catch-up)
1998 1999 Overall -2.25% 1.31% -3.53%
-Public -8.29% 0.58% -8.76%
-Privatized 1.99% 1.26% 0.68%
Always Private -3.92% 2.82% -6.55%
1999 2000 Overall 10.01% 4.39% 5.38%
-Public 9.94% 3.55% 6.15%
-Privatized 10.87% 4.43% 6.16%
Always Private 7.40% 5.77% 1.52%
2000 2001 Overall 7.50% 7.17% 0.33%
-Public 10.08% 6.38% 3.41%
-Privatized 8.71% 7.27% 1.36%
Always Private -0.83% 8.29% -8.38%
2001 2002 Overall 7.73% 9.81% -1.92%
-Public -0.58% 8.92% -8.70%
-Privatized 10.99% 10.06% 0.84%
Always Private 13.17% 10.71% 2.22%
2002 2003 Overall 4.84% 12.47% -6.76%
-Public 6.09% 11.38% -4.69%
-Privatized 7.36% 12.80% -4.82%
Always Private -5.85% 13.36% -16.99%
1998 2003 Overall 30.56% 39.98% -4.14%
-Public 17.06% 34.42% -12.84%
-Privatized 46.48% 40.84% 3.97%
Always Private 9.04% 47.79% -26.26%
Geometric Mean Overall 5.48% 6.96% -0.84%
-Public 3.20% 6.09% -2.71%
-Privatized 7.93% 7.09% 0.78%
Always Private 1.75% 8.13% -5.91%



Table 3-10. Mean Service Quality Indexes
1999 2000 2001 2002 2003
DEC' 23.538 20.090 17.783 18.761 16.724

FEC2 23.222 21.033 18.519 18.240 15.057

Quality 0.062 0.074 0.079 0.074 0.092

Quality by ownership:
Public 0.045 0.048 0.055 0.054 0.061
Privatized 0.060 0.072 0.079 0.077 0.093
Always Private 0.082 0.106 0.106 0.090 0.125
1. Average number of days of service interruption within a year
2. Average number of interruptions within a year
3. Inverse of the average of DEC and FEC











Table 3-11. Vertical and Quality as additional regressors or as mean inefficiency parameters
Variable A B C D
Coeff. Std.Dev. Coeff. Std.Dev. Coeff. Std.Dev. Coeff. Std.Dev.


Lnopex
Lnq
Lnlp
Inmp
Inqxlnlp
Inqxlnmp
Inlpxlnmp
Inqsq
Inlpsq
Inmpsq
cap
len
d2000
d2001
d2002
d2003
Inind
Inresden
Inincome
Inarea
Incusden
undergrd
vertical
Inqlt
Inqltsq
privtzed
alwspriv
lnqlt*privtz
lnqlt*alwspr
Cons
mu
vertical
quality
qltsq
privtzed
alwspriv
qlt*privtz
qlt*alwspr
Cons
lnsig2v
Cons
lnsig2u
q
Cons
Statistics
N
11
chi2
legend:


0.708*** 0.022
0.369*** 0.040
0.392*** 0.073
0.037 0.035
-0.085* 0.046
-0.182 0.138
-0.013 0.017
0.019 0.127
-0.576 0.460
0.072** 0.031
0.581*** 0.073
-0.061* 0.032
-0.096*** 0.036
-0.203*** 0.037
-0.238*** 0.040
0.004 0.039
0.178 0.109
-0.144*** 0.042
0.072*** 0.015
0.501*** 0.070
5.004*** 0.636
0.030 0.028
-0.020 0.028
0.027 0.024





-0.275*** 0.052


0.049 0.031 0.056* 0.029
-0.096 0.085 0.031 0.081
0.030 0.026 -0.017 0.028
0.064 0.065
-0.006 0.053
-0.060*** 0.020
0.013 0.008
0.346** 0.143 0.280*** 0.063


-4.082*** 0.291 -3.914*** 0.088

0.127** 0.055 1.553 1.700
-4.312*** 0.880 -35.512 33.250


255 255
111.86 137.21
18977.36 29613
* p<0.10; ** p<0.05; *** p<0.01


255
108.17
23989.15


255
142.37
19865.29


0.757*** 0.019 0.727*** 0.020 0.766*** 0.018
0.394*** 0.039 0.369*** 0.040 0.352*** 0.037
0.449*** 0.072 0.389*** 0.077 0.483*** 0.069
0.026 0.031 0.016 0.035 0.030 0.030
-0.144*** 0.044 -0.095** 0.046 -0.126*** 0.040
-0.038 0.134 -0.145 0.143 0.003 0.133
0.016 0.013 0.018 0.014 0.024* 0.013
-0.007 0.117 0.098 0.130 0.069 0.117
-0.286 0.441 -0.535 0.477 -0.339 0.431
0.062** 0.030 0.052 0.032 0.044 0.029
0.550*** 0.068 0.545*** 0.074 0.541*** 0.070
-0.057** 0.029 -0.061* 0.032 -0.040 0.029
-0.087*** 0.032 -0.101*** 0.036 -0.068** 0.032
-0.211*** 0.034 -0.210*** 0.037 -0.188*** 0.034
-0.233*** 0.036 -0.253*** 0.040 -0.223*** 0.036
-0.008 0.036 -0.008 0.039 -0.001 0.037
0.157 0.099 0.142 0.111 0.170 0.106
-0.141*** 0.043 -0.151*** 0.045 -0.177*** 0.043
0.066*** 0.014 0.077*** 0.015 0.072*** 0.013
0.482*** 0.064 0.467*** 0.071 0.501*** 0.067
4.370*** 0.594 4.900*** 0.665 4.549*** 0.564
0.007 0.025
0.107*** 0.040
0.058** 0.028
-0.155*** 0.028
0.012 0.040
-0.283*** 0.057
-0.150** 0.060
-0.125*** 0.041 -0.485*** 0.136 -0.499*** 0.047









CHAPTER 4
THE ASSESSMENT OF FIRM'S EFFICIENCY IN PERIODIC TARIFF REVIEWS: AN
EVALUATION OF THE REFERENCE UTILITY APPROACH

Introduction

One of the main tasks in the implementation of a price-cap regime resides in the

establishment of cost-based prices at the scheduled tariff reviews, where the regulator faces

imperfect and asymmetric information regarding firms' cost opportunities. A social welfare

maximizing regulator would face pressures from customers and utility investors, leading to

decisions that are more likely to balance the conflicting interests of powerful stakeholders (so

rulings are likely to reflect the political economy of regulation). Price-caps provide

incentives for efficiency improvements:1 at the rate review, the regulator's intention to extract

part of the firms' rents for the benefit of consumers and society has to be balanced against the

objectives of promoting (1) allocative efficiency (prices that reflect minimum incremental costs),

(2) financial sustainability (meeting each firm's break-even constraint), and (3) further

productivity gains (through strong incentives for cost containment).

A form of yardstick regulation which has been used to tackle the cost benchmark issue

consists of a bottom-up efficiency study based on the engineering knowledge of the industry

process.2 In the model company (or reference utility) approach, prices are set on the basis of the

estimated costs of a hypothetical efficient firm facing the same operating conditions of the

concessionary under the review process. As future prices are not linked to realized costs, the

method has the merit of preserving the efficiency improvement incentives brought by the price-


1 Under a price-cap regime, prices are fixed. The firm and its managers are the residual claimants on production cost
reductions, and bear the disutility of increased managerial effort. It is thus assumed that the conditions and
incentives for efficiency improvement and for the possible achievement of second best prices are settled (Joskow,
2005).
2 The model company approach has been employed to calculate electricity distribution tariffs in Spain and some
Latin-American countries, mainly Chile, Peru, Argentina, El Salvador, and Brazil (Jadresic, 2002).









cap regime.3 4 Other possible advantages include the control for heterogeneity in operating

conditions and the fact that the regulator does not need to base its decisions on cost information

provided by firms.

The approach's usage, however, is not fully endorsed in the literature. Weisman (2000)

asserts that the estimation of efficient costs is an untenable target, given the existing

informational asymmetry between the regulated firm and the regulatory agency (or the

consultants hired to perform the task), and argues that it represents a major retrogression from

the price-cap approach, as the posited efficiency gains need have no foundation in actual market

behavior. Gomez-Lobo and Vargas (2001), on their turn, claim that the method is excessively

detailed, time-consuming, resource intensive and contributes negatively to the transparency and

objectivity of the regulatory process.

It is therefore important to investigate whether the use of the model company methodology

has effectively enabled the attainment of the aforementioned regulator's objectives. The evidence

so far is limited. Serra (2002), and Fisher and Serra (2002) look at the experience in the

telecommunications and electricity distribution sectors in Chile, and consider the findings that

the method's usage led to rate of returns well above the firms' cost of capital as an indication of

a persistent regulatory flaw. Grifell-Tatje and Lovell (2001), on the other hand, examine the

issue in the context of the electricity distribution in Spain and find that the engineering model

was much less costly to operate than the real companies, by virtue of a smaller network and



3 When a price-cap plan links future prices directly to realized costs and the time between schedule reviews is
relatively short, the incentives under a price-cap regime are similar to the ones under rate of return regulation
(Sappington, 2002).
4 Under a model company approach, the efficiency improvement incentives come from the fact that firms
appropriate rents when their actual costs are inferior to the estimated efficient operating costs.
5 See Galetovic and Bustos (2002), and ANEEL (2003).









lower input prices, but did not have their inputs allocated in a cost-efficient manner. The study

indicates that actual companies were more cost efficient than the hypothetical efficient firms, and

concludes that the engineering procedure had understated potential cost savings by nearly one-

third.

The present study contributes to fill the literature gap by analyzing the results obtained

with the use of the model company approach to estimate efficient operating expenditures in the

Brazilian electricity distribution industry periodic tariff review. The implied performance scores

are compared to those obtained using alternative methodologies-Statistical Frontier Analysis

(SFA) and Data Envelopment Analysis (DEA)-under the rationale that some minor divergences

might result from problems in these other methods, but greater discrepancies could reflect

deficiencies in the application of the engineering approach, particularly when they are

independent of the method employed for comparison.

The Brazilian case provides an exceptional opportunity to perform the investigation, as the

number of distribution companies allows the use of sophisticated comparative efficiency

techniques and the consequent computation of efficiency scores and analysis of their evolution

over time. Thus, efficiency estimates and measures of firms' productivity improvements

obtained in a previous benchmarking study portraying the distribution firms' performance in the

period of six years immediately before the tariff review (Silva, 2006b) are employed to examine

the results provided by the engineering approach. Particular attention is given to the degree of

consistency in efficiency estimates and rankings provided by the two methods, the procedure

adopted for firms which experienced the highest-and the lowest-productivity gains in the

period before the review and to the possibility that the regulator's decision might have threatened

the firms' financial sustainability.









In sequence, the study checks for the possible causes of the divergences found. At this

point, the investigation explicitly recognizes that regulatory decisions are taken by a utility

maximizing regulator that operates in a situation of asymmetric information. It is also considered

that the regulator has opportunities to exercise discretion, is potentially influenced by interest

groups, and is subject to direct supervision of its actions. Consequently, the analysis of

regulatory outcomes addresses the possible impact of these factors, in addition to the effects of

the methodology employed.

The study presents evidence that the monitoring of the regulator's activities does not lead

to decisions contrary to the concessionaires' interests. In addition, the investigation shows that

the aforementioned regulator's objectives at the rate review might not have been accomplished in

some situations where the firms' efficiency assessments differed markedly from the ones

suggested by the economic benchmarking approaches. On the one hand, the results indicate that

some firms, mainly the ones serving more affluent consumers, operating in more densely

populated areas and having a lower proportion of electricity delivered to industrial customers,

received substantially lower repositioning indexes than the economic benchmarking methods

would recommend. As a low repositioning index basically serves as a price adjustment that

reduces allowed revenues, the evidence points to a possible violation of firms' break-even

constraints. On the other hand, the findings reveal that significantly higher repositioning indexes

might have been given to companies with the opposite characteristics: managers of firms granted

higher allowed prices can see that cash flows are enhanced, even when efficiency (measured

using other techniques) is not high compared with the performance of other firms. Usually,

incentive systems give weaker-performing firms lower prices since there is scope for efficiency

improvements. Some of the companies benefiting from ANEEL's use of the repositioning index









based on engineering models do not appear in the top ten of the economic benchmarking

efficiency rankings, so weaker performers seem to be rewarded.

The following section describes the methodology adopted by the regulator in the periodic

tariff review and presents the resulting figures obtained. Section 3 explains the methodology and

the data set employed to perform the stochastic frontier approach, presents the corresponding

results, and explores their use to examine the regulator's decisions taken on the basis of the

engineering approach. Section 4 describes the econometric model and presents and interprets the

results. Section 5 explains the robustness check performed with the use of the DEA

methodology. The final section provides concluding observations.

Institutional Background and the ANEEL Model Company Method

The power sector reforms in Brazil began in 1995. While constitutional amendments

abolished the public monopoly over infrastructure industries and allowed foreign companies to

bid for public concessions, the Law 8,987/95 (General Law of Concessions) set the stage for the

beginning of the privatization process, represented by the auctions of Escelsa in 1995 and Light

in 1996. By the end of 2000, a total of 20 distribution companies had been privatized.

In addition, part of the implementation of a new regulatory framework involved the

establishment of an independent regulatory agency (ANEEL) in late 1996 and, in the same year,

the commission of an international consultancy to study and propose a new model for the

electricity sector. The consultant's report was released in 1997, and its proposals were

incorporated into Law 9,648, issued on May of 1998.6 One of the measures introduced by the

approved model was the use of the price-cap regime to regulate distribution tariffs, replacing the

previous cost-of-service system. Price-cap regulation was implemented through the signature of

6 See Ferreira (2000), Mota (2003), and de Oliveira (2003), for detailed descriptions of the new model's
characteristics.









new concession contracts, which took place from 1998 to 2000, and scheduled the first tariff

review for after five (for contracts signed in 1998) or four years. As a result, 61 companies were

submitted to a tariff review process from April 2003 to February 2006.

The Tariff Review Methodology

Despite defining the first regulatory lag, the concession contracts were silent about the cost

methodology that would be applied at the periodic review. The obligation to solve the

methodological vacuum rested with the regulatory agency, which, after a long and heated

discussion process,8 established that the repositioning index for firm i (RIi) would be calculated

as follows:


RR, RS ER, OR
RI R
RDi


(4-1)


where RRi corresponds to the revenue requirement for firm i in the 12-month period after the

tariff review date, RSi stands for firm i's revenue from supply activities, ER, and ORi denote

extra-concession and other revenues of firm i projected for the same period, respectively, and

RDi is the revenue from distribution activities that firm i would obtain if distribution tariffs were

kept the same. The repositioning index represented the percentage increase which would be

applied to firm's tariffs at the rate review. In this case, a higher revenue requirement implies a

higher repositioning index and, consequently, higher prices.



7 Companies Escelsa and Light constituted an exception to this rule. Light was the first to have price-cap regulation
applied, by order of the concession contract signed in November 1996, in which the first tariff review was scheduled
to occur after seven years. Escelsa was submitted to price-cap regulation in August of 1998, and had tariff reviews
every three years thereafter. Except for Escelsa, all companies had the X factor set equal to zero in the first period
prior to the first full review.

8 Peano (2005) provides a detailed description of the process implemented by the regulatory agency to define the
tariff review methodology. Foster and Antmann (" '1 4), and Byatt (2i 4), in turn, discuss the particularities of the
deep controversy that surrounded the regulator's choice of the asset valuation methodology.


i = 1, ... n









The revenue requirement was defined as the revenue needed to cover efficient operating

costs and to provide an adequate return over investments prudently made, corresponding to the

following:

RR = EC +TC +OC +ROC +DEP +T (4-2)

where EC and TC are considered non-controllable costs and stand for the projected costs of

buying energy and paying the tariff charges, respectively, OC is operating costs, ROC is the

return on capital (in monetary units), DEP is depreciation, and T is the firm's taxes.

The return on capital was obtained through the application of a rate of return of 17.07% on

a rate base computed under a Depreciated Optimized Replacement Cost (DORC) methodology.

The operating costs allowed to be passed on to tariffs, on their turn, were given by the sum of the

costs estimated for administration (ADM), commercialization (COM), and operation and

maintenance (O&M) activities performed by a hypothetical efficient firm facing the same

operating conditions of the concessionary under exam. The estimated rate base and efficient

operating costs were, therefore, the main determinants of the authorized price increases at the

rate review.

The methodology employed to come up with the operating costs figures consisted in

determining, for each firm, an optimal organizational structure which would allow the

concessionary to efficiently fulfill its goal of effectively delivering electricity at the required

service quality levels. In order to estimate the costs associated to COM and O&M activities, all

processes and activities (P&A) which the reference utility should perform were identified, along

with the human and equipment resources needed to carry out each P&A. The process/activity's

cost was calculated considering the respective frequency of occurrence, by valuing the required

resources at market prices. Then, the final cost estimates resulted from the application of each









P&A cost to the firm's volume of capital (for O&M) and number of customers (for COM). The

ADM costs, on their turn, were estimated on the basis of firms' volume of capital, number of

customers, and geographical dispersion, with the identified amount of human and equipment

resources required also being valued at market prices.

The tariff review process also included the definition of the X-factor for the time period for

which the formula would be in effect. The analysis of the corresponding methodology employed

is beyond the scope of the present study, but it is important to stress, in this respect, that the X-

factor comprised the productivity gains estimated for the period up to the next tariff review (Xe),

adjusted by a quality factor (Xc), with Xe being given by scale economies resulting from

projected increases in output, net of investments required to meet expected increases in demand.

The rationale, here, was that firms' operating costs were already adjusted to their efficient levels

(with the use of the engineering approach), and no further significant technical efficiency

improvements should be observed thereafter.

It is interesting, at this point, to look at the ANEEL's methodology in the context of an

efficient frontier framework. In theory, the efficient operating cost provided by the engineering

approach should correspond to the point in the efficient frontier associated to the firm under

exam. Thus, if the approach employed effectively enabled the regulator to figure out the firms'

efficiency targets, in spite of the information asymmetry, it follows that the methodology at the

periodic tariff review not only determined the firms' one-time adjustment on their operating

costs, but also assumed that there would not be any frontier shifts in the future. In that case, the

non-adoption of a progressive path towards the efficient target raises concerns over the financial

sustainability constraint of those firms which the regulator's approach revealed as the most

inefficient. Note, here, that the situation gets worse if the model company resulting figure does









not correspond to the firm's efficiency target, and the target is set at a level that is excessively

(and unreasonably) high.

On this respect, it should be noted that the estimated efficient operating costs might not

match the "true" values. As stated before, the engineering model is very detailed and provides

estimates for each process/activity performed by a virtual efficient firm operating in the same

conditions than the company under exam. Given the asymmetric information context, the cost

parameters needed to perform the task are not only difficult to be precisely estimated (taking into

account the specificities of each firm's operating conditions), but also are subject to firm's

misreporting. Consequently, it is indeed possible that the parameters employed to estimate the

efficient costs do not satisfactorily capture the effect of some cost drivers on firms' actual costs.9

Model Company Estimates

The analysis of the results provided by the engineering approach was limited by

availability of data and by the decision to exclude some very small utilities, which deliver less

than 100,000 MWh per year. From the 61 companies subjected to a tariff review process from

April 2003 to February 2006, nine were dropped from the sample due to small size; data for three

others were unavailable. Therefore, the sample includes 49 companies, responsible for 99.24% of

the total electricity delivered in the country in year 2003.

The operating costs estimated for the hypothetical efficient firms are shown in Table 4-1,

along with the realized operating costs reported by the concessionaries and the computed

regulator's efficiency index (ANEELEFF), given by the ratio realized OPEX to estimated OPEX.

Here, two points are worth noting: the wide range observed in the regulator's efficiency indexes



9 This possibility was augmented in the Brazilian experience under exam, as the method and the corresponding
parameters employed were used for the first time and not previously debated with the distribution companies before
the beginning of the tariff review process, increasing the chance of "errors."









and the fact that, under the regulator's view, seven firms were operating more efficiently than the

virtual efficient company.

The variable ANEELEFF varies in the range of 0.848 to 1.986, with mean 1.202. The fact

that the 50% percentile is at 1.180 denotes that the distribution of regulator's efficiency indexes

is skewed to the right, and the mean has been moved up by a few instances where the estimated

efficient operating costs are well below realized costs. It is useful to check whether these firms

identified as highly inefficient using this engineering "Model Company" methodology also show

up in the worst performers grouping of SFA and DEA efficiency rankings.

Note that the engineering method resulted in the allocation of some rents to the seven

companies which had repositioning indexes based on operating costs higher than their realized

costs. 10 In such cases, the comparison with the results provided by the economic benchmarking

approaches takes on special relevance because only the most efficient firms should be allowed to

keep part of their productivity gains, as an incentive for further efficiency improvements. If these

seven firms are not ranked highly by alternative methodologies, the procedure is called into

question. The finding would indicate that the higher tariffs given to these firms constituted an

unjustified benefit, to the detriment of customers and without regard to the efficiency

improvement incentives embedded in the price-cap regime.

Comparative Efficiency Analysis

In comparative efficiency studies, a firm's efficiency is given by a measure of the distance

of the observed practice to the efficient frontier, with the frontier estimation being implemented

10 According to the repositioning index formulae (equations 4-1 and 4-2), the allowed risk-adjusted rate of return of
17.07% would be given to firms operating at the model company's efficient operating costs levels (at the "efficient
frontier"). It follows that returns below 17.07% were assigned to all firms whose estimated efficient costs were
below their actual costs (ANEELEFF >1), with the rate of return being smaller, the greater was the firm's distance to
the regulator's estimated frontier (the greater ANEELEFF was from unity). On the other hand, returns above the
17.07% standard could be earned by the seven firms whose estimated efficient costs were higher than actual
operating costs.









with either a parametric or a non-parametric technique. Non-parametric methods, like Data

Envelopment Analysis (DEA), use mathematical programming techniques and neither require

the specification of production or cost functions nor the imposition of behavioral assumptions.

These methods are generally easy to implement, but carry an implicit restriction in the number of

variables that might be used. Furthermore, they do not allow for random shocks.

Parametric methods, in turn, entail applying an apriori functional form to the frontier,

estimated with econometric tools. They allow for hypothesis testing," enabling the analyst to

investigate the validity of the model specification. Tests of significance can be performed for the

functional form and for the inclusion or exclusion of factors, which are of special relevance for

the electricity distribution industry, where the inclusion of several factors is theoretically

justifiable. Moreover, with a parametric method it is possible to allow for stochastic factors or

measurement errors, which avoids the assumption that all deviations from the best practice

frontier involve inefficiencies. For instance, with Stochastic Frontier Analysis (SFA) a mix of

one-sided and two-sided error terms is employed, with the former capturing the firm's

inefficiency and the latter capturing the effects of random variation in the operating environment.

Ideally, the decision regarding the appropriate method depends on the purposes of the

study and the context under examination. In case, we are interested in investigating the evolution

of efficiency from 1998 to 2003; the investigation is conducted in an environment where random

shocks were present and the inclusion of several variables in the model specification, besides

being theoretically justifiable, is advisable due to the great heterogeneity in operating conditions.

These considerations suggest the use of a stochastic frontier approach, defined in terms of an



11 In non-parametric models, a bootstrap technique may be used to produce confidence intervals around the
estimated individual efficiency and thereby assess statistical properties of the efficiency scores generated (Simar and
Wilson, 1998).









input orientation, given the output exogeneity that characterizes the electricity distribution

industry.

SFA Model and Data

The SFA model employed here is detailed in Silva (2006b). It is based on an unbalanced

panel of 52 companies, responsible for 99.47% of the total electricity delivered in the country in

year 2003, with the data being collected for the period of 1998 to 2003. The model employs a

variable cost specification, reflecting the fact that transformer capacity and network length

constitute capital inputs that are fixed in the short run. Environmental variables are included as

arguments of the variable cost function, instead of as mean inefficiency parameters, to control for

differences in firms' operating conditions; this approach reflects the interest in having efficiency

measures net of factors that impact firms' performance but are beyond the control of the

concessionaries. In addition, in light of the rejection of the null hypothesis of homoskedasticity,

the variance of the inefficiency error component is conditioned on a proxy of firm size, given by

total electricity delivered (Q).

The specification adopted is then the following:

1 1
lnE, =/0 +fy lny, +/,, Inw,,+ +-/Y (lny,, ) +- fnk lnm lnwk, +2 lny Iln~
n 2 2n k n
(4-3)
+/, lnCap, +/,1 InLe n + /,j InZ,,, + jt +/t In y, t + Cn, Inwn,,t +- ,,tt +vt +ut
J n 2

where E and y are the cost and output measures, respectively, w is the vector of factor prices,

Cap stands for transformer capacity, Len represents network length, Z is the vector of

environmental variables, and it is assumed that v,, N(0, o- ) and u,, N+(0, a, ), with a,,

specified as

o-,t = T0 + PQQlt









The modeling of technical change in the way shown in equation 4-3 attempts to obtain

evidence of technological change over the period considered. For the computation of firms'

efficiency indexes and the consequent analysis of efficiency change, however, the study turns to

the use of time-fixed effects, to control for possible changes in macroeconomic factors which

might have affected firms' performance during the period under investigation. It is worth noting

that the use of time-fixed effects explicitly allows the computation of efficiency indexes relative

to yearly-specific frontiers.

The observed technological change (ATC) and technical efficiency change (ATE) are then

combined to provide a more complete picture of the productivity improvements which occurred

in the period under examination, through the computation of Malmquist productivity indices. For

each firm, the Malmquist index of productivity change between two consecutive periods is given

by

MIJ = ATE, ATCJ

where

Eff.Index \/\
ATE= -\\_- -1 +1 and ATC= (1+TC ,).(1+TC )}1/2
Eff.Index,t \J

The dependent variable is given by the operating costs of distribution and retail service

activities (Opex), computed as the sum of labor, materials and third party service contracts

expenses, as reported in the income statement.12,13 Electricity delivered, in MWh (Q), is the



12 The computed labor expenses include firms' contributions to pension funds and to health insurance plans, profit
sharing payments, and management wages. Some firms already report these expenses under the classification of
labor expenses, but most of them do not. The necessary adjustments were made on these cases.
13 In case of vertically integrated companies, the computation of the operating costs of distribution and retail service
activities was made possible by the fact that those companies are required by law to report their expenses separated
by activity.










output measure14 and average wage, calculated as total labor expenditure divided by the number

of employees, is used as a proxy for the price of labor (LP). 15 For the prices of materials (MP)

and third party services (SP), the work uses two price indexes provided by Brazilian Institute of

Statistics (IBGE), and the Caixa Economica Federal (CEF), a public financial institution that is

in charge of most of the social programs of the federal government and provides financing to

housing construction projects. The materials' price index reflects the observed change in the

price of a basket of items used in civil construction, by state, while the third party service's index

portrays the observed change in the salaries paid to an electrician, also by state. The variables

Opex, LP, MP, and SP are expressed in 1998 values, being deflated by a general price index

(IGP-DI).

Transformer capacity is given in MVA, and network length corresponds to the sum of

high-voltage and low-voltage lines, in kilometers. 16 The environmental variables incorporated in

the modeling are the following: customer density (CusDen), given by number of customers

divided by network length; share of electricity delivered to industrial customers (IndShare);

residential density (ResDen), computed as electricity delivered to residential customers divided

by the number of residential customers; service area7 (Area), in Km2; ratio of underground to


'4 The use of two measures of output was prevented by the fact that electricity delivered and number of customers
showed up as highly collinear, with one of them always being dropped by the statistical software employed (Stata).
A better specification was provided by the former, when compared to the latter.
15 Total labor expenditure is employed to compute average wage because it was not possible to obtain information
related to number of employees segregated by sector activity, for the cases of firms that also operate on generation
and transmission.
16 Since these variables showed up as highly correlated with electricity delivered, the variables Cap and Len actually
correspond to the residuals of the regression of transformer capacity on electricity delivered and network length on
electricity delivered, respectively.

17 The Brazilian case justifies the inclusion of both network length and service area in the modeling, and this is
reflected in the statistical significance of both variables as either cost shifters or mean inefficiency parameters.
While some companies have small service areas and relatively high network length (the ones that operate in the
more densely populated states), others have highArea but relatively low Len, because they operate in states which
are more sparsely populated and/or have a high share of the population not being served.









overhead lines (Undergrd); and income per capital, by state (Income), to control for variations in

socio-economic conditions among states. The variables above are included among the most

frequently cost driving factors employed to model electricity distribution, according to Jamasb

and Pollitt (2001) in their survey of the empirical literature on comparative efficiency analysis.

The data were assembled from the regulatory agency, the companies' Web sites, the

financial statements provided to the Sao Paulo Stock Exchange, the Brazilian Association of

Electricity Distribution Companies (ABRADEE), IBGE, and CEF.

SFA Results and Comparison

Descriptive statistics are shown in Table 4-2. The difference between minimum and

maximum values of observations collected for almost all variables employed indicates the

considerable heterogeneity among firms in the sample, in terms of companies' size, system

configuration, and customer structure. The evidenced disparity in firms' indicators corroborates

the need to account for external factors in the comparative efficiency analysis.

Table 4-4 provides the results from the models estimated. The cost function satisfies the

monotonicity condition with respect to output and factor prices at the mean, and the estimated

coefficients have the expected sign, with most of them being significant. The time elasticity

provides a measure of technological change. The evidence shows that there was technological

progress during the sample period, with an annual rate of technological change of around 6.55%,

on average, which denotes that the efficient frontier has shifted considerably from 1998 to 2003.

Firms' efficiency indexes, computed for each year in the period examined, are portrayed in

Appendix B, whereas the Malmquist measures of productivity change are reported in Appendix

C. The results indicate that the Brazilian electricity distribution industry's productivity increased









38.5%, on average, from 1998 to 2003.18 The mean productivity growth rate of 6.73% stands out

when it is compared to the 0.9% mean total factor productivity growth rate of the economy found

by Gomes, Pess6a, and Veloso (2003) for the period of 1992 to 2000.19 Moreover, taking the

results found by Mota (2004) as a proxy for the distribution companies' average productivity

gains in the period of 1994 to 1998 (around 5%),20 it might be concluded that the sector's rate of

productivity growth increased after 1998.

The comparison between the SFA and the Model Company efficiency indexes is limited

to the 49 firms included in Table 4-1. Another restriction comes from the fact that in some cases

the indexes to be compared do not refer to the same period, since ANEELEFF relates to the

month/year the tariff review takes place (April/2003 to November/2005) and our SFA estimates

only go up to 2003. In the analysis that follows, the SFA efficiency indexes obtained for year

2003 are used for comparison (SFA2003), which might introduce distortions to assessments

based on reviews that occurred in late 2004 and in 2005, if the firm performs rather differently

than the others in the period after December/2003.21 The possible distortions, however, should

not be relevant in the context of the present study, as the comparison of efficiency indexes

focuses on the larger discrepancies in the two methods' results.



18 The above mentioned measure of productivity change does not incorporate the scale effect. On the basis of the
computed output elasticity and the actual changes in output from year to year, we have estimated it to be equal to
3.56%, on average, from 1998 to 2003.

19 The authors report that their result is consistent with the findings of two other studies that examined the subject.
20 Mota obtains annual average productivity gains of about 5% in the period of 1994 to 2000 using data only from
these two years and from 14 privatized companies. Considering that the present study provides evidence of
significantly higher privatized firms' productivity gains for the period of 1998 to 2000, Mota's result can be taken as
an upper-bound measure for average productivity gains in the period of 1994 to 1998.
21 Similar distortions may be present for the 17 companies which experienced tariff reviews along the 2003 year. We
examined the issue using for the comparison the average SFA efficiency index for years 2002 and 2003 (SFA0203).
As expected, SFA0203 and SFA2003 were highly correlated (p = .9674). Moreover, the points raised in the analysis
that follows were still present when the SFA0203 efficiency indexes were employed.









The efficiency rankings provided by the SFA and the engineering approaches (Table 4-5)

are not significantly correlated (Spearman's p = 0.0682, p-value = 0.6417). The rankings show

some consistency in terms of the best performers, as Enersul, Coelce, RGE, and CAT-LEO

appear in both top ten extracts. Nonetheless, only Eletropaulo and Jaguari appear in both models

in the bottom ten. Note, in addition, that CEMIG figures in the bottom ten in one method, and in

the top ten in the other.

SFA efficiency estimates are significantly smaller than ANEELEFF,2 varying in the 1.045

to 1.506 interval but concentrated in the 1.045 to 1.127 (75% percentile) range, with mean 1.110.

It follows that the engineering approach has considered firms to be, on average, more inefficient

than indicated by the SFA economic benchmarking technique. The result is not unexpected,

given that one method centers on an ideal context, while the other draws upon actual practice.

Moreover, as the SFA method is based on strong distributional assumptions to disentangle the

effects of inefficiency and random noise, it cannot be ruled out the possibility that some

inefficiency is incorrectly attributed to statistical noise.23 In the present study, however, the

comparison of efficiency indexes is restricted to the cases of higher divergence.

According to the regulator's methodology, Eletropaulo, Light, and CEB are considerably

more inefficient than shown by the benchmarking method, as ANEELEFF exceeds SFA2003 by

0.8305, 0.7604, and 0.6312, respectively. Eletropaulo and Light, however, were the two firms

with the highest productivity improvements in the 1998-2003 period (Appendix C) and,

according to SFA, were not distant from the average performance of other firms, which raises



22 The null of equality of means is rejected at the 1% significance level (p-value (Hi: ANEELEFF > SFA2003) =
.0007).
23 As suggested by the comparison to DEA efficiency measures mentioned later on Section 5, it is also possible that
SFA efficiency indexes are constrained by the half-normal distribution assumed for the inefficiency error term.









serious concerns over their obligation to perform such profound further adjustments24 and points

to the existence of flaws in the application of the engineering approach.25 One possibility is

underscored by the present study. In line with Peano's (2005) claim,26 the results suggest that the

regulator's method might have been biased against firms which operate in more densely

populated areas, since five firms, from the ten which had the highest positive difference between

ANEELEFF and SFA0203, belong to the top ten customer density extract.

The major situations where the implementation of the model company methodology

resulted in firms being considered more efficient than portrayed by SFA were the cases of

Celesc, Coelba, and the seven firms which had ANEELEFF below one (Energipe, Enersul,

Coelce, Cemat, Cemig, Santa Maria and Cat-Leo). The benefit of securing a higher rate of return

over their asset base, given to these nine firms, would only be acceptable if they indeed figured

in the group of best performers and had experienced high productivity improvements in the first

regulatory period, since in this case the regulator would be allowing them to keep part of the

efficiency gains as an incentive for further productivity increments.

Only Cat-Leo, Enersul and Coelce, however, show up in the SFA top ten segment. In the

case of the other six firms, the SFA results indicate that the benefit given was probably

unjustified and harmed customers through higher tariffs. In addition, the rulings did not provide


24 As indicated by the Model Company efficiency index (1.986), Eletropaulo would have to further reduce its
operating expenditures by almost 50% to be able to reap the allowed risk-adjusted rate of return (17.07%).
25 Our previous study of the regulator's performance [Silvai _', In,] had already indicated that the repositioning
index announced at Eletropaulo's tariff review caused a negative surprise in the market, as if it were not a
conceivable number.
26 The author compared the model company's OPEX to actual OPEX of 12 firms submitted to tariff review in year
2003, and identified that the difference between the two measures was inversely related to firm's customer density.
The author argued that the model company method might have been efficient in avoiding reimbursements of over-
investments in more densely populated areas, and allowed an extra return to firms with less densely populated
service areas, which incur higher costs to provide the service at the required quality levels. Peano noticed, however,
that the positive X-factor given at the beginning of the tariff review cycle might impact negatively the incentives for
efficiency improvements.









the appropriate incentives for efficiency improvements in the following regulatory term. The

Cemig's case is emblematic. In spite of the considerable productivity increments in the 1998-

2003 period, the firm still figured as the worst performer according to SFA2003. The fact that the

engineering methodology shows the firm operating rather more efficiently (ANEELEFF -

SFA2003 = -0.5522) may derive from the company's low customer density index, but suggests

the possibility of a differentiated treatment to publicly owned firms, evaluated in the next

section, since Celesc, Celg, CEEE, and Copel also appear in the SFA2003 worst performers'

group but occupy considerably better positions in the Model Company ranking.

It is also worth noting that four out of the five best performers under SFA did not receive

the same benefit of being able to keep part of the rents derived from accomplished productivity

gains. On the contrary, the ANEELEFF indexes of two of them (Manaus and Eletroacre) were

above 1.10, signaling that they would have to further reduce their operating costs by more than

10% to be able to earn the allowed 17.07% rate of return. This apparently tough task, however,

might not be difficult to achieve, as the regulatory scheme set forth for the subsequent regulatory

lag ignored the acute frontier shifts observed in the 1998-2003 period. The expected high

productivity gains brought by technological change should also alleviate the economic situation

of firms which might have been unduly classified as more inefficient. Unfortunately, however,

they will exacerbate the perverse effects of possible over-estimating firms' efficiency.

Econometric Modeling

For each firm, the variable ANEELvsSFA, computed as the ratio of ANEELEFF to

SFA2003, expresses the divergence in the results provided by the two methods. ANEELvsSFA

varies in the range of 0.633 to 1.719, with mean 1.087 (see Table 4-3). The higher the variable is

(relative to unity), the more the firm was considered more inefficient under the regulator's Model

Company approach, when compared to the SFA standard, and the more the firm was harmed by










getting a lower repositioning index (price), assuming that the SFA results are a good

representation of the true values.27 Conversely, when the firm's indicator is smaller (relative to

unity), the firm was considered less inefficient under the regulator's methodology (ANEELEFF<

SFA2003). In this case, the result suggests that the regulator's efficiency index was lower than it

should be. Consequently, the firm, being closer to the efficient frontier than shown by the

economic benchmarking method, was benefited by getting prices higher than recommended by

the SFA standard.28 A ratio of .9, for example, indicates that the firm was considered 10% closer

to the efficient frontier than shown by the SFA model, or that its resulting repositioning index

was augmented by .1 times the operating costs' (OC) participation on the estimated revenue

requirement (see equation 4-2). Logically, the "firm's benefit" gets higher for lower values of

ANEELvsSFA. At the lower bound ratio of 0.633, for instance, the multiplicative factor goes

from the aforementioned .1 to nearly .37.

Differences in the results provided by the Model Company and economic benchmarking

approaches are expected, as the engineering model does not account for substitution possibilities.

Here, though, the analysis focuses on some other possible causes of the observed divergence

between the two indicators. The investigation assumes the existence of a principal-agent

relationship between the Congress (or the Government) and their delegated "representatives" in


27 This assumption draws upon the soundness of the parametric model employed, which controls for heterogeneity in
operating conditions, influence of macroeconomic factors, and random shocks. The rationale, therefore, is that minor
differences in efficiency assessments might be due to eventual SFA inconsistencies, but bigger divergences should
be accounted to problems in the application of the engineering approach. Later in Section 5, in order to ascertain the
validity of this reasoning, the results' consistency is checked using performance indicators provided by an
alternative methodology as the comparison parameter.
28 The relation between the distance to the frontier and the resulting price assigned to the firm is detailed in footnote
10. Since ANEELEFF is given by the ratio of realized OPEX to the Model Company's estimated OPEX, it is
important to stress, at this point, that when the firm shows up closer to the efficient frontier than it should be, it is
because the Model Company's estimated operating costs are overestimated, according to the SFA standard. Under
the same reasoning, the engineering costs would be underestimated in the reverse case where ANEELEFF > 1. The
analysis that follows investigates the possible causes of these over- and underestimations of the Model Company's
operating costs.










regulatory agencies. In this system, interest groups can influence regulatory outcomes. In

addition, information asymmetries raise the possibility that the parameters employed to estimate

the efficient costs do not satisfactorily capture the effect of some cost drivers on firms' actual

expenditures, and affords the regulator some discretion to make choices that maximize its

utility.29

The tariff review process presented two main opportunities for the regulator to exercise its

judgment, possibly reflecting the influence of the industry and its customers: in the definition of

the model company cost parameters and right after the announcement of the efficient operating

cost initial estimates, when deciding upon the acceptance or rejection of firms' revision claims.

Consequently, the investigation follows the line of previous empirical studies which focused on

the determinants of regulatory outcomes, addressing the predictions of the economic theory of

regulation associated with Stigler (1971) and Peltzman (1976) [Nelson, 1982; Primeaux, Filer,

Herren, and Holas, 1984; Naughton, 1989, Nelson and Roberts, 1989; Klein and Sweeney,

1999].

The analysis, however, also accounts for the fact that the regulator's decisions were taken

in the context of incomplete and imperfect information. As a result, the investigation not only

examines the possibility of flaws in the engineering cost parameters employed to estimate the

efficient costs, but also hypothesizes that, in the absence of the information necessary to promote


29 This framework draws upon the contributions of Stigler (1971) and Peltzman (1976), which form the basis of the
economic theory of regulation. The authors posit that stakeholders face costs of organization and information, and
regulators are self-interest maximizers which allocate benefits across interest groups optimally, attempting to equate
political support and opposition at the margin. The authors' contribution helps explain the location of policy in the
competitive price to the monopoly price spectrum. Generally speaking, consumers, being dispersed and having less
at stake, face higher costs of organization than producers and usually do not have the required incentives to spend
the necessary resources to become informed. In case, the prediction is that the producer interest should win the
bidding for the services of a regulatory agency. However, consumers who spend a larger share of their income on a
good have a higher incentive to participate in the regulatory process and should drop more votes for the politician in
response to a price rise. Therefore, goods with a high share in the consumers' budget are more likely to have prices
close to the competitive price.









the desired distribution of productivity gains among stakeholders,30 the regulator might have

employed some of the available data as signals for firms' profitability and cash flow availability.

In addition, the empirical analysis examines a potential external monitoring impact on the

regulator's decisions.

Hence, it is hypothesized that the divergences of the regulator's efficiency measure to the

SFA standard reflect two main factors: (a) possible problems in the cost parameters employed by

the engineering model, due to the difficulties posed by the imperfect information context or to

regulator's decisions; or (b) the regulator's adjustments in the initial operating cost estimate,

made on the basis of its utility function and the available information, in a context of pressure

from interest groups and direct supervision of the regulator's job.

The statistical tests are conducted through two complementary procedures: (a) an OLS

regression of the divergence variable (ANEELvsSFA) on proxies for the explanatory factors

mentioned above; and (b) an examination of the possible determinants of the regulator's

adjustments in the OPEX estimate made during the rate setting, a more direct test for the

influence of interest groups on the regulatory results. In this case, the investigation employs the

disclosed information regarding the initial OPEX estimated via the Model Company engineering

model, the firm's reported OPEX, and the final (adjusted) OPEX. These numbers are used to

compute a measure of firm's bargaining power, which is then regressed on the political variables

(and some other possible explanatory factors). These procedures are detailed below.








30 The Brazilian electricity sector regulator did not employ any kind of economic benchmarking procedure to
estimate the productivity increments experienced by each firm during the regulatory period before the rate review.









Specification and Data

A measure of firm's bargaining power can be expressed by the distance of the final OPEX

to the corresponding initial estimate or, in other words, by the weight (W) given by the regulator

to the initial OPEX estimated by the engineering model. The weight is computed as follows:

OPEXA = W OPEXE + (1- W)OPEXc

SOPEXA -OPEXC
W=
OPEXE -OPEXC

where OPEXI stands for the final operating costs defined by the regulatory agency (ANEEL),

OPEXE stands for the operating costs initially estimated with the application of the engineering

model, and OPEXc is the operating costs reported by the distribution companies. These

variables' values, along with the corresponding WEIGHT computed for each firm's tariff review,

are portrayed in Table 4-1. Note that WEIGHTis inversely related to the bargaining effect, since

the firm's bargaining power can be said to be higher when the variable gets closer to zero.

As the dependent variable is a fraction between zero and one, the estimation follows the

procedure suggested by Papke and Wooldridge (1996) and employs a generalized linear model

(GLM), estimated by the maximum likelihood method, assuming a binomial distribution for W

and a logit link function. The modeling includes, as independent variables, proxies for the

potential influence of interest groups, factors related to possible problems identified in the

application of the engineering model, possible signals employed for firms' profitability and cash

flow availability, and proxies for a potential impact caused by both the external monitoring of the

regulator's activities and a likely learning effect.

These same independent variables are employed in the OLS regression. In this case,

though, an additional variable is included to control for a possible problem resulting from the

labor price measure used in the SFA procedure.









The explanatory variables are the following: income per capital (INCOME),31 share of

electricity delivered to industrial customers (INDSHARE), the log of total electricity delivered, in

GWh (SIZE), customer density (CUSDEN), given by the log of the number of customers divided

by service area, a public company dummy (PUBLIC), percentage growth in residential

consumption per capital from 1998 to 2003 (CONSUMPTION), output growth in the same period

(GROWTH), a categorical variable indicating whether or not the Tribunal de Contas da Uniao

has monitored, or not, the tariff review process (TCU), the number of rate reviews occurring

before the firm's review (LEARNING), and the ratio of SFA labor price to Model Company labor

price (LPDIFF).

The variables INCOME and INDSHARE are proxies for the consumer's participation in the

regulatory process. It is hypothesized that low-income residential customers should exert a

higher opposition to a price increase, when compared to high-income customers. Since the

income elasticity for electricity is less than one, poor families spend a greater share of their

income on electricity and thus have a greater incentive to oppose high prices, assuming the time

cost of political participation is proportional to income.32 As a result, income per capital should

be negatively associated to both WEIGHT and ANEELvsSFA.33 For INDSHARE, however, an

opposite effect is expected, since a rise in the share of electricity delivered to industrial

customers should similarly lead to more opposition to high prices, as the industry has a greater

stake in lobbying for lower electricity prices than residential or commercial customers. Of

31 The state income per capital was employed for 24 companies in the sample (19 state monopolies plus five
concessionaries which serve more than 90% of the consumers in the state). For the remaining 25 companies,
INCOME corresponded to the weighted average income per capital of the ten biggest municipalities in the
company's service area.
32 Knittel (2006), however, hypothesizes that wealth per capital is positively correlated with the degree of residential
interest group activity.

33 The lower the income, the closer should be W to one, denoting a lower firm's bargaining power, and the lower
should be the probability that ANEELEFF is smaller than SFA2003.










course, the price structure (reflecting cost allocation rules across customer categories) may be

more important here than the general price level, but the two are not unrelated.

The SIZE variable is a proxy for the producer's lobby, under the rationale that larger

companies should possess greater ability to influence regulatory decisions. Here, however, as

suggested by Klein and Sweeney (1999), it is conjectured that the expected effect of firm size is

indeterminate, as large utilities are more likely to receive careful scrutiny from the regulatory

34
agency.

The CUSDEN and PUBLIC variables were included in light of the points made earlier: the

Model Company's results might have been biased against firms which operated in more densely

populated areas, and may possibly have favored publicly owned firms. It is thus expected that

WEIGHT and ANEELvsSFA increases with CUSDEN and decreases with PUBLIC.

CONSUMPTION and GROWTH, on their turn, represent possible signals employed by the

regulator for firms' profitability and cash flow availability. The study tests the hypothesis that

the regulator may have wanted to pass on to consumers some of the rents derived from

economies of scale. Here, the expectation is that GROWTH contributes positively to both

WEIGHT and ANEELvsSFA. On the other hand, the residential consumption growth

(CONSUMPTION) captures the rationing effect. Brazil experienced an unforeseen electricity

crisis in 2001, which led to an energy rationing in the period of June 2001 to February 2002.35

The rationing measures significantly reduced the amount of electricity delivered and the average

34 The SIZE variable also controls for a possible endogeneity in the TCU monitoring, detailed below, since the closer
supervision of the regulator's activities during the tariff review process concentrated in the larger firms' cases.
35 The rationing aimed at a 20% reduction in energy consumption, and was implemented through a quota system
where monthly energy consumption targets were established for almost all consumers (poor residential consumers
were exempted). The scheme instituted penalties for non-accomplishments and bonuses for overachievements,
besides allowing the trading of quotas for nonresidential consumers. The quota system met its objectives and
avoided the occurrence of blackouts. Consumption levels from June to December 2001 showed a 20% load
reduction, compared to the previous year's consumption, and a 25% reduction if it is taken into account the new
customers that entered the system in 2001 (Maurer, Pereira, and Rosenblat, 2005).










residential consumption, bringing financial losses to concessionaires that operate in a sector

where the high fixed costs cannot be adjusted (or avoided) to compensate for the reduction in

revenues. As the rationing effect differed among firms,36 the observed change in residential

consumption per capital is used as a proxy for firms' losses and included in the model to test the

hypothesis that the least affected firms (the ones with higher CONSUMPTION) might have had

relatively low price increases (higher WEIGHT and ANEELvsSFA), under the rationale that the

concessionaires which had lower reductions in their cash flows had lower bargaining power in

the tariff review process and, consequently, were the ones most susceptible to rent extraction to

the benefit of consumers. Thus, the expectation is that WEIGHT and ANEELvsSFA vary directly

with CONSUMPTION.

A singular feature of the Brazilian regulatory environment consists of the existence of a

governmental body (Tribunal de Contas da Uniao) which supervises the regulatory agency's

performance.37 This study takes advantage of the fact that TCU closely monitored only some of

the periodic tariff review processes to examine whether this external monitoring has produced an






36 The differentiated effect among firms came mainly from the fact that the rationing measures varied among electric
zones. The quota system was initially applied to the Northeast and Southeast/Center-West sub-markets only, with
more stringent quotas being assigned to the former, when compared to the latter. Subsequently, the rationing was
extended to part of the North region, encompassing softer rules than the ones applied previously. The rationing in
that region was limited to the July to December 2001 period. The South region, in turn, was not included in the
rationing. Compared to the same period of the previous year, energy consumption in the Southeast, Northeast,
Center-West, North, and South sub-markets declined 31%, 28%, 25%, 10%, and 7% in the period of August to
December 2001, respectively (Bardelin, 2004). As stated by Maurer, Pereira, and Rosenblat (2005, p. 72), the
South, despite not being forced to ration, engaged in the load reduction effort as a result of appeals in the media and
for fear of more drastic measures in the upcoming dry season.

7 TCU is an independent organ of the state, which assists in the external control that the Congress possesses over
the whole public administration. The agency audits and reviews administrative decisions of the government to
ascertain that all legal procedures and rules have been followed. TCU is not a court, but the current legislation
attributes to the organ the power to order the review of some procedures undertaken and to impose sanctions and
penalties in cases of strong infractions to the law. TCU exercises an oversight over the regulatory agencies and has
examined both the procedures and substance of regulatory decisions.









effect on regulator's decisions.38 The intention is to shed light on the consequences of having an

institution performing oversight of the regulatory agency's procedures, and possibly contribute to

the literature that focuses on the optimal institutional regulatory framework. The supervision's

expected effect is indeterminate, though. One might conjecture that the monitoring reduces the

regulator's discretion and leads to figures closer to the ones portrayed by SFA, assuming the

SFA results are good representations of the true values. In this case, however, it is not possible to

anticipate the effect on the weight measure. Nonetheless, it is also possible that the external

monitoring has made the regulator exercise a higher scrutiny in the monitored cases, resulting in

lower bargaining power and lower estimates of efficient operational costs (higher WEIGHT and

ANEELvsSFA).

The LEARNING variable is incorporated in the modeling to avoid attributing to firm's

bargaining power (or lack of it) an effect due to "improvements" in the Model Company

method's usage. This was the first time ANEEL employed the Model Company approach. It is

thus reasonable to expect adjustments in the employed engineering cost parameters as more rate

reviews were carried out, resulting in large changes in OPEXF at the first rate reviews, and

progressively smaller changes at the reviews conducted later on. The LEARNING variable should

then be directly related to WEIGHT. It is not possible, however, to anticipate the variable's effect

on ANEELvsSFA. The continuing definition of the engineering cost parameters may either make

the corresponding efficiency estimate converge to the economic benchmarking estimate, or not.

In the computation of LPDIFF, the Model Company labor price was given by total labor

expenses estimated for the reference utility divided by the corresponding number of estimated



38 In the specific cases it decided to closely monitor the tariff review process, TCU requested ANEEL to submit the
correspondent technical notes and all other documents which supported the proposed repositioning indexes right
after they were released, in order to ensure a concomitant supervision of the actions undertaken.









employees.39 The variable is incorporated in the OLS regression only, to check whether the

divergences in results provided by the two approaches are related to the fact that the SFA labor

price was computed on the basis of firms' actual salaries and benefits paid, not accounting for

possible inefficiencies brought by the payment of values above the market price. The higher the

computed variable, the higher should be the upward bias in the firm's efficiency under SFA and,

therefore, the higher should be the ANEELvsSFA measure.

The data were assembled from the same sources employed to perform the SFA study.

Summary statistics are shown in Table 4-3. The TCU monitoring occurred in 12 out of the 49

tariff review processes examined, and 15 of the companies included in the sample are publicly

owned. The average residential consumption growth of-18.3% portrays the rationing effect,

whereas the computed WEIGHT varies in the interval of 0 to 0.998, with mean 0.641.40

As expected, WEIGHTis negatively and significantly (at the 1% level) correlated with the

adjustment made in the initial OPEX estimate (p = -0.6371), expressed by the ratio OPEXA to

OPEXE (see Table 4-1). 41 The adequacy of the variable's usage as a measure of firms'

bargaining power is corroborated by its positive and significant correlation (at the 1% level) to

the divergence measure ANEELvsSFA (p = 0.6324), a result that confirms the association

between higher bargaining power (lower W) and overvaluation of firms' efficiency levels (lower

ANEELvsSFA), and vice-versa.

39 For each of the hypothesized reference utilities, total labor expenses were computed on the basis of market
salaries especially gathered by a commissioned consultancy firm, considering the service area's specificities. In
addition, the estimated figure did not include some benefits actually paid by some concessionaries (additional
vacation in excess of the one-third disposed in the Constitution and profit sharing, for example).
40 In order to be able to run the GLM model, W was set equal to zero in the seven cases where the computed
WEIGHT was negative, shown in Table 4-1. In these cases, OPEXA showed up higher than OPEX essentially by
virtue of revisions implemented by the regulator after the end of the "standard tariff review procedure," under the
claim of offering to firms which had reviews in the beginning of the periodic tariff review (year 2003) the same
treatment (engineering cost parameters) given to firms which had reviews later on (years 2004 and 2005).
41 The higher the adjustment, the lower is W and the higher is the firm's bargaining power.









The variable, however, is not free of problems. There are some cases where W denotes a

high bargaining power in spite of a small percentage change from OPEXE to OPEX4,42 and the

data provides evidence that the mentioned adjustments are related to the "improvements" in the

engineering cost parameters through time, as the average percentage change on OPEXF

decreased steadily as more rate reviews were carried out, from 16.5% (for the first 12 firms in

the periodic tariff review process) to 1.6% (for the last 13 firms). These facts endorse the

inclusion of LEARNING as an additional explanatory variable, and support a research strategy

employing two investigation procedures; the approaches employed in the present study can be

viewed as complementary ways to address a complex topic.

Results Analysis

Regression results are remarkably consistent among the GLM and OLS procedures (Table

4-6). The evidence uncovers four main explanatory factors for the methodologies' divergences in

efficiency assessments. Initially, the positive and significant LEARNING coefficient on both

GLM and OLS models suggests that the firms' order in the periodic tariff review had

implications for regulatory decisions (and thus for the financial well-being of companies). The

first companies that went through the rate setting experienced higher changes in their initial

OPEX estimates and were benefited by the adjustments in the engineering cost parameters made

through time. When LEARNING decreases from the variable's mean value to a value equal to the

mean minus one standard deviation, the predicted WEIGHT decreases from 0.6902 to 0.3621 (-

47.5%), and the divergence measure ANEELvsSFA falls 0.074 points (6.8% of the mean value).

These results indicate that the earlier-reviewed firms ended up obtaining prices higher than



42 In these cases, OPEXE was not far from OPEXc. Thus, the final OPEX (OPEXA), despite representing only a small
percentage increase on OPEXE, was very close to OPEYX, leading to a small W. See, for example, the cases of CAT-
LEO, CELB, and CFLO in Table 4-1.









recommended by the economic benchmarking method. Conversely, later-reviewed firms were

considered more inefficient under the regulator's approach, compared to the SFA standard, as if

the regulator had become stricter after performing more and more rate-making processes.

The results also indicate that the employed proxies for the consumers' participation in

regulatory decisions are statistically significant. In both cases, however, the evidence is contrary

to some interpretations of interest group theory. The INDSHARE's negative coefficients indicate

that firms with a higher proportion of electricity delivered to industrial customers had a higher

bargaining power in the rate review (lower W) and were considered more efficient than shown by

SFA, receiving higher prices. The estimated impact is higher for higher values of the variable.43

Moreover, when INDSHARE increases from the variable's mean value to a value equal to the

mean plus one standard deviation, ANEELvsSFA falls 0.115 points (10.6% of the mean value).

The findings, in case, suggest that industrial demanders may have received higher prices than

would have been approved under a SFA approach to benchmarking.

The positive coefficients on the variable INCOME reveal that companies which serve more

wealthy customers tend to have lower bargaining power in the tariff setting and were harmed by

getting lower prices, i.e., prices seem to be lower (compared to the prices that would arise under

benchmarking using SFA) when customer incomes are higher. A one-standard deviation increase

in INCOME over its mean value shifts the predicted WEIGHT from 0.692 to 0.8151.

Additionally, a one-unit and a one-standard deviation increase in INCOME augments

ANEELvsSFA by 13.2% and 0.074 points (6.8% of the mean value), respectively. The evidence,

here, is consistent with the association between wealth per capital and the degree of residential



43 As reported in Table 4-6, a one-unit increase in INDSHARE over its mean value decreases Wby 7.4%.When the
marginal effect is computed for a value one-standard deviation above the mean, the result indicates that a one-unit
increase in INDSHARE decreases Wby 8.5%.









interest group activity suggested by Knittel (2006), as if the variable were capturing the high

discrepancy in the residential customers' (average) education levels across the different regions

in Brazil, under the rationale that more educated customers face lower costs to become informed

and participate in the tariff review process.

The anticipated impact of customer density on the regulator's results is confirmed as well.

The results indicate that the more densely populated the service area, the lower the company's

bargaining power (here the coefficient estimate is only marginally significant) and the more

harmed was the firm by receiving a repositioning index lower than the SFA benchmarking

procedure would recommend. On the other hand, companies operating in less densely populated

areas were considered more efficient than shown by the SFA method and, consequently,

benefited from higher prices. The estimated impact is nontrivial, as the CUSDEN's change from

the variable's mean value to a value equal to the mean plus or minus one standard deviation

shifts the ANEELvsSFA measure by 0.122 points, roughly 11.2% of its mean. This finding

corroborates Peano's (2005) claim that the regulator may have wanted to provide an extra return

to firms serving less densely populated service areas. However, the result might also indicate a

technical problem in the definition of the cost parameters employed in the engineering model,

which overstated the costs incurred by firms operating under this condition.

Some other results should be highlighted. The aforementioned potential problem

associated to the SFA labor price variable is not confirmed and neither is the conjectured

favorable treatment given to publicly owned firms. In addition, although the regulator knew in

advance which reviews would be closely monitored,44 the results indicate that supervision did

not affect the types of decisions in a systematic way, suggesting that ANEEL was consistent,


44 See footnote 38.









regardless of specific oversight. In particular, the evidence does not support the hypothesis that

the external monitoring may have made the regulatory agency be more strict in its analysis, to

the detriment of the distribution companies. This finding is important, as it corroborates the view

that the TCU's supervision of the regulator's activities does not increase firms' regulatory risk.

Finally, the positive and marginally significant coefficient found for the SIZE variable, in both

GLM and OLS models, does not support the conjectured producer's influence on regulatory

decisions, as it points to larger firms receiving prices lower than would have been approved

under a SFA benchmarking. Here, the evidence suggests that large utilities possibly received

greater scrutiny from the regulatory agency.

Robustness Check: DEA

Since the discussion above rests on the assumption that SFA estimates are good

representations of the true efficiency measures, the results' robustness is checked by performing

the analysis with the use of efficiency estimates provided by an alternative benchmarking

procedure. In case, the same dataset is employed to investigate firms' efficiency levels and their

evolution over time using a DEA technique. Here, the main concern was to use a specification

which could control for exogenous features of the operating environment and be comparable to

the previous parametric modeling. The option was for the use of the approach proposed by Fried,

Schmidt, and Yaisawarng (1999), based on a four-stage procedure to obtain measures of

managerial inefficiency separated from the influence of external operating conditions.

The first stage involves the calculation of an input-oriented DEA frontier under variable

returns to scale (VRS), using electricity delivered (Q) as the output, and Opex, Cap, and Len as

inputs. Specific DEA frontiers are computed for each year in the sample. Therefore, the

procedure provides measures of the relative efficiency of each firm in each period by reference

to yearly-specific frontiers, as well as information on input slacks and output surpluses of each









observation. The efficiency scores obtained at this stage, however, do not account for differences

in the operating environment across production units.

In a second stage, total input slacks are computed as the sum of radial plus non-radial input

slacks of each observation, and expressed as percentages of input quantities, as total slacks may

depend upon external environment as well as unit size.45 The resulting total input slacks

measures are then regressed on the six environmental variables previously mentioned (CusDen,

IndShare, ResDen, Area, Undergrd, and Income), with the purpose of identifying the effect of

external conditions on the excessive use of inputs. Given that input slacks are censored at zero by

definition, three tobit regressions (one for each input) are estimated separately. More formally:

k
TIS = fj(Q ,j,u ), j=1, ,N; k=1,...,K


where TIS1 is unit k's total radial plus non-radial slack for input based on the DEA results


from stage 1, expressed as a percentage of actual input quantity, Qk is the vector of variables

characterizing the operating environment for unit k that may affect the utilization of input /fl


is a vector of coefficients, and uf is a disturbance term.

In a third stage, the regressions' estimated coefficients are used to predict total input slack

for each input and for each unit based on its external variables. The predicted values represent

the "allowable" slack, due to the operating environment.

IS =fj(Q, j) j = 1, ,N; k=,..., K






45 The definition of the total input slack measure in terms of percentage of actual input quantities is recommended by
Fried et al. (1999, note 19) for the situation where the firm size differs significantly among firms in the sample.










These predictions, in turn, are employed to adjust the primary input data for each unity

according to the difference between maximum predicted slack and predicted slack, under the

rationale of establishing a base equal to the least favorable set of external conditions.46

xadj = 1 + axk S fS


In the final stage, the adjusted input variables are employed to re-run the initial input-

oriented DEA VRS model, and generate efficiency scores for each firm in each period net of

factors out of management control (Appendix B). In line with the procedure adopted before, the

DEA efficiency measures obtained for year 2003 (DEA2003) are used for comparison to the SFA

and Model Company results.

DEA efficiency estimates are significantly higher than SFA2003,47 varying in the range of

1 to 2.38, with mean 1.28. This fact, taken together with the evidenced similarity between

DEA2003 and ANEELEFF distributions,48 suggests that either some inefficiency is attributed to

statistical noise in the SFA approach, or the SFA efficiency indexes are constrained by the half-

normal distribution assumed for the inefficiency error term. On the other hand, even though

DEA2003 and SFA2003 efficiency measures and rankings are not significantly correlated,49 there




46 With this procedure, a firm with external variables corresponding to this base level would not have its input vector
adjusted at all, and a firm with external variables generating a lower level of predicted slack would have its input
vector adjusted upward to put in on the same basis as the firm with the least favorable external environment. In other
words, predicted slack below the maximum predicted slack is attributable to external conditions more favorable than
the least favorable conditions prevailing in the sample for that input. By increasing the input vector and leaving the
output vector unchanged, the firm's performance is purged of the external advantage (Fried et al., 1999).

47 The null of equality of means is rejected at the 1% significance level (p-value (Hi: DEA2003 > SFA2003) =
.0001). When the equality of DEA2003 and ANEELEFF means is tested, however, the null is not rejected.
48 Data on the respective mean and standard deviation are provided in Table 4-5. The difference in means is not
statistically significant, as mentioned in the previous note, but DEA2003 's distribution of efficiency indexes is
slightly more spread out than ANEELEFF's.
49 The correlation statistic and the Spearman's rank correlation amount to -0.0579 (p=.6927) and -0.0751 (p=.5966),
respectively.









is some consistency in terms of best and worst performers. Five firms appear in both top ten

extracts, and four firms in both bottom ten (Table 4-5).

The comparison between DEA2003 and ANEELEFF corroborates the indication that the

Model Company approach understated some firms' efficiency levels. Similarly to what was

found in the comparison to SFA2003, ANEELEFF of firms Eletropaulo, Light, CEB, Eletroacre,

Eletrocar, Piratininga, Boavista, and CPFL are considerably higher than DEA2003 in absolute

terms. Additionally, some of the previously mentioned cases of overevaluations of firms'

efficiency are confirmed as well, as the model company efficiency indexes of firms Enersul,

Energipe, Cemig, Coelce, Celesc, Coelba, and Cat-Leo are well below both SFA2003 and

DEA2003. On this respect, the DEA findings provide additional support to the indication that the

benefit given to Energipe, Celesc, and Coelba was unjustified, since these firms do not belong to

the DEA top ten segment either.

The robustness check employs the same OLS model described in the previous section, with

the difference that the dependent variable is now given by a new divergence measure

(ANEELvsDEA), computed as the ratio of ANEELEFF to DEA2003. ANEELvsDEA varies in the

range of 0.459 to 1.986, with mean 0.986 (Table 4-3), and is significantly (1% level) correlated

with the preceding divergence measure (p = 0.7743). As reported in Table 4-6, the fitted model

is not as well specified as before (smaller R2), but the Wald specification test still rejects the null

that the coefficient estimates are all equal zero (p-value = 0.0006). The results confirm the

previously noted effects of Industrial .,\lhi e, Income, and Customer Density variables. The

findings, however, do not support the Learning effect identified before, or the possible impact of

the Size variable.









Concluding Observations

The present study examines the application of the Model Company approach in the

Brazilian electricity distribution sector periodic tariff reviews (April/2003 to February/2006).

The resulting firms' efficiency measures are evaluated with the use of efficiency estimates

obtained from both a parametric and a non-parametric benchmarking model and indices of

productivity changes experienced in the six-year period before the rate review. In the process, the

study tests for possible causes of the identified divergences in efficiency assessments and checks

for potential determinants of firms' bargaining power in the rate setting process.

Despite the criticisms made to its subjectivity and complexity, the Model Company

approach has become increasingly popular for the determination of electricity distribution tariffs

in Latin America (Jadresic, 2002). It is therefore important to verify whether the methodology

has both provided an opportunity for firms to meet their break-even constraints and enabled the

attainment of a welfare maximizing regulator's rate setting objectives: extracting part of the

firms' rents for the benefit of consumers and society, achieving allocative efficiency, and

offering incentives for further productivity improvements.

However, regulatory decisions are made by a regulator operating under information

asymmetries, facing the influence of interest groups and, in the specific case examined here,

subject to direct supervision of its actions. Thus, the analysis of regulatory outcomes addresses

the possible impact of these factors, in addition to the effects of the methodology employed.

The investigation reveals that the regulator's objectives might not have been welfare

maximizing in some situations. On the one hand, some firms were considered to be rather more

inefficient than shown by both SFA and DEA models, resulting in substantially lower price

increases: this result raises concerns over the companies' long-term financial sustainability. On

the other hand, the results point to the existence of firms which the regulator's method









considered to be much more efficient than suggested by the two widely-used benchmarking

methodologies.

The study provides new findings on possible causes for these divergences in the context of

a particular regulatory system. The results indicate that firms with a lower proportion of

electricity delivered to industrial customers, which serve wealthier consumers and operate in

more densely populated areas, had lower bargaining power in the tariff setting and were harmed

by getting prices lower than recommended by the economic benchmarking methods. These

results are consistent with the economic theory of regulation which posits that political influence

affects the level of prices. On the other hand, firms with opposite characteristics had higher

bargaining power and benefited from higher prices. The evidence is consistent with an

association between per capital income and the effectiveness of residential interest group activity.

Moreover, the findings point to a possible inaccuracy of the cost parameters employed in the

engineering Model Company approach; the parameters may inaccurately capture the effect of

consumers' dispersion (customer density) on firms' operating costs, due to either the technical

difficulty in defining the "true" parameter in a context of imperfect and asymmetric information,

or a deliberate intention to avoid compensating investors in utilities operating in areas of higher

consumer concentration, and to provide extra returns to firms working in less densely populated

areas (Peano, 2005).

This benefit given at the beginning of the tariff review cycle impacted negatively the

incentives for efficiency improvements provided to firms which do not appear in the top ten

segments of SFA and DEA efficiency ranking. The same disincentive was received by four of

the top five firms in the SFA ranking, which could not keep part of the rents brought by their

productivity improvements. In sum, the regulator's methodology imposed on firms a one-time









adjustment to the virtual company's efficient operating costs, which in some cases were rather

different than the ones estimated by the benchmarking methods. Moreover, the rulings (and

associated price trajectories) ignored the significant frontier shifts of almost 7% per year

revealed by the parametric modeling, a point that would exacerbate the perverse effects of the

hypothesized over-evaluations of firms' efficiency.

Interestingly, the findings do not provide support to the hypothesis that the monitoring of

the regulator's activities may lead to decisions contrary to firms' interests and increase firms'

regulatory risk, one of the possible effects of having an institution supervise the regulator's job.

Regulator's decisions were not affected in a systematic way by special oversight. Despite its

specificity, the result adds to the literature on the optimal regulatory framework design.

It should be stressed that the results outlined above are robust to the choice of

benchmarking methodology (SFA or DEA) to employ as a comparison parameter. Moreover, the

results do not support those who are concerned with possible limitations of the SFA

methodology. However, for those who are more hesitant to abandon engineering models, at a

minimum the present investigation presents a way to promote greater transparency to the process

and credibility for the results obtained with the application of the Model Company method. Once

the divergences in efficiency assessments are identified, and possible explanatory factors are

uncovered, it remains the regulator's job to justify the choices made or demonstrate that the

divergences do not come from deficiencies in the application of a particular methodology.

The proposed joint use of a comparative efficiency analysis technique benefits all

stakeholders including the regulator; the agency could employ other benchmark techniques to

alleviate potential adverse selection problems and consequently come up with more reliable

approximations of firms' break-even points. It would then be possible to better exploit the price-










cap incentives for efficiency improvements50 and promote the desired allocation of productivity

gains among stakeholders. Note, on this topic, that the redistribution of rents should ideally be

based on information regarding the productivity increments of each firm during the whole

regulatory period prior to the next review, information that the model company approach alone

cannot provide.

The experience so far on setting price caps has indicated that quantitative benchmarking

techniques may at least serve as an additional tool to the regulator, whose importance is

underscored by information contained in comparisons available from having a large number of

companies in the regulated industry.51 Thus, there appears to be no reason for not using them in

the Brazilian electricity distribution industry.
























50 The efficiency improvement incentives associated to the price-cap method are often hindered by the regulator's
uncertainties about firms' inherent costs, which usually lead to tariffs set at a too high level, given the fear of
violating firms' financial sustainability constraint. Comparative efficiency analysis helps reduce the regulator's
informational disadvantage and enables the definition of better participation constraints, thereby allowing a more
fruitful use of high-powered incentive mechanisms.
51 The use of benchmarking techniques on regulatory price reviews is discussed by Rossi and Ruzzier (2000), Pollitt
(2005), Ster (2005), and Dassler, Parker, and Saal (2006), among others.












Table 4-1. Initially Estimated OPEX, Final OPEX, and Firm's Reported OPEX


SAELPA
SANTA CRUZ
SANTA MARIA
V. PARANAP.
CSPE
DMEPC
ELETROACRE
ELETROCAR
JAGUARI
MOCOCA
XANXERE


185,395,425
40,012,199
16,498,783
31,929,120
17,823,015
16,750,348
31,603,551
11,932,292
10,677,128
12,246,733
10,165,730


1. Regulator's Efficiency Index: ratio OPEXc over OPEXA
2. The Weight shows how close the Final OPEX (OPEXA) is to the Engineering Estimated OPEX (OPEXE).
3. Firm's order in the tariff review process.


COMPANY OPEXE(A)
AES-SUL 152,378,578
BANDEIRANTE 178,748,147
BOA VISTA 19,299,948
BRAGANTINA 22,762,688
CAIUA 34,405,902
CAT-LEO 69,812,982
CEAL 145,326,676
CEB 127,222,668
CEEE 231,085,795
CELB 26,293,985
CELESC 411,731,525
CELG 455,583,137
CELPA 229,717,734
CELPE 374,778,267
CELTINS 74,420,106
CEMAR 212,939,949
CEMAT 172,964,954
CEMIG 808,746,752
CENF 19,073,913
CEPISA 140,241,902
CERJ 278,164,419
CERON 122,533,378
CFLO 10,983,263
COELBA 341,063,413
COELCE 244,517,894
COPEL 588,545,532
COSERN 97,792,392
CPFL 328,589,815
ELEKTRO 323,531,823
ELETROPAULO 588,395,853
MANAUS 87,650,951
ENERGIPE 68,983,023
ENERSUL 112,343,069
ESCELSA 209,658,844
LIGHT 463,351,823
NACIONAL 19,052,515
PIRATININGA 170,825,329
RGE 157,117,648


OPEXA(B)
168,526,897
200,857,495
19,312,706
27,599,813
39,736,410
73,173,281
146,266,520
145,601,583
235,718,790
27,302,875
440,713,597
483,121,893
269,031,550
379,210,684
82,065,329
217,204,197
197,274,615
936,572,499
19,748,356
141,016,014
297,502,578
122,743,263
11,658,505
431,347,472
282,727,424
606,611,885
113,400,305
421,760,792
348,509,294
645,184,235
87,948,585
82,571,280
130,154,623
217,182,804
516,334,111
22,337,700
191,017,669
170,367,818
190,428,585
44,081,288
19,771,653
37,622,908
19,150,347
17,466,270
32,045,408
11,958,308
11,157,355
13,167,147
11,223,622


OPEXc (C)
213,700,000
254,995,034
26,152,813
35,211,066
48,111,601
73,047,000
175,019,103
257,359,512
266,328,472
29,812,883
520,128,813
580,123,409
310,500,443
476,485,325
94,330,458
258,866,541
187,200,000
893,609,000
22,158,000
160,151,652
391,983,902
130,521,640
12,462,079
437,000,000
260,000,000
688,548,640
136,100,000
549,100,000
436,873,603
1,281,200,000
102,481,654
70,000,000
113,300,000
275,672,979
944,760,674
29,134,454
265,380,252
174,089,900
214,242,000
48,887,827
19,299,425
45,725,014
23,078,303
19,635,723
35,519,561
16,844,993
15,783,997
16,895,193
12,416,298


ANEELEFF1
1.2680
1.2695
1.3542
1.2758
1.2108
0.9983
1.1966
1.7676
1.1299
1.0919
1.1802
1.2008
1.1541
1.2565
1.1495
1.1918
0.9489
0.9541
1.1220
1.1357
1.3176
1.0634
1.0689
1.0131
0.9196
1.1351
1.2002
1.3019
1.2535
1.9858
1.1652
0.8478
0.8705
1.2693
1.8297
1.3043
1.3893
1.0218
1.1251
1.1090
0.9761
1.2154
1.2051
1.1242
1.1084
1.4086
1.4147
1.2831
1.1063


WEIGHT2
0.7367
0.7100
0.9981
0.6114
0.6111
-0.0390
0.9683
0.8588
0.8685
0.7133
0.7326
0.7789
0.5133
0.9564
0.6160
0.9072
-0.7077
-0.5063
0.7813
0.9611
0.8301
0.9737
0.5434
0.0589
-1.4680
0.8193
0.5926
0.5775
0.7796
0.9180
0.9799
-12.3614
-17.6132
0.8860
0.8899
0.6742
0.7864
0.2193
0.8255
0.5415
-0.1686
0.5873
0.7474
0.7519
0.8872
0.9947
0.9060
0.8020
0.5299


ORDER3
2
6
24
9
9
12
22
16
17
19
15
23
5
20
14
22
1
1
12
22
8
25
9
3
3
13
3
1
5
4
24
3
1
15
7
9
6
2
22
9
10
9
9
13
25
21
9
9
15


(B) / (A)
10.60%
12.37%
0.07%
21.25%
15.49%
4.81%
0.65%
14.45%
2.00%
3.84%
7.04%
6.04%
17.11%
1.18%
10.27%
2.00%
14.05%
15.81%
3.54%
0.55%
6.95%
0.17%
6.15%
26.47%
15.63%
3.07%
15.96%
28.35%
7.72%
9.65%
0.34%
19.70%
15.85%
3.59%
11.43%
17.24%
11.82%
8.43%
2.71%
10.17%
19.84%
17.83%
7.45%
4.27%
1.40%
0.22%
4.50%
7.52%
10.41%


(A) / (C)
71.30%
70.10%
73.80%
64.65%
71.51%
95.57%
83.03%
49.43%
86.77%
88.20%
79.16%
78.53%
73.98%
78.65%
78.89%
82.26%
92.40%
90.50%
86.08%
87.57%
70.96%
93.88%
88.13%
78.05%
94.05%
85.48%
71.85%
59.84%
74.06%
45.93%
85.53%
98.55%
99.16%
76.05%
49.04%
65.40%
64.37%
90.25%
86.54%
81.84%
85.49%
69.83%
77.23%
85.31%
88.98%
70.84%
67.65%
72.49%
10.41%













Table 4-2. SFA Descriptive Statistics
Variable 1998 1999 2000 2001 2002 2003 1998-2003 Range


98,905 85,773 84,953 74,258 70,455
(132857) (111025) (113274) (100497) (97838)


70,134
(97273)


5,074,129 5,260,394 5,520,603 4,790,657 5,063,016 5,110,973
(8442352) (8346455) (8719154) (7649132) (7569014) (7404106)

38.9052 32.4873 35.9164 34.0834 39.2159 41.9181
(18 9536) (147348) (18517) (147472) (223052) (21 2112)

78.6138 72.8605 70.946 70.9966 68.4405 68.595
(67104) (4419) (39627) (35699) (3 2352) (34071)

74.0168 66.802 64.4161 61.4267 53.5491 58.3822
(179825) (184658) (169462) (14435) (11 4705) (127854)

25.7095 26.6959 27.9056 28.7718 30.8484 32.0821
(18 6995) (19 1257) (200373) (204782) (21 8955) (224005)

0.2959 0.2980 0.3068 0.3132 0.3308 0.3257
(0 1461) (0 1434) (0 1432) (0 1413) (0 1498) (0 1568)

2.1026 2.0789 2.0028 1.7162 1.6803 1.6774
(06267) (06282) (05139) (04687) (04625) (04167)

129,178 129,210 129,203 131,495 126,671 126,725
(242029) (239567) (239564) (241463) (237902) (237882)

828,166 879,502 919,894 934,543 979,891 1,012,766
(1099440) (1134211) (1188028) (1228822) (1255942) (1287816)

5,769.74 5,086.45 5,160.16 4,996.71 4,386.60 4,642.68
(280422) (2351 86) (23793) (2272 11) (1880 11) (198973)

3,218.57 3,269.12 3,269.12 3,142.07 3,206.25 3,206.25
(4908) (487248) (487248) (4835 87) (4751 46) (4751 46)

41,998.10 42,957.20 42,957.20 42,959.70 42,131.10 42,131.10
(65700 6) (65399 9) (65399 9) (66063 9) (64894 5) (64894 5)


UNDERGRD 0.006592 0.006462 0.006462 0.005940 0.006338 0.006338
(0246) (0244) (0244) (0244) (0241) (0241)


OPEX


[2490, 559072]


80,640
(108952)

5,137,639
(7970465)

37.1144
(18 7994)

71.701
(5 5173)

63.022
(16 7229)

28.6965
(20 4544)

0.3119
(0 1463)

1.8749
(0 5537)

128,723
(237747)

926,545
(1193257)

5,001.43
(2317 13)

3,218.73
(4792 04)

42,520.10
(64850 3)

0.006356
(0241)


[0, .1391]


OBSERVE.


50 51 51 50 52 52


Mean values reported for each year and for the period 1998-2003. Standard deviation in parentheses.


[103191, 37540051]


[6.5398, 128.4681]


[60.008, 96.620]


[29.434, 98.120]


[6.747, 137.093]


[.0333, .6438]


[.663, 4.572]


[252, 1253165]


[19625, 5744178]


[1060.012, 12747]


[.1, 22728.4]


[720.3, 379518.58]


CUSDEN


INDSHARE


RESDEN


AREA


NUMCUST


INCOME












Table 4-3. GLM and OLS Descriptive Statistics
Continuous Variables Mean Std. Deviation Range Categorical Variables Value Frequency


WEIGHT 0.641 0.323 (0, 0.998) PUBLIC 0 34

ANEELVSSFA 1.087 0.202 (0.633, 1.719) 1 15

ANEELVSDEA 0.986 0.298 (0.459, 1.986) TCU 0 37

CUSDEN(LN) 2.825 1.511 (0.250, 7.019) 1 12

LPDIFF 0.978 0.414 (0.301, 2.474)

INCOME (LN) 2.065 0.558 (0.856, 3.795)

INDSHARE 0.322 0.158 (0.039, 0.644)

SIZE (LN) 0.702 1.567 (-2.083, 3.491)

CONSUMPTION -18.297 10.474 (-65.217, 6.820)

GROWTH 12.194 13.607 (-22.525, 37.771)

LEARNING 22.694 14.585 (0, 47)













Table 4-4. Stochastic Cost Frontier Results
Variable Time-trend formulations Time Fixed-Effects formulations
A B C D E F


LnOpex
LnQ

LnLP

InAMP

Cap

Len

InlndShare

InResDen

Inlncome

InArea

InCusDen

Undergrd

T

InQ*t

lnLP*t

InMP*t

Tsq

Private

Private*t

Privtzed

Alwspriv

D1999

D2000

D2001

D2002

D2003

Cons

Insig2v
Cons

Insig2u
Q

Cons

Statistics
N
L1
Chi?


0.771***
(.025)
0.442***
(.062)
0.364***
(.116)
0.096***
(.027)
0.561***
(.064)
-0.007
(.034)
0.169*
(.089)
-0.179***
(.035)
0.072***
(.012)
0.500***
(.061)
4.765***
(.589)
-0.054**
(.027)
-0.015**
(.005)
-0.012
(.017)
0.001
(.029)
-0.004
(.007)




















-0.150**
(.070)

-3.964***
(.307)

0.109*
(.064)
-4.423***
(1.193)


0.781***
(.023)
0.403***
(.064)
0.374***
(.112)
0.108***
(.027)
0.561***
(.063)
0.009
(.033)
0.157*
(.088)
-0.145***
(.037)
0.074***
(.011)
0.496***
(.061)
4.480***
(.582)
-0.052*
(0.026)
-0.014***
(.005)
-0.010
(.017)
0.001
(.027)
-0.004
(.007)
-0.110**
(.047)
0.006
(.012)















0.001
(.052)

-3.734***
(.080)

0.606
(.761)
-15.104
(14.702)


0.803***
(.035)
0.409***
(.066)
0.315***
(.114)
0.058*
(.034)
0.522***
(.064)
-0.013
(.035)
0.143
(.092)
-0.186***
(.039)
0.066***
(.013)
0.466***
(.064)
4.486***
(.563)
-0.051**
(.025)
-0.007
(.005)
-0.014
(.017)
0.047
(.030)
-0.005
(.007)




-0.094*
(.052)
-0.081
(.087)











-0.142
(.089)

-4.242***
(.486)

-0.063
(.102)
-3.590***
(1.029)


0.708***
(.018)
0.395***
(.034)
0.381***
(.068)
0.103***
(.028)
0.534***
(.063)
-0.008
(.033)
0.131
(.089)
-0.168***
(.037)
0.074***
(.012)
0.469***
(.062)
4.830***
(.602)




















-0.039
(.033)
-0.106***
(.033)
-0.150***
(.034)
-0.268***
(.035)
-0.303***
(.036)
-0.200***
(.052)

-3.898***
(.242)

0.131**
(.051)
-4.502***
(.956)


0.739***
(.017)
0.348***
(.035)
0.401***
(.069)
0.102***
(.030)
0.525***
(.066)
-0.003
(.034)
0.099
(.092)
-0.139'**
(.039)
0.073***
(.013)
0.462***
(0.064)
4.465***
(.611)




















0.049
(.183)
0.048
(.294)
0.029
(.346)
-0.075
(.355)
-0.137
(.337)
-0.581
(.367)


306 306 306 306 306
128.429 136.29 144.157 120.077 124.98343
21833.984 23964.368 17857.736 21535.49 28332.588


Legend: p<0.10; ** p<0.05; *** p,
terms are omitted.


0.756***
(.017)
0.366***
(.034)
0.480***
(.069)
0.080**
(.028)
0.516***
(.063)
0.009
(.033)
0.145*
(.087)
-0.185***
(.040)
0.073***
(.012)
0.465***
(.061)
4.601***
(.583)




















-0.044
(.032)
-0.102***
(.033)
-0.147***
(.033)
-0.273***
(.036)
-0.308***
(.035)
-0.437)***
(.105)


306
134.31818
17974.398


:0.01 Standard deviation in parenthesis. Coefficients on translog squared and interaction













Table 4-5. Efficiency Rankings and Indexes
Ranking ANEELEFF SFA2003 DEA2003


1


CEMAR
CEAL
COSERN
CELG
CSPE
CAIUA
V. PARANAPANEMA
ELEKTRO
CELPE
AES-SUL
ESCELSA
BANDEIRANTE 2
BRAGANTINA
MOCOCA
CPFL
NATIONAL
CERJ
BOA VISTA
PIRATININGA
ELETROCAR
JAGUARI
CEB
LIGHT
ELETROPAULO


Mean
Std. Deviation
25% Percentile
75% Percentile


Company
ENERGIPE
ENERSUL
COELCE
CEMAT
CEMIG
SANTA MARIA
CAT-LEO
COELBA
RGE
CERON
CFLO
CELB
XANXERE
ELETROACRE
SANTA CRUZ
CENF
DMEPC
SAELPA
CEEE
COPEL
CEPISA
CELTINS
CELPA
ELN/AM (MANAUS)
CELESC


Eff. Index
0.848
0.871
0.920
0.949
0.954
0.976
0.998
1.013
1.022
1.063
1.069
1.092
1.106
1.108
1.109
1.122
1.124
1.125
1.130
1.135
1.136
1.149
1.154
1.165
1.180
1.192
1.197
1.200
1.201
1.205
1.211
1.215
1.254
1.257
1.268
1.269
1.270
1.276
1.283
1.302
1.304
1.318
1.354
1.389
1.409
1.415
1.768
1.830
1.986

1.202
0.217
1.106
1.270


Company
RGE
CAT-LEO
CELB
ELETROACRE
ELN/AM (MANAUS)
ENERSUL
COELCE
COSERN
LIGHT
CEAL
CEMAR
CENF
ESCELSA
BANDEIRANTE 2
CSPE
BOA VISTA
PIRATININGA
ELEKTRO
MOCOCA
COELBA
SANTA CRUZ
ELETROCAR
CELPE
CPFL
ENERGIPE
AES-SUL
CEPISA
CERON
CELTINS
DMEPC
SAELPA
CFLO
SANTA MARIA
CEB
NATIONAL
CERJ
CEMAT
CELPA
COPEL
CEEE
CELG
JAGUARI
CAIUA
BRAGANTINA
ELETROPAULO
V. PARANAPANEMA
XANXERE
CELESC
CEMIG

Mean
Std. Deviation
25% Percentile
75% Percentile


Eff. Index
1.045
1.057
1.059
1.060
1.062
1.065
1.068
1.069
1.069
1.070
1.072
1.072
1.074
1.074
1.077
1.080
1.081
1.082
1.083
1.084
1.088
1.090
1.090
1.092
1.093
1.093
1.093
1.094
1.095
1.096
1.097
1.101
1.101
1.106
1.113
1.115
1.127
1.129
1.132
1.133
1.140
1.147
1.148
1.151
1.155
1.179
1.196
1.283
1.506

1.110
0.072
1.074
1.127


Company
AES-SUL
CEMIG
COELCE
CPFL
ELETROPAULO
ELN/AM (MANAUS)
LIGHT
PIRATININGA
RGE
ELETROACRE
ELETROCAR
JAGUARI
DMEPC
MOCOCA
BANDEIRANTE 2
XANXERE
CELB
CFLO
COSERN
CELPA
CELESC
NATIONAL
ELEKTRO
CSPE
BOA VISTA
ENERGIPE
CELPE
SANTA MARIA
CERON
SAELPA
CENF
ESCELSA
COELBA
CEEE
CEMAR
CEB
CEMAT
CEAL
COPEL
CERJ
CEPISA
BRAGANTINA
ENERSUL
V. PARANAPANEMA
SANTA CRUZ
CELG
CAIUA
CAT-LEO
CELTINS

Mean
Std. Deviation
25% Percentile
75% Percentile


Eff. Index
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.024
1.073
1.096
1.116
1.120
1.138
1.144
1.153
1.170
1.171
1.183
1.206
1.217
1.221
1.233
1.233
1.258
1.259
1.266
1.274
1.284
1.337
1.340
1.379
1.437
1.449
1.499
1.522
1.577
1.600
1.618
1.647
1.656
1.692
1.733
2.174
2.381


1.283
0.302
1.024
1.437











Table 4-6. Regression Results
GLM OLS
Variable Coefficient Marginal Effect ANEELvsSFA ANEELvsDEA


Industrial share



Income (In)



Size (In)



Consumption



Growth



Customer density (In)



TCU monitoring



Public company



Learning



Labor price effect



Intercept



Statistics
N
Log pseudolikelihood
R2


-3.466**
(1.560)

1.223***
(0.468)

0.313*
(0.190)

0.011
(0.018)

0.003
(0.016)

0.303*
(0.176)

-0.694
(0.560)

0.144
(0.488)

0.094***
(0.020)


-3.515***
(1.232)



49
-18.376


Robust standard errors in parentheses
Legend: p<.l; ** p<.05; *** p<.01


-0.561**
(0.271)

0.151*
(0.075)

0.039
(0.028)

0.003
(0.002)

0.001
(0.003)

0.124***
(0.040)


-0.741**
(0.339)

0.262**
(0.103)

0.067
(0.041)

0.002
(0.004)

0.001
(0.003)

0.065*
(0.038)


-0.156
(0.132)

0.030
(0.102)


0.020***
(0.005)


-0.728***
(0.156)

0.132**
(0.050)

0.048*
(0.027)

0.001
(0.002)

-0.001
(0.002)

0.081***
(0.023)

-0.058
(0.067)

-0.075
(0.063)

0.005**
(0.002)


-0.078
(0.104)

0.078
(0.085)

-0.001
(0.003)


-0.029
(0.097)


0.756***
(0.143)


0.533***
(0.172)


0.634


0.522









CHAPTER 5
SUMMARY AND CONCLUSIONS

This dissertation is composed of three empirical essays (chapters) on the reforms

accomplished in the Brazilian electricity sector. Initially, an event-study methodology is

employed to examine the actual behavior of the created autonomous regulatory body, vis a vis

the predictions of the theories regarding the pattern of the government intervention in business.

In sequence, a parametric benchmarking approach is used to investigate whether the price-cap

incentive mechanism effectively lead to performance improvements in the electricity distribution

sector, checking for difference in performance between public and private firms, and looking at

the possibility of efficiency catch-up. Then, the SFA efficiency estimates and indexes of

productivity change, along with efficiency measures provided by a non-parametric

benchmarking technique, are utilized to evaluate whether the use of the model company

approach in the distribution companies' first periodic tariff review enabled the attainment of the

welfare maximizer regulator's rate setting objectives.

The event-study examines the regulator's performance in the Brazilian electricity sector

using a methodology specially designed to a regulatory context, which explicitly accounts for the

possibility of event anticipation. Despite the more pronounced interests that characterize a

developing country's regulatory environment, the results indicate that the regulator has acted

relatively independently, with its decisions not favoring a single interest group. The findings are

similar to the ones obtained in previous studies that focused in the United Kingdom's context,

and do not support the claim that Brazilian regulatory agencies are captured by the industry.

On the contrary, the evidenced unpredictability of regulatory agency's decisions suggests

that the regulatory agency has favored different interest groups at different times, supporting the

claim that the utility maximizing regulator will not exclusively serve a single economic interest.









The observed unpredictability, on the other hand, reinforces the need of improvements in the

regulatory discussion process, with the adoption of measures to increase the transparency and to

promote more substantive public hearings.

The study suggests that the need to provide incentives for new investments has had a

significant role in the regulatory process. In addition, electricity companies have been

compensated for the regulatory risk they face. The estimates indicate that regulatory decisions

led to an increase in firms' market value over the period examined and account for part of the

difference of the sample securities' performance with respect to the market.

Some specific findings are also worth noting. The adoption of the asset base repositioning

cost methodology was not that harmful to distribution companies, as one would anticipate in

light of the press coverage related to the issue, and the evidence raises the concern over the

objectivity and transparency of the methodology employed in the distribution companies'

periodic tariff review, suggesting the need of improvements. Moreover, the results indicate that

the Government's proposal to review the regulatory agencies' responsibilities and performance

was seen as a step back in the electricity sector regulation and increased the regulatory risk.

The second study confirms the theoretical predictions regarding the impact of incentive

regulation on firms' performance. Brazilian electricity distribution companies have experienced

high productivity growth rates after the sector reforms, above what was found in a previous study

for the period before the reforms. The productivity increment relates to the closing of the

efficiency gap present in 1998, and is driven by the performance of privatized and public

companies.

Privatized firms responded more aggressively than public firms to the new incentives

brought by price-cap regulation, denoting that incentives were higher to profit-oriented managers









operating under a shareholders' pressure to quickly recoup the investments made. The study's

estimate of privatized firms' incremental annual productivity growth rate (3.36%), on the other

hand, brings about the need to tailor specific efficiency improvement incentives to public firms,

since it represents not implemented-but-achievable productivity gains, which could have resulted

in lower tariffs to customers.

The subset of firms privately owned before the reforms shows up as more efficient, on

average, than other firms in the beginning of the period examined. Its productivity growth rate

evidenced in the present study, therefore, is consistent with a limited space for efficiency

improvements on operating and maintenance expenses on the more efficient firms subject to a

rate of return regulation scheme. Given the sensibly higher productivity growth rates experienced

by other firms, "always private" firms face a decline in their efficiency levels over the period.

This research provides another possible explanation. It shows that the observed decline in these

firms' mean efficiency level derives, fundamentally, from their low productivity growth in 2003,

which might therefore indicate a possible strategic behavior of some of these firms, associated to

the periodic aspect of the price-cap incentive regulation scheme.

The results suggest a possible occurrence of strategic behavior of another sort as well. In

spite of plausible economies of scope, vertically integrated distribution companies show up as

more inefficient than other firms, raising the possibility of cost shifting. Stricter rules regarding

cost allocation and/or a closer look at these companies' accounting numbers may be appropriate.

Interestingly, the study reveals that the high performance improvement experienced by

privatized firms in the period does not come from mere reductions in costs brought by

deterioration in the quality of service provided, a result that also indicates the effectiveness of the

quality regulation instruments implemented by the regulator.









All these findings ultimately provide a better understanding of the cost opportunities faced

by each firm, and consequently enable the establishment of prices conducive to a greater social

welfare. The regulator has had the opportunity to define new electricity distribution prices in the

periodic tariff review that started in 2003. On that opportunity, the choice was for the use of the

model company approach to estimate each firm's efficient operational costs. This paper's

findings provide the basis not only for evaluating the regulator's decisions in those

circumstances, notably with respect to their consequences in terms of both distribution of

productivity gains among stakeholders and incentives for further efficiency improvements, but

also for examining the model company approach itself. The approach's usage is not pacific in the

theory and its implementation in the Brazilian context has been disputed among the parties

involved.

The performed evaluation of the results obtained with the use of the model company

approach reveals that the regulator's objectives might not have been welfare maximizing in some

situations.

On the one hand, some firms were considered to be rather more inefficient than shown by

both SFA and DEA models, resulting in substantially lower price increases: this result raises

concerns over the companies' long-term financial sustainability. On the other hand, the results

point to the existence of firms which the regulator's method considered to be much more

efficient than suggested by the two widely-used benchmarking methodologies.

The study provides new findings on possible causes for these divergences in the context of

a particular regulatory system. The results indicate that firms with a lower proportion of

electricity delivered to industrial customers, which serve wealthier consumers and operate in

more densely populated areas, had lower bargaining power in the tariff setting and were harmed









by getting prices lower than recommended by the economic benchmarking methods. These

results are consistent with the economic theory of regulation which posits that political influence

affects the level of prices. On the other hand, firms with opposite characteristics had higher

bargaining power and benefited from higher prices. The evidence is consistent with an

association between per capital income and the effectiveness of residential interest group activity.

Moreover, the findings point to a possible inaccuracy of the cost parameters employed in the

engineering Model Company approach; the parameters may inaccurately capture the effect of

consumers' dispersion (customer density) on firms' operating costs, due to either the technical

difficulty in defining the "true" parameter in a context of imperfect and asymmetric information,

or a deliberate intention to avoid compensating investors in utilities operating in areas of higher

consumer concentration, and to provide extra returns to firms working in less densely populated

areas.

This benefit given at the beginning of the tariff review cycle impacted negatively the

incentives for efficiency improvements provided to firms which do not appear in the top ten

segments of SFA and DEA efficiency ranking. The same disincentive was received by four of

the top five firms in the SFA ranking, which could not keep part of the rents brought by their

productivity improvements. In sum, the regulator's methodology imposed on firms a one-time

adjustment to the virtual company's efficient operating costs, which in some cases were rather

different than the ones estimated by the benchmarking methods. Moreover, the rulings (and

associated price trajectories) ignored the significant frontier shifts of almost 7% per year

revealed by the parametric modeling, a point that would exacerbate the perverse effects of the

hypothesized over-evaluations of firms' efficiency.









Interestingly, the findings do not provide support to the hypothesis that the monitoring of

the regulator's activities may lead to decisions contrary to firms' interests and increase firms'

regulatory risk, one of the possible effects of having an institution supervise the regulator's job.

Regulator's decisions were not affected in a systematic way by special oversight. Despite its

specificity, the result adds to the literature on the optimal regulatory framework design.

It should be stressed that the results outlined above are robust to the choice of

benchmarking methodology (SFA or DEA) to employ as a comparison parameter. Moreover, the

results do not support those who are concerned with possible limitations of the SFA

methodology. However, for those who are more hesitant to abandon engineering models, at a

minimum the present investigation presents a way to promote greater transparency to the process

and credibility for the results obtained with the application of the Model Company method. Once

the divergences in efficiency assessments are identified, and possible explanatory factors are

uncovered, it remains the regulator's job to justify the choices made or demonstrate that the

divergences do not come from deficiencies in the application of a particular methodology.

The proposed joint use of a comparative efficiency analysis technique benefits all

stakeholders including the regulator; the agency could employ other benchmark techniques to

alleviate potential adverse selection problems and consequently come up with more reliable

approximations of firms' break-even points. It would then be possible to better exploit the price-

cap incentives for efficiency improvements and promote the desired allocation of productivity

gains among stakeholders. Note, on this topic, that the redistribution of rents should ideally be

based on information regarding the productivity increments of each firm during the whole

regulatory period prior to the next review, information that the model company approach alone

cannot provide.









The experience so far on setting price caps has indicated that quantitative benchmarking

techniques may at least serve as an additional tool to the regulator, whose importance is

underscored by information contained in comparisons available from having a large number of

companies in the regulated industry. Thus, there appears to be no reason for not using them in

the Brazilian electricity distribution industry.









APPENDIX A
DETAILS ON EVENT STUDY'S SAMPLE AND DATA












Table A-1. Missing Observation Problem (Number of computed stock returns missing)
COPEL ELETR ELETB CEMIG CESP LIGHT CELES TRACT EMAE FCATA COELCE TRANP CERJ PAULFL


Total missing


0 0 0 0 2 1 4 52 58 119 266 334 382 552


% missing 0.00 0.00 0.00 0.00 0.15 0.07 0.29 3.79 4.23 8.67 19.39 24.34 27.84 40.23


Missing 1998

Missing 1999

Missing 2000

Missing 2001

Missing 2002

Missing 2003


0 0 0 0 0 1 2 52 13 38 46 196

0 0 0 0 0 0 2 0 4 67 51 138


6 2

0 8


0 0 0 0 2 0 0 0 10 8 46 0 55 34

0 0 0 0 0 0 0 0 2 2 85 0 104 124

0 0 0 0 0 0 0 0 27 0 24 0 157 226

0 0 0 0 0 0 0 0 2 4 14 0 60 158


* 1: COELCE was not included in the sample because it has a high number of missing observations in all years examined. It was checked the possibility of its
inclusion in the regressions for year 2003, but the fact that this security has 38 observations missing in the 2002-2003 period was reducing significantly the total
number of observations and preventing the analysis of the significance of some announcements.











Table A-2. Companies in the Sample


COMPANY STO


COPEL PNB

ELETROPAULO PN

ELETROBRAS PNB

CEMIG PN

CESP PN

LIGHT ON

CELESC PNB

TRACTEBEL ON

EMAE PN

FCATAGUAZES PNA

TRANSM. PAULISTAPN

CERJ ON


ENERGY
DELIVERED TO
CK SECTOR OWNERSHIP I
FINAL
M Pu CONSUMERS (GWh)
M Public 17,629.1


Private

Public

Public

Public

Private

Public

Private

Public

Private

Public

Private


37,424.0

22,911.7*2

37,542.1

2,122.7

23,783.9

12,006.3

21.9


1,005.9


7,325.9


Total


PARTIC. IN ENERGY
DISTRIB. GENER.
MARKET (GWh)

5.8% 16,825


12.2%

7.5%

12.3%

0.7%

7.8%

3.9%




0.3%


2.4%

52.9%


165,022*3

32,561

32,505

4,144

374

18,605

2,614

233


241


PARTIC. IN
TOTAL
ENERGY
GENER.
4.6%


45.1%

8.9%

8.9%

1.1%

0.1%

5.1%

0.7%

0.1%


0.1%

74.7%


TOTAL ENERGY
ENERGY BILLED
SOLD (GWH) (R$1,000)
22,648.1 2,028,705

37,424.0 4,732,541

211,889.4*4 12,222,302

42,479.1 3,668,206

31,526.7 1,416,175

23,802.3 3,087,772

12,203.6 1,233,600

22.178,7 826,093

3,689.3 167,985

1,036.7 132,745

650,148

7.656.5 1,085,935


Source: Setor Eletrico Ranking 2001 Cadernos de Infra-Estrutura BNDES.
- "D" stands for Distribution; "G" for Generation; "T" for Transmission; and "M" for mixed companies (vertically integrated).
1: Although the energy delivered to final customers is very representative, it accounts for only 10.8% of the total energy sold by the company.
*2: Energy delivered by Chesf (7,546.3), Eletronorte (14,963.8) and Eletronuclear (401,6).
*3: It does not include the energy from Eletronuclear and CGTEE.
*4: It does not include the energy from Eletronuclear and CGTEE. It incorporates, however, the energy from Itaipu.













APPENDIX B
SFA AND DEA EFFICIENCY INDEXES


Mean
StdDeviation


COMPANY
AES-SUL
BANDEIRANTE
BANDEIRANTE 2
BOA VISTA
BRA GANTINA
CAIUA
CAT-LEO
CEAL
CEB
CEEE
CELB
CELESC
CELG
CELPA
CELPE
CELTINS
CEMAR
CEMAT
CEMIG
CENF
CEPISA
CERJ
CERON
CFLO
COELBA
COELCE
COPEL
COSERN
CPFL
ELEKTRO
ELETROPAULO
MANA US
ENERGIPE
ENERSUL
ESCELSA
LIGHT
NATIONAL
PIRATININGA
RGE
SAELPA
SANTA CRUZ
SANTA MARIA
. PARANAP.
COCEL
CSPE
DMEPC
ELETROACRE
ELETROCAR
JAGUAR
MOCOCA
SULGIPE
CPEE
X4AYERE


SFA DEA
1998 1999 2000 2001 2002 2003 1998 1999 2000 2001 2002 2003
1.062 1.056 1.076 1.078 1.101 1.093 1.000 1.000 1.000 1.000 1.000 1.000
1.378 1.205 1.146 1.000 1.000 1.000
1.077 1.074 1.000 1.096
1.071 1.101 1.057 1.046 1.085 1.080 1.238 1.381 1.333 1.196 1.205 1.217
1.079 1.091 1.121 1.120 1.128 1.151 1.325 1.294 1.575 1.623 1.629 1.600
1.101 1.107 1.106 1.149 1.132 1.148 1.397 1.364 1.517 1.639 1.692 1.733
1.073 1.066 1.078 1.075 1.066 1.057 1.869 1.582 1.873 2.024 1.946 2.174
1.090 1.079 1.084 1.073 1.071 1.070 1.575 1.464 1.555 1.656 1.664 1.449
1.113 1.096 1.103 1.105 1.097 1.106 1.443 1.376 1.359 1.295 1.393 1.379
1.154 1.121 1.113 1.161 1.174 1.133 1.520 1.447 1.451 1.372 1.397 1.337
1.110 1.094 1.083 1.065 1.063 1.059 1.460 1.357 1.361 1.481 1.418 1.120
1.424 1.358 1.353 1.304 1.512 1.283 1.572 1.447 1.420 1.218 1.261 1.170
1.065 1.135 1.171 1.181 1.126 1.140 1.536 1.515 1.828 1.984 1.761 1.692
1.092 1.165 1.176 1.188 1.151 1.129 1.412 1.333 1.239 1.136 1.217 1.153
1.227 1.202 1.124 1.098 1.102 1.090 1.381 1.304 1.307 1.387 1.399 1.233
1.089 1.096 1.085 1.118 1.100 1.095 2.410 2.058 2.257 2.801 2.525 2.381
1.083 1.108 1.123 1.118 1.081 1.072 1.377 1.460 1.451 1.441 1.433 1.340
1.063 1.114 1.135 1.145 1.112 1.127 1.818 1.445 1.618 1.678 1.575 1.437
1.422 1.620 1.593 1.493 1.498 1.506 1.000 1.000 1.000 1.000 1.000 1.000
1.157 1.124 1.129 1.135 1.090 1.072 1.309 1.309 1.316 1.403 1.441 1.266
1.111 1.139 1.138 1.094 1.129 1.093 1.748 1.706 1.642 1.692 1.724 1.577
1.144 1.126 1.116 1.125 1.126 1.115 1.437 1.342 1.534 1.490 1.570 1.522
1.111 1.118 1.130 1.089 1.095 1.094 1.486 1.406 1.451 1.178 1.277 1.258
1.048 1.059 1.079 1.071 1.084 1.101 1.047 1.227 1.107 1.075 1.124 1.138
1.104 1.075 1.091 1.121 1.080 1.084 1.486 1.109 1.383 1.709 1.439 1.284
1.114 1.097 1.103 1.081 1.068 1.068 1.115 1.060 1.100 1.120 1.096 1.000
1.144 1.131 1.148 1.057 1.142 1.132 1.453 1.192 1.330 1.064 1.302 1.499
1.081 1.056 1.062 1.064 1.072 1.069 1.300 1.009 1.235 1.326 1.318 1.144
1.216 1.185 1.132 1.127 1.129 1.092 1.073 1.000 1.127 1.000 1.004 1.000
1.122 1.109 1.081 1.070 1.082 1.174 1.164 1.193 1.004 1.183
1.489 1.289 1.390 1.210 1.131 1.155 1.000 1.000 1.000 1.000 1.000 1.000
1.051 1.098 1.058 1.074 1.061 1.062 1.000 1.160 1.136 1.000 1.000 1.000
1.091 1.070 1.098 1.111 1.112 1.093 1.342 1.062 1.316 1.404 1.387 1.221
1.074 1.080 1.085 1.068 1.066 1.065 1.773 1.475 1.748 1.783 1.712 1.618
1.108 1.135 1.114 1.095 1.078 1.074 1.208 1.074 1.215 1.295 1.171 1.274
1.233 1.179 1.148 1.131 1.105 1.069 1.000 1.000 1.055 1.000 1.000 1.000
1.065 1.075 1.086 1.063 1.088 1.113 1.000 1.088 1.112 1.116 1.189 1.171
1.127 1.081 1.000 1.000
1.049 1.059 1.064 1.055 1.057 1.045 1.499 1.000 1.142 1.133 1.092 1.000
1.088 1.083 1.099 1.100 1.103 1.097 1.546 1.264 1.502 1.616 1.553 1.259
1.080 1.071 1.078 1.091 1.077 1.088 1.328 1.222 1.565 1.799 1.773 1.656
1.091 1.081 1.080 1.103 1.083 1.101 1.182 1.067 1.182 1.282 1.224 1.233
1.090 1.116 1.118 1.147 1.138 1.179 1.295 1.321 1.493 1.567 1.647 1.647
1.115 1.087 1.083 1.092 1.100 1.132 1.000 1.000 1.000 1.000 1.000 1.000
1.093 1.063 1.052 1.050 1.053 1.077 1.473 1.323 1.222 1.112 1.151 1.206
1.146 1.112 1.115 1.119 1.106 1.096 1.217 1.285 1.292 1.299 1.233 1.024
1.038 1.062 1.071 1.065 1.088 1.060 1.000 1.000 1.000 1.000 1.000 1.000
1.076 1.054 1.058 1.075 1.087 1.090 1.000 1.000 1.000 1.000 1.000 1.000
1.135 1.081 1.068 1.091 1.099 1.147 1.000 1.000 1.000 1.000 1.000 1.000
1.056 1.044 1.045 1.053 1.060 1.083 1.000 1.000 1.000 1.000 1.093 1.073
1.047 1.066 1.071 1.096 1.103 1.146 1.000 1.181 1.235 1.292 1.321 1.471
1.148 1.100 1.055 1.066 1.071 1.101 1.686 1.479 1.064 1.014 1.041 1.145
1.103 1.098 1.148 1.143 1.141 1.196 1.000 1.052 1.202 1.172 1.131 1.116

1.128 1.121 1.121 1.113 1.113 1.111 1.327 1.243 1.314 1.341 1.318 1.279
0.100 0.092 0.092 0.072 0.084 0.070 0.299 0.224 0.269 0.354 0.314 0.298












APPENDIX C
MALMQUIST TFP INDEXES


1999/1998 2000/1999 2001/2000 2002/2001 2003/ '2r'
8.48% 6.74% 8.45% 6.64% 10.27%


23.83%


COMPANY
AES-SUL
BANDEIRANTE
BANDEIRANTE 2
BOA VISTA
BRAGANTINA
CAIUA
CAT-LEO
CEAL
CEB
CEEE
CELB
CELESC
CELG
CELPA
CELPE
CELTINS
CEALAR
CEMA T
CEMIG
CENF
CEPISA
CERJ
CERON
CFLO
COELBA
COELCE
COPEL
COSERN
CPFL
ELEKTRO
ELETROPAULO
ELN/AM (ANA US)
ENERGIPE
ENERSUL
ESCELSA
LIGHT
NATIONAL
PIRATININGA
RGE
SAELPA
SANTA CRUZ
SANTA MARIA
V. PARANAPANEMA4
COCEL
CSPE
DMAEPC
ELETROACRE
ELETROCAR
JAGUAR
MOCOCA
SULGIPE
CPEE
XANXERE

MAean
Cumulative Index


15.79%


-0.44%
2.32%
3.30%
5.25%
6.88%
9.12%
11.98%
5.51%
13.76%
0.90%
0.00%
10.24%
2.61%
3.60%
1.68%
-4.75%
5.56%
3.00%
9.74%
5.19%
1.00%
11.03%
9.61%
10.56%
9.47%
12.26%

25.34%
1.37%
8.40%
6.10%
5.43%
14.44%
1.98%

6.43%
6.36%
4.20%
3.05%
0.94%
4.13%
5.49%
5.55%
0.94%
2.89%
7.74%
2.54%
-0.91%
6.48%
1.64%

6.04%
6.04%


7.60%
1.18%
4.37%
3.89%
5.65%
7.23%
10.12%
5.95%
9.52%
5.12%
6.25%
15.50%
4.84%
5.31%
4.97%
12.86%
2.47%
6.23%
9.64%
4.97%
0.76%
7.22%
7.74%
8.11%
6.72%
14.95%
10.16%
2.29%
11.13%
3.84%
6.87%
10.50%
13.09%
2.46%

7.15%
4.74%
3.26%
2.53%
3.77%
2.47%
4.12%
2.73%
2.98%
0.93%
4.61%
1.80%
0.90%
6.64%
-3.16%

6.03%
12.44%


5.24%
4.77%
0.63%
5.64%
7.59%
8.28%
5.06%
7.00%
13.50%
8.18%
6.69%
11.40%
1.26%
7.83%
6.52%
18.56%
2.73%
10.55%
8.16%
10.43%
3.90%
6.18%
11.11%
19.19%
7.51%
11.01%
12.03%
26.20%
6.16%
5.76%
9.36%
10.62%
12.34%
6.13%

9.23%
6.79%
3.38%
0.59%
1.74%
1.62%
3.57%
2.65%
4.75%
0.20%
1.53%
1.65%
-0.62%
1.59%
2.58%

6.94%
20.25%


0.73%
4.55%
6.68%
6.83%
7.27%
9.83%
8.85%
5.65%
-6.77%
14.80%
11.63%
8.97%
6.67%
11.25%
11.00%
11.59%
7.92%
3.60%
9.28%
6.60%
2.44%
13.60%
10.62%
2.18%
7.48%
10.75%
10.81%
19.81%
9.46%
7.43%
8.47%
11.05%
13.60%
1.93%

8.76%
7.18%
6.68%
5.23%
5.89%
2.22%
3.35%
4.60%
2.26%
1.46%
3.54%
2.59%
1.47%
2.55%
2.83%

6.88%
28.52%


10.78%
5.80%
3.71%
4.36%
7.46%
7.97%
8.73%
14.32%
6.27%
28.56%
9.06%
10.90%
11.25%
6.05%
9.02%
7.04%
11.91%
6.17%
11.25%
11.00%
7.83%
2.64%
9.89%
9.98%
12.59%
9.19%
15.22%
8.90%
10.63%
8.89%
9.84%
9.01%
10.24%
15.04%
2.55%
15.38%
10.82%
8.34%
4.80%
2.26%
1.79%
0.32%
1.86%
4.68%
7.44%
2.87%
0.16%
1.47%
-1.21%
0.71%
-1.85%

7.77%
38.50%


2003/1998
47.67%
43.39%
10.78%
20.16%
17.61%
20.78%
32.62%
40.72%
51.31%
61.21%
34.29%
69.48%
43.65%
40.32%
71.96%
23.23%
42.70%
35.08%
59.16%
27.31%
39.42%
57.86%
40.16%
11.17%
57.77%
59.63%
63.90%
47.40%
82.79%
48.91%
114.45%
42.55%
40.48%
46.63%
57.77%
90.02%
15.92%
15.38%
50.14%
38.17%
24.36%
14.36%
14.86%
11.20%
19.76%
21.86%
19.63%
8.61%
18.67%
10.45%
-0.40%
19.13%
1.90%









REFERENCE LIST


Ai, C., Martinez S., & Sappington, D. (2004), "Incentive Regulation and Telecommunications
Service Quality", Journal ofRegulatory Economics 26, 263-285.

Amihud, Y., & Mendelson, H. (1986), "Asset Pricing and the Bid-Ask Spread", Journal of
Financial Economics, 17, 223-249.

ANEEL (2003), "Metodologia e Calculo da "Empresa de Refer6ncia" relative a area de
concessdo da AES SUL", Nota Tecnica N 047/2003/SRE/ANEEL, Anexo I.

Antoniou, A., & Pescetto, G. (1977), "The Effect of Regulatory Announcements on the Cost of
Equity Capital of British Telecom", Journal of Business Finance and Accounting, 24, 1-
25.

Armstrong, M., & Sappington, D. (2003), "Recent Developments in the Theory of Regulation",
Handbook ofIndustrial Organization (Vol. III), M. Armstrong and R. Porter, eds., Elsevier
Science Publishers, forthcoming.

Atkinson, S. E., Comwell, C., & Honerkamp, O. (2003), "Measuring and Decomposing
Productivity Change: Stochastic Distance Function Estimation Versus Data Envelopment
Analysis", Journal of Business & Economic Statistics 21, 284-294.

Bagdadioglu, N., Waddams Price, C. M., & Weyman-Jones, T. G. (1996), "Efficiency and
Ownership in Electricity Distribution: A Non-Parametric model of the Turkish
Experience", Energy Economics 18, 1-23.

Bardelin, C.E.A. (2004), "Os efeitos do Racionamento de Energia Eletrica ocorrido no Brasil em
2001 e 2002 com Enfase no Consumo de Energia Eletrica", Dissertacgo de Mestrado,
Universidade de Sdo Paulo.

Beneish, M. (1991), "The Effect of Regulatory Changes in the Airline Industry on Shareholders'
Wealth", Journal of Law andEconomics, 98, 395-430.

Beneish, M., & Whaley, R. (1996), "An anatomy of the S&P game: The effects of changing the
rules", Journal ofFinance, 51, 1909-1930.

Berg, S., Lin, C., & Tsaplin, V. (2005), "Regulation of State-Owned and Privatized Utilities:
Ukraine Electricity Distribution Company Performance", Journal ofRegulatory
Economics 28, 259-287.

Binder, J.J. (1985a), "On the use of the Multivariate Regression Model in Event Studies",
Journal of Accounting Research, 23(1), 370-383.

Binder, J.J. (1985b), "Measuring the effects of regulation with stock price data", Rand Journal of
Economics, 16(2), 167-221.









Binder J.J. (1998), "The Event Study Methodology Since 1969", Review of Quantitative Finance
andAccounting, 11, 111-137.

BNDES (2001), Cadernos de Infra-Estrutura, Setor Eletrico Ranking 2001, Vol. 1.

Botasso, A., & Conti, M. (2003), "Cost Inefficiency in the English and Welsh Water Industry:
An Heteroskedastic Stochastic Cost Frontier Approach", Working Paper.

Brown, A., & de Paula, E. (2002), "Strengthening of the Institutional and Regulatory Structure of
the Brazilian Power Sector", Report, The World Bank.

Brown, A., & de Paula, E. (2004), "PPIAF Phase 2 Project for Brazil Power Sector, Task 4:
Strengthening of the Institutional and Regulatory Structure of the Brazilian Power Sector",
Report, The World Bank.

Brown, S.J., & Warner, J.B. (1985), "Using Daily Stock Returns: The Case of Event Studies",
Journal ofFinancial Economics, 14, 3-31.

Buckland, R., & Fraser, P. (2001), "Political and Regulatory Risk: Beta Sensitivity in UK
Electricity Distribution", Journal ofRegulatory Economics, 19(1), 5-25.

Burns, P., & Weyman-Jones, T. G. (1996), "Cost Functions and Cost Efficiency in Electricity
Distribution: A Stochastic Frontier Approach", Bulletin of Economic Research 48, 41-64.

Byatt, I. (2004), "Regulation and Asset Valuation", Report, The World Bank.

Carroll, K.A., & Landim, D.J. (1993), "Measuring market response to regulation of the cable tv
industry", Journal ofRegulatory Economics, 5, 385-399.

Chordia, T., Shivakumar, L., & Subrahmanyam, A. (2004), "Liquidity Dynamics Across Small
and Large Firms", Economic Notes by Banca Monte dei Paschi di Siena SpA, 33(1), 111-
143.

Coelli, T., Perelman, S., & Romano, E. (1999), "Accounting for Environmental Influences in
Stochastic Frontier Models: With Application to International Airlines", Journal of
Productivity Analysis 11, 251-273.

Coelli, T., Prasada Rao, D., & Battese, G. (1998), An Introduction to Efficiency and Productivity
Analysis. Boston, MA: Kluwer Academic Publishers.

Cubbin, J., & Tzanidakis, G. (1998), "Regression versus data envelopment analysis for
efficiency measurement: an application to the England and Wales regulated water
industry", Utilities Policy 7, 75-85.

Dassler, T., Parker, D., & Saal, D.S. (2006), "Methods and trends of performance benchmarking
in UK utility regulation", Utilities Policy 14, 166-174.









de Oliveira, A. (2003), "The Political Economy of the Brazilian Power Industry Reform",
Working paper.

Di Tella, R., & Dyck, A. (2002), "Cost Reductions, Cost Padding and Stock Market Prices: the
Chilean Experience with Price Cap Regulation", Working Paper No. 03-050, Harvard
Business School.

Dnes, A.W., Kodwani, D.G., Seaton, J.S., & Wood, D. (1998), "The Regulation of the United
Kingdom Electricity Industry: An Event Study of Price-Capping Measures", Journal of
Regulatory Economics, 13, 207-225.

Dnes, A.W., & Seaton, J.S. (1999), "The regulation of electricity: results from an event study",
Applied Economics, 31, 609-618.

Elyasiani, E., Hauser, S., & Lauterbach, B. (2000), "Market response to liquidity improvements:
Evidence from exchange listings", Financial Review, 41, 1-14.

Estache, A., Perelman, S., & Trujillo, L. (2005), "Infrastructure Performance and Reform in
Developing and Transition Economies: Evidence from a Survey of Productive Measures",
World Bank Policy Research Working Paper, n. 3514

Estache, A., Rossi, M. A., & Ruzzier, C. A. (2004), "The Case for International Coordination of
Electricity Regulation: Evidence from the Measurement of Efficiency in South America",
Journal ofRegulatory Economics, 25(3), 271-295.

Fama, E. F. (1976), Foundations ofFinance (Basic Books, New York).

Ferreira, C.K.L. (2000), "Privatizacao do Setor Eletrico no Brasil", in Pinheiro, A. C. and K.
Fukisaku (eds.), A privatizacao no Brasil: o caso dos services de utilidade public, Rio de
Janeiro: BNDES / OCDE.

Filippini, M., & Wild, J. (2001), "Regional differences in electricity distribution costs and their
consequences for yardstick regulation of access prices", Energy Economics 23, 477-488.

Forsund, F., & Kittelsen, S. (1998), "Productivity Development of Norwegian Electricity
Distribution Utilities", Resource andEnergy Economics 20(3), 207-224.

Foster, V., & Antmann, P. (2004), "The Regulatory Challenge of Asset Valuation: A Case Study
from the Brazilian Electricity Distribution Sector", ENERGY Working Notes, 2, World
Bank.

Galetovic, A., & Bustos, A. (2002), "Regulaci6n pro empresa eficiente: ,quien es realmente
usted?", Estudios Publicos, 86, 145-182.

Goldstein, A. (1999), "Brazilian privatization in international perspective: the rocky path from
state capitalism to regulatory capitalism", Industrial and Corporate Change, 8, 673-711.









Gomes, V., Pess6a, S. A., & Veloso, F. A. (2003), "Evolucgo da Produtividade Total dos Fatores
na Economia Brasileira: Uma Analise Comparativa", mimeo.

Gomez-Lobo, A., & Vargas, M. (2001), "La regulaci6n de las empresas sanitarias en Chile: una
revision del caso de EMOS y una propuesta de reform regulatoria", Documento de
Trabajo N 177, Departamento de Economia, Universidad de Chile.

Greene, W. (2005a), "Reconsidering heterogeneity in panel data estimators of the stochastic
frontier model", Journal ofEconometrics 16, 269-303.

Greene, W. (2005b), "Fixed and Random Effects in Stochastic Frontier Models", Journal of
Productivity Analysis 23, 7-32.

Grifell-Tatje, E., & Lovell, C.A.K. (2003), "The Managers versus the Consultants",
Scandinavian Journal of Economics, 105, 119-138.

Hattori, T. (2002), "Relative Performance of U.S. and Japanese Electricity Distribution: An
Application of Stochastic Frontier Analysis", Journal ofProductivity Analysis 18, 269-
284.

Hattori, T., Jamasb, T., & Pollitt, M. G. (2003), "A Comparison of UK and Japanese electricity
distribution performance 1985-1998: lessons for incentive regulation", DAE Working
Paper WP 0212, University of Cambridge.

Irastorza, V. (2003), "Benchmarking for Distribution Utilities: A Problematic Approach to
Defining Efficiency", The Electricity Journal 16(10), 30-38.

Jadresic, A. (2002), "The Model Company Approach for Tariff Regulation in Electricity
Distribution", Report, The World Bank.

Jamasb, T., & Pollitt, M. G. (2001), "Benchmarking and Regulation: International Electricity
Experience", Utilities Policy 9, 107-130.

Jamasb, T., Nillesen, P., & Pollitt, M. G. (2003), "Gaming the Regulator: A Survey", The
Electricity Journal XX, 68-80.

Joskow, P. L. (2005), "Incentive Regulation in Theory and Practice: Electricity Distribution and
Transmission Networks", mimeo.

Klein, C.C., & Sweeney, G. H. (1999), "Regulator preferences and utility prices: evidence from
natural gas distribution utilities", Energy Economics, 21, 1-15.

Knittel, C.R. (2006), "The Adoption of State Electricity Regulation: The Role of Interest
Groups", The Journal ofIndustrial Economics, 54(2), 201-222.

Kridel, D., Sappington, D., & Weisman, D. (1996), "The Effects of Incentive Regulation in the
Telecommunications Industry: A Survey", Journal ofRegulatory Economics, 9, 269-306.









Kumbhakar, S. C., & Hjalmarsson, L. (1998), "Relative Performance of Public and Private
Ownership under Yardstick Competition: Electricity Retail Distribution", European
Economic Review 42, 97-122.

Kumbhakar, S. C., & Lovell, C. A. K. (2000), Stochastic Frontier Analysis. Cambridge
University Press.

Laffont, J-J., & Tirole, J. (1993), A Theory ofIncentives in Procurement and Regulation,
Cambridge, MA: MIT Press.

Lamdin, D.J. (1999), "Event studies of regulation and new results on the effect of the cigarette
advertising ban", Journal ofRegulatory Economics, 16, 187-201.

Lowry, M. N., & Kaufmann, L. (1999), The Cost Structure of Power Distribution. Pacific
Economics Group, Madison, WI.

Lynch, A., & Mendenhall, R. (1997). "New evidence on stock price effects associated with
changes in the S&P 500 index", Journal ofBusiness, 70, 351-383.

Maurer, L., Pereira, M., & Rosenblatt, J. (2005), "Implementing Power Rationing in a Sensible
Way: Lessons Learned and International Best Practices", Report 305/05, The World Bank.

Morana, C., & Sawkins, J.W. (2002), "Stock Market Reaction to Regulatory Price Reviews in
the English and Welsh Water Industry", Journal ofRegulatory Economics, 22(2), 185-204.

Mota, R. L. (2003), "The Restructuring and Privatisation of Electricity Distribution and Supply
Businesses in Brazil: A Social Cost-Benefit Analysis". Working Paper, Churchill College,
University of Cambridge.

Mota, R. L. (2004), "Comparing Brazil and U.S.A. Electricity Distribution Performance: What
was the Impact of Privatisation?", Working Paper, Churchill College, University of
Cambridge.

Naughton, M.C. (1989), "Regulatory preferences and two-part tariffs: the case of electricity",
Southern Economic Journal, 55, 743-758.

Nelson, R.A. (1982), "An empirical test of the Ramsey theory and Stigler-Peltzman theory of
public utility regulation", Economic Inquiry, 20, 277-290.

Nelson, J.P., & Roberts, M.J. (1989), "Ramsey numbers and the role of competing interest
groups in electric utility regulation", Quarterly Review of Economics and Business, 29, 21-
42.

Neuberg, L. (1977), "Two Issues in the Municipal Ownership of Electric Power Distribution
Systems", Bell Journal of Economics 8, 303-323.

Oliveira, A., & Pires, J.C.L. (1994), "Setor Eletrico Brasileiro: diagn6stico e perspectives". Rio
de Janeiro: IE/UFRJ, mimeo.









Papke, L.E., & Wooldrige, J. (1996), "Econometric methods for fractional response variables
with an application to 401(k) plan participation rates", Journal ofAppliedEconometrics
11,619-632.

Peano, C.R. (2005), "Regulacdo Tarifaria do Setor de Distribuicgo de Energia Eletrica no Brasil:
Uma Analise da Metodologia de Revisdo Tarifaria Adotada pela Aneel", Dissertacgo de
Mestrado, Universidade de Sdo Paulo.

Peltzman, S. (1976), "Toward a More General Theory of Regulation", Journal of Law and
Economics, 19, 211-40.

Pires, J.C.L., & Goldstein, A. (2001), "Agnncias Reguladoras Brasileiras: Avaliacgo e Desafios",
Revista do BNDES, 8 (16), 3-42.

Pollitt, M. G. (2005), "The Role of Efficiency Estimates in Regulatory Price Reviews: Ofgem's
Approach to Benchmarking Electricity Networks", Utilities Policy 13, 279-288.

Prager, R.A. (1992), "The effects of deregulating cable television: evidence from the financial
markets", Journal ofRegulatory Economics, 4, 347-363.

Primeaux, Jr. W.J, Filer, J.E., Herren, R.S., & Hollas, D.R. (1984), "Determinants of regulatory
policies toward competition in the electric utility industry", Public Choice, 43, 173-186.

Roberts, M. J. (1986), "Economies of Density and Size in the Production and Delivery of
Electric Power", LandEconomics 62, 378-387.

Rose, N.L. (1985), "The incidence of regulatory rents in the motor carrier industry", Rand
Journal ofEconomics, 16, 299-318.

Rossi, M.A., & Ruzzier, C.A. (2000), "On the regulatory application of efficiency measures",
Utilities Policy 9, 81-92.

Salinger, M. (1992), "Standard Errors in Event Studies", Journal ofFinancial and Quantitative
Analysis, 27(1), 39-53.

Salvanes, K. G., & Tjotta, S. (1994), "Productivity Differences in Multiple Output Industries: An
Application to Electric Distribution", Journal ofProductivity Analysis 5, 23-43.

Sappington, D. (2002), "Price Regulation", in The Handbook of Telecommunications Economics.
Volume I: Structure, Regulation, and Competition, edited by M. Cave, S. Majumdar, and I.
Vogelsang. Amsterdam: Elsevier Science Publishers, 225-293.

Sarafidis, V. (2002), "An Assessment of Comparative Efficiency Measurement Techniques",
Europe Economics Occasional Paper 2, London.

Sawkins, J.W. (1996), "Balancing multiple interests in regulation: an event study of the English
and Welsh water industry", Journal ofRegulatory Economics, 9, 249-268.









Schipper, K., & Thompson, R. (1983), "The Impact of Merger-Related Regulations on the
Shareholders of Acquiring Firms", Journal ofAccounting Research, 21, 184-221.

Schwert, G.W. (1981), "Measuring the Effects of Regulation: Evidence from the Capital
Markets", Journal ofLaw and Economics, 24, 121-145.

Serra, P. (2002), "Regulaci6n del sector electric chileno", Perspectivas, 6, 11-43.

Shuttleworth, G. (2003), "Firm-Specific Productive Efficiency: A Response", The Electricity
Journal 16(3), 42-50.

Silva, H.C.D. (2006a), "An Empirical Assessment of the Regulator's Performance in the
Brazilian Electricity Sector", University of Florida, mimeo.

Silva, H.C.D. (2006b), "Privatization, Incentive Regulation, and Efficiency Improvements in the
Brazilian Electricity Distribution Industry", University of Florida, mimeo.

Simar, L., & Wilson, P. (1998). "Sensitivity Analysis of Efficiency Scores: How to Bootstrap in
Nonparametric Frontier Models." Management Science 44: 49-61.

Stem, J. (2005), "UK regulatory price reviews ant the role of efficiency estimates", Utilities
Policy 13, 273-278.

Stigler, G.J. (1971), "The Theory of Economic Regulation", Bell Journal of Economics and
Management Science, 2, 3-21.

Wang, H. (2002), "Heteroskedasticity and Non-Monotonic Efficiency Effects of a Stochastic
Frontier Model", Journal ofProductivity Analysis 18, 241-53.

Weisman, D.L. (2000), "The inefficiencyy of the "efficient-firm" cost standard", The Antitrust
Bulletin, Spring, 195-211.









BIOGRAPHICAL SKETCH

Hamilton Caputo Delfino Silva is a Senior Analyst at Tribunal de Contas da Uniao TCU

(Federal Audit Office), an independent organ of the state, which assists in the external control

that the Congress possesses over the whole public administration.

Mr. Silva was awarded the Bachelor in Business Administration degree in 1986, from the

Universidade de Brasilia-Brazil. In 1994, he was awarded a Master in Business Administration

degree from Fundacao Getulio Vargas in Sao Paulo, Brazil.

Mr. Silva joined the University of Florida as a doctoral student in 2001. After completing

the course requirements and being admitted to candidacy, Mr. Silva moved back to Brazil in

December of 2004, where he split his time between working at TCU and performing the job

needed to finish his dissertation. Mr. Silva was awarded the Doctor of Philosophy degree in

Economics in August of 2007.





PAGE 1

1 EMPIRICAL ESSAYS IN TH E ECONOMICS OF REGULATION By HAMILTON CAPUTO DELFINO SILVA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2007

PAGE 2

2 2007 Hamilton Caputo Delfino Silva

PAGE 3

3 To Ethevaldo and Arlete Rita, Leonardo, and Fernanda

PAGE 4

4 ACKNOWLEDGMENTS The enduring process of obtaining a doctorate degree involves sacrif ices, encouragement, and support from several sources. I am deeply grateful to my parents. Their unconditional love, encouragement, and example have always been the major inspiration to achie ve my goals, and to them I owe my education. I am extremely indebted to my wife Rita. Th e decision to pursue the doctorate, in a foreign country and without any help fr om my organization in Brazil, imposed an extraordinary burden on her. I owe this accomplishment to her strength and courage, as well as to her resilience in the not so rare difficult moments. I also thank my children, Leonardo and Fernanda, for being very accommodating and understanding the importance of this project for their dad. The support provided by my brothers Emilson, Edilene, and Anderson was decisive as well. Their words of encouragement were ofte n remembered when the strength and motivation became to fade. In addition, Emilsons help was critical for the starting of my doctorate. I would like to express my grat itude to my advisor Sanford Berg for his support since the first day of doctoral studies. His back-up and dedi cation to this project were decisive. A special thanks also goes to Professor Larry Kenny, for hi s commitment and very helpful comments and suggestions. Finally, I am grateful for the financial support of the Public Utility Research Center and the Florida/Brazil Institute. This project would not have even st arted without the help of these organizations.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 ABSTRACT....................................................................................................................... ..............9 CHAPTER 1 INTRODUCTION..................................................................................................................12 2 AN EMPIRICAL ASSESSMENT OF THE REGULATORS PERFORMANCE IN THE BRAZILIAN ELECTRICITY SECTOR.......................................................................16 Introduction................................................................................................................... ..........16 Institutional Background....................................................................................................... .20 Methodology.................................................................................................................... .......23 Data........................................................................................................................... ..............28 Findings....................................................................................................................... ...........32 Robustness Checks.............................................................................................................. ...43 Conclusions.................................................................................................................... .........44 3 PRIVATIZATION, INCENTIVE RE GULATION, AND EFFICIENCY IMPROVEMENTS IN THE BRAZILI AN ELECTRICITY DISTRIBUTION INDUSTRY....................................................................................................................... .....61 Introduction................................................................................................................... ..........61 Institutional Background....................................................................................................... .64 Methodology.................................................................................................................... .......66 The Electricity Distribution Technology.........................................................................66 Comparative Efficiency Studies......................................................................................69 Stochastic Cost Frontier and Treatme nt of Environmental Variables.............................72 Specification and Data......................................................................................................... ...77 Findings....................................................................................................................... ...........84 Service Quality and Economies of Vertical Integration.........................................................94 Conclusion..................................................................................................................... .........99 4 THE ASSESSMENT OF FIRMS EFFICIEN CY IN PERIODIC TARIFF REVIEWS: AN EVALUATION OF THE REFE RENCE UTILITY APPROACH................................109 Introduction................................................................................................................... ........109 Institutional Background and the ANEEL Model Company Method..................................113 The Tariff Review Methodology...................................................................................114 Model Company Estimates............................................................................................117 Comparative Efficiency Analysis.........................................................................................118

PAGE 6

6 SFA Model and Data.....................................................................................................120 SFA Results and Comparison........................................................................................123 Econometric Modeling.........................................................................................................127 Specification and Data...................................................................................................131 Results Analysis............................................................................................................137 Robustness Check: DEA.......................................................................................................140 Concluding Observations......................................................................................................144 5 SUMMARY AND CONCLUSIONS...................................................................................154 APPENDIX A DETAILS ON EVENT STUDYS SAMPLE AND DATA................................................161 B SFA AND DEA EFFICIENCY INDEXES..........................................................................164 C MALMQUIST TFP INDEXES............................................................................................165 REFERENCE LIST................................................................................................................. ....166 BIOGRAPHICAL SKETCH.......................................................................................................173

PAGE 7

7 LIST OF TABLES Table page 2-1 Events List................................................................................................................ .........46 2-2 Descriptive Statistics..................................................................................................... .....48 2-3 Correlation Matrix......................................................................................................... ....50 2-4 Securities Returns, Stock Market Return s and Exchange Rate Variation, by Year..........51 2-5 Individual Regressions Results..........................................................................................52 2-6 Hypothesis Tests Results for Events in Year 1999............................................................53 2-7 Hypothesis Tests Results for Events in Year 2000............................................................54 2-8 Hypothesis Tests Results for Events in Year 2001............................................................55 2-9 Hypothesis Tests Results for Events in Year 2002............................................................56 2-10 Hypothesis Tests Results for Events in Year 2003............................................................57 2-11 Hypothesis Tests Results for CARs Before Significant Events.........................................58 2-12 Hypothesis Tests' Results for Events' Overall Effect.........................................................58 2-13 Significant Announcements Categorization. Direction and Estimated Magnitude of Regulatory Announcements Effect on Security Returns..................................................59 2-14 Random Events' Results.................................................................................................... .60 3-1 Descriptive Statistics..................................................................................................... ...102 3-2 Stochastic Cost Frontier Results......................................................................................103 3-3 Elasticities............................................................................................................... .........104 3-4 Technological Change by Year and Ownership..............................................................104 3-5 Efficiency Evolution....................................................................................................... .105 3-6 Decomposition of Privatized Firms' Efficiency Evolution..............................................105 3-7 Productivity Growth Rate and Decomposititon...............................................................106 3-8 Productivity Growth Rate and Decomposition by Ownership Type...............................106

PAGE 8

8 3-9 Productivity Growth Rate and Decomposition by Ownership Type Firms with Q > 400,000 MWh/year..........................................................................................................107 3-10 Mean Service Quality Indexes.........................................................................................107 3-11 Vertical and Quality as additional regressors or as mean inefficiency parameters.........108 4-1 Initially Estimated OPEX, Final OPEX, and Firms Reported OPEX............................148 4-2 SFA Descriptive Statistics...............................................................................................149 4-3 GLM and OLS Descriptive Statistics..............................................................................150 4-4 Stochastic Cost Frontier Results......................................................................................151 4-5 Efficiency Rankings and Indexes.....................................................................................152 4-6 Regression Results......................................................................................................... ..153 A-1 Missing Observation Problem (Number of computed stock returns missing).................162 A-2 Companies in the Sample.................................................................................................163

PAGE 9

9 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy EMPIRICAL ESSAYS IN THE ECONOMICS OF REGULATION By Hamilton Caputo Delfino Silva August 2007 Chair: Sanford Berg Major: Economics Reform processes undertaken on networ k industries have encompassed unbundling, privatization, introduction of market-oriented regimes for their competitive segments, and implementation of a new regulatory framework for the remaining segments with natural monopoly characteristics. The new regulator y framework has involved the creation of independent regulatory bodies a nd the incorporation of theoreti cal advances from the economics literature on incentive regulation, under the ultimate objective of providing the conditions and incentives for efficiency improvement and for th e possible achievement of second best prices. This dissertation adds to the literature on the im pacts of the restructuring measures implemented and contains three empirical essays on the reforms accomplished in the Brazilian electricity sector. In Chapter 2, information on stock market reactions to regulatory announcements are employed to investigate the level of consistency and predictability of decisions taken by the Brazilian electricity sector regul ator and to identify the extent the aforementioned regulatory body has been able to act indepe ndently and to balance multiple interests in re gulation. The findings indicate that the regulat ors decisions have not favored a single interest group. The evidenced unpredictability of the regulatory agencys decisions suggests the need for

PAGE 10

10 improvements in the regulatory di scussion process, with the adopti on of measures to increase the transparency and to promote more substantive public hearings. Additi onally, the study shows that the need to provide incentives for new inve stments has played a significant role in the regulatory process. The estimates indicate that regulatory decisions accounted for roughly 12% of the difference in performance of the sample securities with respect to the market index. The Chapter 3 examines the impact of pr ivatization and incentive regulation on firms performance in the period of 1998 to 2003. Evid ence indicates performance improvement after the implementation of sector reforms, with bot h privatized and public companies reducing the efficiency gap with respect to companies that were privately owned before the reforms. The results show that privatized firms responded mo re aggressively than public firms to the new incentives brought by price-cap regulation, and s uggest that the high performance improvement experienced by privatized firms did not come from mere reduction in costs brought by deterioration in the quality of service. The findi ngs also indicate a possible strategic behavior associated with the periodic aspect of pric e-cap regulation, as well as to cost shifting implemented by companies that operate in the electricity generation segment. The Chapter 4 is motivated by the increasing use of the model company approach to determine electricity distribution tariffs in Latin America, despite the criticisms made to the methods subjectivity and obscurity. The study ex amines whether the use of the engineering approach in the Brazilian elec tricity distribution s ector periodic tariff review enabled the attainment of the welfare maximizer regulato rs rate setting object ives, by comparing the methods implied performance scores to e fficiency measures provided by economic benchmarking approaches. Results s how that some firms, mainly th e ones serving more affluent consumers, operating in more densely populat ed areas and having a lower proportion of

PAGE 11

11 electricity delivered to industria l customers, received substantia lly lower repositioning indexes than the economic benchmarking methods would recommend, pointi ng to a possible violation of firms break-even constraints. The findings al so reveal that signifi cantly higher repositioning indexes might have been given to companie s with the opposite characteristics, negatively affecting the incentives for further productivity imp rovements, as some of the possibly benefited companies do not appear in the top ten of the benchmarking efficiency rankings.

PAGE 12

12 CHAPTER 1 INTRODUCTION Network industry reforms recently impleme nted have involved unbundling, privatization, introduction of market-oriented regimes for their competitive segments, and implementation of a new regulatory framework for the remaining segmen ts with natural monopoly characteristics. One of the features of the new regulator y framework has been the emergence of autonomous regulatory bodies. The associated in crease in regulatory discretion, however, has raised concerns over the possible influence of interest groups on regulatory outcomes, strengthening the debate over th e pattern of government interven tion in business, represented by two main opposing theories: the public interest approach, whic h asserts that the government acts to lessen or eliminate the inefficiencies e ngendered by market failure; and the capture or interest group approach, which argues that the government acts to tr ansfer wealth between interest groups in an effort to maximize poli tical support. Examining the actual behavior of regulatory agencies, in this context, constitu tes an empirical question, addressed by Sawkins (1996), Dnes, Kodwani, Seaton, and Wood ( 1998), and Dnes and Seaton (1999) on the perspective of the energy and wate r industries in Un ited Kingdom. The new regulatory design has also comprised the incorporation of theoretical advances from the economics literature on incentive regul ation, notably the implementation of incentive mechanisms, such as the price-cap method. In case, the firm and its managers are the residual claimants on production cost reduc tions, and bear the disutility of increased managerial effort (Joskow, 2005). As a result, the conditions and incen tives for efficiency improvement and for the possible achievement of second best prices are settled. However, whethe r price-cap regulation effectively leads to efficiency improvement also constitutes an empirical question.

PAGE 13

13 This dissertation examines thes e questions in the context of the reforms accomplished in the Brazilian electricity sect or. The reforms began in 1995. While constitutional amendments abolished the public monopoly over infrastructure industries and allowed foreign companies to bid for public concessions, the Law 8,987/95 (General Law of Concessions) set the stage for the beginning of the privatization process, represen ted by the auctions of Escelsa in 1995 and Light in 1996. By the end of 2000, a total of 20 dist ribution companies had been privatized. In addition, the implementation of a new regul atory framework involved the establishment of an independent regulatory agency (ANEEL) in late 1996 and the institution of a new model for the electricity sector in 1998. The mode l focused on privatization and unbundling of generation, transmission and distribution assets, gradual transition to a competitive generation environment in nine years, creation of a whol esale power market, opera tion of the transmission network by an independent operato r, and use of the price-cap re gime to regulate distribution tariffs, replacing the previous cost of servic e system. Price-cap regulation was implemented through the signature of new concession cont racts, which took place from 1998 to 2000, and scheduled the first tariff review fo r after five (for contracts signed in 1998) or four years. As a result, 61 companies were submitted to a tariff review process from April/2003 to February/2006. The analysis is performed in Chapters 2 to 4. The Chapter 2 (An Empirical Assessment of the Regulators Performance in the Brazilian El ectricity Sector) uses information on stock market reactions to regulatory announcements to investigate the level of consistency and predictability of decisions taken by the Brazilian electricity sector regulat or, and to identify the extent the aforementioned regulatory body has been able to act independently and to balance multiple interests in regulation. The investiga tion improves upon previous studies by focusing in

PAGE 14

14 the context of a developing country, where the interests are presumably more pronounced, and by employing an identification strategy that expl icitly accounts for the possibility of event anticipation. In additi on, the study sheds light on the debate over the possible capture of Brazilian regulatory agencies. The Chapter 3 (Privatization, Incentive Regul ation, and Efficiency Improvements in the Brazilian Electricity Distribution Industry) concentrates on the effectiveness of the price-cap incentive mechanism issue. The study examines the impact of priva tization and incentive regulation on performance of Brazi lian electricity distribution comp anies in the period of 1998 to 2003, employing a Stochastic Frontier Approach (SFA) that controls for heterogeneity in operating conditions, influence of macro economic factors, and random shocks. The investigation evaluates th e efficiency evolution and the productivity gains that occurred in the period, checks for difference in performance between public and private firms, and looks at the possibilit y of efficiency catch-up. It also exam ines whether vertically integrated firms might be behaving strategical ly, shifting costs from unregulat ed to regulated activities, and whether efficiency changes are associated with variations in service quality. The Chapter 4, entitled The Assessment of Fi rms Efficiency in Periodic Tariff Reviews: An Evaluation of the Reference Utility Approach, is an extension of the previous one. The obtained SFA efficiency estimates and measur es of firms productivity improvements, along with efficiency measures provided by an alte rnative benchmarking technique (Data Envelopment Analysis), are employed to examine the results derived from the use of the Reference Utility approach at the distribution compan ies first periodic tariff review. The study is motivated by the increasing use of the reference utility model to determine electricity distribution tariffs in Latin America, despite the cr iticisms made to the methods

PAGE 15

15 subjectivity and obscurity. The analysis ch ecks whether the methodology has enabled the attainment of the welfare maxi mizer regulators rate setting ob jectives, and is based upon the reasoning that substantial (and consistent) divergences in the Model Companys implied performance scores, relative to efficiency measures provided by economic benchmarking approaches, reflect deficiencies in the engineering methods application. In sequence, the investigati on checks for the possible causes of the divergences found, not only exploring the predictions of the same inte rest group theory of regulation employed in the second chapter, but also accounting for the fact th at the regulators decisions were taken in an incomplete and imperfect information context. In cas e, it is examined the possibility of flaws in the engineering cost parameters employed to esti mate the efficient costs and the potential use of some of the available data as signals for firms profitability and cash flow availability, as a subsidy for the regulators decisions regard ing the distribution of productivity gains among stakeholders. The three chapters mentioned above are self-contained, presen ting the results obtained and the corresponding conclusions. Nonetheless, the ma in results and conclusions evidenced in the three studies that compound this dissertation are outlined in Chapter 5, along with directions for future research.

PAGE 16

16 CHAPTER 2 AN EMPIRICAL ASSESSMENT OF THE RE GULATORS PERFORMANCE IN THE BRAZILIAN ELECTRICITY SECTOR Introduction The analysis of regulatory impacts provides insi ghts into the level of predictability and consistency of regulatory actions, and is esse ntial if we are to understand how regulatory processes and regulatory decisions affect infrastructure perfor mance. The subjects importance has increased with the emergence of autonomou s regulatory bodies, in th e context of network industry reforms recently implemented, as the associ ated boost in regulatory discretion has raised concerns over the possible influe nce of interest groups on regulato ry outcomes, although the role of special interests within pred ecessor government ministries wa s not insignificant, and probably less transparent. The issues raised by regulation continue a debate over the pattern of government intervention in business, which ha s long been carried out in the fi elds of economics and political science and resulted in two main opposing theories: the public in terest approach, which asserts that the government acts to lessen or eliminate the inefficiencies engendered by market failure, serving as an impartial referee that aims to maximize social welfare; and the capture or interest group approach, which argues that the government acts to tr ansfer wealth between interest groups in an effort to maximize political support.1 Recent agency theoretical models 1 According to Stigler (1971), regulators are self-interest maximizers and stakeholders face costs of organization and information. Consumers, being disperse d and having less at stake, face higher costs of organization than producers and usually do not have the required incentives to spend the necessary resources to become informed. Consequently, the prediction was that the producer interest would win the bidding for the services of a regulatory agency. Stiglers formulation brought theoretical foundation to the producer pr otection view that character izes the capture theory of regulation, based on the accumulating evidence of empirical research done before the 1970s. The Stiglers argument, however, was further developed by Peltzman (1976), who posited that regulatory agencies would not exclusively serve a single economic interest. Utility-maximizing regul ators would allocate benefits across interest groups optimally, attempting to equate political support and opposition at the margin. Peltzmans contribution helps explain the location of policy in the competitive price to the mon opoly price spectrum. Consumers who spend a larger share of their income on a good have a higher incentive to participate in the regulatory process and should drop more

PAGE 17

17 further developed the interest group theory by explicitly recognizing both the existence of informational asymmetries and the principalagent relationship which exists between the Congress (or the Government) and their delegates in regulatory agencies.2 The possibility that the regulator may favor sp ecific interest groups underscores the need to devise regulatory frameworks incorporating a sy stem of checks and balances. However, to determine how regulatory processes and decisions affect sector pe rformance, it is necessary to assess the actual behavior of the regulatory agencies. The pr esent study uses information on stock market reactions to regulatory announcements to investigate the level of consistency and predictability of decisions taken by the Brazilian electricity sector regulat or, and to identify the extent the aforementioned regulatory body has been able to act independently and to balance multiple (and often conflicting) interests in regulation. With respect to network industrie s, the empirical research con ducted to date focused in the United Kingdoms context. While Sawkins (1996) an alyzed the performance of the water sector regulator (Ofwat), both Dnes, Kodwani, Seat on, and Wood (1998) and Dnes and Seaton (1999) concentrated on the behavior of th e electricity industrys regulato ry agency (Offer). In all these studies, the similar findings that no interest group had been systematically favored led to the conclusion that the regulatory bodies were performing their duties reasonably well. The conclusion of balanced decision-ma king, however, was drawn on the basis of securities abnormal returns evid enced at the moment of the regulatory announcements. Studies neglected the possibility that the findings were actually re flecting a change in market votes for the politician in response to a price rise. Therefor e, goods with a high share in the consumers budget are more likely to have prices close to the competitive price. 2 See Laffont and Tirole (1993), and Armstrong and Sappington (2003), for contributions on the subject.

PAGE 18

18 expectations.3 The present research attempts to corr ect for this flaw by also looking at the significance of the cumulative abnor mal returns in the period before the events disclosure. Thus, when there is evidence of an early incor poration of the events effect, the regulatory announcements impact is based on a computed measure of the even ts overall effect. The present investigation also distinguishes from the prev ious ones by focusing in the context of a developing country, wh ere the different interests in re gulatory outcomes seem to be more pronounced. In addition to both Brazilian fi rms and consumers having a higher incentive to participate in the regulatory process th an their counterparts in Great Britain,4 it may be argued that the Brazilian Government ex erts a higher pressure on the regul ator to keep tariffs at low levels than a Government in a developed count ry, by virtue of the memories of the recent hyperinflation period and the consequent still strong concern with the effect of tariff increases on inflation. Moreover, Brazil faces an acute need to expand network access (provide universal service) and to increase the amount of energy dist ributed, to avoid energy shortages in the near future and promote economic growth. The resulti ng greater necessity to provide incentives for (private) investments in the sector should be refl ected in a higher pressu re on the regulator along these lines in Brazil than in Great Britain. 3 A finding of negative and significant abnormal returns at the exact moment a regulatory announcement is made, for example, may not indicate that the decision affected firms value negativ ely, but, on the contrary, that the decision was not as good to firms as the market was exp ecting. Thus, when a significant abnormal return is found, one has to check for the possibility of event anticipation. If this possibility is confirmed, the conclusion regarding the events impact has to be taken on the basis of the events overall effect. 4 Brazilian firms face higher regulatory and country risks than firms in Great Britain, which im pacts the necessity to mitigate the possibility of not getting a rate of return th at covers their correspondent higher cost of capital. The affirmative regarding Brazilian consumers, on the other hand comes from the fact that they are poorer. Since the income elasticity for electricity is less than one, electricity is a greater share of the budget of poor people than of rich people. Thus, in case of a tariff increase, poor consumers e xperience a higher percentage change in their cost of living than those who are more affluent. Therefore, poor consumers (or consumers in poor countries) have a greater incentive than rich consumers (or consumers in rich countri es) to become informed and participate in the regulatory process.

PAGE 19

19 The Brazilian electricity sectors regulatory framework has been cr itically explored and analyzed relative to international experience,5 resulting in some policy changes. Nevertheless, there remains a debate over the regulat ors possible capture by the industry,6 indicating a need for further investigations of the regulatory agencys performance.7 The present study attempts to fill this gap. The papers contributions to the literature ar e thus the following: (a) it extends the eventstudy methodology employed in previous studies, by explicitly recognizing the need to account for the possibility of event anti cipation; (b) it is, to our knowledge the first empirical assessment of a regulatory agency performance in the cont ext of a developing count ry, where the interests are presumably more pronounced; (c) it examines th e level of consistency and predictability of decisions taken by the Brazilian elect ricity sector regulator; and (d) it sheds light on the debate over the possible capture of Brazilian regulatory agencies. The following section presents the characterist ics of the Brazilian el ectricity industry and describes the reforms implemented in recent years. Section 3 explains the methodology employed and Section 4 outlines the data set. Section 5 presents and interprets the results 5 The main initiative on this respect was represented by Brown and de Paulas reports (2002, 2004) prepared under the World Banks PPIAF Project for Brazil Power Sector, Task 4: Str engthening of the Institutional and Regulatory Structure of the Brazilian Power Sector. The first repo rt, issued in December 2002, analyzed the state of institutional arrangements in electricity regulation ri ght after the energy crisis and proposed specific recommendations for strengthening the regulatory structur e. The second report was prepared in July 2004 and reviewed the previous recommendations in light of the regu latory changes occurred in the period, detailed in the following section. The vast majority of the recommendations made were effectively implemented by the Government (Ministry of Mines and Energy) and the regulatory agency (ANEEL). 6 The debate was reinforced when the Government submitted a proposal to the Congress to better define the role of the regulatory agencies, in September of 2003. The propo sal, which is still pending in the Congress, contains elements that limit regulatory agencies independence, such as establishing performance contracts and defining ombudsman duties. The justification that accompanies the proposal refers to the need of social control over the job done by all infrastructure regulatory agencies. 7 Based on a literature review, Pires and Goldstein (2001) appears to be a unique study on the subject. The authors evaluate the performance of the regulatory agencies responsible for the regulation of the telecommunications, electricity and oil sectors. It is not an empirical study, however.

PAGE 20

20 obtained, while Section 6 portrays the robustnes s checks performed. The final section provides concluding observations. Institutional Background In event studies, the interpretation of resu lts requires a reasonab le understanding of the context in which the regulatory initiatives were undertaken.8 In Brazil, fluvial basins with an enormous hydroelectric potential and a large territory account for the configuration of the electricity sector, wh ere hydropower generation9 is linked to a larg e transmission network. Transmission lines deliver electric ity at high voltage to, for the most part, regional distribution monopolies. Most of the investments on hydr opower facilities were made in the period of 1965 to 1990. These investments were primarily carried out by Eletrobras, a state-owned enterprise which was a holding company comprised of four concession aries engaged in generation and transmission (Furnas, Chesf, Eletronorte, a nd Eletronuclear) and two regional distribution utilities (Light and Escelsa).10 Nevertheless, there occurre d a progressive deteriorati on of the public sectors capacity to invest, caused by the deepening in m acroeconomic instability in the 1980s, as well as a widespread inefficiency emerged from a ve rsion of rate-of-return regulation. The system utilized a single nationwide uniform tariff acco mpanied by a compensation scheme to equalize price and cost differentials among firms. The resulting weak sector performance brought about the need of structural reform.11 8 The following is partly based on Goldstein (1999) and Mota (2003). 9 At the end of 2000, hydroelectric power plants accou nted for 88.4% of the tota l capacity of 67,713 MW. 10 By the early 1990s, Eletrobras acc ounted for 57% of total generating capacity, with the remaining 43% being mostly provided by Itaipu, the binational enterprise between Brazil and Paraguai, and vertically integrated utilities owned by the states of Sao Paulo (CESP), Minas Gerais (CEMIG), Parana (COPEL), and Rio Grande do Sul (CEEE). 11 For details regarding the determinants of structural reform in the electricity sector, s ee Oliveira and Pires (1994), and Ferreira (2000).

PAGE 21

21 The power sector reforms began in 1995. While constitutional amendments abolished the public monopoly over infrastructure industries and allowed forei gn companies to bid for public concessions, the Law 8,987/95 (General Law of C oncessions) set the stag e for the beginning of the privatization proce ss, represented by the auctions of Escelsa in 1995 and Light in 1996. In addition, part of the implementation of a new regulatory framework involved the establishment of an independent regulatory agen cy (ANEEL) in late 1996 and, in the same year, the commission of an international consulta ncy to study and propose a new model for the electricity sector. The consu ltants report was released in 1997 and its proposals were incorporated into Law 9,648, issued on May of 1998.12 In essence, the approved model focused on privatization and unbundling of generation, tr ansmission and distribution assets, gradual transition to a competitive generation environment in nine years, creation of a wholesale power market, operation of the transmission network by an independent operator, and use of the pricecap regime to regulate distribu tion tariffs, replacing the previ ous cost of service system. Two main points emerge from this reform pro cess. First, the market was opened to private investors while the restructuring was still under study, and before cr ucial points such as pricing regulation and tariff review procedures were defi ned. Second, as a consequence of this uncertain context in which it was born, the regulatory agen cy, in its first years of existence, devoted its resources to refining elements incompletely spec ified in the law. Staff were engaged in the process of discussing and resolving issues, in cluding developing specifics for the new sector model still to be implemented.13 12 See Ferreira (2000), Mota (2003), and de Oliveira (2003), for detailed descriptions of the new models characteristics. 13 The events identified for years 1999 and 2000 are symptomatic of that (Table 2-3).

PAGE 22

22 By the end of 2000, the privatization process had concentrated on distribution companies,14 reflecting a deliberate intent of privatizing generation only once some key elements of the new model had been established, such as the wholesal e power market, and a significant delay in the implementation process itself. Nonetheless, the pros pect of an imminent privatization resulted in generation companies controlled by Eletrobras dramatically reducing new investments. The slow pace of the new models implementa tion, the failure in designing an arrangement that could make investments in thermo plants economically viable and a severe drought in the end of 2000 were the main factors behind the Braz ilian energy crisis, culminated by the rationing measures announced in May of 2001. The rationi ng imposed severe losses on both generation and distribution companies, ultimately recognize d and authorized to be compensated for by the Government in the so-called Acordo Geral do Setor (Sector General Agreement).15 The energy crisis also led to a re-evaluation of some of th e foundations of the model that had been implemented. In January of 2002, the Gove rnment decided to post pone the flotation of the generation companies Furnas, Chesf and Elet ronorte, and announced that it would continue establishing the prices of energy supplied by state-owned generators.16, 17 Then the new Government introduced extensive modifications in sect or arrangements, which were officially 14 A total of 23 companies had been privatized, with 19 being distributors and 4 generators. In 2000, the private sector participation in the distribution and generation markets was 60% and 20%, respectively. 15 The Agreement established that the losses would be covered by special loans that distribution and generation companies would contract with BNDES, the Brazilian Development Bank. Moreover, it determined that the loans would be paid back with the proceeds from the extr aordinary tariff increase (2.9% for residential and rural consumers, and 7.9% for the other consumers). 16 The Government decided to abolish the disposition (approved by Law 9,648/98) which stipulated that the energy supplied by the state-owned generators would be progressively sold in the free market, at the proportion of one fourth per year, starting in January of 2003. The argument employed was the need to prevent a boom in electricity prices. However, given the negative reaction to its initiative, later on 01/31/02 the Government partly revised its position, defining that the energy would be sold in public auctions, with the minimum price being the one that the state-owned generator had been charging on its contracts. 17 As a result of these initiatives, the Governments of the States of Sao Paulo and Parana reconsidered the privatization of their electricity companies as well (Cesp and Copel).

PAGE 23

23 announced in July 2003: the formation of a ge neration pool, discontinua tion of the wholesale power market, prohibition of self-dealing, and the commercialization of energy through long term contracts (20 years), to be signed be tween all generators and distributors. These developments provided the back drop for ra te review announcements affecting individual distribution utilities. Methodology The event-study technique consists in testing for the existence of significant changes in prices (abnormal returns) of firm s securities at the moment that the event was initially released or announced. The methodology relies on the assump tion that stock markets operate efficiently, so that any unanticipated event that has an imp act on firms value will be immediately reflected in security prices, with the price change being an unbiased estimate of the change in firms future cash flows. As a result of this feature, the methodology has been used to assess the impact of regulatory initiatives in several industries.18 In the identification strategy employed in the present study, the standard event-study technique is initially used to check for the existence of abnorma l returns at the moment of the regulatory events announcement. The investigation, at this point, is accomplished by performing both a joint hypothesis test a nd a constrained regression pr ocedure, where the events coefficients are forced to be the same across securities, in order to gather evidence on the direction of the events impacts. The sequen ce explicitly recognizes that any conclusion regarding a regulatory events impact has to ac count for the possibility of event anticipation. Thus, identified cases of significant abnormal re turns are checked for the existence of abnormal returns in the 5-day peri od before the events announcement. If the anticipation is confirmed, the 18 Some examples are Rose (1985), Bene ish (1991), Prager (1992), Carrol and Landim (1993), and Landim (1999).

PAGE 24

24 events impact is given by a computed measure of the events total (or net) abnormal returns. These procedures are detailed below. Event studies in general usua lly adopt the statistical met hod where abnormal returns are modeled as the prediction errors of the regressi on of security returns on market returns, and hypothesis tests are conducted under the assumpti on that the residuals are independent and identically distributed (the market model). In the context of a re gulatory event-study, though, some problems arise. Here, it is quite common to have all or most of the firms in the sample belonging to the same industry, or the regulatory events occurri ng during the same calendar time period for all firms, violating the independen ce assumption. Moreover, there is evidence that market model residual variances differ across fi rms, as a result of the positive relationship between market model residual variance and systema tic risk (as well as between the variance of returns and systematic risk).19 To account for these potential problems, this study employs the approach developed by Binder (1985b) and Schipper and Thompson (1983), which assumes the following returngenerating process for a firm i : t j it jt ij t i t i mt i i itD LIQUID DOLLAR R R1 t = 1, T (2-1) where Rit is the return on the securi ty of firm i in period t, Rmt is the equally weighted return on the market portfolio in that period ( IBOVESPA ),20 DOLLAR is the percentage variation in the exchange rate in the period t,21 LIQUID stands for a liquidity measure, given by the ratio of 19 See Fama (1976, pp. 121-124). 20 Regressions were performed using either the equally weighted (IBOVESPA) or the value-weighted (IBX) indexes of market returns. Better specifications were obtained when the equally weighted index was used. 21 DOLLAR is given by 100 11 t tER ER. The exchange rate expresses the amount of reais needed to buy one dollar, with its increase reflecting an appreciation of the dollar.

PAGE 25

25 negotiated volume at time t to companys market value, Djt is a dummy variable equal to 1 if event j occurred at time t and 0 otherwise, and it is a random disturbance. Note, for instance, that i and i are the market model parameters, and ij is a measure of the abnormal return associated with event j for firm i.22 The DOLLAR s incorporation in the model was motivat ed by the belief that the oscillation of the exchange rate might have caused differe ntial impacts on companies in the electricity sector. In the period investigat ed, electricity firms were more exposed to exchange rate fluctuations than the average firm in stock mark et, since most of them either bought energy from Itaipu (whose tariff is quoted in dollars) or were more indebted in foreign currency than firms from other sectors. Given that changes in th e exchange rate and stock market returns are negatively correlated, a negativ e sign is predicted for DOLLAR. The expectation is that an increase in the exchange rate produces a highe r (negative) impact on electricity companies,23 compared to the impact on the average firm in the market (represented by the stock market index). The LIQUID s inclusion, on its turn, is due to th e existing evidence that liquidity helps explain time-series varia tions in stock returns.24 In case, the prediction is that LIQUID has a positive sign, since it is anticipated that an increase in liquidity over time leads to an increase in 22 It was tested an alternative specification, where abnorm al returns were measured as CAPM prediction errors. In case, the risk-free asset return (given by the SELI C daily variation) was subtracted from both Rit and Rmt. However, lower R2 was observed for all individual regressions performed. 23 Except for the cases of Eletrobrs (which holds a participation of 50% in the Itaipu Enterprise), and the transmission company Transp. 24 Beneish and Whaley (1996), Lynch and Mendenhall (1997), and Elyasiani, Hauser, and Hauterbach (2000) find evidence of permanent excess returns associated to improvements in the stocks liquidity.

PAGE 26

26 stock returns, as a result of adjustments in st ock prices to incorporat e the corresponding change (reduction) in transaction costs.25 An equation of the above form is specified for each firm in the sample, resulting in a system of N equations for N firms, estimated w ith a seemingly unrelated regressions procedure. Given that the individual return equations are estimated jointly with generalized least squares, the procedure has the advantage that heterosked asticity across equations and contemporaneous dependence of the disturbances ar e explicitly incorporated into the hypothesis tests. Moreover, the approach makes it possible to test the joint hypothe sis that all dummy variable coefficients (i.e. the abnormal returns of all firms) for a given event equal zer o, which has particular relevance in the context of regulatory events, where it is quite comm on to have differential impacts (including in sign) among firms in the industry.26 The analysis is complemented by another se t of hypothesis tests, where the system of equations is estimated with the event coeffici ents being constrained to be the same across equations (securities). The estimates obtained from this procedure are equal to the estimates which would be obtained from a single regression run on a weighted portfolio of the original securities, where the weights are proportional to the inverse of the estimated covariance matrix of residuals used in the joint GLS estimation (Schipper and Thompson, 1985).27 The constrained regression procedure provides estimates of the re gulatory event parameters and, as such, enables 25 Investors value liquidity, the quick execution of their orders at the lowest cost. Liquidity, therefore, represents the ease of buying or selling the stock, which can be measured by the bid-ask spread. The more illiquid is the stock, the higher is the spread, or the transaction costs to the invest or, and the lower should be th e stocks price, which adjusts to incorporate the higher transaction cost s. This return-spread relation is taken as a rational response by an efficient market to the existence of the spread (trading friction and tran saction costs), rather than an indication of market inefficiency (Amihud and Mendelson, 1986). 26 When the abnormal returns differ in sign across firm s this will frequently be a more powerful test of the hypothesis that the event affects security holder wealth th an the test that the average abnormal return equals zero (Binder, 1998). 27 The estimates represent variance-minimizing weighted averages in which greater weights are given to observations with low variance and low or negative cova riance with the other observations (Salinger, 1992).

PAGE 27

27 to get evidence on the direction of the events impacts and to draw inference in terms of economic significance. In addition, the sample is composed of companies that operate in the generation, transmission, and distribution sectors, but some of the events are es sentially distribution events, in the sense that they are expected to have an impact on distribution companies only. Since the incorporation of gene ration and transmission companies abnormal returns in these cases might end up obscuring the events effects on distribution firms, hypothesis tests were performed considering two groups of firms, one consisting of all companies in the sample, and another consisting only of the distribution companies. Hence, the investigation on the existence of a bnormal returns at the moment of the events announcement is carried out by te sting the following hypotheses: H1: The regulatory event parameters for each firm in the sample are all equal zero; H2: The regulatory event parameters for each distribution firm in the sample are all equal zero; H3: Under the assumption that the event parame ters are the same for each firm in the sample, the abnormal return during the event window equals zero; H4: Under the assumption that the even t parameters are the same for each distribution firm in the sample, the abnormal return during the event window equals zero. For the specific events where the null hypotheses H1 or H2 are rejected, th e possibility of event anticipation is checked. The same joint hypothesis and c onstrained estimation tests are applied to the cumulative abnormal returns (CAR) earned by each security in the period of five trading days before the events announcement.28 If there is evidence of event anticipation, a 28The test for cumulative abnormal returns followed the procedure suggested by Salinger (1992, p.42), which consists in the inclusion of dummies for each day t of the event-window as explanatory variables. In the

PAGE 28

28 conclusion regarding the event impact on security returns is taken on the basis of the event overall effect, given by the results of a ne w set of hypothesis tests where the event period includes the announcement and the (f ive) pre-announcement days. Both stock returns and stock market returns are measured on a daily basis. Consequently, the base specification employs a 1-day event window, which means that it is being tested whether or not the regulatory event had an imp act on firms value in the day it was initially released or announced.29 Nonetheless, in order to account for the possibility of information leakage in the day before, as well as for the ch ance that an announcement was made after the closing of the stock market, the study also makes use of 2-day and 3-day event windows.30 Due to the missing observation problem (detai led in Section 4 below), the present study investigates the significance of regulatory events of each year, from 1999 to 2003, using the observations from the year under examination and the previous year. This procedure prevented a further reduction in sample size, by enabling the inclusion of stock returns from firms that have a missing observation problem in some years, but not in others. Data The sample was limited by availability of data and by the fact that only 34 electricity companies had stocks traded at the Sao Pa ulo Stock ExchangeBOVESPA, the main Stock Exchange in Brazil. observation for period t (t>1), the dummy for period t takes on the value 1 and the dummy for period t-1 takes on the value 29 Brown and Warner (1985) provide evidence that shorter event windows increase the power of the tests. 30 In the 2-day window, the event dummy was set equal to 1 in the event day and in the day immediately after, while in the 3-day case the same was done for the three-day window centered on the event day. On the few occasions where events were close together, the du mmies were truncated to prevent overlap.

PAGE 29

29 The information obtained from BOVESPA include d daily stock price data related to 31 companies,31 for the period of 03/16/1998 to 09/30/2003. The data, however, presented a severe missing observation problem. Seventeen companies had more than 40% of observations missing,32 and were consequently dropped from the sample. Daily stock returns were computed for the remaining 14 firms, resulting in a total of 1372 observations for a company with no missing data. The computation of stock returns incorporated the necessary adjustments for share splits and rights issues occurred, stock dividends gi ven, and all forms of cash payments made to stockholders.33 Nonetheless, the transforma tion of share prices into stock returns increased the number of missing observations. The final sample was then defined considering the number of missing returns in the period of two years of data (yeart and yeart-1) employed to run the individual regressions for years 1999 to 2003 (Appendix A). Ultimately, stock returns from a total of 12 companies in the electri city sector were used to perform the statistical tests, with the exact number of companies stock returns em ployed in each year varying from 9 to 11. Detailed information regarding these 12 companies is provided in Appendix A. Importantly, the numbers for participation in the distribution and generati on markets indicate that the sample is composed of some of the large companies in the electricity sector.34 These numbers also show that, except for a few cases, it is not possible to come up with a pure classification of the sample companies in terms of belonging to distribution or ge neration sectors. The 31 The data received from BOVESPA did not include daily stock price of firms CPFL Piratininga, Energipe and VBC Energia. 32 From the mentioned 17 firms, 09 had more than 80% of missing data and 15 were in the above 60% range. 33 In order to ascertain the correctness of the adjustments performed, the computed daily returns were aggregated into monthly returns and compared to the information on monthly stock performance provided by BOVESPA. 34 As of December/2000, Eletropaulo, Cemig, Light, Copel, and Celesc were among the eight biggest distribution companies in the country. On the other hand, Eletrobras, Cemig, Cesp, Tractebel and Copel were, in this order, the five biggest generation companies (Setor Eltrico Ranking 2001, Vol. 1 BNDES).

PAGE 30

30 classification reported on the third column, which will be utilized throughout the paper, comes from the comparative analysis of the data regarding electricity delivered to final customers and electricity generated. The sample implications to the present st udy are twofold. First, the paper assesses the regulatory events impact on the la rge players, which implies that the results cannot be taken to draw conclusions with respect to effects in the electricity sect or as a whole. Second, given the existing evidence of an inverse relationship between firm size and systematic risk, as well as the already mentioned relation between systematic risk and variance of returns, the use of larger companies should strengthen the po wer of the statistical tests. The identification of the regulatory events wa s made through the careful analysis of the following material: (a) resolutions, resolution proposals submitted to public hearings, and press releases issued by ANEEL during the period of March 1999 to September 2003; (b) reports in Gazeta Mercantil, the main financial newspaper in Brazil, related to economic regulation of the electricity sector, released in the period of March 1999 to June 2002; and (c) daily reports in the electronic newsletter IFE, join tly provided by Eletrobras and the Economics Institute of the Federal University of Rio de Janeiro (UFRJ) over the period of November 2000 to September 2003.35 The 65 events listed in Table 21 were selected on the basis of their relevance in terms of the new regulatory framework that was being implemented, the expected magnitude of the impact to the stakeholders involved and the repe rcussion in the media. Note that the events selection has the date 03/16/1999 as its starting point, given the d ecision to use at least one year of observations (250 trading days) as the estimation period. 35 The electronic newsletter IFE collects reports related to electricity regulation released in the following newspapers/magazines: Correio Braziliense, Dirio Catarinens e, Dirio do Grande ABC, Dirio do Nordeste, Folha de Pernambuco, Folha de So Paulo, Gazeta Mercantil, Inve stNews/GM, Jornal do Bras il, Jornal do Commercio, O Estado do Paran, O Estado de So Paulo, O Globo, Zero Hora, Valor, and Canal Energia.

PAGE 31

31 The descriptive statistics provided in Table 22 depicts a higher variability of stock returns in 1998 and 1999, compared to the other years under examination. Both the range and the computed standard deviation of stock market and securities daily returns decrease sharply from 1998 to 2003. The higher volatility in stock returns in the first two years examined should make it more difficult to reject the nul l hypotheses associated with the 1999s events. Note also that the exchange rate varies more in years 1999 and 2002, implying an expected higher impact of this variable on security returns in these years. The mean liquidity measures indicate that so me securities (Fcata, Tract, and Cerj) are not actively traded. Here, the main conc ern resides in the fact that stoc ks with irregular transactions may not have incorporated in their prices the ch ange in firms value br ought by an unanticipated event. The correlation matrix (Table 2-3) indica tes that this seems to occur with respect to macroeconomic events that affect the market as a whole, since the return s from the less traded securities are the ones which have the smallest correlation with stock market returns. Table 2-3 also reveals, as expected, a significant direct association between vol ume negotiated and stock market returns (with the logical exce ption of the least traded stocks)36 and a negative and significant correlation between market return s and variations in the exchange rate.37 Notice that some of the sample firms stock returns di splay a quite high negative correlation with the exchange rate, which corroborates the need to incl ude this variable in th e model specification. 36 For some of the not actively traded securities, there is at least an indication that they were more negotiated in days of announcements specific to these firms (their liquidity measure is significantly correlated with their own returns, but not with the market returns). This, however, does not o ccur with Fcata (the least trad ed stock), which is also the only security whose daily returns are not significantly corre lated with the variation in the exchange rate. Thus, the fact of not being regularly negotiated can be taken as the main reason for the bad specification of Fcata individual regressions, mentioned in Section 5 below. 37 Due to the evidenced correlation between these variables, we have also performed statistical tests with the DOLLAR variable being defined by the residuals of the regression of exchange rate variation on stock market returns. Practically no changes were observed in relation to the event parameter estimates shown in the present study.

PAGE 32

32 The annual exchange rate variation and the a nnual performance of each stock are displayed in Table 2-4. The exchange rate depreciated hi ghly in years 1999 and 2002, which is in line with the picture shown by Table 2-2. On the other ha nd, the comparison among sample firms average stock returns and market index re turns reveals that the stocks from the electricity companies included in the sample outperformed the market in the period examined,38 in spite of the rather poorer performance in year 1999. The present study investigates to what extent the difference in performance can be explained by regulatory decisions. Findings The regressions incorporated the necessary adjustments for correlation in security returns over time. The individual regressions estima ted parameters are presented in Table 2-5.39 The findings indicate that the m odel is well specified (R2 in the range of .30 to .80), with a few exceptions in the cases of the three le ast traded securities already mentioned.40 As expected, the IBOVESPA s coefficients are positive and strongly significant. The average betas estimated for each firm are in line with the sector of activity examined,41 whereas the sample average beta shows a U-shaped evolution in the period. The decr ease in firms systematic risk from 1998 to 2000 reflects the reduction in uncertainty brought by the progressive definition of some crucial points of the model being implemented. On the other hand, the betas in crease after 2001 might 38 The same holds when the Electricity Sector Index (IEE) reported by BOVESPA is compared to the market index. As expected, the sample firms average stock return and the IEE were highly correlated ( = 0.6956). 39 For reasons of space, we do not show the event dummies estimated coefficients of each individual regression. 40 The main concern was with respect to Fcata, included in the sample of years 2001 to 2003, whose regressions showed an R2 in the range of .06 to .11. The results obtained wh en this security was excluded from the sample were similar to the ones reported in this paper, though. The three cases in which the hypothesis tests provided stronger results are mentioned later on in the text. 41 For some firms, the low negotiated volume might have resulted in an underestimation of their beta values.

PAGE 33

33 be indicating that the model revi sion carried out after the energy crisis amplified the regulatory risk.42 The DOLLAR coefficient is negative and significant in about half of the individual stock returns regressions, denoting that variations in the exchange ra te did provoke a differentiated impact on some firms in the electricity sector, co mpared to other firms in the stock market. The impact occurs in practically all years examined, and is not more pronounced in years 1999 and 2002, as it was expected. Additionally, the results indi cate that liquidity contributes to explaining variations in firms stock returns, and support the prediction regarding th e variables sign, since the majority of the LIQUID s estimated coefficients is positive and significant. Here, however, as it is possible to anticipate an increase in tradi ng activities in days of important announcements, it should be noted that, in case of events that impact firms value positively, the LIQUID s estimated coefficient might be capturing part of the estimated announcement effect, leading to an over rejection of the null and to an under estimation of the a bnormal returns associated to these events.43 Tables 2-6 to 2-10 report the results from th e hypotheses tests performed on the sixty-five regulatory events. Before proceeding to their an alysis, though, the peculiarity of the regulatory process must be emphasized. In a regulatory cont ext, where the main issues follow the ritual initial proposal public heari ng final decision, the events information content is often leaked or flagged before the re gulators final decision, with th e corresponding impact on firms value being progressively incorporated into st ock prices before the official announcement is 42 It is beyond the scope of this paper the exam of the regulatory events impact on firms systematic risk. For studies that address the issue, see Antoniou and Pesce tto (1997), Buckland and Fraser (2001), and Morana and Sawkins (2002). 43 This problem is possibly reduced by th e fact that the statistically significan t LIQUIDs coefficients were, in their majority, the ones related to larger (and with higher trading volume) firms, given that Chordia, Shivakumar, and Subrahmanyam (2004) have found that these firms liquidity are the least affected by information shocks.

PAGE 34

34 made. As a result, a finding of no significant effect may occur for events that impacted firms value. It follows that it is ha rder to reject the null hypotheses H1 and H2,44 when the decision is taken after the implementation of the mentioned process.45 On the other hand, the occurrence (or not) of significant abnormal retu rns on rulings set forth after the ritual constitutes evidence regarding the level of cons istency and predictability of regulatory decisions. The peculiarities of the regulatory process, associated to the high volatility that characterized the Brazilian stock market in years 1998 and 1999, may e xplain the non-rejection of the null hypotheses in any of the 1999s events. In contrast, in years 2000 to 2003 there is evidence of significant abnormal returns for a to tal of 23 regulatory announcements. Here, the surprise verified in more than one third of the in vestigated events might be attributed to two main factors: a) the ruling deviated from what the ma rket had anticipated; or b) there were still too much uncertainty regarding the final decision, du e to a possibly wide spectrum in terms of regulatory choices and methodologies or simply to the fact that no (or ve ry little) information was provided to the market previously, particularly in cases of rulings not submitted to the mentioned regulatory process. We will return to these points later. For all of these 23 events, the analysis that follows focused on the outcomes of the tests of hypothesis H2 and H4 in case of a distribution event and incorporated the results from the 44 It also contributed to tougher joint hypothesis tests the incorporation of the LIQUID variable in the model, due to the already mentioned expected higher traded volume in days of important announcements, and the fact that most of the firms in the sample were included in the portfolio employed to compute the market index. Consequently, the IBOVESPA captured part of the variatio n in the sample firms daily stock prices caused by events specific to the electricity sector (as of December/2001, electricity companies accounted for 12 % of the market index, in terms of market capitalization value). 45 The ritual initial proposal public hearing final d ecision is a practice adopted by the regulatory agency. Therefore, all rulings in wh ich the decision took place at the Governme nt (Ministerial) level did not follow this process. The rulings of this nature amount to 23 out of the 65 events listed in Table 2-1. In addition, some of the regulatory agencys announcements did not follow this process either, mainly because they were just the initial proposal regarding the issue. The ritual occurred in 22 of the (42) ANEELs announcements listed in Table 2-1. The decisions which do not follow the ritual should convey more information to the market at the moment of their announcement. It is thus expected that they lead to a higher proportion of stock price reactions than the decisions that follow.

PAGE 35

35 constrained regression procedure (columns H3 and H4 in Tables 2-6 to 2-10), the cumulative abnormal returns hypothesis tests (Table 2-11), and the event overall effect (Table 2-12), when applicable. Moreover, the examination of th e results took into account the findings in terms of each securitys abnormal returns at each of the 23 significant events, especially when the evidence indicated the occurrence of differential impact among the sample securities. The same findings, on the other hand, were employed to examin e the possibility of th e results being driven by a factor other than the regulat ory announcement under investigation.46 In particular, the analysis revealed that in two opportunities ( E012700 and E010901 ) the events abnormal returns cannot be attributed to the regulatory announcemen ts made on those days, but to some specific factors that affected Light and Fcata, respectively.47 It is worth noting that the evidence conf irmed the importance of checking for event anticipation, given the finding of significant abno rmal returns in the 5-day period before the announcement of 7 out of the 23 si gnificant events (Table 2-11). The need to perform this additional analysis is also illustrated by the results found for E021700 and E040803 which denote that the abnormal returns evidenced in the announcement day are actually an adjustment in market expectations, since the total eff ect is not significantly different from zero.48 46 The focus on the results from the tests of hypothesis H2 and H4 in case of a distribution event is already a way to control for that. Note, for instance, the results for the distribution event E080503. The rejection of H1 only (in the 3day window specification) indicates th e occurrence of significant abnormal returns on generation companies securities, and consequently cannot be credited to the regulatory announcement under exam. 47 In the case of E012700, the null hypotheses H1 and H2 are not rejected when Light is dropped from the sample ( 2 = 6.00 (p=. 539), and 2 = 5.14 (p=. 274), respectively). Moreover, the events coefficient on Lights individual regression is highly positive, in contrast with the even ts expected effect. The same non-rejection of the null hypotheses H1 and H2 occurs in E010901 when Fcata is excluded from the sample ( 2 = 9.32 (p=. 502), and 2 = 5.85 (p=. 322), respectively). 48 Take, for example, the announcement of the Emergency Plan for the Short-Run (E040803). In case, while the positive CAR indicates an early incorporation into stock prices of the adopted measures (early on March 31st some newspapers had disclosed the information that the Government would provide resources to help capitalize firms in the sector), the negative abnormal retu rns on the announcement day reflects the markets surprise with the decision to postpone the incorporation of increases in non-controllable costs (higher price of energy bought from Itaipu, caused by the dollar appreciation) into tariffs, taken to alleviat e the impact of tariff adjustments on inflation. In sum,

PAGE 36

36 Information regarding the remaining 19 signifi cant announcements is summarized in Table 2-13. The majority of the significant events ca used a positiveas opposed to negativeeffect on firms value, which is in line with the alr eady mentioned better perf ormance of the sample firms securities over the period, compared to th e market portfolio. On the other hand, the results show that a higher proportion of decisions taken at the Ministeria l level turn out significant, when contrasted to regulatory agencys decisi ons, confirming the prediction that Government announcements should convey more information to the market.49 As previously stated, finding abnormal retu rns at Government announcements should be credited more to uncertainty regarding the final decision, than to deviation in market expectations. The distinction is relevant because in the last ca se the events estimated impact cannot be taken as a direct measure of the decision effect on firms value, since the market prior is unknown.50 In this context, it is interesting to note that the estimated impacts on security returns portrayed in Table 2-13 are, as a general rule, related to the nature of the Government announcements. While a negative impact is evidenced at the disclosure of the rationing plan measures ( E051801 )51, reflecting their deleterious conse quences on companies future cash flows, positive effects are found for the subse quent compensatory measures undertaken by the the non-rejection of the null hypothesis of no event overall e ffect suggests that the unforeseen tariff decision led to a reevaluation of markets previous expectations that firms would be benefited by the Government measures. 49 43% (10/23) of Governments rulings turn out sign ificant, against 21% (9/42) of ANEELs rulings. As anticipated, the difference might be attribut ed to the fact that Ministerial decisions are less susceptible to information leakage or anticipation, since they do not follow the regulat ory process. This result, ho wever, may also reflect the difference in the nature of the decisions taken at the Mini sterial and the regulatory agency levels. Since Ministerial rulings focus on the broad definition of the model for the sector, it is expected that they have a higher impact in firms value than agencys rulings, which in some cases only detail what has been previously defined by the Ministry. 50 This point is detailed later on in the text. 51 For both E051801 and E080701, the results are stronger when Fcata is excluded from the sample ( 2=18.61 (p=.046), and 2 =10.85 (p= .054), respectively).

PAGE 37

37 Government ( E090601, E102501, E09080352, E091603 ). Additionally, the finding of differential impact across firms is consistent with the deci sion to control the price of energy supplied by state-owned generation companies ( E010902 ), a typical within sect or redistribution policy, whereas the positive abnormal returns found for E013102 are in line with the Governments reconsideration of its previous position, materialized by the determ ination that the state-provided energy would be sold in public auctions.53 The negative reaction to the G overnments proposed revision in the regulatory agencies job ( E092303 )54 is according to expectations as well, si nce the initiative to limit the agencies independence constitutes a step back in the electr icity sector regulation an d induces the fear of a political use of the regulator, in creasing the regulatory risk. Note, however, that in two occasions this alignment between the nature of the Gove rnment announcement and the estimated effect on security returns is not observe d. First, negative abnormal return s are evidenced in case of a compensatory measure ( E081602 ),55 reflecting the decisions impli cations for Eletrobras and indicating that the market was expecting a be tter compensation system for the distribution companies. Second, differential impact across fi rms in the sample is shown for a change in ANEELs board of directors ( E050201) as if the new composition would be beneficial to some firms, but not to others, when no abnormal retu rns were expected for this event. The new directors were indicated by the same Government that had designated th e previous ones, what 52 The settlement between BNDES and AES in 09/08/03 solved a dispute that had lasted for more that one year, and was interpreted as a signal that the Government was concerned with electricity companies solvency. 53 The estimates of this announcements impact on each of the samples secu rities indicated that one distribution company (Eletropaulo) was negatively af fected by the Government decision. 54 See footnote 6. 55 The Government decided that resources from the Reserva Global de Reversao would be used to cover the reduction in concessionaries revenues brought by the ne w criteria for low-income cust omer (see footnote 57). The negative impact on Eletrobras security should come from the fact that the mentioned resources were being used by this company to capitalize the seven highly indebted former state-owned distribution companies that were transferred to the Federal Government.

PAGE 38

38 should not signal a change in the way the regulato r would rule on regulatory issues in the future. Here, though, the result is not conc lusive. The individual securities coefficient estimates are not consistent by firm type and th e CARs findings suggest the possi ble early incorporation of the rationing plan measures. Among the nine regulatory agen cys initiatives displayed in Table 2-13, some consist of decisions taken after the implementation of th e discussion process previously mentioned. In these cases, the observed effect on security return s denotes that the final decision deviated from market expectations formed along the discussi on process, what poses difficulties to an interpretation of the estimated effects on firms value in the absence of information regarding the market prior.56 The result found for the ruling defi ning the basic transmission network ( E111000 ), for example, should be revealing that the action taken towards the effective implementation of the model for the electricit y sector reduced regulatory uncertainty, but it cannot be ignored the po ssibility that the positive impact is just a review of (negative) expectations formed on the basis of all the informati on released or flagge d along the regulatory procedure. One possibility to overcome the problem is to consider the observed reaction to the agencys proposal concerning the issu e as a proxy for the market prior.57 ANEELs initial proposition on the ruling imposing restrictions on agents participation in the market ( E041700 ) had an impact on stock returns. In case, the estimated event parameter is not significant, indicating the occurrence of di fferential effects among firms, with some benefiting and others 56 Regulatory discussion processes usually take two to three months. Thus, eventual abnormal returns occurred along this period, as a result of a progressive incorporation into stock prices of expectations concerning the rulings impact on firms value, will not be captured by the CAR check em ployed in the present study, limited to the 5-day period before the final announcement. 57 This possibility relies on the assumption that the expectatio ns initially formed on the basis of the regulators initial proposition do not change in the period that goes up to the announcement of the final decision concerning the issue.

PAGE 39

39 harmed by the proposal. By taking these effects as the market expectations concerning the rulings impact on firms value, one may conclude that the positive abnormal returns associated to the final decision ( E072100 ) result from adjustments made by the regulator on its original position, making it more aligned with companies interests. Similar analysis might be applied to the events related to Escelsas periodic tariff review. The announcements were embedded with a high si gnaling power, given the difficulties faced by concessionaries in the rationing pe riod and the fact that it preced ed the rate reviews of other companies. Under the generalized uncertainty regarding methodological choices and methods to be used, the positive impact evidenced for the regulators initial proposal ( E062101 ) provides indication of its robustness and soundness. And the finding of anot her positive impact when the regulator takes its final decision (E080701)58 denotes that ANEEL made further adjustments in the methodology, interpreted as favoring compan ies in the sector, or revised upward the repositioning index initially proposed, as if the re gulator indeed wanted to signal that was aware of the problems imposed on distribution companie s by the rationing plan and would compensate them in the future for that.59 Conversely, the results found for the norm i ssued to define low-income customer ( E050202 ) suggest that ANEELs decisi on on this topic impacted firms value negatively.60 In this case, however, there is no proxy for the mark et prior, since the pr evious Congress decision regarding the issue ( E041002 ) caused no reaction on stock prices. Nonetheless, the fact that a 58 See footnote 51. 59 When asked about compensatory measures to the electric ity companies, the President of BNDES affirmed that the Government had been signaling that it would solve the questions pending in the sector, giving as an example the recent tariff increase given to Escelsa (F olha de So Paulo, 10/19/2001). 60 The prediction was that the new criteria for low-income customer would increase the number of households under this classification from 10.3 million to 18 million, or 36% of the total mark et (Gazeta Mercantil, May 16th, 2002). The negative impact on firms returns should come from th e fact that low-income cu stomers were exempt from paying the extraordinary tariff increase set fo rth by the Sector General Agreement.

PAGE 40

40 negative surprise is found for E050202 shows that the regulatory ag ency had some discretion in defining the concept, and used the flexibility to set up a ruling which was not as favorable to firms as the market was expecting.61 The findings related to ANEELs decisions regarding the asset base valuation62 were unexpected. The voluminous press coverage related to the issue, which invariably reported the distribution companies disappointment with th e proposed reposition cost methodology, led to an expectation of a strong negative imp act of ANEELs initial proposition ( E062102 ), not confirmed by the hypothesis tests results. The fi ndings provide evidence that abnormal returns did occur,63 but suggest a differential effect across firms, as if the market had been able to realize that the methodology would harm recently privati zed companies, by not considering the price paid for the corresponding assets, but benefit the concessionaries th at had invested in facilities promoting universal service a nd built a large asset base.64 Additionally, the ev idence concerning ANEELs final decision ( E090402 ), which kept the same methodology, does not provide support to the alleged surprise that this decision produced on concessionaries.65 Hence, if the distribution companies expected that the regu lator would reconsider its initia l proposal, the findings indicate that the market did not. This result, it must be stressed, is consistent with a market belief regarding the regulator s impartiality. 61 ANEELs decision may be interpreted as favoring consumers if it is assumed that the market did not expect a ruling which would favor electricity companies. 62 Foster and Antmann (2004) provide a detailed overview of the discussion process that resulted in the adoption of the repositioning cost methodology by ANEEL. 63 The event is significant at the 5% level when Fcata is excluded from the sample ( 2=11.63, p-value = .040). 64 Positive coefficient estimates were found for Cemig, Light and Celescs securities. For Eletropaulo, on the other hand, the results suggest a negative impact. 65 The Brazilian Electricity Distributors Association (ABRAD EE) issued a note, released on 09/04/2002, expressing its surprise with the decision taken by the regulatory agency. According to the Association, there was an ongoing negotiation process with Aneel and other sectors of th e Government in order to revise the reposition cost methodology initially proposed (Canal Energia, 09/04/02).

PAGE 41

41 The results related to the announced repositioni ng indexes, on their turn, provide insights into the level of objectivity and transparency of the methodology employed in the periodic tariff review.66 Here, the evidence is mixed. The no impact on stock returns found for the repositioning indexes proposed initially ( E021703, E030703, E031103) denotes that the numbers released were close to market expectations. A significant impact is observed for E041703 though, due to the positive abnormal returns caused by the regu latory announcement on Eletropaulo and Light securities.67 The final indexes announced on this day mu st have incorporated adjustments made by the regulator, and consequently led to a reevaluation of the e xpectations concerning these two firms future tariff reviews. However, the market expectations were not confirmed for Eletropaulo ( E052603 ), since its stock prices fell consider ably when its own repositioning index was disclosed.68 This negative surprise suggests the need of improvements in the periodic tariff review methodology, moving to greater predictability. The results obtained by the present study, when taken together, show that the regulatory decisions which impacted positively electricity firms value were essentially represented by initiatives taken to effectiv ely implement the model approv ed by Law 9,648/98 and by measures adopted to compensate firms for the losses im posed by the energy crisis of 2001. These results denote that the need to provide the proper e nvironment and the suitab le incentives to the 66 The rationale is that, in case of a transparent and objective methodology, th e regulators disclosure of the first repositioning indexes, along with a detailed description of how they were computed, is sufficient for the accurate prediction of the other firms indexes. The two main points of the methodology employed in the periodic tariff review were the asset base valuation by the repositioning cost and the estimation of efficient operational costs using the reference company approach prev iously adopted in Spain and Chile. 67 The estimates for the abnormal returns caused by the announcement on Eletropaulo and Light securities were 5.70 (p=.069) and 8.70 (p=.003), respectively. The event is signif icant at the 5% level when Fcata is excluded from the sample ( 2=11.15, p-value = .049). 68 The estimated abnormal returns caused by E052603 on Eletropaulos security was 5.96 (p=. 057).

PAGE 42

42 realization of the so desired investments in el ectricity generation, transmission and distribution sectors had a high weight in th e regulators utility function69 over the period examined. Interestingly, the evidence reveals that most of these favorable regulatory decisions were discussed and deliberated at the Ministerial or the Congressional levels, preserving the regulatory agencys impartiality, which is indeed supported by the results. This inves tigation suggests that ANEEL is seen as a neutral ins titution and acts as a relatively independent organization. While some of the ANEELs decisions which significantly aff ected security returns led to an increase in electricity firms value, some provoked a di fferential effect among firms, and some were effectively contrary to companies interests. Some caution on this interpretation must be exercised, though, given the fact that every event-study restricts the analysis to decisions which could not be correctly an ticipated by the market, a problem that strongly affects regulato ry agencys rulings which follow the ritual initial proposalpublic hearingf inal decision, compared to Ministerial rulings. On this respect, it must be stressed that finding 5 stoc k price reactions in 22 cases where the information was flagged or leaked in the regulatory proce ss exceeds what would be expected. Actually, this result constitutes a sign of unpredictabil ity of regulatory agencys decisions. This finding also provides insi ghts into the applicability to the Brazilian electricity sector regulation of the two opposing theories concerni ng the pattern of government intervention in business. As a higher predictability should be ex pected if the regulatory agency consistently serves as an impartial referee that aims to maximize social welfare or consistently favors a specific interest group (Government, industry, or customers), the evidenced unpredictability indicates that the regulatory agen cy has favored different intere st groups at di fferent times, 69 The word regulator, here, is used in a broad sense. It encompasses not only the regulatory agency, but also other actors with the power to adopt regulatory initiatives, su ch as the Government (Ministry) and the Congress.

PAGE 43

43 confirming Peltzmans (1976) prediction that the utility maximizi ng regulator will not exclusively serve a single economic interest. In terms of economic significance, the result s found in the present study indicate that the regulatory decisions examined were responsible for an increase of around 8% in sample firms market value (Table 2-13). As previously stat ed, the equally-weighted sa mple firms portfolio outperformed the market in the period investigat ed. A variety of factors might have contributed for these extra-returns, including efficiency improvements resulting from privatization and implementation of incentive regulation.70 This research, however, suggests that regulatory decisions account for roughly 12% of the difference in performance of the sample securities with respect to the market index. Robustness Checks Two procedures were adopted to examine the robustness of the results found in the present study. First, it was verified the possibility of a sma ll sample bias. In light of the existing evidence that the use of a statisticsuch as the chisquaredwhose distribution is only asymptotically known may lead to an over rejection of the null,71 new hypotheses tests were conducted using Raos F exact statistic. The findings were practi cally identical to the ones obtained previously. Consequently, the likelihood of a sm all sample bias was disregarded. Secondly, a counterfactual anal ysis was implemented to chec k the extent to which the significant abnormal returns eviden ced in this study may effec tively be attributed to the regulatory announcements investigated or simply result from the Brazilian stock markets volatility. The analysis consisted in the application of the same hypotheses tests to a random set 70 We are currently addressing the issue in another paper. The results obtained do indicate that privatization and incentive regulation led to efficiency improvements in the Brazilian electricity distribution sector. 71 Binder (1985a) presents evidence that joint hypothesis te sts in the multivariate regression model, using the chisquared statistics, are biased against th e null hypothesis when ther e are 60, or in some cases even 250 observations, per equation.

PAGE 44

44 of 69 dates.72 The results displayed in Table 2-14 show the rejection of th e null of no abnormal returns in 8 random events. In thr ee of them, however, the same anal ysis of individual securities coefficient estimates applied before revealed that the abnormal returns are due to specific factors that affected CERJ ( E090199 ) and Eletrobras ( E112601 and E031303 ).73 Given that one would expect 5% of the random events to be significant, finding significant abnormal returns in 5 of the 69 random events is not out of line. In addition, the result may be due to information leakage, very common in a regulatory context, and thus reflect an early incorporation of regulatory announcements made later on. Nonetheless, it should be emphasized the high disparity between the numbers of signi ficant random events and significant regulatory events, which gives credence to th e results provided by this study. Conclusions The paper examines the regulators performance in the Brazilian electricity sector using an event-study methodology specially designed to a re gulatory context, whic h explicitly accounts for the possibility of event anticipation. Despite the more pronounced interests that characterize a developing countrys regulatory en vironment, the results indicate that the regulator has acted relatively independently, with its decisions not favoring a si ngle interest group. The findings are similar to the ones obtained in previous studies that focused in the United Kingdoms context, and do not support the claim that Brazilian regula tory agencies are captured by the industry. On the contrary, the evidenced unpredictability of regulatory agencys decisions suggests that the regulatory agency has favored different interest groups at different times, supporting the claim that the utility maximizing regulator will not exclusively serve a single economic interest. 72 75 dates were randomly generated in the period of 03/16/1999 to 09/30/2003, but six of them had to be dropped because they were already included in the Events list. 73 The test statistics were the following: 2=4.03 (p=.854), for E090199 without Cerj; 2=8.66 (p=.565), for E112601 without Eletrobras; and 2=3.76 (p=.927), for E031303 without Eletrobras.

PAGE 45

45 The observed unpredictability, on the other hand, reinforces the need of improvements in the regulatory discussion process, with the adoption of measures to in crease the transparency and to promote more substantive public hearings, as recommended by Brow n and de Paula (2002, 2004). The study suggests that the need to provide incentives for new investments has had a significant role in the regulat ory process. In addition, elect ricity companies have been compensated for the regulatory risk they face. The estimates indicate that regulatory decisions led to an increase in firms market value over the period examined and a ccount for part of the difference of the sample securities performan ce with respect to the market. Some specific findings are also worth noting. The adoption of the a sset base repositioning cost methodology was not that harmful to distri bution companies, as one would anticipate in light of the press coverage rela ted to the issue, and the eviden ce raises the concern over the objectivity and transparency of the methodol ogy employed in the distribution companies periodic tariff review, suggesting the need of improvements. Moreove r, the results indicate that the Governments proposal to review the regulat ory agencies responsibi lities and performance was seen as a step back in the electricity s ector regulation and increased the regulatory risk.

PAGE 46

46 Table 2-1. Events List Event Initiative Description E032399 Aneel Resolution proposal: access to the transmission system E051099 Aneel Resolution proposal: definition of the Normative Values E061099* Aneel Extraordinary tariff review E070199 Aneel Resolution: regulates the commercialization of energy not previously contracted E072999 Aneel Resolution: defines the Normative Values E080399* Aneel Resolution: Escelsas peri odic price review final numbers E092199* Aneel Resolution proposal: quality of service E092399* Aneel Resolution proposal: defines righ ts and duties of consumers and utilities E100199 Aneel Resolution: rules for acc ess to the transmission system E121499* Aneel The regulator denies the requests for an extraordin ary tariff review presented by 08 distribution companies E012700* Aneel Resolution: quality of service E021700 Gov CELPEs privatization E041700 Aneel Resolution proposal: limits for particip ation in the market (m arket concentration) E042400 Aneel Resolution proposal: rules for the wholesale power market E050300* Aneel Resolution: defines the quality parameter for the X factor in Escelsas tariff review E061500 Gov CEMARs privatization E072100 Aneel Resolution: limits fo r participation in the market / market concentration E080400 Aneel Resolution: rules fo r the wholesale power market E111000 Aneel Resolution: defines the basic transmission network E112900 Aneel/Gov Resolution: rights and duties of c onsumers and utilities / SAELPAs privatization E120700* Aneel Resolution: new quality st andards for distribution companies E010901* Aneel Resolution proposal: defines procedures for or dinary, extraordinary and periodic tariff reviews E042001 Aneel Intervention in the wholesale power market E050201 Gov Change in Aneels board of directors E051801 Gov Announcement of the rationing plan measures E062101* Aneel Regulators proposal for Escelsas periodic tariff review E080701* Aneel Resolution: final numbers for Escelsas periodic tariff review E090601* Gov Compensation for increases in non-controll able costs of distributi on companies (MP 2,227/01) E102501* Gov Authorizes that variations in distribution companies non-controllable co sts be adjusted by the Selic interest rate, until the annual tariff re view date (Port. 296/01) E111301* Gov Agreement with distribution companies regarding the compensation of rationing losses E112101* Aneel Resolution: detail s measures implem ented by MP 2,227/01 and Port. 296/01 E112301 Gov Agreement with generation companies re garding the compensation of rationing losses E121101 Aneel Decision favoring distribution companies in thei r dispute against Eletrobras concerning the energy from Itaipu E121201 Gov Change in Aneels board of directors E121701 Gov General Sector Agreement E010902 Gov Revitalizati on Plan for the Electricity Sector E013102 Gov Announcement that energy supp lied by state-owned generators w ould be sold in public auctions E020102 Gov Details of the Revitalizatio n Plan for the Electricity Sector E021902 Gov President announces the end of rationing period E041002 Congress General Se ctor Agreement is approved, incorporating a new criterion for low-income consumer. E042502 Aneel New criteria makes it more attractive to invest in electricity transmission networks E050202* Aneel Resolution: new criter ia for low-income consumer

PAGE 47

47 E060402 Gov Revitalization Committee announces additional meas ures of the Revitalization Plan for the Electricity Secto r E062102* Aneel Resolution proposal: met hodology for asset base valuation E081602 Gov Decision to use resources from the RGR to cover revenue reductions brought by low-income consumers criteria (Decree 4,336/02) E083002 Aneel Resolutions: detail Decree 4, 336/02 measures; define normative values for contracts with thermo-electric plants; allow distribution companies to contract re pairs in the transmission network below 230KV with transmission companies E090402* Aneel Resolution: methodology for asset base valuation E101502 Aneel Resolution: credits from pending operations in the wholesale power market must be adjusted for inflation E103002* Aneel Resolution proposal: X-factor methodology E110702 Aneel Resolution: reviews the previous definition that credits in the wholesale power market would be adjusted for inflation E011403* Aneel Adoption of the model firm approach in the distributors periodic tariff review E021703* Aneel Repositioning indexes proposed for the periodic tariff review of Cemig, Cemat, Enersul and CPFL E030703* Aneel Repositioning indexes proposed for RGE and AES SUL E031103* Aneel Repositioning indexes proposed for Coelce, Cosern, Energipe and Coelba E040803 Gov Emergency Plan for the Short-Run E041703* Aneel Final repositioning indexes for RGE, AES SUL, Coelce, Cose rn, Energipe, Coelba E043003* Aneel Resolution: Electricity Universal Plan E052603* Aneel Repositioning index proposed for Eletropaulo E070303* Aneel Final repositioning index for Eletropaulo E072103 Gov General guidelines of the Ne w Model for the Electricity Sector E080503* Gov Emergency Program to help the distribu tion companies (MP 127/03 general guidelines) E090803 Gov BNDES and AES settle an agreement E091603* Gov Emergency Program to help the distribu tion companies announcement of its conditions E092303 Gov Law project: revision in regulatory agencies job E092503* Aneel Repositioning index proposed for Light

PAGE 48

48Table 2-2. Descriptive Statistics 1998 1999 2000 2001 2002 2003 Mean Range Mean Range Mean Range Mean Range Mean Range Mean Range RetIbov -0.196 [-15.8, 18.7] 0.421 [-9.9, 33.4] -0.024 [-6.4, 5.0] -0.025 [-9.1, 7.6] -0.053 [-6.5, 6.3] 0.200 [-3.8, 3.6] 3.960 3.123 2.074 2.140 2.070 1.576 RetCopel -0.196 [-15.3, 31.4] 0.382 [-16, 40.9] 0.017 [-6.3, 9.4] 0.092 [-10.3, 9.5] -0.148 [-8.4, 14] -0.008 [-9.7, 6.9] 5.735 4.171 2.858 3.222 3.200 2.970 RetEletr -0.492 [-31.9, 29.1] 0.376 [-16.5, 12.1] -0.003 [-10.9, 11.4] 0.004 [-9.7, 11.4] -0.310 [-12.1, 21] 0.271 [-8.5, 13.2] 5.822 3.890 3.168 3.299 4.598 4.046 RetEletb -0.293 [-14.1, 25.1] 0.356 [-13.4, 38.5] -0.005 [-9.6, 19.8] 0.034 [-12.1, 11.6] -0.024 [-8.8, 11.5] 0.176 [-8.1, 7.7] 5.767 4.287 2.947 3.466 3.227 3.013 RetCemig -0.209 [-24.3, 28.4] 0.335 [-14.6, 30.2] -0.091 [-7.3, 9.6] 0.111 [-9.0, 8.8] -0.005 [-10.1, 10] 0.166 [-6.5, 7] 5.616 4.072 2.764 2.819 3.171 2.305 RetCesp -0.177 [-15.6, 17.6] 0.294 [-17, 11.1] 0.362 [-10.3, 25.5] 0.041 [-20.1, 20.3] -0.236 [-10.1, 12.1] 0.214 [-7.8, 12.7] 5.062 4.020 4.649 4.738 3.528 3.371 RetLight -0.418 [-23.1, 16.7] 0.197 [-18.1, 15.6] 0.099 [-12.5, 14.5] -0.169 [-18.0, 16.6] -0.281 [-10.1, 8.7] 0.058 [-11, 15.9] 4.836 3.845 3.212 3.874 3.193 4.117 RetCeles -0.296 [-18.9, 19.5] 0.267 [-14.3, 16. 7] -0.108 [-8.1, 10.2] -0.033 [-7.6, 20] 0.100 [-7.2, 10.4] 0.070 [-5, 10.4] 5.212 3.897 3.173 3.524 2.697 2.541 RetEmae -0.326 [-31.7, 20.5] 0.739 [-12.6, 34.7] 0.185 [-12.7, 15.9] 0.294 [-10, 20.5] -0.122 [-11.5, 16.5] 0.032 [-5.8, 13.4] 8.415 5.859 4.177 3.621 3.251 2.765 RetFcata -0.138 [-10, 11.1] 0.307 [-17.2, 20.8] 0.178 [-8.4, 20] 0.097 [-11.5, 22.9] -0.052 [-12.3, 13.6] 0.105 [-12.1, 9.3] 2.750 4.639 3.639 3.623 3.426 3.163 RetCoelc 0.115 [-15.8, 16.8] 0.811 [-11.5, 37.5] 0.201 [-10.4, 9.1] -0.104 [-7.1, 4.1] -0.076 [-6.8, 6.5] 0.316 [-5.2, 7.5] 4.698 4.383 2.639 1.669 2.050 2.060 RetTranp 0.714 [-15.5, 19.5] 0.201 [-13.4, 35.9] 0.079 [13.4, 14.4] 0.066 [-10.3, 8] 0.373 [-9.5, 6.6] 5.720 5.119 3.845 3.105 2.388 RetCerj -0.083 [-14.3, 34.6] 0.261 [-33.3, 20] 0.228 [-19.1, 15.2] 0.710 [-14.6, 20] -0.202 [-13.4, 16] 0.218 [-37, 30.4] 5.346 5.050 4.210 5.480 4.026 7.144 RetTract -0.028 [-18.7, 17.9] 0.117 [-13.9, 21.7] 0.235 [-7.5, 14.4] 0.209 [-7.7, 12.1] -0 .012 [-12, 8.5] 0.455 [-11.6, 23.9] 4.899 3.405 2.989 4.057 2.849 4.087 LiqCopel 0.272 [.006, 6.59] 0.233 [.001, 1.80] 0.201 [.009, 1.53] 0.269 [.004, 1.46] 0.370 [.066, 2.31] 0.656 [.053, 3.63] 0.609 0.213 0.163 0.182 0.256 0.555 LiqEletr 0.150 [.001, 1.37] 0.193 [.001,1.30] 0. 110 [.008, .712] 0.157 [.004, .963] 0. 207 [.014, .855] 0.329 [.075, 1.35] 0.185 0.184 0.096 0.142 0.160 0.198 LiqEletb 0.476 [.074, 1.77] 0.652 [.001, 2.12] 0.478 [.088, 2.05] 0.515 [.081, 2.48] 0.653 [.124, 3.61] 0.683 [.169, 2.21] 0.261 0.338 0.265 0.281 0.371 0.323 LiqCemig 0.462 [.053, 1.65] 0.372 [.001, 2.01] 0.541 [.120, 4.27] 0.459 [.034, 1.65] 0.487 [.083, 1.79] 0.629 [.166, 2.14] 0.262 0.244 0.472 0.215 0.246 0.306

PAGE 49

49LiqCesp 0.591 [.037, 2.43] 0.572 [.002, 2.42] 0.354 [.030, 2.13] 0.335 [.045, 1.57] 0.210 [.017, 1.15] 0.314 [.035, 1.51] 0.424 0.487 0.289 0.203 0.139 0.279 LiqLight 0.201 [.008, 3.17] 0.239 [0, 1.74] 0.184 [.008, 3.14] 0. 069 [.003, .404] 0.036 [.001, .515] 0.027 [.001, .221] 0.282 0.228 0.350 0.060 0.051 0.028 LiqCeles 0.643 [.008, 3.83] 0.893 [.090, 5.28] 0.736 [.078, 5.43] 0.534 [.052, 2.89] 0.612 [.032, 3.16] 0.786 [.071, 6.80] 0.610 0.703 0.634 0.408 0.509 0.855 LiqEmae 0.202 [.001, 2.33] 0.162 [.001, 1.83] 0.124 [.001, 1.63] 0.182 [.001, 2.73] 0.049 [.001, .702] 0.104 [.001, .836] 0.296 0.267 0.216 0.306 0.080 0.120 LiqFcata 0.058 [.001, 2.81] 0.026 [.001, .227] 0.123 [.001, 4.16] 0.017 [.001, .427] 0.012 [.001, .362] 0.004 [.001, .040] 0.225 0.039 0.418 0.037 0.035 0.007 LiqTranp 0.343 [.028, 1.88] 0.386 [.040, 2.90] 0.433 [. 063, 17.15] 0.161 [.008, .645] 0.153 [.019, .448] 0.302 0.405 1.118 0.102 0.091 LiqCerj 0.041 [.001, .668] 0.113 [.001, 4.57] 0.187 [.001, 3.42] 0.004 [.001, .060] 0.002 [.001, .030] 0.001 [.001, .013] 0.066 0.303 0.390 0.007 0.005 0.002 LiqTract 0.123 [.002, .740] 0.080 [.001, 1.65] 0.053 [.001, .542] 0.047 [.001, 2.83] 0.036 [.002, .239] 0.039 [.001, .488] 0.123 0.143 0.072 0.183 0.036 0.067 Dolar 0.033 [-.413, .254] 0.174 [-8.44, 11.10] 0. 037 [-1.36, 1.81] 0.075 [-3.15, 3.44] 0. 181 [-8.93, 4.87] -0.097 [-2.98, 2.93] 0.078 1.742 0.511 1.022 1.553 0.967 Security Returns' Mean Standard Deviation, by Year: 5.346 4.403 3.503 3.633 3.255 3.382 Standard deviations are info rmed below the mean values.

PAGE 50

50Table 2-3. Correlation Matrix RetIbov RetCopel RetEletr RetEletb RetC emig RetCesp RetLight Re tCeles RetEmae Re tFcata RetTranp RetCerj RetTract RetIbov 1.000 0.7484* 0.5621* 0.8192* 0.7755* 0.6090* 0. 5143* 0.6006* 0.3797* 0.1480* 0.5363* 0.2697* 0.3987* (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) LiqCopel 0.0670* 0.0889* (.013) (.001) LiqEletr 0.1040* 0.1780* (.000) (.000) LiqEletb 0.1117* 0.1552* (.000) (.000) LiqCemig 0.0874* 0.0780* (.001) (.004) LiqCesp 0.0917* 0.1474* (.001) (.000) LiqLight 0.0278 0.0990* (.304) (.000) LiqCeles 0.1009* 0.2016* (.000) (.000) LiqEmae 0.0677* 0.3490* (.013) (.000) LiqFcata -0.0425 0.0377 (.126) (.233) LiqTranp 0.0147 0.1233* (.638) (.000) LiqCerj 0.0861* 0.0937* (.004) (.003) LiqTract 0.0211 0.1737* (.444) (.000) Dolar -0.0578* -0.0910* -0.1897* -0.0723* -0.1190* -0.1577* -0. 1660* -0.0876* -0.1077* -0.0503 -0.0942* -0.0827* -0.0672* (.032) (.001) (.000) (.007) (.000) (.000) (.000) (.001) (.000) (.075) (.002) (.009) (.015) The indicates significance at the 5% level. P-value in parenthesis.

PAGE 51

51 Table 2-4. Securities Returns, Stock Market Re turns and Exchange Rate Variation, by Year. 1998 1999 20002001200220031998-2003 1999-2003 COPEL -50.29 109.56 -5.64 10.50 39.12 -9.30 -40.02 20.66 ELETR -73.08 108.93 -12.37 -11.69 -64.46 42.73 -77.92 -17.98 ELETB -58.94 94.09 -11.06 -6.29 -17.22 27.73 -29.77 71.03 CEMIG -51.16 87.00 -27.40 19.32 -12.83 29.87 -10.43 83.39 CESP -45.06 -34.68 88.17 -16.20 -52.40 34.45 -63.78 -34.08 LIGHT -64.85 35.37 12.56 -45.31 -6 .57 -4.57 -73.88 -25.70 CELES -56.85 59.76 -32.50 -20. 47 17.39 7.41 -53.34 8.14 EMAE -71.59 303.85 26.71 75.38 32.11 -1.01 71.35 503.17 FCATA -24.23 43.04 31.40 8.41 -24.05 10.64 29.74 71.22 TRANP 81.82 21.26 1.37 4.49 90.46 344.79 344.79 CERJ -34.23 37.50 30.92 122.13 22.96 -5.23 92.02 191.95 TRACT -19.17 16.03 60.77 37.11 -12.28 101.43 265.28 351.92 AVGRET -48.49 97.08 19.54 20.80 -18.33 32.24 58.32 207.37 IBOVESPA -41.52 151.93 -10.72 -11. 02 -17.01 42.08 38.01 135.99 DOLAR 6.60 48.06 9.30 18.67 52. 27 -17.26 157.91 141.94 AVGRET displays the returns of the equally-weighted portfolio composed by the securities included in the sample.

PAGE 52

52Table 2-5. Individual Regressions Results Year1 Variable RetCopel RetEletr RetEletb RetCemig RetCesp Re tLight RetCeles RetEmae RetCerj RetTract RetFcata RetTranp Avg Beta2 1999 Ibovespa 1.024*** 0.770*** 1.291*** 1.087*** 0.977*** 0.730*** 0.828*** 0.814*** 0.559*** 0.898 Dolar 0.070 -0.453*** -0.036 -0.225** -0.632*** -0.425*** -0.207 -0.521** -0.454** Liquid 0.124 3.974*** 0.601* -0.471 1.253*** 4.184*** 0.461* 8.264*** -0.155 R-sq 0.6695 0.4730 0.8003 0.7010 0.5246 0.4788 0.4261 0.2985 0.1529 N 411 411 411 411 411 411 411 411 411 2000 Ibovespa 0.854*** 0.786*** 1.096*** 0.919*** 1.015*** 0.659*** 0.958*** 0.684*** 0.489*** 0.829 Dolar 0.027 -0.490*** -0.023 -0.200** -0.566*** -0.437*** -0.175* -0.547*** -0.042 Liquid 1.781*** 3.148*** 0.805*** 0.194 0.853** 2.709*** 0.472*** 7.745*** 8.385*** R-sq 0.6127 0.3560 0.6783 0.5524 0.3671 0.3277 0.4553 0.2976 0.3193 N 475 475 475 475 475 475 475 475 475 2001 Ibovespa 0.871*** 0.750*** 1.054*** 0.893*** 1.102*** 0.827*** 1.015*** 0.699*** 0.661*** 0.335*** 1.156*** 0.903 Dolar -0.200 -0.451*** 0.134 0.036 -0.095 -0.156 0.180 -0.419** 0.226 -0.192 0.152 Liquid 2.231*** 3.307*** 0.736* 0.425* 0.177 1.862*** 0.402** 4.415*** 1.168 0.340 0.386** R-sq 0.4847 0.3416 0.5060 0.5312 0.3604 0.3967 0.4684 0.3133 0.2211 0.1054 0.3392 N 468 468 468 468 468 468 468 468 468 468 468 2002 Ibovespa 0.997*** 1.026*** 1.189*** 1.026*** 1.164*** 0.952*** 0.720*** 0.804*** 0.154** 1.036*** Dolar -0.250*** -0.398*** -0.060 -0.191*** -0.066 0.075 -0.005 0.229** -0.228** 0.042 0.990 Liquid 1.297*** 1.316 0.432 0.237 -0.004 -5.055** 1.379*** 0.107 -2.231 -0.037 R-sq 0.5635 0.4621 0.6182 0.5654 0.4716 0.4342 0.3749 0.2666 0.0905 0.4705 N 487 487 487 487 487 487 487 487 487 487 2003 Ibovespa 1.082*** 1.179*** 1.265*** 1.107*** 1.186*** 0.887*** 0.487*** 0.605*** 0.134 0.917*** 0.968 Dolar -0.288*** -0.449*** -0.154** -0.224*** -0.081 0.040 -0.085 0.095 -0.070 -0.011 Liquid 0.204 2.848*** 0.505* 0.092 2.933*** 0.407 1.262*** 6.122* -3.616 1.012 R-sq 0.5920 0.4940 0.6733 0.6184 0.5587 0.3604 0.3544 0.1748 0.0611 0.4577 N 430 430 430 430 430 430 430 430 430 430 Avg Beta 0.966 0.902 1.179 1.006 1.089 0.811 0.802 0.732 0.559 0.640 0.208 1.036 0.918 Legend: p < 0.10; ** p < 0.05; *** p < 0.01 1. For each Year t the regressions include data from periods t and t-1 2. The computed annual sample average beta s do not incorporate Fcatas measures.

PAGE 53

53Table 2-6. Hypothesis Tests Resu lts for Events in Year 1999 EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW H1 H2 H3 H4 H1 H2 H3 H4 H1 H2 H3 H4 Coef. Coef. Coef. Coef. Coef. Coef. E032399 4.02 1.75 0.22 -0.46 3.80 1.60 -0. 30 -0.89 2.81 2.42 -0.83 -0.92 (.910) (.941) (.855) (.760) (.923) (.952) (.732) (.384) (.971) (.877) (.237) (.272) E051099 9.35 7.82 -1.94 -1.87 4.16 2.13 -1 .18 -0.81 5.81 3.51 -0.93 -0.41 (.405) (.251) (.114) (.213) (.900) (.907) (.176) (.433) (.758) (.742) (.191) (.627) E061099* 5.89 4.73 1.63 1.47 7.00 4.24 1.88** 1.66 7.28 4.67 1.32* 1.15 (.750) (.579) (.183) (.326) (.636) (.643) (.030) (.105) (.608) (.587) (.060) (.170) E070199 5.72 4.27 -0.23 0.39 5.22 3.43 -0. 31 -0.27 3.67 3.05 -0.30 -0.22 (.767) (.640) (.848) (.792) (.815) (.753) (.721) (.790) (.931) (.802) (.667) (.797) E072999 4.30 3.68 0.38 0.76 2.98 2.30 0.00 -0.45 6.05 5.11 -0.11 -0.63 (.890) (.719) (.757) (.614) (.965) (.890) (.999) (.664) (.734) (.529) (.877) (.453) E080399* 3.07 1.98 -0.72 -0.02 6.09 3.10 -0 .39 0.24 3.97 1.89 -0.92 -0.74 (.961) (.921) (.555) (.989) (.730) (.796) (.655) (.814) (.913) (.929) (.193) (.377) E092199* 2.08 0.26 -0.75 0.17 3.79 3.31 0.65 1.51 3.49 2.36 0.43 0.96 (.990) (.999) (.539) (.911) (.924) (.768) (.452) (.142) (.941) (.884) (.537) (.252) E092399* 8.50 4.09 -0.85 -0.73 5.57 1.84 0.26 -0.37 5.22 1.73 0.27 -0.43 (.485) (.663) (.487) (.624) (.782) (.933) (.766) (.721) (.814) (.942) (.756) (.675) E100199 2.14 1.60 -0.14 0.27 1.63 0.85 0.10 -0.06 4.71 4.16 0.12 0.00 (.989) (.952) (.908) (.857) (.996) (.990) (.906) (.953) (.859) (.655) (.867) (.999) E121499* 3.07 1.84 0.26 0.40 3.14 2.16 -0. 60 -0.30 2.99 1.91 -0.67 -0.43 (.961) (.933) (.832) (.788) (.958) (.904) (.485) (.768) (.964) (.928) (.344) (.611) # Securities 9 9 9 9 9 9 9 9 9 9 9 9 # Observ. 411 411 411 411 411 411 411 411 411 411 411 411 P-value in parenthesis. The denotes a distribution event. Legend: p < 0.10; ** p < 0.05; *** p < 0.01.

PAGE 54

54Table 2-7. Hypothesis Tests Resu lts for Events in Year 2000 EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW H1 H2 H3 H4 H1 H2 H3 H4 H1 H2 H3 H4 Coef. Coef. Coef. Coef. Coef. Coef. E012700* 23.66*** 20.35*** -0.08 1.31 11.87 9.03 -0. 76 -0.05 6.46 4.37 -1.09* -1.13 (.005) (.001) (.938) (.330) (.220) (.107) (.322) (.959) (.693) (.497) (.083) (.130) E021700 19.39** 7.66 0.58 1.37 18.26** 8.89 0.13 0.91 20.61** 11.57** 0.51 1.49** (.022) (.176) (.595) (.309) (.032) (.113) (.862) (.338) (.014) (.041) (.413) (.048) E041700 21.33** 13.90** -1.26 0.43 13.86 10.30* -1. 27* -1.08 11.82 9.34* -1.27** -1.34* (.011) (.016) (.246) (.751) (.127) (.067) (.099) (.257) (.223) (.096) (.044) (.075) E042400 2.54 0.88 0.11 -0.01 3.92 2.65 0.71 0.46 3.86 2.68 0.70 -0.20 (.979) (.971) (.921) (.991) (.916) (.753) (.355) (.629) (.920) (.748) (.360) (.791) E050300* 9.49 5.50 0.66 0.28 8.80 4.44 0.59 1.47 8.73 4.46 0.54 0.77 (.393) (.358) (.546) (.835) (.456) (.487) (.585) (.121) (.462) (.485) (.618) (.304) E061500 13.41 5.74 2.38** 1.61 12.49 7.58 1.87** 1.37 16.51* 5.13 1.25** 0.43 (.145) (.332) (.029) (.230) (.187) (.180) (.015) (.147) (.056) (.400) (.046) (.569) E072100 25.78*** 9.41* 1.98* 0.94 26.77*** 9.11 1.61** 1.00 34.26*** 12.87** 1.55** 1.07 (.002) (.093) (.069) (.484) (.001) (.104) (.037) (.290) (.000) (.024) (.014) (.155) E080400 5.90 2.38 -1.77 -1.80 8.11 3.20 -1 .54** -1.03 2.98 0.49 -0.16 -0.06 (.750) (.794) (.104) (.181) (.523) (.669) (.046) (.279) (.965) (.992) (.796) (.936) E111000 9.22 7.92 0.66 1.20 7.14 4.58 1.15 1.53 16.69* 14.96** 1.21* 1.87** (.417) (.161) (.545) (.374) (.622) (.469) (.136) (.107) (.053) (.011) (.053) (.012) E112900 1.04 0.15 0.01 -0.39 1.32 0.61 0.35 0.17 6.03 5.27 1.01 0.98 (.999) (.999) (.994) (.771) (.998) (.987) (.652) (.854) (.736) (.384) (.108) (.190) E120700* 6.57 5.33 1.31 1.12 9.40 9.08 -0. 35 -0.60 11.75 7.99 -0.40 -0.46 (.682) (.376) (.228) (.406) (.401) (.105) (.652) (.531) (.227) (.157) (.520) (.539) # Securities 9 9 9999999999 # Observ. 475 475 475 475 475 475 475 475 475 475 475 475 P-value in parenthesis. The denotes a distribution event. Legend: p < 0.10; ** p < 0.05; *** p < 0.01.

PAGE 55

55Table 2-8. Hypothesis Tests Resu lts for Events in Year 2001 EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW H1 H2 H3H4H1H2H3H4H1H2H3H4 Coef. Coef. Coef. Coef. Coef. Coef. E010901* 18.19* 13.97** 0.59 0.12 11.05 9.39 0.28 0.10 5.52 4.81 1.13** 1.20* (.077) (.030) (.565) (.918) (.439) (.152) (.693) (.906) (.903) (.568) (.048) (.071) E042001 5.59 4.17 -0.22 -0.25 16.51 9.89 -0 .99 -0.96 15.64 10.50 -1.17** -1.16* (.899) (.653) (.832) (.837) (.123) (.129) (.163) (.245) (.154) (.105) (.042) (.084) E050201 13.94 9.70 -0.41 0.11 21.53** 10.00 -0 .10 0.24 12.12 4.93 -0.19 -0.40 (.236) (.137) (.686) (.925) (.028) (.124) (.885) (.775) (.354) (.552) (.740) (.545) E051801 16.90 11.52* -0.94 -1.04 18.63* 10.20 -1 .77** -1.41* 12.78 5.21 -0.86 -0.71 (.111) (.073) (.360) (.377) (.068) (.116) (.012) (.086) (.308) (.516) (.133) (.287) E062101* 19.79** 12.92** 2.08** 2.42** 30.05*** 9.62 1.59** 1.08 46.36*** 12.73** 1.53*** 1.03 (.048) (.044) (.044) (.042) (.001) (.141) (.030) (.204) (.000) (.047) (.009) (.128) E080701* 5.47 2.81 0.74 1.29 11.76 11.03* 1.33* 2.01** 8.33 6.90 0.70 1.16* (.906) (.832) (.471) (.274) (.381) (.087) (.060) (.014) (.683) (.330) (.222) (.083) E090601* 23.10** 13.77** 0.32 0.72 30.08*** 16.20** 0.53 0.56 27.34*** 13.71** 1.42** 1.59** (.017) (.032) (.757) (.543) (.001) (.012) (.451) (.499) (.004) (.033) (.014) (.017) E102501* 27.01*** 19.62*** -0.20 -0.14 33.58*** 28.72*** -0 .18 -0.31 25.73*** 17.71*** -0.54 -0.52 (.004) (.003) (.846) (.908) (.000) (.000) (.804) (.708) (.007) (.007) (.345) (.436) E111301* 5.06 1.92 0.68 0.54 4.72 1.60 0.91 0.55 5.20 2.79 0.33 0.07 (.928) (.927) (.507) (.647) (.944) (.952) (.196) (.499) (.921) (.834) (.560) (.912) E112101* 16.16 9.23 1.06 0.84 14.14 9.72 1.28* 1.38* 11.30 8.17 0.72 0.76 (.135) (.161) (0.300) (.473) (.225) (.137) (.069) (.093) (.418) (.226) (.208) (.252) E112301 4.33 2.45 0.61 0.71 15.67 7.20 0.36 1.15 16.61 7.40 0.38 1.18 (.959) (.873) (.548) (.545) (.153) (.302) (.610) (.162) (.120) (.285) (.586) (.150) E121101 6.87 3.38 0.46 0.46 7.28 3.43 0.46 0.47 17.73* 4.03 0.80 0.60 (.809) (.759) (.653) (.695) (.776) (.753) (.646) (.683) (.088) (.673) (.259) (.465) E121201 17.12 8.70 -0.61 -2.03* 16.24 11.69* -0 .68 -1.26 16.13 11.33* -0.68 -1.26 (.104) (.191) (.552) (.084) (.132) (.069) (.334) (.124) (.136) (.078) (.332) (.124) E121701 8.99 7.48 0.05 0.45 14.28 8.64 -0 .07 0.34 10.98 5.99 -0.12 0.31 (.623) (.278) (.961) (.700) (.217) (.194) (.924) (.675) (.445) (.424) (.836) (.640) # Securities 11 11 11111111111111111111 # Observ. 468 468 468468468468468468468468468468 P-value in parenthesis. The denotes a distribution event. Legend: p < 0.10; ** p < 0.05; *** p < 0.01.

PAGE 56

56Table 2-9. Hypothesis Tests Resu lts for Events in Year 2002 EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW H1 H2 H3 H4 H1 H2 H3 H4 H1 H2 H3 H4 Coef. Coef. Coef. Coef. Coef. Coef. E010902 9.29 7.45 0.33 0.96 27.77*** 17.00*** -0. 92 -0.15 18.66** 11.59* -0.49 0.12 (.504) (.281) (.744) (.406) (.002) (.009) (.202) (.851) (.045) (.071) (.396) (.852) E013102 33.22*** 19.27*** 1.97* 0.97 30.18*** 18.22*** 2.03** 0.95 28.67*** 23.41*** 2.14*** 1.89** (.000) (.004) (.056) (.402) (.001) (.006) (.046) (.406) (.001) (.001) (.003) (.019) E020102 6.53 4.53 0.82 0.82 8.25 6.16 0.73 0.83 8.01 6.07 0.73 0.83 (.769) (.604) (.422) (.478) (.604) (.405) (.303) (.302) (.627) (.415) (.298) (.303) E021902 5.44 3.06 -0.64 -0.71 10.87 9.40 -0 .08 -0.35 12.43 10.64 -0.03 -0.24 (.859) (.801) (.528) (.537) (.367) (.152) (.915) (.666) (.257) (.100) (.962) (.717) E041002 6.48 6.14 1.65 1.83 6.05 2.95 0.90 0.74 2.37 1.99 0.41 0.39 (.773) (.407) (.107) (.114) (.810) (.815) (.206) (.361) (.992) (.920) (.474) (.549) E042502 10.95 9.70 1.07 1.66 11.35 10.09 1.05 1.69 9.08 5.54 0.97 1.20 (.361) (.138) (.296) (.152) (.331) (.120) (.298) (.140) (.524) (.476) (.169) (.134) E050202* 16.45* 14.14** -2.05** -2.34** 7.72 4.63 -1 .05 -1.04 6.18 4.75 -0.78 -0.75 (.087) (.028) (.046) (.043) (.656) (.592) (.141) (.197) (.800) (.576) (.179) (.257) E060402 5.00 2.62 0.59 1.55 2.42 0.89 -0. 28 -0.21 1.44 0.85 -0.35 -0.34 (.891) (.854) (.561) (.178) (.992) (.989) (.690) (.798) (.999) (.990) (.543) (.607) E062102* 13.43 12.16* 0.78 0.88 5.52 4.15 -0. 68 -0.58 7.45 6.52 -0.21 -0.08 (.200) (.058) (.447) (.451) (.853) (.656) (.340) (.477) (.682) (.367) (.714) (.905) E081602 22.16** 10.38 0.82 1.35 16.67* 7.01 0.63 1.39* 13.35 4.47 0.01 0.77 (.014) (.109) (.426) (.241) (.082) (.319) (.380) (.084) (.204) (.612) (.981) (.243) E083002 30.91*** 21.10*** -1.02 -1.51 10.78 9.06 -0.71 -1.14 4.00 2.20 0.12 0.19 (.000) (.002) (.323) (.194) (.374) (.170) (.320) (.157) (.947) (.900) (.841) (.775) E090402* 4.56 3.86 -1.27 -1.52 8.23 6.65 -0 .65 -0.90 9.21 6.19 -0.94 -1.12* (.918) (.695) (.213) (.188) (.606) (.354) (.359) (.263) (.512) (.402) (.103) (.088) E101502 9.95 7.11 -0.90 -0.95 10.32 8.74 -1 .46** -1.75** 6.34 4.37 -1.20** -1.30** (.444) (.310) (.380) (.408) (.412) (.188) (0.040) (.030) (.786) (.626) (.038) (.048) E103002* 5.12 3.35 -1.18 -1.42 10.93 8.42 -1 .05 -1.23 7.80 6.98 -0.50 -0.43 (.883) (.763) (.252) (.220) (.362) (.208) (.144) (.130) (.648) (.323) (.393) (.516) E110702 10.37 5.01 -1.65 -1.07 6.46 3.17 -1 .03 -0.68 6.44 3.01 -1.04 -0.66 (.408) (.542) (.106) (.354) (.775) (.786) (.148) (.396) (.776) (.807) (.142) (.411) # Securities 10 10 10101010101010101010# Observ. 487 487 487487487487487487487487487487 P-value in parenthesis. The denotes a distribution event. Legend: p < 0.10; ** p < 0.05; *** p < 0.01.

PAGE 57

57Table 2-10. Hypothesis Tests Resu lts for Events in Year 2003 EVENT 1-DAY WINDOW 2-DAY WINDOW 3-DAY WINDOW H1 H2 H3 H4 H1 H2 H3 H4 H1 H2 H3 H4 Coef. Coef. Coef. Coef. Coef. Coef. E011403* 9.93 6.28 -0.04 -0.36 6.63 5.14 -0 .23 -0.53 1.39 1.19 -0.19 -0.27 (.446) (.392) (.968) (.744) (.760) (.525) (.738) (.515) (.999) (.977) (.726) (.681) E021703* 7.85 6.00 -0.18 -0.05 7.37 3.54 -0 .64 -0.39 5.42 2.01 -0.37 -0.37 (.643) (.423) (.851) (.962) (.690) (.738) (.344) (.626) (.861) (.918) (.501) (.576) E030703* 11.08 7.21 1.40 2.34** 9.51 4.39 -0 .24 0.64 3.90 2.67 0.05 0.49 (.351) (.301) (.141) (.036) (.484) (.623) (.723) (.426) (.951) (.849) (.929) (.458) E031103* 9.00 3.97 -0.54 -1.23 11.56 6.64 -0 .12 -1.03 10.79 6.09 -0.09 -0.98 (.532) (.681) (.565) (.271) (.315) (.355) (.854) (.201) (.373) (.413) (.895) (.221) E040803 22.65** 17.46*** -1.81* -3.54*** 46.45*** 25.85*** -1 .13* -2.65*** 44.90*** 24.81*** -1.18* -2.66*** (.012) (.008) (.059) (.002) (.000) (.000) (.098) (.001) (.000) (.000) (.079) (.001) E041703* 14.71 12.27* 0.78 1.09 14.76 3.03 0.00 -0.22 23.74*** 6.19 -0.24 -0.62 (.143) (.056) (.413) (.330) (.141) (.804) (.998) (.782) (.008) (.402) (.665) (.341) E043003* 18.25* 6.67 1.31 1.61 11.86 8.10 0. 32 0.39 10.91 6.64 -0.03 -0.33 (.051) (.352) (.169) (.151) (.294) (.231) (.635) (.631) (.364) (.355) (.956) (.614) E052603* 7.73 6.08 -1.40 -2.17* 9.39 8.82 -0 .50 -0.58 6.81 6.54 -0.24 -0.54 (.654) (.414) (.140) (.053) (.495) (.183) (.464) (.474) (.743) (.365) (.659) (.413) E070303* 8.28 5.80 -0.09 -0.44 13.04 2.71 0.68 -0.03 13.05 1.87 0.98* 0.40 (.601) (.445) (.925) (.697) (.221) (.843) (.315) (.974) (.220) (.931) (.073) (.539) E072103 2.43 1.37 -0.22 -0.49 4.89 2.20 0. 07 -0.10 3.24 1.35 -0.29 -0.35 (.991) (.967) (.815) (.664) (.898) (.900) (.922) (.903) (.975) (.968) (.602) (.597) E080503* 7.71 2.88 -0.10 0.81 7.46 3.58 -0 .94 -0.68 20.69** 4.08 0.12 -0.44 (.657) (.824) (.919) (.469) (.681) (.732) (.166) (.397) (.023) (.666) (.823) (.497) E090803 11.92 9.44 0.16 0.64 47.52*** 44.84*** 0.94 1.97** 39.44*** 36.50*** 1.08* 2.00*** (.290) (.150) (.866) (.567) (.000) (.000) (.167) (.015) (.000) (.000) (.051) (.003) E091603* 20.55** 7.23 1.22 1.56 11.99 4.11 0.57 0.60 18.72** 5.89 0.75 0.43 (.024) (.300) (.199) (.165) (.285) (.661) (.399) (.458) (.044) (.435) (.176) (.511) E092303 9.90 9.32 -0.33 -0.33 20.44** 18.74*** -0 .40 -0.42 14.07 11.58* -0.62 -0.44 (.449) (.156) (.731) (.772) (.025) (.004) (.557) (.600) (.170) (.072) (.258) (.505) E092503* 13.60 7.69 -0.47 -1.73 8.03 3.53 0.07 -0.86 8.13 3.58 0.07 -0.83 (.192) (.261) (.619) (.123) (.626) (.740) (.922) (.285) (.616) (.733) (.922) (.297) # Securities 10 10 10101010101010101010 # Observ. 430 430 430430430430430430430430430430 P-value in parenthesis. The denotes a distribution event. Legend: p < 0.10; ** p < 0.05; *** p < 0.01.

PAGE 58

58 Table 2-11. Hypothesis Te sts Results for CARs Before Significant Events EVENT H 1 H 2 H 3 H 4 Coef. Coef. E012700 5.29 3.84-0.15-0.61 (.808) (.573) (.951) (.830) E021700 27.03*** 12.69**-1.62-1.57 (.001) (.026) (.499) (.581) E041700 4.82 1.711.391.05 (.777) (.887) (.588) (.724) E072100 3.33 1.56-0.42-2.05 (.950) (.906) (.862) (.466) E111000 6.13 5.453.013.66 (.727) (.364) (.207) (.195) E010901 21.69** 18.44***2.414.25* (.027) (.005) (.265) (.080) E050201 15.89 10.79*-3.54-6.03** (.145) (.095) (.103) (.013) E051801 12.23 7.31-1.12-2.50 (.346) (.293) (.604) (.302) E062101 2.79 2.301.631.54 (.993) (.889) (.452) (.524) E080701 4.67 1.741.240.94 (.946) (.942) (.567) (.696) E090601 7.99 6.563.503.55 (.714) (.363) (.104) (.142) E102501* 32.46*** 26.88***4.95**6.05** (.001) (.001) (.022) (.013) E010902 4.56 1.60-0.82-0.36 (.918) (.952) (.709) (.883) E013102 17.24* 6.93-0.38-3.24 (.069) (.327) (.863) (.190) E050202 3.22 1.120.271.53 (.975) (.980) (.904) (.537) E062102 7.86 7.614.45**5.29** (.642) (.268) (.044) (.034) E 081602 22.71** 21.55***-6.37***-8.82*** (.012) (.002) (.004) (.000) E083002 8.62 1.86-0.890.65 (.568) (.932) (.686) (.792) E040803 33.71*** 18.53***2.745.65** (.001) (.005) (.196) (.025) E041703* 38.71*** 4.222.010.74 (.000) (.647) (.322) (.757) E090803 14.88 12.51*1.814.43* (.136) (.052) (.375) (.068) E091603* 46.22*** 29.26***5.08***3.59 (.000) (.001) (.007) (.113) E092303 25.99*** 18.60***-4.37**-3.71* (.004) (.005) (.016) (.086) P-value in parenthesis. The denotes a distribution event. Legend: p < 0.10; ** p < 0.05; *** p < 0.01. Table 2-12. Hypothesis Tests' Resu lts for Events' Overall Effect EVENT H1 H2 H3 H4 Coef. Coef. E021700 11.48 6.47-0.692.69 (.244) (.263) (.797) (.399) E102501* 29.85*** 24.19***4.52*5.55* (.002) (.001) (.067) (.065) E081602 17.71* 16.39**-5.60**-7.52*** (.060) (.012) (.026) (.008) E040803 11.98 6.820.482.14 (.286) (.338) (.847) (.472) E091603* 42.16*** 25.80***6.09***4.54* (.000) (.000) (.005) (.081) E092303 16.09* 11.14*-5.25**-4.94** (.097) (.084) (.012) (.049) P-value in parenthesis. The denotes a distribution event The event overall effect was computed only for significant even ts whose hypothesis tests rejected the null that the CARs in the 5-day period before the announcem ent was not different from zero (in case of E010901 the analysis showed that the observed abnormal returns could not be attributed to the announcement examined). Legend: p < 0.10; ** p < 0.05; *** p < 0.01.

PAGE 59

59Table 2-13. Significant Announcements Catego rization. Direction and Estim ated Magnitude of Regulatory Announcements Effect on Security Returns Positive Impact Negative Impact Differential Impact Across Firms in the Sample Type of Announcement Event Initiative Estimated Abnormal Returns Type of Announcement Event Initiative Estimated Abnormal Returns Type of Announcement Event Initiative Initiative to implement the model for the electricity sector (final ruling on agents participation in the market) E072100 Aneel 1.98 Rationing Plan E051801 Gov -1.77 Initiative to implement the model for the electricity sector (proposed ruling on agents participation in the market) E041700 Aneel Initiative to implement the model for the electricity sector (definition of the basic transmission network) E111000 Aneel 1.21 Initiative to implement a Congress decision (definition of low-income customers concept) E050202 Aneel -2.34 Change in ANEELs board of directors E050201 Gov Rate decision (initial proposal) E062101 Aneel 2.08 Compensatory measure E081602 Gov -5.60 Within sector redistribution policy E010902 Gov Rate decision (final decision) E080701 Aneel 2.01 Revision in regulatory agencies job E092303 Gov -5.25 Initiative to implement the model for the electricity sector (proposed ruling on asset base valuation methodology) E062102 Aneel Compensatory Measure E090601 Gov 1.42 Several rulings issued in the same day E083002 Aneel Compensatory Measure E102501 Gov 4.52 Partial revision of a within sector redistribution policy previously adopted E013102 Gov 2.14 Compensatory Measure E090803 Gov 1.08 Compensatory Measure E091603 Gov 6.09 Rate decision (final decision) E041703 Aneel 1.09 Includes only the events for which the null hypotheses H1 or H2 were rejected. Events E012700 E010901 E021700 and E040803 were dropped, however, because a deeper analysis revealed that either the abnormal returns could not be attributed to the re gulatory announcements investigated or the impact observed in the announcement day was actually a review in market expectations concerning the events effect.

PAGE 60

60Table 2-14. Random Events' Results 1999 2000200120022003 Event H1 H2 Event H1 H2 EventH1 H2 EventH1 H2 EventH1 H2 E041499 12.87 10.82 E012400 8.54 4.96 E011001 14.29 8.62 E012302 6.40 4.35 E021003 9.54 6.19 (.168) (.094) (.481) (.420) (.217) (.196) (.780) (.629) (.481) (.402) E042399 16.38* 14.22 E021000 13.74 1.69 E012901 5.48 3.27 E030102 5.95 1.79 E030503 8.89 4.21 (.059) (.027) (.132) (.890) (.905) (.774) (.745) (.937) (.542) (.648) E051899 10.94 7.71 E022500 4.14 3.09 E021501 6.11 2.97 E032702 1.78 1.00 E031303 19.24** 1.70 (.280) (.260) (.902) (.686) (.865) (.812) (.997) (.985) (.037) (.944) E060199 4.99 2.95 E032000 10.48 8.55 E042701 13.71 8.80 E040102 18.33** 16.79*** E032503 6.76 4.80 (.835) (.815) (.313) (.128) (.249) (.185) (.049) (.010) (.747) (.569) E090199 21.03** 20.27*** E032300 6.49 5.74 E070201 17.14 12.46* E042302 5.32 3.64 E032703 6.48 6.16 (.013) (.003) (.690) (.332) (.103) (.052) (.868) (.725) (.773) (.405) E092099 3.85 1.44 E041400 11.08 2.51 E070501 26.33*** 20.00*** E042902 8.35 4.96 E052803 4.08 3.63 (.921) (.963) (.270) (.775) (.006) (.003) (.594) (.548) (.943) (.727) E110399 4.49 1.76 E041900 5.03 4.36 E080601 6.51 3.62 E050902 11.92 9.33 E070403 13.15 3.62 (.876) (.940) (.754) (.499) (.837) (.728) (.290) (.155) (.215) (.727) E110899 3.63 2.99 E042000 22.02*** 19.26*** E083101 9.99 6.24 E060502 12.16 6.46 (.934) (.811) (.005) (.002) (.531) (.397) (.274) (.374) E120999 7.76 7.51 E052200 2.38 1.31 E100401 11.57 3.83 E070202 3.70 1.03 (.559) (.276) (.983) (.933) (.396) (.700) (.960) (.984) E121099 8.57 4.21 E062600 5.57 3.74 E102401 13.89 5.94 E072502 11.32 5.45 (.478) (.649) (.781) (.588) (.238) (.429) (.333) (.487) E072000 13.21 5.32 E102901 18.78 8.43 E080702 7.40 4.74 (.153) (.377) (.065) (.208) (.686) (.577) E090600 5.55 1.03 E103001 8.05 7.09 E080802 27.70*** 19.81*** (.784) (.959) (.708) (.312) (.002) (.003) E091300 3.09 2.58 E112201 6.03 5.14 E093002 19.67** 15.18** (.960) (.764) (.871) (.525) (.032) (.019) E092100 4.89 4.47 E112601 21.79** 6.32 E101102 2.84 1.57 (.844) (.484) (.026) (.388) (.984) (.954) E101000 4.38 1.08 E113001 18.4* 5.54 E101802 11.29 4.42 (.884) (.955) (.072) (.476) (.335) (.619) E102300 5.61 4.84 E102202 13.65 11.33* (.778) (.435) (.189) (.078) E103100 5.28 3.55 E111102 4.50 3.96 (.809) (.615) (.875) (.682) E111600 6.86 4.91 E111402 5.61 1.86 (.651) (.427) (.847) (.931) E120100 10.14 6.49 (.339) (.261) P-value in parenthesis. Legend: p < 0.10; ** p < 0.05; *** p < 0.01.

PAGE 61

61 CHAPTER 3 PRIVATIZATION, INCENTIVE REGULATION, AND EFFICIENCY IMPROVEMENTS IN THE BRAZILIAN ELECTRICITY DISTRIBUTION INDUSTRY Introduction Network industries have experienced a remarkab le change in the last twenty years. Once characterized by state-owned, vertically-i ntegrated companies, electricity power, telecommunications, natural gas, water & sewerage, railroads, a nd ports industries from several countries have gone through a reform proce ss which encompasses unbundling, privatization, introduction of market-oriented regimes for their competitive segments, and implementation of a new regulatory framework for the remaining segm ents with natural mon opoly characteristics. The new regulatory framework has involved the creation of i ndependent regulatory bodies and the incorporation of theoretical advances from the economics literature on incentive regulation. Here, the regulator is seen as a social welfare maxi mizer operating in a context of imperfect and asymmetric information regard ing firms demand, cost opportunities, and managerial effort. The regulator seeks to limit the regulated firms rents and to allocate some of these rents to consumers, subject to a firm break-even constraint.1 In this context, some incentive mechanisms emerge as an alternative to the customarily employed cost-of-service or rate of return regulation. Under a price-cap regime2, the most popular regulati on, prices are fixed. The firm and its managers are the residual claimant s on production cost reductions, and bear the 1 For a comprehensive review of the theoretical literature on incentive regulation, see Laffont and Tirole (1993), and Armstrong and Sappington (2003). 2 Sappington (2002) provides details on the design and implementation of price-cap regulation, as well as of other forms of incentive regulation.

PAGE 62

62 disutility of increased managerial effort (Jos kow, 2005). Thus, the conditions and incentives for efficiency improvement3 and for the possible achievement of second best prices are settled. Whether price-cap regulation effectively le ads to efficiency improvements, however, constitutes an empirical question, given the di fferent ways in which it is implemented in practice.4 Benchmarking, or comparative efficiency anal ysis, is a technique used to address the issue of relative performance since it enables the computation of efficiency scores and the analysis of their evolution over time. More than that, by provid ing information on each firms inherent cost opportunities, the benchmark exercise helps alleviate the pote ntial adverse selection problem faced by the regulator, consequently allo wing the establishment of cost-effective prices at the scheduled tariff review. From the researchers perspect ive, though, it allows an ex-post evaluation of the new prices the regulator has set. The present study uses a benchmarking methodology to assess the impact of privatization and incentive regulation on firms performance. The related literature on the electricity distribution sector is limited, possibly due to a reduced number of observations and data availability constraints.5 Nonetheless, for the empirical studies that have addressed the topic, the general finding is that privatization has been associated with improvements in efficiency, 3 The theory states that price-cap regulation provides incentives to improvements in performance in other dimensions as well, such as innovation, efficient choice of operating technology, and even service quality. In this paper, however, I focus on the efficien cy improvement possi ble impact only. 4 The length of time between schedule reviews and the degree of association of prices to realized costs, for example, may mitigate the efficiency improvements incentives brought by price-cap regulation. Wh en a price-cap plan links future prices directly to realized costs and the time between schedule reviews is relatively short, the incentives under a price-cap regime are similar to the ones under rate of return regulation (Sappington, 2002). 5 Given the natural monopoly characteristic of electricity distribution, there is a small number of firms in most countries which have undertaken sector reforms. In addition, there does not exist, up to this moment, a widespread understanding among regulators of the need to collect and keep the longitudinal data necessary to perform studies of this nature. This seems to occur even in Great Britain (see footnote 17).

PAGE 63

63 although only when accompanied by incentive regulation mechanisms (Estache, Perelman, and Trujillo, 2005, p. 8).6 Mota (2004), in contrast, compares the perfor mance of 14 Brazilian privatized electricity distribution companies to the average performance of 72 U.S. investor-own ed electric utilities, using data from 1994 and 2000, and finds that priva tization had no statistica lly significant impact on efficiency when operating costs are used as an input, but resulted in a strong drop in efficiency for the models that used total costs. Motas study finds that the Brazilian distribution companies experienced annual average producti vity gains of around 5% during the period. The present study of 52 Brazilian electric ity distribution companies evaluates the efficiency evolution and the productivity gain s that occurred in the sector from 1998 to 2003, checks for difference in performance between public and private firms, and examines the possibility of efficiency catch-up. It also investigates whether ve rtically integrated firms might be behaving strategically shifting costs from unregulated to regulated activities, and whether efficiency changes are associated wi th variations in service quality. The investigation provides evid ence of performance improveme nt after the implementation of sector reforms, and finds that both priv atized and public compan ies have reduced the efficiency gap with respect to companies that were privately owned before the reforms. The results show that privatized firms responded mo re aggressively than public firms to the new incentives brought by price-cap regulation. The fi ndings also indicate a possible strategic behavior associated with the peri odic aspect of price cap regulati on, as well as to cost shifting 6 Berg, Lin and Tsaplin (2005) also find that privatized firms respond differen tly to incentives than public firms. Their empirical analysis of 24 Ukrain e electricity distribution companies from 1998 to 2002 indicates that privatelyowned firms not only respond to incentives that add to net cash flows, but also respond more aggressively than do state-owned distribution utilities to cost-plus regulatory ince ntives that increase profits but decrease efficiency. The authors point out that comparisons of public and private utility performance need to be explicit about the incentive regimes facing both ownership types.

PAGE 64

64 implemented by companies that operate in the electricity generation segment. Moreover, the paper suggests that the high performance improve ment experienced by privatized firms in the period comes essentially from a more efficient op eration of their units, in line with what was expected under an incen tive regulation scheme, and not from mere reductions in costs brought by deterioration in the quality of service. Some of the results implications for policy are highlighted, and it is underline d the possible use of the papers findings to evaluate the regulators decisions taken and methodology employed at the peri odic tariff review. The following section briefly describes the refo rms undertaken in the Brazilian electricity sector. Section 3 discusses the technology of th e electricity distributi on industry, presents an overview of the different benchmarking techniques provides a detailed pict ure of the stochastic cost frontier approach, and shows how exogenous factors that influence producers performance may be incorporated into the analysis. The model specification and the data set are described in Section 4. Section 5 presents and interprets the results obtained, while Section 6 explores the possibility of strategic behavi or and the relationship between observed efficiency changes and variations in service qua lity. The final section provides c oncluding observations and directions for future research. Institutional Background The power sector reforms in Brazil bega n in 1995. While constitutional amendments abolished the public monopoly over infrastructure industries and allowed foreign companies to bid for public concessions, the Law 8,987/95 (General Law of Concessions) set the stage for the

PAGE 65

65 beginning of the privatization process, represen ted by the auctions of Escelsa in 1995 and Light in 1996. By the end of 2000, a total of 20 dist ribution companies had been privatized.7 In addition, part of the implementation of a new regulatory framework involved the establishment of an independent regulatory agen cy (ANEEL) in late 1996 and, in the same year, the commission of an international consulta ncy to study and propose a new model for the electricity sector. The consu ltants report was released in 1997 and its proposals were incorporated into Law 9,648, issued on May of 1998.8 In essence, the approved model focused on privatization and unbundling of generation, tr ansmission and distribution assets, gradual transition to a competitive generation environment in nine years, creation of a wholesale power market, operation of the transmission network by an independent operator, and use of the pricecap regime to regulate distribu tion tariffs, replacing the previ ous cost of service system.9 This paper focuses on the distribution segment only. Fo r the corresponding conce ssionaries, therefore, the sector reform changes which might have a ffected performance were privatization and the implementation of incentive regulation.10 7 Privatization concentrated on years 1997 and 1998, when nine and five firms where pr ivatized, respectively. Only three firms had been privati zed up to the end of 1996. 8 See Ferreira (2000), Mota (2003), and de Oliveira (2003), for detailed descriptions of the new models characteristics. 9 With exception to companies Escelsa and Light, price cap regulation was implemented through the signature of new concession contracts, which took place from 1998 to 2000, and had their first tariff review scheduled for after five (for contracts signed in 1998) or four years. Light was the first to have price-cap regulation applied, by order of the concession contract signed in Nove mber/1996, in which the first tariff revi ew was scheduled to occur after seven years. Escelsa was submitted to pricecap regulation in August of 1998, and had tariff reviews every three years thereafter. Except for Escelsa, a ll companies had the X factor set equal to zero in the first period prior to the first full review. 10 Competition in electricity supply to high-voltage customers began only in 2004. Therefore, its possible effects on distribution firms performance are not captured by the present study. On the other hand, distribution companies were affected by the unforeseen el ectricity crisis in 2001, caused by severe drought conditions and under investments in generation and transmission. The subsequent rationing measures proposed by the Government reduced significantly electricity consumption and firms re venues in that year, and n eed to be controlled for in the analysis.

PAGE 66

66 Methodology The Electricity Distribution Technology The modeling of electricity distribution tec hnology is not straightforward. Many factors influence electricity distribution costs and, with respect to some of them, a controversy exists over their exogeneity. Neuberg (1977) provided the theoretical foundati ons for the four factors that have been considered the main distribution costs drivers. The higher the amount of k ilowatt-hours sold, the greater the wear and tear on tr ansformers, while increases in the number of customers induce higher meter-reading and billing expenses. The pr obability of a wire-outage is assumed as a monotonically increasing functi on of network length, and geogr aphically dispersed areas, encompassing several cities, imply higher repair co sts because of the greate r labor input required by increased repair labor travel time, along with higher meterreading and billing expenses. Roberts (1986), on the other hand, emphasized the role of demand density (demand per unit of area), as a factor affecting scale ec onomies and, consequently, the average cost of delivering a unit of power.11 After noting that demand density can change if either the demands of existing customers change or new customers move into the service area, with the latter requiring customer-specific investments for the deliv ery of the product, the author argued for the inclusion of both factors (output density and customer density) in the model specification. Roberts work also pointed out the multiproduct nature of the electricity distribution activity, with the consequent importance of treating lowvoltage and high-voltage deliveries of power as 11 The study finds that average distribution cost falls as output per customer increases, a result consistent with previous empirical studies findings. According to Roberts, output density (output per customer) is useful for explaining not only differences in efficiency across co mpanies but also differences within firms across time.

PAGE 67

67 separate products, as well as the need to cons ider the percentage of more costly underground facilities on firms total distribution equipment. Some other distribution costs determinants are noted by Bu rns and Weyman-Jones (1996). Maximum demand determines the overall capacity of the system and at individual nodes, and together with energy delivered aff ects system load factor (reflecti ng the extent of peak use in the system). In addition, transformer capacity a ffects network losses. The type of customer, measured by the share of industria l KWh delivered, for example, de termines the extent to which power lines operate at different cap acities at different times, or th e effect of delivering energy at different voltages. This same effect is someti mes captured by the load factor, used by Filippini and Wild (2001) in place of maximum demand, and defined as the ratio of average over maximum demand. The higher the load factor, the lower the average distribution costs, given the smaller fluctuations of electricity demand over time. The way these factors should be incorporated in the analysis depends, initially, on the objective for the environment under studyoutput maximization or cost minimization. An important characteristic of th e electricity distribution industr y is that, in general, the concessionaries are required to provide service at specified tari ffs. As a consequence, output is demand driven, with firms maximizing their prof its by minimizing the cost of producing a given level of output. In this cont ext, a cost function is the a ppropriate approach to deriving performance comparisons. This approach also has the advantages of bein g able to accommodate the multiproduct nature of the electricity distri bution activity, and to treat variable and quasifixed inputs differently. The knowle dge that some inputs are not variable during the period under study can be exploited by replaci ng a cost function with a variable cost f unction (Khumbakar and Lovell, 2000).

PAGE 68

68 However, in a cost function approach, a meas ure of cost is modeled as a function of output(s) and input prices, leavi ng the question as to how all th e above-mentioned factors that affect distribution costs can be incorporated into the function. Neuberg (1977), taking the separate marketability of components as a necessa ry property of a vector of outputs, argues that only number of customers and electricity delivered might be considered as outputs. Nevertheless, the other factors could be included in a cost function if they were ta ken as exogenous factors reflecting differences in distri bution system from firm to fi rm (environmental variables). The exogenous characteristics of some of the cost determinants listed above are subject to debate. Frsund and Kittelsen (1998) question the exogeneity of energy deliv ered and number of customers in a context where firms can decide upon their prices, which fortunately is not the rule on this regulated industry. The major concern in volves network length a nd transformer capacity, which Fillipini and Wild (2001), among others, do not use as explanatory va riables, arguing that they are capital inputs endogenous to the firm. Th ese same factors are sometimes considered as fixed capital inputs in the short run, and consequently included as explanatory variables in a variable cost specification, such as those us ed by Salvanes and Tj tta (1994), Burns and Weyman-Jones (1996) and Botasso and Conti (2003). Moreover, Neuberg (1977) had already ra ised the point that the network lengths exogeneity might come from taki ng it as a proxy for a linear meas ure of the territory, assuming a fixed geographic distribution of customers, whose importance is emphasized by Kumbhakar and Hjalmarsson (1998) in the contex t of sparsely populated count ries, where, according to the authors, the amount of capital in the form of network reflects the geographical dispersion of customers rather than differen ces in productive efficiency.

PAGE 69

69 An additional insight is provided by the empiri cal literature on comparative efficiency analysis. Jamasb and Pollitt (2001) survey this literature and report the frequency with which different input and output variables are used to model electricity distribution. Mota (2004) performs a similar task, but concentrates only on academic studies. Both surveys find that the most frequently used inputs are network length, transformer capacity and number of employees, and the most widely used outputs are units of energy delivered (with a high proportion of the studies decomposing it into high-voltage and low-voltage sales), number of customers and service area. Mota (2004) also reports that load factor, cust omer density, and output density are used as environmental variab les in the studies surveyed. Comparative Efficiency Studies The increased emphasis on efficiency analysis has its origin in the implementation of incentive regulation. The adoption of incentive m echanisms aimed at improving the performance of companies formerly subject to rate-of-return regulation br ought the need to measure the expected efficiency gains at the firm level, to be reflected in the X factor of a price-cap regime with an RPI-X rule.12 In addition, the policy ch ange stimulated research on whether the incentive mechanisms have effectively attained th eir performance improvement objectives. In comparative efficiency studies, the estimati on of an efficient frontier is the shared goal of the different methodologies. In all benchmar king methods, firms efficiency is given by a measure of the distance of the obs erved practice to the efficient frontier. What differs is the technique for estimating the efficient frontier. 12 The use of a comparative efficiency study with this purp ose is subject to criticisms, however. Shuttleworth (2003) and Irastorza (2003), among others, ar gue that it provides misleading results, by confusing inefficiency with heterogeneity, and therefore should not be used.

PAGE 70

70 A first possibility is to perform a botto m-up efficiency study, where the theoretical yardstick comes from the engi neering knowledge of the indust ry process. This model (or theoretical) firm approach has not been the rule, but it was used in countries like Spain, Chile, Peru, and more recently by the elect ricity regulator in Brazil. Acad emic researchers have focused on estimating the efficient frontier based on an em pirical estimate using observed data, with its estimation being implemented with either a parametric or a non-parametric technique.13 Non-parametric methods, like Data Envelopment Analysis (DEA), use mathematical programming techniques and do not require specification of producti on or cost functions nor the imposition of behavioral assumptions. These methods are generally easy to implement, but carry an implicit restriction in the number of variable s that might be used, and do not allow for random shocks. Parametric methods, in turn, entail applying an a priori functional form to the frontier, estimated with econometric tools. They allow for hypothesis testing,14 enabling the analyst to investigate the validity of the model specification. Tests of significance can be performed for the functional form and for the inclusio n or exclusion of factors, which is of special relevance for the electricity distribution in dustry, where the inclusion of several factors is theoretically justifiable. Moreover, with a parametric meth od it is possible to allow for stoc hastic factors or measurement errors, which avoids the assumption that all devi ations from the best practice frontier involve inefficiencies. 13 For a detailed description of the different methods to perform efficiency analysis and an assessment of the strengths and weaknesses of each, see Kumbhakar and Love ll (2000), Coelli, Prasada Rao, and Battese (1998), Cubbin and Tzanidakis (1998), Sarafidis (2002), and Atkinson et al (2003). 14 In non-parametric models, a bootstrap technique may be used to produce confidence intervals around the estimated individual efficiency and thereby assess statistical properties of the efficiency scores generated (Simar and Wilson, 1998).

PAGE 71

71 Thus, parametric methods can be deterministic or stochastic. A deterministic approach, like the Corrected Ordinary Least Square (COLS), does not allow for random shocks of elements beyond management control, which might have also contributed (positively or negatively) to the discrepancy between the individual firm performance and the frontier.15 This problem can be addressed by using the so-called stochastic front iers (Stochastic Frontier Analysis [SFA]), which use a mix of one-sided and tw o-sided error terms, with the former capturing the firms inefficiency and the latter capturing the e ffects of random variation in the operating environment.16 Jamasb and Pollitt (2001), Mota (2004), a nd Estache, Perelman, and Trujillo (2005) perform comprehensive reviews of the compara tive efficiency literature on the electricity industry, providing several examples of the us e of the methods mentioned above to examine firms performance. In some of the existing studies, however, th e choice of method was determined by ease of use or limited by sample size or data restrictions.17 Ideally, the decision re garding the appropriate method depends on the purposes of the study and the context under examination. This study aims to investigate efficiency e volution through the period of 1998 to 2003, the decomposing of productivity growth for each firm into techni cal change and efficiency change, looking separately at public and private firms, and the degree of convergence in efficiency scores. The investigation, in turn, is conducte d in an environment where random shocks were present and the 15 Another drawback of COLS is that the structure of best practice produc tion technology is the same as the structure of the central tendency pr oduction technology, since the estima ted frontier is parallel to the OLS regression. Thus, the frontier does not reflect the production technology of the most efficient producers, but the one from producers down in the middle of the data (Khumbhakar and Lovell, 2000). 16 The stochastic frontier approach is described in Section 2.3. 17 For example, Pollitt (2005) reports that sample size and data limitations have restricted Ofgems methodological choices and prevented the successful implementation of SFA and the incorporation of stochastic factors into the analysis of efficiency.

PAGE 72

72 inclusion of several variables in the model specification, besides bei ng theoretically justifiable, is advisable due to the high heterogeneity in operati ng conditions. These considerations lead to the use of a stochastic frontier approach, defined in terms of an input orie ntation, given the output exogeneity that characterizes the electricity distri bution industry. Stochastic Cost Frontier and Trea tment of Environmental Variables18 A cost frontier can be expressed as Ei c(yi, wi; ), i = 1, I, (3-1) where Ei = T iwxi = nwni xni is the expenditure incurred by producer i, yi = (y1 i, yM i) 0 is a vector of outputs produced by producer i, wi = (w1 i, wN i) > 0 is a vector of input prices faced by producer i, c(yi, wi; ) is the deterministic cost fron tier common to all producers, and is a vector of technology parameters to be estimated. When the formulation above incorporates the fact that expenditure may be affected by random shocks not under the control of producers, the following stochastic cost frontier is obtained: Ei c(yi, wi; ) exp{vi} (3-2) Thus, the stochastic cost frontier consis ts of two parts: a deterministic part c(yi, wi; ) common to all producers and a prod ucer-specific random part exp{vi}, which captures the effects of random shocks on each producer. In this context, a measure of cost efficiency of producer i is given by CEi = i i i iE v w y c } exp{ ) ; ( (3-3) 18 This section draws upon Kumbhakar and Lovell (2000).

PAGE 73

73 which defines cost efficiency as the ratio of minimum cost attainable in an environment characterized by exp{vi} to observed expenditure. CEi 1, with CEi = 1 if, and only if, Ei = c(yi, wi; ) exp{vi}. If it is assumed that the deterministic part c(yi, wi; ) of a single-output cost frontier takes the log-linear Cobb-Douglas functional form, the stochastic cost frontier can then be written as i n ni n i y iv w y E ln ln ln0 (3-4) i i n ni n i y iu v w y E ln ln ln0 (3-5) where vi is the two-sided random-noise component, and ui is the nonnegative co st inefficiency component of the composed error term i = vi + ui. The noise component vi is assumed to be iid and symmetric, distributed independently of ui. Thus, the error term i = vi + ui is asymmetric, being positively skewed since ui 0. Under the above representation, a measure of cost efficiency of each producer i is provided by CEi = exp{-ui}. (3-6) Estimates of the production technology parameters, as well as of the cost efficiency of each producer, can be obtained with the maximum likelihood method, which requires that some distributional assumptions be made. While vi and ui must be assumed as distributed independently of each other and of the regressors, estimation of CEi requires that separate estimates of statistical noise vi and cost inefficiency ui be extracted from estimates of i for each producer, which, in turn, calls for distributi onal assumptions on the two error components. Therefore, by assuming that the two-sided rando m-noise component is nor mally distributed, and that the nonnegative cost inefficiency compone nt follows a half-normal, an exponential, a

PAGE 74

74 truncated normal, or a gamma distribution, CEi is obtained from the conditional distribution of ui given i.19 More formally: CEi = E(exp{-ui}| i) (3-7) It is worth noting that additional information might be provided when repeated observations on each producer are available. In a panel formulation, evidence of cost efficiency change can be obtained by including time as a m ean inefficiency parameter, when a truncated normal distribution is assumed for ui. On the other hand, the effects of technical change can be captured as well if time is included in the determin istic kernel of the stochastic cost frontier. When a translog functional form is adopted, that would amount to the inclusion of a time-trend (t) and its square (t2) as additional regressors, obtaining: i i tt t ni i n yn n ki k ni nk i yy n ni n i y iu v t t w y w w y w y E 2 2 02 1 ln ln ln ln 2 1 ln 2 1 ln ln ln (3-8) Technical progress will be evidenced by a negative partial derivative of observed expenditure with respect to time. It should be stressed, however, th at the inclusion of time in the manner depicted in Equation 38 accounts for what is known as Hicks-neutral technical change. This essentially implies that the cost frontiers (as production isoquants) ar e shifting each year but their slopes (e.g. the MRTS) do not change (Coelli, Prasada Rao, and Battese, 1998). Nonneutral technical change is obtained by also including terms involving the intera ctions of the other regressors and time.20 19 See Kumbhakar and Lovell (2000, pp. 141-2) for the likelihood function and CEi point estimator expressions. 20 Likelihood Ratio tests might be employed to guide the decision upon which formulation should be used to account for technical change (neu tral or non-neutral).

PAGE 75

75 The stochastic frontier approach provides di fferent ways to incorporate environmental variables, which exert an influence on producer performance in spite of not being inputs or outputs of the production process. The point deserves special atten tion in an efficiency analysis context, where it is essential to control for variation in producer performance due to variation in exogenous variables characterizing the environment in which produc tion takes place. The way to proceed, however, depends on a previous judgm ent about how the interference of exogenous factors occurs. The environmen tal variables may influence th e structure of the technology by which conventional inputs are conv erted to outputs, or they may influence the efficiency with which inputs are converted to outputs. In the firs t case, environmental factors should be included directly in the production or co st frontiers as regressors, produ cing efficiency scores which are net of environmental influences In the second, these factors s hould be modeled so that they directly influence the inefficiency term.21 For this last case, one possibility is to a ssume a truncated normal distribution for the inefficiency error term and relax its consta nt-mean property, by allowing the mean to be a function of the e xogenous variables (zi). More specifically, ui ~ N+(it, 2u), with it specified as M j it j j itz1 0 (3-9) Another possibility is to rela x not the constant-mean but the constant-variance property of the truncated (or half) normal distribution for the inefficiency error term, by allowing the 21 This approach yields efficiency scores which incorpor ate the environmental effects. Coelli, Perelman and Romano (1999) call them gro ss efficiency scores The study proposes the substitution of M j it j jz1 ,in equation (3-9) for min[ M j it j jz1 ,], in order to obtain net efficiency scores. It is argued that the modificati on enables the efficiency measures to be estimated in a context where all firms ar e assumed to face identical conditions (i.e., the most favorable).

PAGE 76

76 variance to be a function of the exogenous variables. According to Kumbhakar and Lovell (2000), this procedure makes it possible not only to incorporate exogenous influences on efficiency but also to correct for one possible source of heteroskedastic ity. More specifically, ui ~ N+(0,2 uit), with 2 uit specified as M j it j j uitz1 0 2 (3-10) In spite of being sometimes neglected in practice, the possibl e violation of the homoskedastic assumption requires special attenti on in parametric effici ency studies, since the consequences of heteroskedasticity are potentially more severe in stochastic frontier models, than in a classical linear regression model.22 Heteroskedasticity can appear in either error component, as long as the sources of noise and/or inefficiency vary with companies size, what is quite possible. While unmodeled heteroskedasticity in the symmetric noise error component (vi) leads to biased estimates of technical efficiency, unmodeled heteroskedasti city in the one-sided inefficiency error component (ui) leads to bias in both estimates of the parameters of the cost frontier and estimates of technical efficiency.23 The empirical literature on efficiency analys is provides examples of the use of these different approaches to account for environmental variables. Burns and Weyman-Jones (1996) and Estache, Rossi, and Ruzzier (2 004) include these variables as additional regressors in the functions employed, Hattori, Jamasb, and Po llitt (2003) and Mota (2004) utilize the environmental factors to mode l the mean inefficiency, and Botasso and Conti (2003) employ them to model the variance of the inefficiency error term. There ar e also some studi es that apply 22 If the error term is heteroskedastic in a classical linear regression model, estimators are unbiased and consistent, but not efficient. 23 See Kumbhakar and Lovell (2000, pp. 115-122), for details on the direction of bias on estimates of technical efficiency caused by heteroskedasticity in either of the error components.

PAGE 77

77 more than one of these approaches. Wang (2002) parameterizes both the mean and the variance of the inefficiency error term to accommodate non-monotonic e fficiency effects of exogenous variables. Coelli, Perelman and Romano (1999) and Hattori (2002), on th e other hand, utilize models where environmental variables are firs t included as arguments of the input distance function and then as parameters of the mean in efficiency, comparing the specifications employed on the basis of the likelihood-ratio test. Both studies find that the sizes of the estimated efficiency measures differ significantly w ith the model selected, denoting that the methodological decision is a key point to be addressed in an efficiency analysis.24, 25 Specification and Data In the present study, the sample was limited by availability of data and by the decision to exclude some very small concessionaries, wh ich deliver less than 100,000 MWh per year. From the total of 64 electricity di stribution companies in the count ry, nine were dropped from the sample due to small size and data for three others were unavailable. The unavailability of data also prevented the incorporation of the period be fore 1998 into the study. Therefore, the sample includes 52 companies, responsible for 99.47% of the total electricity delive red in the country in year 2003, with the data being co llected for the period of 1998 to 2003. The data were assembled from the regulatory agency, the companies websites, the financial statements provided to the Sao Paulo Stock Exchange, the Brazilian Associ ation of Electricity Distribution Companies (ABRADEE), the Brazilian Institu te of Statistics (IBGE), a nd the Caixa Economica Federal 24 Coelli, Perelman and Romano (1999) argue that, in the ab sence of a strong preference for one approach over the other, it is advisable to turn to the data for guidance. 25 No matter which approach is adopted to control for environmental variables, some unobserved heterogeneity might still be embedded in the estimated efficiency meas ures. To address this poin t, Greene (2005a, 2005b) proposes a true fixed effects model, in which firm fixed e ffects are included either as regressors or as parameters of the mean inefficiency, along with some environmental variables or not. This approach, though, might result in biased results if the number of repeated observations on each producer is small. In a ddition, the method forces any time-invariant inefficiency to be absorbed by the firm specific constant term, resulting in an underestimation of inefficiency.

PAGE 78

78 CEF, a public financial institution that is in charge of most of the social programs of the federal government and provides financing to house construction projects. One of the companies in the sample (Elektro) resulted from a split that occurred in January of 1998. Consequently, its data fo r that year were not used. Another company (Bandeirante) was split in 2001. The companys data for the year 2001 was disregarded, and the resulting two companies were treated as new firms in the samp le (after 2002). As a result, the sample has 50 firms in 1998, 51 in 1999 and 2000, 50 in 2001, and 52 firms in 2002 and 2003. Among the sample companies, 23 were publicly owned in the beginning of 1998. Seven of them were privatized during the period examin ed (four in 1998, one in 1999, and two in 2000). For the purposes of this study, a pr ivatized company was considered Private in the same year it was sold only if the privatization occurred before June 30th. Therefore, there were 21 publicly owned distribution companies in 1998, 19 in 1999, 17 in 2000, and 16 from 2001 to 2003. We use a variable cost specification, give n our belief that transformer capacity and network length constitute capital inputs that are fixed in the short r un, and provide important information in terms of system configuration. In addition, environmenta l variables are included as arguments of the variable cost function, instead of as mean inefficiency parameters, as we are interested in having efficiency indexes net of factors exogenous to the firms.26 We turn to the stochastic frontiers hypothe sis testing capabilities to guide our decision regarding the functional form, th e incorporation of technological change, and the distributional assumption for the inefficiency error term. Li kelihood ratio tests reje cted not only the CobbDouglas functional form in favor of the transl og, but also the Hicks-ne utral formulation of 26 It should also be stressed that the specification em ployed provided a better fit than the one where the environmental variables entered as mean inefficiency parameters, according to the Likelihood-ratio tests that were performed comparing these two alterna tive ways of treating environmental variables against a nested model.

PAGE 79

79 technical change in favor of the non-neutral formulation. Moreove r, the tests supported the halfnormal distribution for the inefficiency erro r term, compared to the truncated-normal distribution.27 In light of the pronounced heterogeneity among the companies in the sample, in terms of size and customer structure,28 we checked for the presence of heteroskedasticity on the two error components. The null of homoskedasticity was supported for the two-sided noise component (vi), but rejected for the one-sided inefficiency error term (ui). As a result, in our model the variance of inefficiency error component is conditioned on a proxy of firm size, given by total electricity delivered (Q). Hence, the specification adopted is shown in equation 3-1129 it it tt jn nit nt it yt t jit j i l i c nit it n yn n kit k nit nk it yy n nit n it y itu v t t w t y t Z Len Cap w y w w y w y E 2 2 02 1 ln ln ln ln ln ln ln ln ln 2 1 ln 2 1 ln ln ln (3-11) where E and y are the cost and output measures, respectively, W is the vector of factor prices, Cap stands for transformer capacity, Len represents network lengt h, Z is the vector of environmental variables, and it is assumed that vit ~ N(0, 2 v) and uit ~ N+(0, 2 uit), with 2 uit specified as 2 uit = it QQ 0 27 The null hypothesis that the mean of th e inefficiency error term distribution is equal zero was not rejected, even at the 10% significance level. 28 This point is detailed in the interpretation of the descriptive statistics, presented below. 29 The decision to not use a multiple output formulation, where outputs would be defined in terms of electricity delivered to low-voltage customers and to high-voltage cust omers, was due to the inclus ion of share of electricity delivered to industrial customers as an environmental variable. The alternativ e specifications were compared, with the one employed in the present study providing a better fit.

PAGE 80

80 The modeling of technical change in the wa y shown in equation 3-11 is used to get evidence of technological change over the period considered. Fo r the computation of firms efficiency indexes and the consequent analysis of efficiency change, however, we turn to the use of time fixed effects, to contro l for possible changes in macroec onomic factors that might have affected firms performance during the period unde r investigation. In the present study, it is important to avoid attributing to efficiency change variations in cost caused by other phenomena, such as industry-wide technologica l advances and change s in interest rates, exchange rates or electricity sector policy. Note that the use of tim e fixed effects explicitly allows the computation of efficiency indexes relative to yearly-specific frontiers. The evolution of firms performance (efficien cy change), the possibility of differential performance between private and publicly owned firms, and the existence of a privatization effect on firms efficiency are investigated with a conditional mean specification, where the following variables are included as parameters of th e mean inefficiency error term: (a) indicators of time (time trend and its s quare); (b) a private dummy ( PRIVATE ); and (c) two other indicator variables representing companies that were privatized ( PRIVTZED ) and the ones that were already private before the beginni ng of the privatization process ( ALWSPRIV ). Additional insights on efficiency evolution ar e supplied by the analysis of av erage inefficiency scores and mean relative cost inefficiency measures, comp uted for each year in the period of 1998 to 2003, which also enables the investigation of the possi bility of efficiency catch-up over the period. The observed technological change ( TC ) and technical efficiency change ( TE ) are then combined to provide a more complete picture of the productivity improvements occurred in the period under examination. This is done thr ough the computation of Malmquist productivity indices, following the methodology proposed by Coelli, Prasada Rao, and Battese (1998) for

PAGE 81

81 stochastic frontier methods, adapted to a cost frontier context. For each firm, the Malmquist index of productivity change between two consecutive periods is given by j j jTC TE MI where 1 1 ., 1 t j t jIndex Eff Index Eff TE and 2 / 1 1 ,1 1 t j t jTC TC TC The resulting measures of productivity ch ange, along with their decomposition into productivity catch-up ( TE) and frontier shift ( TC) components, are employed to compare the performance of firms by ownership type. In our model (equation 3-11 above), the depende nt variable is give n by the operating costs of distribution and retail servic e activities (Opex), computed as the sum of labor, materials and third party service contracts expenses, as reported in the income statement.30, 31 Electricity delivered, in MWh (Q), is the output measure32 and average wage, calculated as total labor expenditure divided by the number of employees, is used as a proxy for th e price of labor (LP).33 For the prices of materials (MP) and third party services (SP), the work uses two price indexes provided by IBGE and CEF. The mate rials price index reflects th e observed change in the price of a basket of items used in civil construction, by State, whil e the third party services index 30 The computed labor expenses include firms contributions to pension funds and to health insurance plans, profit sharing payments, and management wages. Some firms alr eady report these expenses under the classification of labor expenses, but most of them do not. The necessary adjustments were made on these cases. 31 In case of vertically integrated companies, the computa tion of the operating costs of distribution and retail service activities was made possible by the fact that those companie s are required by law to report their expenses separated by activity. 32 The use of two measures of output was prevented by the fact that electricity deliver ed and number of customers showed up as highly collinear, with one of them being alwa ys dropped by the statistical software employed (Stata). A better specification was provided by the former, when compared to the latter. 33 Total labor expenditure is employed to compute average wage because it was not possible to obtain information related to number of employees segregat ed by sector activity, for the cases of firms that also operate on generation and transmission.

PAGE 82

82 portrays the observed change in the salaries paid to an electr ician, also by State. The variables Opex, LP, MP, and SP are expressed in 1998 valu es, being deflated by a general price index (IGP-DI). The regulatory agency does not keep track of changes in transformer capacity and network length of each distribution co mpany over time. However, information on these variables was needed for the periodic tariff review, and gather ed in year 2002. It is worth noting that by employing these data on the present study we are implicitly assuming that both variables remained constant during the whole period under examination, despite th e observed increases in output and in the number of connections.34 Notwithstanding the possi ble stringency of this assumption, the variables were kept in the mode l due to their observed importance in explaining variation in operating costs among firms, as is mentioned in the next section. Transformer capacity is given in MVA, and network length corresponds to the sum of high-voltage and lowvoltage lines, in kilometers. Since these variables showed up as highly correl ated with electricity delivered, the variables Cap and Len in our spec ification (equation 3-11) actually correspond to the residuals of the regression of transformer capacity on electric ity delivered and network length on electricity delivered, respectively. Particular attention was give n to incorporating exogenous f actors that could control for differences in firms operating conditions, give n the heterogeneity th at characterizes the Brazilian electricity dist ribution industry, as well as our interest in having efficiency measures net of factors that impact firms performance but are out of contro l of the concessionaries. After checking the significance of their influence on fi rms technology (as cost frontier shifters) and on 34 The companies in the sample experienced an increase of 23% in the number of customers, from 1998 to 2003. Electricity delivered by these firms, on the other hand, increases 7.6% from 1998 to 2000, and 2.1% from 1998 to 2003.

PAGE 83

83 firms efficiency (as mean inefficiency parameters)35, the following environmental variables were included in the modeling: customer de nsity (CusDen), given by number of customers divided by network length; share of electricity delivered to industrial customers (IndShare); residential density (ResDen), computed as electricity delivered to residential customers divided by the number of residentia l customers; service area36 (Area), in Km2; ratio of underground to overhead lines (Undergrd); and income per capita, by State (Income), to cont rol for variations in socio-economic conditions among States. Descriptive statistics are shown in Tabl e 3-1. The difference between minimum and maximum values of observations collected for almo st all variables used i ndicate the considerable heterogeneity among firms in the sample, in terms of companies size, syst em configuration, and customer structure. Electricity delivered, for example, varies from 103,191 to 37,540,051 MWh, while the share of elec tricity delivered to industrial cu stomers ranges from 3 to 64%, and transformer capacity fluctuates in the interval of 120 to 22,728 M VA. The evidenced disparity in firms indicators corroborates the need to acc ount for external factors in the comparative efficiency analysis. It is observed, from the evolution of the va riables mean values, that operating costs drop around 26%, in real terms, from 1998 to 2003.37, 38 This reduction is partly due to the falling 35 The six environmental variables incl uded in our specification showed up as statistically significant mean inefficiency parameters. The results indicated that firms efficiency increases with IndShare, ResDen, and Income, and decreases with CustDen, Area, and Undergrd. 36 The Brazilian case justifies the inclusion of both netw ork length and service area in the modeling and this is reflected in the statistical si gnificance of both variables as either cost shifters or mean inefficiency parameters. While some companies have small service areas and relativel y high network length (the ones that operate in the more densely populated states), others have high Area but relatively low Len, because they operate in states which are more sparsely populated and/or have a high share of the population not being served. 37 The indicated variation, as well as all others based on the numbers portrayed in Table 3-1 and mentioned in this section, is adjusted for the change in the numbe r of firms in the sample from year to year.

PAGE 84

84 prices of materials and third party services, which fell 12.9% and 21.9% respectively, in the same period. It is also interesting to note th e dramatic reduction in electricity delivered that occurred in 2001 (-7.6%), as a consequence of the rationing imposed by the government. The volume of electricity delivered increased in 2002 and 2003, but it st ill was, in 2003, smaller than the volume delivered in 1998, and 5.1% less than the amount delivered in 2000. The same rationing effect is observed in the average residential consumption, which decreased 13.8% in year 2001. This indicator, however, does not recover in years 2002 and 2003 (it is even slightly smaller on these years). Th is fact, which may be due to higher electricity prices in the post-rationing peri od, might also indicate a shift in the residential customers demand for electricity, as a result of a change of habits induced by the rationing measures.39 Findings Prior to estimation, all variables were normalized by their sample median values. Additionally, in order to ensure homogeneity of degree one in pric es, the dependent variable and the factor prices were normalized by the price of third party services. The models are estimated by maximum likelihood, using the Stata 9 software. Table 3-2 provides the results from the models estimated, while the elasticities of operating costs with respect to outpu t, factor prices, and time are reporte d in Table 3-3. It is observed that the estimated cost function satisfies the monotonicity condition with respect to output and factor prices at the mean. Moreover, as was expect ed, operating costs are negatively affected by 38 The higher drops occur from 1998 to 1999 (-14.1%), and from 2000 to 2001 (-9.5%), year in which there was the energy rationing. Mean operating costs are also reduced by 5.7% from 2001 to 2002. 39 From 2001 to 2003, the number of residential customers of the sample firms increases from 38,160,276 to 42,237,897, with the amount of electricity delivered to this group of customers going from 66,399,207 MWh to 70,824,904 MWh. This volume of electricity was still sma ller than the one observed in 2000 (75,283,460 MWh), when there were 36,816,473 residential customers. Thus, the already existing residential customers in year 2000 consumed, in that year, more electricity than in year 2003. The numbers abov e are adjusted for the change in the number of firms in the sample from year to year.

PAGE 85

85 increases in the share of electr icity delivered to industrial customers, and positively affected by increases in electricity delivered service area, factor prices, sh are of underground facilities, and customer density. Note also that transformer cap acity and network length show up as statistically significant (at the 1% level) va riables explaining variations in operating costs among firms. As capital proxies, their positive coefficients met our expectation. The income per capita coefficient estimate is negative and significant, which might be reflecting higher maintenance costs incurred in lo w-income areas and, possibly, the impact of work force qualifications on firms productivity Furthermore, the positive and significant coefficient on Q as a variance parameter indicates that va riations in firms inefficiency levels increase with firm size. The time elasticity provides a measure of technol ogical change. The results show that there was technological progress during the sample period, with an annual rate of technological change of around 6.55%, on average, which denotes that the efficient fron tier has shifted considerably from 1998 to 2003. Note also that the estimated coefficient on the squared term of time is negative, indicating that the obs erved rate of technological pr ogress has increased through the period examined. We investigate the possibili ties of differences in tec hnology and in the rate of technological change between public and privat e firms through the use of a private dummy and an interaction term private*time. The negative and significant private dummy coefficient (see Model B in Table 3-2) indica tes that private firms have had a better technology. This conclusion, however, should be ta ken with caution, since the dumm y captures the effect of any systematic difference between the firms groups which might include a difference in firms performance.

PAGE 86

86 The results also show a slightly higher rate of technological change for public firms, when compared to their private counterparts. Nevert heless, when private firms are separated into privatized and not privatized fi rms (called always private), it is observed that the above mentioned relationship holds only for the comparison of public and always private firms, since the privatized companies actually had a higher ra te of technological chan ge than the public companies (Model C in Table 3-2). These result s are corroborated by th e information provided in Table 3-4, where the technological change obs erved in the period is segregated by year and firms ownership type. Firms efficiency indexes are computed on the basis of the time fixedeffects specification. An indication of efficiency e volution throughout the period and of comparative performance of public and private companies is provided by the results of the conditional mean inefficiency model (Models E and F in Table 3-2). The coe fficient on the time variable is negative, but not significant, suggesting that efficiency impr oves only slightly over the period. On the other hand, the results indicate that private firms, sp ecifically the privatized ones, are significantly more efficient than public firms in the period examined. A more detailed picture is seen in Table 3-5, which reports average inefficiency scores and mean relative cost inefficiency measures, comp uted for each year in the period of 1998 to 2003 and discriminated by firms ownership type. Th e relative cost inefficiency was obtained by dividing each firms inefficiency score by the minimum score observed in each year. The efficiency improvement is reflected in th e reduction of average inefficiency scores in the period. On average, efficien cy increases 1.51% from 1998 to 2003. The efficiency gains are followed by a concomitant reduction in dispersion of efficiency scores, with the standard deviation going from .0999 in 1998 to .0697 in 2003, constituting evidence of a catch-up effect.

PAGE 87

87 This evidence is corroborated by the mean relative inefficiency measures. While in 1998 the distribution companies were, on aver age, 8.7% more inefficient th an the most efficient company in that year, in 2003 they were only 6.3% more inefficient than the most efficient distribution utility. This indicates that the observed effi ciency improvement comes, essentially, from companies that were relatively ineffi cient in the beginning of the period. Additional insights are obtained when one looks at the evolution of these measures for public and private firms separately. Here, to avoid possible distortions in th e efficiency evolution analysis resulting from the privatizations occurred in the period40, the average inefficiency and the mean relative inefficiency measures comput ed for public firms reflect only the performance of the 16 firms that remained public up to 2003. Notice first that in the beginning of the pe riod the group of private companies which did not go through a privatization pro cess is more efficient, on averag e, than both the public and the privatized companies groups.41 In addition, the public firms mean inefficiency measure is slightly inferior to the privatized firms measure42, indicating that the privatization process did not concentrate initially on the most efficient public firms.43 40 For example, if the public companies that turn out to be privatized in years 1998 to 2000 performed differently than the average public companies in that period, an an alysis that do not take these firms into account would compare average public firms scores for years 2001 to 2003 to distorted average public firms scores for years 1998 to 2000. 41 According to one-sided mean comparison tests performed, always private firms mean inefficiency in 1998 is significantly lower than both public firms mean ineffi ciency (p-value = .0291) an d privatized firms mean inefficiency (p-value = .0192). 42 The null of no difference in mean inefficiency of public and privatized firms was not rejected, even at the 10% level. The p-value of the alternative hypothesis that privatiz ed firms mean inefficiency is greater than public firms mean inefficiency was equal to .3517. 43 This is the same to say that the hypothesis that it is efficiency that leads to priv atization, and not the reverse (which Bagdadioglu, Waddams Price, and Weyman-Jones (1996) claim to have found some support for in the Turkish case), is not supported in the present context. Th is fact, however, does not rule out the possibility of endogeneity of privatization decisions a point that we address later.

PAGE 88

88 In line with what was mentioned before, the information in Table 3-5 shows that both the efficiency improvement occurred in the sect or and the identified catch-up effect come, essentially, from the performan ce of privatized firms, which experience an increase in mean efficiency of 5.58% from 1998 to 2003, associated with an expressive redu ction in dispersion of efficiency scores. Public firms mean efficiency increases only 0.35% in the same period, while the same measure for always private firms decreases almost 3%. We have identified, among the privatized comp anies, the ones that were privatized after June/1998, since their corresponding 1998 efficiency scores are free of any privatization effect. It is interesting to observe that these firms expe rience an average efficiency increment of 7.8% from 1998 to 200344 (Table 3-6), much higher than the obtained by the ones that remained publicly owned. Furthermore, it is worth noting that the decrease in always private mean efficiency in the period under exam comes, esse ntially, from the deteri oration on these firms performance from 2002 to 2003, confirmed by a one-s ided mean comparison test, which rejected the null hypothesis of equality of mean inefficiency scores on the two periods, at the 1% significance level.45 We will return to these points later. The rates of technological change and of efficiency ch ange, observed between two consecutive years, are combined to provide a measure of productivity change, given by the Malmquist indexes reported in Table 3-7. The re sults indicate that the Brazilian electricity 44 Some companies were privatized in 1999 and 2000, but their efficiency improvement in the period before the privatization was not representative. It corresponded to only 8.4% of the ab ove mentioned efficiency increment in the period of 1998 to 2003. 45 The same null hypothesis of equality of mean inefficiency scores (of always private firms) was also rejected for years 2000 and 2001, at the 5% significance level, and supported for all other consecutive periods.

PAGE 89

89 distribution industrys produc tivity increased 38.5%, on av erage, from 1998 to 2003.46 In addition, productivity growth ra tes are increasing th roughout the period and come, essentially, from frontier shifts (37%), with a relatively sma ll catch-up effect for the sector as a whole (1%). The computed mean productivity growth rate of 6.73% stands out when it is compared to the 0.9% mean total factor pr oductivity growth rate of the economy found by Gomes, Pessa, and Veloso (2003) for the period of 1992 to 2000.47 Moreover, taking the results found by Mota (2004) as a proxy for the distri bution companies average produc tivity gains in the period of 1994 to 1998 (around 5%)48, it might be concluded that the sect ors rate of productivity growth increased after 1998. The observed increment in the sectors pr oductivity comes from the performance of privatized and public firms, which experi ence productivity growths of 57.2% and 39.2%, respectively, from 1998 to 2003, against a produc tivity growth of only 16.1% of the always private firms (Table 3-8). These results, taken together with the ones previously mentioned, constitute evidence that the high productivity gain s observed in the period of 1998 to 2003 are associated to the closing of the efficiency gap present in 1998. Both the firms that were privatized and the ones that remained public ly owned have decreased the difference in performance with respect to the always private firms group, with privatized firms actually surpassing the always private firms efficiency level. 46 The above mentioned measure of productivity change does not incorporate the scale e ffect. On the basis of the computed output elasticity and the actual changes in output from year to year, we have estimated it to be equal to 3.56%, on average, from 1998 to 2003. 47 The authors report that their result is consistent with the findings of two other studies that examined the subject. 48 Mota obtains annual average productivity gains of about 5% in the period of 1994 to 2000 using data only from these two years and from 14 privatized companies. Considering that the present study provides evidence of significantly higher privatized firms productivity gains for the period of 1998 to 2000, Motas result can be taken as an upper-bound measure for average productivity gains in the period of 1994 to 1998.

PAGE 90

90 Note that public firms efficiency has i ndeed improved significantly from 1998 to 2003. The performance improvement, however, is not re flected in better efficiency indexes (public firms mean efficiency increase only 0.35% in th e same period), or in a positive catch-up effect (Table 3-8), due to the higher improvement of priv atized firms performance. Efficiency analysis is comparative and dynamic. Public firms have e ssentially kept the same distance to a frontier that has shifted considerably during the period, mainly in response to productivity gains of privatized companies. Looking at possible motivating factors for these findings, and in particul ar at privatization and the implementation of incentive regulation, it mu st be stressed, at first, that in the present study it is not possible to precisely assess the fr action of productivity improvement due to the adoption of the price-cap regime, and neither if th ere is effectively a causality relationship on this respect, given the lack of data from the peri od pre-reforms and the absence of a control group. Furthermore, it must be recognized that, despite the use of time fixed effects, other concurrent factors may have impacted the results, by aff ecting some firms and not others. Distribution companies with a high exposur e to financing in dollar49, mainly some of the privatized ones, might have had to cut costs and be more produc tive to compensate the higher expenses brought by the real devaluation in 1999, whereas some public companies may have had to do the same to put an end in the series of negative profits whic h became unsustainable in a context of increasing State budget constraints. However, no matter the specific reason that has induced each firm to expend some effort to improve its performance, it might be conjecture d that this action only took place because the 49 The real devaluation also impacted the amount distribution companies had to pay for the energy bought from Itaipu. However, given that these expenses were considered non-controllable costs and, consequently, entirely passed on to the tariffs, I do not consider them as an efficiency improvement inducing factor.

PAGE 91

91 price-cap regime provided the cond itions and incentives for it to happen, which would not be present in a rate-of-return regulation context.50 In this sense, the observed increase in firms productivity growth, over the mean productivity gr owth rate found by Mota (2004) for the period before 1998, could be associated with th e implementation of incentive regulation. This association is corrobor ated by the finding that productiv ity gains increase throughout the period, which is consistent with the progr essive implementation of incentive regulation in Brazil.51 The increasing productivity gains may also re sult from the fact that the labor force reductions implemented by many firms which bega n to operate under incentive regulation may actually increase short term labor costs. Labor fo rce reductions are typically achieved by paying employees to terminate their employment voluntarily, and the associated expenditures are recorded as short term costs (Kri del, Sappington, and Weisman, 1996). It should also be noted that most of those firms regarded as more inefficient at the beginning of the incentive regulation scheme have experienced very high productivity gains, in line with the regulators expectations. Given the similarity in the two firm groups efficiency levels in the beginning of the period, the perf ormance of firms which remained publicly owned can be taken as a proxy for the performance that privatized firms would have had if they had remained public. Consequently, the difference of almost 18% in productivity growth from 1998 to 2003, which is equivalent to a (geometric) me an productivity growth rate of 3.36% per year, constitutes a measure of the privatization effect. It also provides evidence that privatized firms have responded more aggressively than public firms to the efficiency improvement incentives 50 This assertion comes from the fact that in rate-of-return regulation revenues are linked to realized costs. Hence, in the usual situation of a one-year regula tory lag, it is expected that the regulated company will not have an incentive to make a costly effort to reduce costs because possible higher profits in the current period have to be balanced against lower revenues in all subsequent periods. 51 See footnote 9.

PAGE 92

92 brought by the price-cap regulati on, implying that incentives we re higher for profit-oriented managers operating under a shareholders pr essure to quickly recoup investments. The analysis, however, has to account for the possibility of endogeneity of privatization decisions. The mean comparison tests performed reve aled that privatized firms, on average, have a bigger size and a higher share of in dustrial customers than public firms52, which is consistent with the hypothesis that companies with a hi gher potential for performance improvement were the ones selected to be privat ized. To address the topic, a new performance comparison was implemented using a more homogeneous group of firms, given by the ones with Q higher than 400,000 KWh per year.53 With the new sample, although th e computed Malmquist indexes indicate a smaller productivity improvement for all firms over the period, compared to the one obtained previously (Table 3-8) the difference in productivity growth between privatized and public firms is even higher than the one found before (29% agai nst 18%). This result corroborates the conclusion that pr ivatized firms responded more a ggressively than public firms to the price-cap incentives. Conversely, the difference in pr oductivity growth rate of these two firm groups provides an indication of the additional improvement in perf ormance that public firms might have obtained. In light of the distribu tional consequences of these not-imp lemented-but-achievable productivity gains, in the sense that they could have resulted in lower tariffs to customers, the issue is of 52 Privatized and Public firms were compared in terms of any variable used in the estimation procedure. The null of equality of means was rejected for the variables Q, NumCust, Cap, and IndShare, with privatizeds mean showing up as greater than publics mean in all these cases. 53 Five out of the fifteen biggest distribution companies ar e publicly owned. Due to the new criteria, fifteen firms were dropped from the initial sample (5 public, 1 privatized and 9 always private). And when public and privatized firms were compared using data from the new sample, th e null of equality of means was rejected only for the variables LP (Public > Privatized) and IndShare (Public < Privatized). The two groups of firms do not differ in size anymore.

PAGE 93

93 particular relevance for the regulator, who n eeds to target the appropriate efficiency improvement incentives to this group of firms. The electricity regulator has ha d the opportunity to act on this matter in the periodic tariff review that started in April 2003. It is of interest, therefore, to evaluate the regulators decisions on this occasion, notably with respect to their consequences in terms of both distribution of productivity gains among stakeholders and incentiv es for further efficiency improvements. Such evaluation should not be restricted to the specific case of public firms, though, since its insights are equally valuable in the case of the other dist ribution companies. It is important to note what was done with respect to companies that have ex perienced lower productivity gains in the period, such as the always private ones, and, by virtue of its signaling effect to all concessionaries, whether privatized firms productivity gains in the period were completely passed on to the new tariffs.54 Always private firms were more efficient, on average, than th e other distribution companies in 1998, and have experienced, from 1998 to 2003, a productivity growth higher than the mean total factor produc tivity growth rate of the ec onomy found by Gomes, Pessa, and Veloso (2003) for the period of 1992 to 2000 (3.0 4% and 0.9%, respectively). Hence, their performance in the period might be taken as cons istent with what was expected under a price-cap regime in view of their initial efficiency lead over other distribution companies and of the fact that the efficiency analysis performed in th is study does not incorporate changes in capital expenditures.55 54 I am currently working on these points in another paper. 55 Under rate-of-return regulation, firms have an incentive to overinvest in capital, moving away from the optimal capital/labor ratio. Note that these firms will probably show up as relatively efficient if the performance analysis is based on operating expenses only, as is the case in the present study. In this context, when price-cap regulation is implemented, the prospect for efficien cy improvements in operating and maintenance expenses is limited.

PAGE 94

94 The results found for this group of firms might also be related to the f act that they operate at a lower scale, with lower poten tial for efficiency improvements.56 Note, however, that the observed decline of these firms mean efficiency levels57 derives from their low productivity growth in 2003, which is 23% less than the averag e of their productivity grow th in the four years before (Table 3-9). The deterioration in perfor mance might have been caused by the need to accomplish some regulatory requirements, such as the ones related to service quality (a point addressed in the next section). N onetheless, it cannot be ruled out a possible st rategic behavior of some of these firms, which could have inflated cost s near the time of the periodic tariff review in order to obtain higher repositioning indexes and, consequently, higher allowed revenues for the subsequent tariff period.58, 59 Service Quality and Economie s of Vertical Integration Two additional points are investigated. First, in view of the well-known concern that pricecap regulation may provide incentives to reductions in quality of service, it is examined if there is an association between effici ency and service quality in the Brazilian electricity distribution 56 When always private firms were compared to other firms, the results from the mean comp arison tests revealed that Always Privates mean values were significantly higher th an Public or Privatizeds mean values for the variables SP and Income. On the other hand, Public or Privatizeds mean valu es showed up as significantly higher than Always Privates for the variables Opex, Q, NumCust, LP, Area, Cap, Len, Undergrd, and CusDen. In essence, Always Private firms are of a considerable smaller size than other firms, and operate in smaller service areas. 57 As mentioned before, always private mean efficiency score decreases almost 3% in the period, with the most significant change occurring from 2002 to 2003, according to the results of the mean comparison tests performed. 58 Although firms own operating costs were not consider ed directly in the definition of the repositioning index, under the reference company approach adop ted by Aneel, a firm that artificially inflated its costs in the period near the tariff review would be acting in order to increase its ba rgaining power in the tariff review process. The rationale is that the probability of a firm re quest to increase its estimated effici ent operational costs be accepted by the regulatory agency increases with the difference between estimated and real (informed by the firms) costs. 59 This type of regulation game, associated with the periodic aspect of incentive-based regulation, is known to regulators and was reported in the survey conducted by Jamasb, Nillesen, and Pollitt (2003). Evidence of strategic behavior of the same sort was found by Di Tella and Dyck (2002), in a study of the Chilean electricity distribution utilities. The study reports U-shaped cost reductions associat ed with the introduction of price cap regulation, with strong initial cost reductions reversing every four years, coinciding with regulatory reviews.

PAGE 95

95 sector. 60 Beyond that, it is checked whether the variati ons in efficiency leve ls identified in the present study may result from ch anges in service quality. Secondly, given that some distribution companies in the sample also operate in generation and transmission, it is investigated the presence of economies of vertical integration and, as a corollary, th e possible occurrence of firms strategic behavior. Vertically-integrated companies generally e xperience economies of scope, which would help them be more efficient than other concessionaries.61 They also have the opportunity to behave strategically, classifying expenses incurr ed in their generation and transmission activities as costs pertaining to the re gulated distribution activity.62 In this context, a finding that vertically integrated firms are significantly more ineffici ent than other distribut ion companies, despite plausible economies of scope, would indica te that cost shiftin g may be occurring. The 14 companies that generated at least 10% of the electricity they delivered to final customers in 2003 are consider ed vertically integrated.63 Service quality is measured on the basis of two continuity of service indexes implemented by the regulator, DEC and FEC, associated with duration and frequency of service interruptions, respectively. While the DEC index 60 Actually, as it is stated by Ai, Martinez, and Sappingt on (2004), economic theory does not provide unequivocal predictions regarding the effects of in centive regulation on service quality. When authorized revenues are not tied directly to realized costs, a regulated firm may be temp ted to reduce service quality in order to reduce costs, and thereby increase profit. On the other hand, because in centive regulation can allow a firms revenue to rise substantially above realized costs, incentive regulation may motivate the firm to improve service quality in order to enhance the demand for its products and thereby increase revenue. Nonetheless, given that in the electricity distribution sector the demand is essentially exogenous to the firm, especially when there is no competition, it is possible to argue that the first effect should pr evail over the second in this specific context. 61 See Lowry and Kaufmann (1999), for details about economies of vertical integration in the power distribution sector. 62 The rationale for this behavior is pr ovided by Josk ow (2005). In light of the uncertainties the regulator faces about the firms inherent cost opportunities, the firm would like to convince the regulator that it is a higher cost firm than it actually is, in the hope that the regulator will then set higher prices for the services it provides as it satisfies the firms long run participation constraint, increasing the regulated firms profits, creating dead-weight losses from prices that are too high, and allowing the firm to capture surplus from consumers. 63 Among the vertically-integrated firms, six are publicly owned and five are privatized companies.

PAGE 96

96 measures the number of days of service interrupt ion within a period, the FEC measures the total number of service interruptions within a period. Note that service quality is inversely related to DEC and FEC values. The service quality variable ( Quality ), therefore, is gi ven by the inverse of the equally weighted average of th e continuity of service indexes.64, 65 A severe missing observation problem was identi fied with respect to DEC and FEC data for 1998. Thus, Quality measures for each firm in the sample were computed for the period of 1999 to 2003. The mean values of DEC and FE C indexes, as well as of the resulting Quality variable, are displayed in Table 3-10, which shows that the con tinuity of service indicators improved substantially over the period, resulting in an increase of almost 50% in the computed Quality variable from 1999 to 2003. The numbers portrayed provide an indication of the effectiveness of quality regulation instruments implemented by the regulator. The Quality decomposition by firm group reveals that the mentioned increase in service quality was driven by privatized and always private firms, whose Quality measures raised 54.6% and 51.6%, respectively. It also shows that th e lower service quality improvement experienced by public firms (35.7%) resulted in an increase in the distance of these firms to the others in terms of quality of service provided to customers. It must be stressed, however, that in light of the rather different operating conditions experien ced by these groups of firms, the continuity of service indicators incorporate fa ctors out of firms control and, consequently, cannot be taken 64 I recognize that some important dimensions of service quality are not incorporated in the Quality measure, such as those related to customer satisfaction, the speed with which reported problems are resolved and the intrinsic quality of the product offered to customers, an inverse function of variations in electricity tension levels. The Quality measure, however, captures the adequacy of system maintenance, the amount of human and capital resources allocated to network restoration and re pair, and the existing facilities to recupe rate the system after each interruption (vehicles, communication, qualifica tion of the work force, etc.). 65 In the Brazilian electricity distribution sector, quality regulation and enforcement have essentially been done through the definition of target DEC and FEC indexes for each year, associated with the imposition of penalties for non-accomplishment.

PAGE 97

97 conclusively to compare the resources allocated and the effort spent by each firm group in service quality provision. To address the point, it was assumed that the different operating conditions are incorporated in the target continuity of servi ce indexes set forth by the regulator. Ratios target over actual indexes were computed for each firm, and the resulting relative DEC and FEC measures were used to derive an equally weighted adjusted quality measure.66 The corresponding values found for each firm group in 2003 confirmed the service quality ranking shown in Table 3-10.67 Public firms do provide lower servi ce quality than other firms, a finding that rules out a possible explanation for their lo wer efficiency level evidenced in the present study. Always private firms, on the other hand, s how an adjusted quality measure that increases over the period and is well above unity in 2003, whic h denotes the provision of service quality at levels well above what has been required by the re gulator. It is investigat ed below whether this overinvestment in quality of service might be one of the reasons for the deterioration in performance over the period experienced by these firms. Two procedures are employed to examine the relationship between efficiency and service quality, overall and for each firm group, as we ll as the possible existence of economies of vertical integration. First, the Quality measure, interaction terms Quality*ownership dummies and a vertically-integrated dummy ( Vertical ) are included as additional arguments of the variable cost function, with the resulting efficiency scores being compared to the ones obtained 66 The adjusted quality measure is given by: DECrelative + FECrelative, where the relative measure equals Target (DEC or FEC) over Actual (DEC or FEC). Since service quality is inversely related to DEC and FEC values, the higher the relative measures, the higher the service quality. In addition, the higher is the computed adjusted quality measure from one, the greater is the distance from the actual service quality provided to the target defined by the regulator. 67 The computed adjusted quality measures for year 2003 were the following: 1.566, 1.731, and 2.183, for public, privatized, and always private firms, respectively.

PAGE 98

98 previously.68 The rationale here is to consider service quality as an additional output electricity distribution companies must provide in orde r to accomplish the regulatory requirements. Secondly, the same variables are included as parame ters of the mean ineffi ciency error term, in a conditional mean specification.69 The results obtained are reporte d in Table 3-11. Initially, the evidence tells that, after controlling for other factors, servi ce quality and the fact of being vertically integrated or not do not contribute significan tly to explain variations in opera ting costs among Brazilian electricity distribution companies (Model A). Both Vertical and Quality coefficients have signs contrary to what was expected and are not significant, wh ile all other coefficients remain practically the same. Likewise, the resulting effi ciency indexes are quite similar to the ones obtained previously (correlation of .9941). Note, however, that the se rvice quality impact on operating costs differs with firms ownership type (Model B). While costs increase exponentially with service quality levels for public firms, as it was expected, for privatized companies a negative relationship holds.70 Among the privatized, the ones that provide better service quality have lower operating costs.71 The conditional mean specification (Models C and D in Tabl e 3-11) reveals an unexpected positive, and marginally signifi cant (p-value = .055), coefficient on the Vertical 68 Since service quality is costly, it is expected that firms that provide better service quality are penalized when this variable is not accounted for in the analysis, by showing up as more inef ficient than the ones that provide lower service quality. Hence, the control for this factor should increase efficiency indexes of firms that provide higher quality. 69 The ownership dummies themselves were also incorporated in the specifications that included the interaction terms Quality*ownership dummies, to prevent attributing to service quality an effect due to the firms ownership type. 70 Although the coefficient on the interaction term lnqlt*alwspr is negative and significant, the computed elasticity of always private firms service quality was close to zero (-.0077). 71 In light of the longitudinal nature of the data, additiona l tests (random effects, between effects, and fixed effects regressions) were performed in order to ascertain that the negative relationship found between Quality and Opex comes from variations in Quality between firms.

PAGE 99

99 dummy. Despite probable economies of scope, vert ically integrated firms show up as more inefficient than other distribution companies. The point deserves the regulators attention, since cost shifting is one of its possible explanatory factors. The findings also show that qua lity of service providedas it is measured in the present studydoes not help explain variations in firm s efficiency levels. The coefficients on Quality and its squared term indicate that serv ice quality increases inefficiency when Quality is higher than 1.6, a value slightly higher than the 75% percentile (1.52). Here, again, the impact differs with firms ownership type (Model D). For public and always private firms the Quality coefficients are positive, but not significant, a result that does not confirm the raised possibility that the always private firms overinvestment in quality of se rvice was one of the reasons for their evidenced deterioration in performance. On the other hand, the findings do indicate that, among privatized companies, the more efficient on es also provide better quality of service. At this point, it is important to note that the evidence not only supports the effectiveness of quality regulation instruments implemented by the regulator, but also suggests that the expressive privatized companies efficiency im provement identified in the present study comes effectively from a more efficien t operation of their units, in line with what was expected under an incentive regulation scheme, a nd not from mere reductions in costs brought by deterioration in the quality of service provided. Conclusion This paper confirms the theoretical predictions regarding the impact of incentive regulation on firms performance. Brazili an electricity distribution co mpanies have experienced high productivity growth rates after th e sector reforms, above what was found in a previous study for the period before the reforms. The productivity incr ement relates to the closing of the efficiency gap present in 1998, and is driven by the perfor mance of privatized and public companies.

PAGE 100

100 Privatized firms responded more aggressively than public firms to the new incentives brought by price-cap regulation, de noting that incentives were highe r to profit-oriented managers operating under a shareholders pressure to qu ickly recoup the invest ments made. The studys estimate of privatized firms in cremental annual productivity grow th rate (3.36%), on the other hand, brings about the need to tailor specific efficiency improveme nt incentives to public firms, since it represents not implemen ted-but-achievable productivity ga ins, which could have resulted in lower tariffs to customers. The subset of firms privately owned before the reforms shows up as more efficient, on average, than other firms in the beginning of th e period examined. Its pr oductivity growth rate evidenced in the present study, therefore, is consistent with a limited space for efficiency improvementson operating and maintenance expens eson the more efficient firms subject to a rate of return regulation sc heme. Given the sensibly high er productivity growth rates experienced by other firms, always private firms face a decline in their efficiency levels over the period. This research provi des another possible explanation. It shows that the observed decline in these firms mean efficiency level derives, funda mentally, from their low productivity growth in 2003, which might therefore indicate a possible strategic behavi or of some of these firms, associated to the periodic aspect of the price-cap incentive regulation scheme. The results suggest a possible occurrence of stra tegic behavior of anot her sort as well. In spite of plausible economi es of scope, vertically integrated distribution companies show up as more inefficient than other firms, raising the possi bility of cost shifting. Stricter rules regarding cost allocation and/or a closer look at these companies accoun ting numbers may be appropriate. Interestingly, the paper reveals that the high performance improvement experienced by privatized firms in the period does not come from mere reductions in costs brought by

PAGE 101

101 deterioration in the quality of service provided, a result that also indicates the effectiveness of the quality regulation instruments implemented by the regulator. All these findings ultimately provide a better understanding of the cost opportunities faced by each firm, and consequently enable the establis hment of prices conducive to a greater social welfare. The regulator has had the opportunity to define new electricity di stribution prices in the periodic tariff review that started in 2003 and is still in place for some fi rms. On that opportunity, the choice was for the use of the model compa ny approach to estimate each firms efficient operational costs. This papers fi ndings provide the ba sis not only for evalua ting the regulators decisions in those circumstances, notably with re spect to their consequences in terms of both distribution of productivity gain s among stakeholders and incentives for further efficiency improvements, but also for examining the model company approach itself. The approachs usage is not pacific in the theory and its implementa tion in the Brazilian context has been disputed among the parties involved.

PAGE 102

102 Table 3-1. Descriptive Statistics Variable 1998 1999 2000 2001 2002 2003 1998-2003 Range OPEX 98,905 85,773 84,953 74,258 70,455 70,134 80,640 [2490, 559072] (132857) (111025) (113274) (100497) (97838) (97273) (108952) Q 5,074,129 5,260,394 5,520,603 4,790,657 5,063,016 5,110,973 5,137,639 [103191, 37540051] (8442352) (8346455) (8719154) (7649132) (7569014) (7404106) (7970465) LP 38.9052 32.4873 35.9164 34.0834 39.2159 41.9181 37.1144 [6.5398, 128.4681] (18.9536) (14.7348) (18.517) (14.7472) (22.3052) (21.2112) (18.7994) MP 78.6138 72.8605 70.946 70.9966 68.4405 68.595 71.701 [60.008, 96.620] (6.7104) (4.419) (3.9627) (3.5699) (3.2352) (3.4071) (5.5173) SP 74.0168 66.802 64.4161 61.4267 53.5491 58.3822 63.022 [29.434, 98.120] (17.9825) (18.4658) (16.9462) (14.435) (11.4705) (12.7854) (16.7229) CUSDEN 25.7095 26.6959 27.9056 28.7718 30.8484 32.0821 28.6965 [6.747, 137.093] (18.6995) (19.1257) (20.0373) (20.4782) (21.8955) (22.4005) (20.4544) INDSHARE 0.2959 0.2980 0.3068 0.3132 0.3308 0.3257 0.3119 [.0333, .6438] (0.1461) (0.1434) (0.1432) (0.1413) (0.1498) (0.1568) (0.1463) RESDEN 2.1026 2.0789 2.0028 1.7162 1.6803 1.6774 1.8749 [.663, 4.572] (0.6267) (0.6282) (0.5139) (0.4687) (0.4625) (0.4167) (0.5537) AREA 129,178 129,210 129,203 131,495 126,671 126,725 128,723 [252, 1253165] (242029) (239567) (239564) (241463) (237902) (237882) (237747) NUMCUST 828,166 879,502 919,894 934,543 979,891 1,012,766 926,545 [19625, 5744178] (1099440) (1134211) (1188028) (1228822) (1255942) (1287816) (1193257) INCOME 5,769.74 5,086.45 5,160.16 4,996.71 4,386.60 4,642.68 5,001.43 [1060.012, 12747] (2804.22) (2351.86) (2379.3) (2272.11) (1880.11) (1989.73) (2317.13) CAP 3,218.57 3,269.12 3,269.12 3,142.07 3,206.25 3,206.25 3,218.73 [.1, 22728.4] (4908) (4872.48) (4872.48) (4835.87) (4751.46) (4751.46) (4792.04) LEN 41,998.10 42,957.20 42,957.20 42,959.70 42,131.10 42,131.10 42,520.10 [720.3, 379518.58] (65700.6) (65399.9) (65399.9) (66063.9) (64894.5) (64894.5) (64850.3) UNDERGRD 0.006592 0.006462 0.006462 0.005940 0.006338 0.006338 0.006356 [0, .1391] (.0246) (.0244) (.0 244) (.0244) (. 0241) (.0241 ) (.0241) # OBSERV. 50 51 51 50 52 52 306 Mean values reported for each year and for the period 1998-2003. Standard deviation in parentheses.

PAGE 103

103 Table 3-2. Stochastic Cost Frontier Results Variable Time-trend formulations Time Fixed-Effects formulations A B C D E F ln Opex ln Q 0.771*** 0.781*** 0.803*** 0.708*** 0.739*** 0.756*** (.025) (.023) (.035) (.018) (.017) (.017) ln LP 0.442*** 0.403***0.409***0.395*** 0.348*** 0.366*** (.062) (.064) (.066) (.034) (.035) (.034) ln MP 0.364*** 0.374***0.315***0.381*** 0.401*** 0.480*** (.116) (.112) (.114) (.068) (.069) (.069) Ca p 0.096*** 0.108***0.058*0.103*** 0.102*** 0.080** (.027) (.027) (.034) (.028) (.030) (.028) L en 0.561*** 0.561***0.522***0.534*** 0.525*** 0.516*** (.064) (.063) (.064) (.063) (.066) (.063) ln I ndShare -0.007 0.009-0.013-0.008-0.003 0.009 (.034) (.033) (.035) (.033) (.034) (.033) ln R esDen 0.169* 0.157*0.1430.131 0.099 0.145* (.089) (.088) (.092) (.089) (.092) (.087) ln I ncome -0.179*** -0.145***-0.186***-0.168*** -0.139*** -0.185*** (.035) (.037) (.039) (.037) (.039) (.040) ln A rea 0.072*** 0.074***0.066***0.074*** 0.073*** 0.073*** (.012) (.011) (.013) (.012) (.013) (.012) ln CusDen 0.500*** 0.496***0.466***0.469*** 0.462*** 0.465*** (.061) (.061) (.064) (.062) (0.064) (.061) Under g rd 4.765*** 4.480***4.486***4.830*** 4.465*** 4.601*** (.589) (.582) (.563) (.602) (.611) (.583) T -0.054** -0.052*-0.051** (.027) (0.026) (.025) ln Q t -0.015*** -0.014***-0.007 (.005) (.005) (.005) ln LP t -0.012 -0.010-0.014 (.017) (.017) (.017) ln MP t 0.001 0.0010.047 (.029) (.027) (.030) Ts q -0.004 -0.004-0.005 (.007) (.007) (.007) Private -0.110** (.047) Private t 0.006 (.012) Privtzed -0.094* ( .052 ) A lws p riv -0.081 ( .087 ) Privtzed t -0.013 (0.013) Alws p riv* t 0.029 (0.018) D1999 -0.0390.049 -0.044 ( .033 ) ( .183 ) ( .032 ) D2000 -0.106*** 0.048 -0.102*** ( .033 ) ( .294 ) ( .033 ) D2001 -0.150*** 0.029 -0.147*** ( .034 ) ( .346 ) ( .033 ) D2002 -0.268*** -0.075 -0.273*** ( .035 ) ( .355 ) ( .036 ) D2003 -0.303*** -0.137 -0.308*** ( .036 ) ( .337 ) ( .035 ) Cons -0.150** 0.001-0.142-0.200*** -0.581 -0.437)*** ( .070 ) ( .052 ) ( .089 ) ( .052 ) ( .367 ) ( .105 ) Mu Private -0.100*** ( .027 ) T -0.140 ( .269 ) Times q 0.015 ( .032 ) Privtzed -0.155*** ( .032 ) Alws p riv 0.022 ( .036 ) Cons 0.649 0.359*** ( .577 ) ( .102 ) lnsig2u Q 0.109* 0.606-0.0630.131** ( .064 ) ( .761 ) ( .102 ) ( .051 ) Cons -4.423*** -15.104-3.590***-4.502*** ( 1.193 ) ( 14.702 ) ( 1.029 ) ( .956 ) N 306 306306306 306 306 Ll 128.429 136.29144.157120.077 124.98343 134.31818chi2 21833.984 23964.36817857.73621535.4928332.588 17974.398 Legend: p<0.10; ** p<0.05; *** p<0.01 Standard deviation in parenthesis. Coefficients on translog squared and interaction terms are omitted.

PAGE 104

104 Table 3-3. Elasticities Mean Std. Deviation Range At variables' median values Output 0.7077 0.0227 [ .6394 .7530 ] 0.7092 Labor Price 0.3908 0.0602 [ .2335 .5319 ] 0.3962 Materials' Price 0.3943 0.3246 [ -.2229 1.0662 ] 0.3911 Time -0.0655 0.0300 [ -.1331 -.0058 ] -0.0694 Table 3-4. Technological Cha nge by Year and Ownership 1998 1999 2000 2001 2002 2003 1998 2003 Geom. Mean Overall 5.31% 5.75% 6.35% 6.58% 7.39% 7.83% 46.2% 6.54% Public 5.62% 6.05% 6.66% 7.01% 7.75% 8.15% 49.0% 6.87% Private 5.17% 5.62% 6.21% 6.37% 7.23% 7.69% 44.9% 6.38% Privatized 7.37% 7.68% 8.36% 8.59% 9.37% 9.82% 63.4% 8.53% Always Private 2.46% 2.89% 3.43% 3.81% 4.48% 4.96% 24.1% 3.67%

PAGE 105

105 Table 3-5. Efficiency Evolution 1998 1999 2000 2001 2002 2003 Mean Inefficiency Score 1.1279 1.1206 1.1205 1.1133 1.1134 1.1109 (.0999) (.0917) (.0917) (.0722) (.0837) (.0697) Public Firms' Mean Inefficiency1 1.1384 1.1513 1.1499 1.1342 1.1533 1.1344 (.1161) (.1422) (.1378) (.1145) (.1401) (.1125) Private Firms' Mean Inefficiency 1.1229 1.1065 1.1071 1.1034 1.0956 1.1004 (.0928) (.0529) (.0583) (.0383) (.0272) (.0358) Privatized Firms' Mean Ineff.2 1.1516 1.1258 1.1240 1.1095 1.0962 1.0873 (.1148) (.0616) (.0693) (.0415) (.0268) (.0261) "Always Private" Firms' Mean Ineff.3 1.0866 1.0809 1.0847 1.0958 1.0948 1.1189 (.0285) (.0201) (.0281) (.0335) (.0287) (.0400) Mean Relative Inefficiency 1.0865 1.0738 1.0727 1.0645 1.0569 1.0629 (.0963) (.0879) (.0878) (.0690) (.0794) (.0666) Public Firms' Mean Rel. Ineff.1 1.0967 1.1032 1.1008 1.0845 1.0948 1.0854 (.1119) (.1363) (.1319) (.1094) (.1330) (.1077) Private Firms' Mean Rel. Ineff. 1.0818 1.0603 1.0599 1.0551 1.0400 1.0529 (.0894) (.0507) (.0558) (.0366) (.0258) (.0342) Privatized Firms' Mean Rel. Ineff.2 1.1094 1.0788 1.0760 1.0609 1.0406 1.0403 (.1106) (.0591) (.0663) (.0397) (.0254) (.0249) "Always Private" Firms' Mean Rel. Ineff.3 1.0467 1.0357 1.0384 1.0478 1.0393 1.0705 (.0275) (.0193) (.0269) (.0320) (.0273) (.0383) Standard deviation in parenthesis. 1. Includes only the 16 firms that remained publicly owned during the whole period 2. Firms privatized from 1995 to 2003 3. Firms already private in 1995 Table 3-6. Decomposition of Privat ized Firms' Efficiency Evolution 1998 1999 2000 2001 2002 2003 Mean. Ineff. Privatized Before1 1.1418 1.1175 1.1244 1.1105 1.0948 1.0872 (.1143) (.0649) (.0811) (.0410) (.0250) (.0287) Mean Ineff. Privatized After2 1.1790 1.1451 1.1229 1.1066 1.0988 1.0875 (.1249) (.0533) (.0338) (.0478) (.0320) (.0218) Mean Rel. Ineff. Privatized Before1 1.1000 1.0708 1.0764 1.0619 1.0393 1.0402 (.1101) (.0622) (.0776) (.0392) (.0237) (.0275) Mean Rel. Ineff. Privatized After2 1.1358 1.0973 1.0750 1.0582 1.0431 1.0405 (.1204) (.0511) (.0323) (.0457) (.0304) (.0209) Standard deviation in parenthesis. 1. Firms privatized from 1995 to June/1998 2. Firms privatized from June/1998 to 2003

PAGE 106

106 Table 3-7. Productivity Growth Rate and Decomposititon Malmquist Index Technological Chan ge Technical Effi ciency Change (Frontier Shift) (Catch-up) 1998 1999 6.04% 5.50% 0.51% 1999 2000 6.03% 6.05% -0.02% 2000 2001 6.94% 6.42% 0.46% 2001 2002 6.88% 6.92% -0.04% 2002 2003 7.77% 7.61% 0.11% 1998 2003 38.50% 37.00% 1.01% Geometric Mean 6.73% 6.50% 0.20% Table 3-8. Productivity Growth Rate and Decomposition by Ownership Type Malmquist Index Technological Chan ge Technical Effi ciency Change (Frontier Shift) (Catch-up) 1998 1999 Public 4.67% 5.83% -1.06% Private 6.69% 5.35% 1.24% Privatized 9.47% 7.45% 1.84% Always Private 3.18% 2.68% 0.49% 1999 2000 Public 6.43% 6.35% 0.08% Private 5.84% 5.91% -0.07% Privatized 8.13% 7.97% 0.14% Always Private 2.80% 3.16% -0.35% 2000 2001 Public 8.11% 6.83% 1.16% Private 6.40% 6.23% 0.12% Privatized 9.43% 8.28% 1.03% Always Private 2.56% 3.62% -1.03% 2001 2002 Public 5.62% 7.38% -1.60% Private 7.47% 6.70% 0.70% Privatized 10.03% 8.73% 1.20% Always Private 4.22% 4.14% 0.07% 2002 2003 Public 9.45% 7.95% 1.36% Private 7.03% 7.46% -0.44% Privatized 10.29% 9.42% 0.80% Always Private 2.45% 4.72% -2.18% 1998 2003 Public 39.24% 39.38% -0.10% Private 38.19% 35.90% 1.55% Privatized 57.19% 49.46% 5.10% Always Private 16.14% 19.70% -2.99% Geometric Mean Public 6.84% 6.87% -0.02% Private 6.68% 6.33% 0.31% Privatized 9.47% 8.37% 1.00% Always Private 3.04% 3.66% -0.60%

PAGE 107

107 Table 3-9. Productivity Growth Rate and Deco mposition by Ownership Type Firms with Q > 400,000 MWh/year Malmquist Index Technological Chan ge Technical Effi ciency Change (Frontier Shift) (Catch-up) 1998 1999 Overall -2.25% 1.31% -3.53% Public -8.29% 0.58% -8.76% Privatized 1.99% 1.26% 0.68% Always Private -3.92% 2.82% -6.55% 1999 2000 Overall 10.01% 4.39% 5.38% Public 9.94% 3.55% 6.15% Privatized 10.87% 4.43% 6.16% Always Private 7.40% 5.77% 1.52% 2000 2001 Overall 7.50% 7.17% 0.33% Public 10.08% 6.38% 3.41% Privatized 8.71% 7.27% 1.36% Always Private -0.83% 8.29% -8.38% 2001 2002 Overall 7.73% 9.81% -1.92% Public -0.58% 8.92% -8.70% Privatized 10.99% 10.06% 0.84% Always Private 13.17% 10.71% 2.22% 2002 2003 Overall 4.84% 12.47% -6.76% Public 6.09% 11.38% -4.69% Privatized 7.36% 12.80% -4.82% Always Private -5.85% 13.36% -16.99% 1998 2003 Overall 30.56% 39.98% -4.14% Public 17.06% 34.42% -12.84% Privatized 46.48% 40.84% 3.97% Always Private 9.04% 47.79% -26.26% Geometric Mean Overall 5.48% 6.96% -0.84% Public 3.20% 6.09% -2.71% Privatized 7.93% 7.09% 0.78% Always Private 1.75% 8.13% -5.91% Table 3-10. Mean Service Quality Indexes 1999 2000 2001 2002 2003 DEC1 23.538 20.090 17.783 18.761 16.724 FEC2 23.222 21.033 18.519 18.240 15.057 Quality3 0.062 0.074 0.079 0.074 0.092 Quality by ownership: Public 0.045 0.048 0.055 0.054 0.061 Privatized 0.060 0.072 0.079 0.077 0.093 Always Private 0.082 0.106 0.106 0.090 0.125 1. Average number of days of service interruption within a year 2. Average number of interruptions within a year 3. Inverse of the average of DEC and FEC

PAGE 108

108 Table 3-11. Vertical and Quality as additional regressors or as mean inefficiency parameters Variable A B C D Coeff. Std.Dev. Coeff. Std.Dev. Coeff. Std. Dev. Coeff. Std.Dev. Lnopex Lnq 0.708*** 0.022 0.757*** 0.019 0.727*** 0.020 0.766*** 0.018 Lnlp 0.369*** 0.040 0.394*** 0.039 0.369*** 0.040 0.352*** 0.037 lnmp 0.392*** 0.073 0.449*** 0.072 0.389*** 0.077 0.483*** 0.069 lnqxlnlp 0.037 0.035 0.026 0.031 0.016 0.035 0.030 0.030 lnqxlnmp -0.085* 0.046 -0.144*** 0.044 -0.095** 0.046 -0.126*** 0.040 lnlpxlnmp -0.182 0.138 -0.038 0.134 -0.145 0.143 0.003 0.133 lnqsq -0.013 0.017 0.016 0.013 0.018 0.014 0.024* 0.013 lnlpsq 0.019 0.127 -0.007 0.117 0.098 0.130 0.069 0.117 lnmpsq -0.576 0.460 -0.286 0.441 -0.535 0.477 -0.339 0.431 cap 0.072** 0.031 0.062** 0.030 0.052 0.032 0.044 0.029 len 0.581*** 0.073 0.550*** 0.068 0.545*** 0.074 0.541*** 0.070 d2000 -0.061* 0.032 -0.057** 0.029 -0.061* 0.032 -0.040 0.029 d2001 -0.096*** 0.036 -0.087*** 0.032 -0.101*** 0.036 -0.068** 0.032 d2002 -0.203*** 0.037 -0.211*** 0.034 -0.210*** 0.037 -0.188*** 0.034 d2003 -0.238*** 0.040 -0.233*** 0.036 -0.253*** 0.040 -0.223*** 0.036 lnind 0.004 0.039 -0.008 0.036 -0.008 0.039 -0.001 0.037 lnresden 0.178 0.109 0.157 0.099 0.142 0.111 0.170 0.106 lnincome -0.144*** 0.042 -0.141*** 0.043 -0.151*** 0.045 -0.177*** 0.043 lnarea 0.072*** 0.015 0.066*** 0.014 0.077*** 0.015 0.072*** 0.013 lncusden 0.501*** 0.070 0.482*** 0.064 0.467*** 0.071 0.501*** 0.067 undergrd 5.004*** 0.636 4.370*** 0.594 4.900*** 0.665 4.549*** 0.564 vertical 0.030 0.028 0.007 0.025 lnqlt -0.020 0.028 0.107*** 0.040 lnqltsq 0.027 0.024 0.058** 0.028 privtzed -0.155*** 0.028 alwspriv 0.012 0.040 lnqlt*privtz -0.283*** 0.057 lnqlt*alwspr -0.150** 0.060 Cons -0.275*** 0.052 -0.125*** 0.041 -0.485*** 0.136 -0.499*** 0.047 mu vertical 0.049 0.031 0.056* 0.029 quality -0.096 0.085 0.031 0.081 qltsq 0.030 0.026 -0.017 0.028 privtzed 0.064 0.065 alwspriv -0.006 0.053 qlt*privtz -0.060*** 0.020 qlt*alwspr 0.013 0.008 Cons 0.346** 0.143 0.280*** 0.063 lnsig2v Cons -4.082*** 0.291 -3.914*** 0.088 lnsig2u q 0.127** 0.055 1.553 1.700 Cons -4.312*** 0.880 -35.512 33.250 Statistics N 255 255 255 255 ll 111.86 137.21 108.17 142.37 chi2 18977.36 29613 23989.15 19865.29 legend: p<0.10; ** p<0.05; *** p<0.01

PAGE 109

109 CHAPTER 4 THE ASSESSMENT OF FIRMS EFFICIEN CY IN PERIODIC TARIFF REVIEWS: AN EVALUATION OF THE REFERENCE UTILITY APPROACH Introduction One of the main tasks in the implementation of a price-cap regime resides in the establishment of cost-based prices at the sche duled tariff reviews, where the regulator faces imperfect and asymmetric information regarding firms' cost opportunities. A social welfare maximizing regulator would face pressures from customers and utility investors, leading to decisions that are more likely to balance the c onflicting interests of pow erful stakeholders (so rulings are likely to reflect the political economy of regulat ion). Price-caps provide incentives for efficiency improvements:1 at the rate review, the regulator's intention to extract part of the firms' rents for the benefit of consumers and society ha s to be balanced against the objectives of promoting (1) allocative efficiency (prices that reflect minimum incremental costs), (2) financial sustainability (meeting each firm 's break-even constraint), and (3) further productivity gains (through strong in centives for cost containment). A form of yardstick regulation which has been used to tackle the cost benchmark issue consists of a bottom-up efficiency study base d on the engineering knowledge of the industry process.2 In the model company (or reference utility) approach, prices are se t on the basis of the estimated costs of a hypothetical efficient firm facing the same operating conditions of the concessionary under the review proc ess. As future prices are not linked to realized costs, the method has the merit of preserving the efficiency improvement incentives brought by the price1 Under a price-cap regime, prices are fixed. The firm and its managers are the residual claimants on production cost reductions, and bear the disutility of increased managerial effort. It is thus assumed that the conditions and incentives for efficiency improvement and for the possible achievement of second best prices are settled (Joskow, 2005). 2 The model company approach has been employed to calcu late electricity distribution tariffs in Spain and some Latin-American countries, mainly Chile, Peru, Arge ntina, El Salvador, and Brazil (Jadresic, 2002).

PAGE 110

110 cap regime.3, 4 Other possible advantages include the control for heterogeneity in operating conditions and the fact that the regulator does not need to base its decisions on cost information provided by firms.5 The approachs usage, however, is not fully endorsed in the literature. Weisman (2000) asserts that the estimation of efficient cost s is an untenable target, given the existing informational asymmetry between the regulated firm and the regulatory agency (or the consultants hired to perform the task), and argu es that it represents a major retrogression from the price-cap approach, as the posited efficiency gains need ha ve no foundation in actual market behavior. Gomez-Lobo and Vargas (2001), on their turn, claim that the method is excessively detailed, time-consuming, resource intensive and contributes negatively to the transparency and objectivity of the regulatory process. It is therefore important to investigate wh ether the use of the model company methodology has effectively enabled the attainment of the af orementioned regulators objectives. The evidence so far is limited. Serra (2002), and Fisher and Serra (2002) look at the experience in the telecommunications and electricity distribution sectors in Chile, and consider the findings that the methods usage led to rate of returns well above the firms cost of capital as an indication of a persistent regulatory flaw. Gr ifell-Tatj and Lovell (2001), on the other hand, examine the issue in the context of the elec tricity distribution in Spain and find that the engineering model was much less costly to operate than the real companies, by virtue of a smaller network and 3 When a price-cap plan links future prices directly to realized costs and the time between schedule reviews is relatively short, the incentives under a price-cap regime are similar to the ones under rate of return regulation (Sappington, 2002). 4 Under a model company approach, the efficiency impr ovement incentives come fr om the fact that firms appropriate rents when their actual costs are inferior to the esti mated efficient operating costs. 5 See Galetovic and Bustos (2002), and ANEEL (2003).

PAGE 111

111 lower input prices, but did not have their inputs allocated in a cost-efficient manner. The study indicates that actual companies we re more cost efficient than th e hypothetical efficient firms, and concludes that the engineering procedure had understat ed potential cost sa vings by nearly onethird. The present study contributes to fill the literature gap by an alyzing the results obtained with the use of the model company approach to estimate efficient operating expenditures in the Brazilian electricity dist ribution industry periodic tariff review. The implied performance scores are compared to those obtained using alternat ive methodologiesStatisti cal Frontier Analysis (SFA) and Data Envelopment Analysis (DEA)unde r the rationale that so me minor divergences might result from problems in these other met hods, but greater discrepancies could reflect deficiencies in the application of the engine ering approach, particul arly when they are independent of the method employed for comparison. The Brazilian case provides an exceptional opport unity to perform the investigation, as the number of distribution companies allows the use of sophisticated comparative efficiency techniques and the consequent computation of effi ciency scores and analysis of their evolution over time. Thus, efficiency estimates and m easures of firms productivity improvements obtained in a previous benchmar king study portraying the distributi on firms performance in the period of six years immediately before the tari ff review (Silva, 2006b) are employed to examine the results provided by the engineer ing approach. Particular atten tion is given to the degree of consistency in efficiency estimates and ranki ngs provided by the two methods, the procedure adopted for firms which experienced the highe stand the lowestprodu ctivity gains in the period before the review and to the possibility that the regulators decision might have threatened the firms financial sustainability.

PAGE 112

112 In sequence, the study checks fo r the possible causes of the divergences found. At this point, the investigation explicitly recognizes that regulatory d ecisions are taken by a utility maximizing regulator that operates in a situation of asymmetric information. It is also considered that the regulator has op portunities to exercise discretion, is potentially influenced by interest groups, and is subject to direct supervision of its actions. C onsequently, the analysis of regulatory outcomes addresses the possible impact of these factors, in addition to the effects of the methodology employed. The study presents evidence that the monitoring of the regulat ors activities does not lead to decisions contrary to the concessionaires inte rests. In addition, the investigation shows that the aforementioned regulators objectives at the ra te review might not have been accomplished in some situations where the firms efficiency assessments differed markedly from the ones suggested by the economic benchmarking approaches On the one hand, the results indicate that some firms, mainly the ones serving more a ffluent consumers, operating in more densely populated areas and having a lower proportion of el ectricity delivered to industrial customers, received substantially lower repositioning i ndexes than the economic benchmarking methods would recommend. As a low reposi tioning index basically serves as a price adjustment that reduces allowed revenues, the evidence points to a possible vi olation of firms break-even constraints. On the other hand, the findings rev eal that significantly hi gher repositioning indexes might have been given to companies with the oppos ite characteristics: managers of firms granted higher allowed prices can see th at cash flows are enhanced, even when efficiency (measured using other techniques) is not high compared with the performance of other firms. Usually, incentive systems give weaker-performing firms lo wer prices since there is scope for efficiency improvements. Some of the companies benefiti ng from ANEELs use of the repositioning index

PAGE 113

113 based on engineering models do not appear in the top ten of the economic benchmarking efficiency rankings, so weaker perf ormers seem to be rewarded. The following section describes the methodology adopted by the regulator in the periodic tariff review and presents the resulting figures obtained. Sect ion 3 explains the methodology and the data set employed to perform the stochastic frontier approach, pres ents the corresponding results, and explores their use to examine the regulators decisions ta ken on the basis of the engineering approach. Section 4 de scribes the econometric model a nd presents and interprets the results. Section 5 explains the robustness check performed with the use of the DEA methodology. The final section pr ovides concluding observations. Institutional Background and the ANEEL Model Company Method The power sector reforms in Brazil began in 1995. While constitutional amendments abolished the public monopoly over infrastructure industries and allowed foreign companies to bid for public concessions, the Law 8,987/95 (Gener al Law of Concessions) set the stage for the beginning of the privatization process, represen ted by the auctions of Escelsa in 1995 and Light in 1996. By the end of 2000, a total of 20 di stribution companies had been privatized. In addition, part of the implementation of a new regulatory framework involved the establishment of an independent regulatory agency (ANEEL) in late 1996 and, in the same year, the commission of an international consulta ncy to study and propose a new model for the electricity sector. The consu ltants report was released in 1997, and its proposals were incorporated into Law 9,648, issued on May of 1998.6 One of the measures introduced by the approved model was the use of the price-cap regime to regulate distributio n tariffs, replacing the previous cost-of-service system Price-cap regulation was implem ented through the signature of 6 See Ferreira (2000), Mota (2003), and de Oliveira (2003), for detailed descriptions of the new models characteristics.

PAGE 114

114 new concession contracts, which took place fr om 1998 to 2000, and scheduled the first tariff review for after five (for contra cts signed in 1998) or four years.7 As a result, 61 companies were submitted to a tariff review process from April 2003 to February 2006. The Tariff Review Methodology Despite defining the first regulat ory lag, the concession contract s were silent about the cost methodology that would be applied at the peri odic review. The oblig ation to solve the methodological vacuum rested with the regulat ory agency, which, after a long and heated discussion process,8 established that the re positioning index for firm i (RIi) would be calculated as follows: i i i i i iRD OR ER RS RR RI i = 1, n (4-1) where RRi corresponds to the revenue requirement for firm i in the 12-month period after the tariff review date, RSi stands for firm i 's revenue from supply activities, ERi and ORi denote extra-concession and ot her revenues of firm i projected for the same period, respectively, and RDi is the revenue from distri bution activities that firm i would obtain if dist ribution tariffs were kept the same. The repositioning index represented the percentage increase which would be applied to firms tariffs at the rate review. In this case, a higher revenue requirement implies a higher repositioning index and, consequently, higher prices. 7 Companies Escelsa and Light constituted an exception to this rule. Light was the first to have price-cap regulation applied, by order of the concession contract signed in November 1996, in which the first tariff review was scheduled to occur after seven years. Escelsa was submitted to price-cap regulation in August of 1998, and had tariff reviews every three years thereafter. Ex cept for Escelsa, all companies had the X factor set equal to zero in the first period prior to the first full review. 8 Peano (2005) provides a detailed description of the process implemented by the regulatory agency to define the tariff review methodology. Foster and Antmann (2004), and Byatt (2004), in turn, discuss the particularities of the deep controversy that surrounded the regulators choice of the asset valuation methodology.

PAGE 115

115 The revenue requirement was defined as the revenue needed to cover efficient operating costs and to provide an adequa te return over investments prude ntly made, corresponding to the following: T DEP ROC OC TC EC RR (4-2) where EC and TC are considered non-controllab le costs and stand for the projected costs of buying energy and paying the tariff charges, respec tively, OC is operating costs, ROC is the return on capital (in monetary units), DEP is depreciation, and T is the firms taxes. The return on capital was obtained through the ap plication of a rate of return of 17.07% on a rate base computed under a Depreciated Op timized Replacement Co st (DORC) methodology. The operating costs allowed to be passed on to ta riffs, on their turn, were given by the sum of the costs estimated for administration (ADM), commercialization (COM), and operation and maintenance (O&M) activities performed by a hypothetical effi cient firm facing the same operating conditions of the concessionary under ex am. The estimated rate base and efficient operating costs were, therefore, th e main determinants of the aut horized price increases at the rate review. The methodology employed to come up with the operating costs figures consisted in determining, for each firm, an optimal orga nizational structure which would allow the concessionary to efficiently fulfill its goal of ef fectively delivering electricity at the required service quality levels. In order to estimate the costs associated to COM and O&M activities, all processes and activities (P&A) which the referenc e utility should perform were identified, along with the human and equipment resources needed to carry out each P&A. The process/activitys cost was calculated considering the respective fr equency of occurrence, by valuing the required resources at market prices. Then, the final cost estimates resulted from the application of each

PAGE 116

116 P&A cost to the firms volume of capital (for O&M) and number of customers (for COM). The ADM costs, on their turn, were estimated on the ba sis of firms volume of capital, number of customers, and geographical dispersion, with the identified amount of human and equipment resources required also being valued at market prices. The tariff review process also included the definition of the X-factor for the time period for which the formula would be in effect. The analysis of the corresponding methodology employed is beyond the scope of the present st udy, but it is important to stress in this respect, that the Xfactor comprised the productivity gains estimated for the period up to the next tariff review ( Xe), adjusted by a quality factor ( Xc), with Xe being given by scale economies resulting from projected increases in output, ne t of investments required to meet expected increases in demand. The rationale, here, was that firms operating costs were already adjusted to their efficient levels (with the use of the engineeri ng approach), and no further si gnificant technical efficiency improvements should be observed thereafter. It is interesting, at this point, to look at the ANEELs methodology in the context of an efficient frontier framework. In theory, the effi cient operating cost prov ided by the engineering approach should correspond to th e point in the efficient frontier associated to the firm under exam. Thus, if the approach employed effectively enabled the regulator to figure out the firms efficiency targets, in spite of the informati on asymmetry, it follows that the methodology at the periodic tariff review not only determined the firms one-time adjustment on their operating costs, but also assumed that there would not be a ny frontier shifts in the future. In that case, the non-adoption of a progressive path towards the effi cient target raises con cerns over the financial sustainability constraint of thos e firms which the regulators a pproach revealed as the most inefficient. Note, here, that the situation gets worse if the model company resulting figure does

PAGE 117

117 not correspond to the firms efficiency target, and the target is set at a level that is excessively (and unreasonably) high. On this respect, it should be noted that th e estimated efficient ope rating costs might not match the true values. As stated before, the en gineering model is very detailed and provides estimates for each process/activity performed by a virtual efficient firm operating in the same conditions than the company under exam. Given th e asymmetric information context, the cost parameters needed to perform the task are not only difficult to be precisely estimated (taking into account the specificities of each firms operating conditions), but also are subject to firms misreporting. Consequently, it is indeed possibl e that the parameters employed to estimate the efficient costs do not satisfactoril y capture the effect of some co st drivers on firms actual costs.9 Model Company Estimates The analysis of the results provided by the engineering approach was limited by availability of data and by the decision to exclud e some very small utilities, which deliver less than 100,000 MWh per year. From the 61 companies subjected to a tariff review process from April 2003 to February 2006, nine were dropped from the sample due to small size; data for three others were unavailable. Therefore, the sample includes 49 companies, responsible for 99.24% of the total electricity delivered in the country in year 2003. The operating costs estimated for the hypothetica l efficient firms are shown in Table 4-1, along with the realized operati ng costs reported by the concessionaries and the computed regulators efficiency index ( ANEELEFF ), given by the ratio realized OPEX to estimated OPEX. Here, two points are worth noting: the wide range observed in the regulators efficiency indexes 9 This possibility was augmented in the Brazilian experience under exam, as the method and the corresponding parameters employed were used for the first time and not previously debated with the distribution companies before the beginning of the tariff review proce ss, increasing the ch ance of errors.

PAGE 118

118 and the fact that, under the regulators view, seve n firms were operating more efficiently than the virtual efficient company. The variable ANEELEFF varies in the range of 0.848 to 1.986, with mean 1.202. The fact that the 50% percentile is at 1.180 denotes that th e distribution of regulato rs efficiency indexes is skewed to the right, and the mean has been moved up by a few instances where the estimated efficient operating costs are well be low realized costs. It is useful to check whether these firms identified as highly inefficien t using this engineering Model Company methodology also show up in the worst performers grouping of SFA and DEA efficiency rankings. Note that the engineering me thod resulted in the allocation of some rents to the seven companies which had repositioning indexes based on operating costs higher than their realized costs.10 In such cases, the comparison with the re sults provided by the economic benchmarking approaches takes on special relevance because only the most efficient firms should be allowed to keep part of their productivity gain s, as an incentive for further efficiency improvements. If these seven firms are not ranked highl y by alternative methodologies, th e procedure is called into question. The finding would indicate that the higher tariffs given to these firms constituted an unjustified benefit, to the detriment of cust omers and without regard to the efficiency improvement incentives embedded in the price-cap regime. Comparative Efficiency Analysis In comparative efficiency studies, a firms e fficiency is given by a measure of the distance of the observed practice to the e fficient frontier, with the frontier estimation being implemented 10 According to the repositioning index formulae (equations 4-1 and 4-2), the allowed risk-adjusted rate of return of 17.07% would be given to firms operating at the model companys efficient operating costs levels (at the efficient frontier). It follows that returns below 17.07% were a ssigned to all firms whose estimated efficient costs were below their actual costs (ANEELEFF >1), with the rate of return being sma ller, the greater was the firms distance to the regulators estimated frontier (the greater ANEELEFF was from unity). On the other hand, returns above the 17.07% standard could be earned by the seven firms whose estimated efficient costs were higher than actual operating costs.

PAGE 119

119 with either a parametric or a non-parametric technique. Non-parametric methods, like Data Envelopment Analysis (DEA), use mathematical programming techniques and neither require the specification of production or cost functions nor the impos ition of behavioral assumptions. These methods are generally easy to implement, but carry an implicit restric tion in the number of variables that might be used. Furthermore, they do not allow for random shocks. Parametric methods, in turn, entail applying an a priori functional form to the frontier, estimated with econometric tools. They allow for hypothesis testing,11 enabling the analyst to investigate the validity of the model specification. Tests of significance can be performed for the functional form and for the inclus ion or exclusion of factors, wh ich are of special relevance for the electricity distribution indus try, where the inclusion of se veral factors is theoretically justifiable. Moreover, with a parametric method it is possible to allow for stochastic factors or measurement errors, which avoids the assumption that all deviations from the best practice frontier involve inefficiencies. For instance, with Stochastic Frontier Analysis (SFA) a mix of one-sided and two-sided error terms is employed, with the former capturing the firms inefficiency and the latter captu ring the effects of random variati on in the operating environment. Ideally, the decision regarding the appropria te method depends on the purposes of the study and the context under examina tion. In case, we are interested in investigating the evolution of efficiency from 1998 to 2003; the investigation is conducted in an environment where random shocks were present and the incl usion of several variables in the model specification, besides being theoretically ju stifiable, is advisable due to the gr eat heterogeneity in operating conditions. These considerations suggest the use of a stochastic frontier approach, defi ned in terms of an 11 In non-parametric models, a bootstrap technique may be used to produce confidence intervals around the estimated individual efficiency and thereby assess statistical properties of the efficiency scores generated (Simar and Wilson, 1998).

PAGE 120

120 input orientation, given the out put exogeneity that characterizes the electricity distribution industry. SFA Model and Data The SFA model employed here is detailed in Silva (2006b). It is based on an unbalanced panel of 52 companies, responsible for 99.47% of th e total electricity delive red in the country in year 2003, with the data bei ng collected for the period of 1998 to 2003. The model employs a variable cost specification, reflecting the fact that transformer capacity and network length constitute capital inputs that ar e fixed in the short run. Environmental variables are included as arguments of the variable cost function, instead of as mean inefficiency parameters, to control for differences in firms operating conditions; this appr oach reflects the interest in having efficiency measures net of factors that impact firms performance but are beyond the control of the concessionaries. In addition, in light of the rejection of the null hypothesis of homoskedasticity, the variance of the inefficiency error component is conditioned on a proxy of firm size, given by total electricity delivered ( Q ). The specification adopted is then the following: it it tt jn nit nt it yt t jit j i l i c nit it n yn n kit k nit nk it yy n nit n it y itu v t t w t y t Z Len Cap w y w w y w y E 2 2 02 1 ln ln ln ln ln ln ln ln ln 2 1 ln 2 1 ln ln ln (4-3) where E and y are the cost and output measures, respectively, w is the vector of factor prices, Cap stands for transformer capacity, Len represents network length, Z is the vector of environmental variables, and it is assumed that vit ~ N(0, 2 v) and uit ~ N+(0, 2 uit), with 2 uit specified as 2 uit = it QQ 0

PAGE 121

121 The modeling of technical change in the wa y shown in equation 4-3 attempts to obtain evidence of technological change over the period considered. Fo r the computation of firms efficiency indexes and the consequent analysis of efficiency change, however, the study turns to the use of time-fixed effects, to control for possible changes in macroeconomic factors which might have affected firms performance during th e period under investigati on. It is worth noting that the use of time-fixed effects explicitly allows the computation of efficiency indexes relative to yearly-specific frontiers. The observed technological change ( TC ) and technical efficiency change ( TE ) are then combined to provide a more complete picture of the productivity improvements which occurred in the period under examination, through the computation of Malm quist productivity indices. For each firm, the Malmquist index of productivity ch ange between two consecutive periods is given by j j jTC TE MI where 1 1 ., 1 t j t jIndex Eff Index Eff TE and 2 / 1 1 ,1 1 t j t jTC TC TC The dependent variable is given by the opera ting costs of distribut ion and retail service activities ( Opex ), computed as the sum of labor, mate rials and third party service contracts expenses, as reported in the income statement.12,13 Electricity delivered, in MWh ( Q ), is the 12 The computed labor expenses include firms contributions to pension funds and to health insurance plans, profit sharing payments, and management wages. Some firms alr eady report these expenses under the classification of labor expenses, but most of them do not. The necessary adjustments were made on these cases. 13 In case of vertically integrated companies, the computa tion of the operating costs of distribution and retail service activities was made possible by the fact that those companie s are required by law to report their expenses separated by activity.

PAGE 122

122 output measure14 and average wage, calculated as tota l labor expenditure divided by the number of employees, is used as a proxy for the price of labor ( LP ).15 For the prices of materials ( MP ) and third party services ( SP ), the work uses two price indexe s provided by Brazilian Institute of Statistics (IBGE), and the Caixa Economica Federal (CEF), a public financial institution that is in charge of most of the social programs of the federal government and provides financing to housing construction projects. The materials pric e index reflects the observed change in the price of a basket of items used in civil construc tion, by state, while the th ird party services index portrays the observed change in the salaries paid to an electr ician, also by state. The variables Opex LP MP and SP are expressed in 1998 values, being deflated by a general price index (IGP-DI). Transformer capacity is given in MVA, and network length corresponds to the sum of high-voltage and low-voltage lines, in kilometers.16 The environmental variables incorporated in the modeling are the follow ing: customer density ( CusDen ), given by number of customers divided by network length; share of electric ity delivered to industrial customers ( IndShare ); residential density ( ResDen ), computed as electricity delivered to residential customers divided by the number of residentia l customers; service area17 ( Area ), in Km2; ratio of underground to 14 The use of two measures of output was prevented by the fact that electricity deliver ed and number of customers showed up as highly collinear, with one of them always be ing dropped by the statistical software employed (Stata). A better specification was provided by the former, when compared to the latter. 15 Total labor expenditure is employed to compute average wage because it was not possible to obtain information related to number of employees segregat ed by sector activity, for the cases of firms that also operate on generation and transmission. 16 Since these variables showed up as highly correl ated with electricity delivered, the variables Cap and Len actually correspond to the residuals of the regression of transformer capacity on electricity delivered and network length on electricity delivered, respectively. 17 The Brazilian case justifies the inclusion of both networ k length and service area in the modeling, and this is reflected in the statistical si gnificance of both variables as either cost shifters or mean inefficiency parameters. While some companies have small service areas and relativel y high network length (the ones that operate in the more densely populated states), others have high Area but relatively low Len, because they operate in states which are more sparsely populated and/or have a high share of the population not being served.

PAGE 123

123 overhead lines ( Undergrd ); and income per capita, by state ( Income ), to control for variations in socio-economic conditions among states. The variables above are included among the most frequently cost driving factors employed to mode l electricity distribution, according to Jamasb and Pollitt (2001) in their survey of the empirical literature on comparative efficiency analysis. The data were assembled from the regulatory agency, the companies Web sites, the financial statements provided to the Sao Paulo Stock Exchange, the Brazilian Association of Electricity Distribution Compan ies (ABRADEE), IBGE, and CEF. SFA Results and Comparison Descriptive statistics are shown in Tabl e 4-2. The difference between minimum and maximum values of observations collected for almost all variables employed indicates the considerable heterogeneity among firms in the sa mple, in terms of companies size, system configuration, and customer structure. The evid enced disparity in firms indicators corroborates the need to account for external factors in the comparative efficiency analysis. Table 4-4 provides the results from the models estimated. The cost function satisfies the monotonicity condition with respec t to output and factor prices at the mean, and the estimated coefficients have the expected sign, with most of them being significant. The time elasticity provides a measure of technological change. The evidence shows that there was technological progress during the sample period, with an annua l rate of technological change of around 6.55%, on average, which denotes that the efficient fron tier has shifted considerab ly from 1998 to 2003. Firms efficiency indexes, computed for each year in the period examined, are portrayed in Appendix B, whereas the Malmquist measures of productivity change ar e reported in Appendix C. The results indicate that the Brazilian electricity distribution industrys productivity increased

PAGE 124

124 38.5%, on average, from 1998 to 2003.18 The mean productivity growth rate of 6.73% stands out when it is compared to the 0.9% mean total factor productivity growth rate of the economy found by Gomes, Pessa, and Veloso ( 2003) for the period of 1992 to 2000.19 Moreover, taking the results found by Mota (2004) as a proxy for the distribution companies average productivity gains in the period of 1994 to 1998 (around 5%),20 it might be concluded that the sectors rate of productivity growth in creased after 1998. The comparison between the SFA and the M odel Company efficiency indexes is limited to the 49 firms included in Table 4-1. Another rest riction comes from the fact that in some cases the indexes to be compared do not refer to the same period, since ANEELEFF relates to the month/year the tariff review ta kes place (April/2003 to Novemb er/2005) and our SFA estimates only go up to 2003. In the analysis that follows, the SFA efficiency indexes obtained for year 2003 are used for comparison ( SFA2003 ), which might introduce di stortions to assessments based on reviews that occurred in late 2004 and in 2005, if the firm performs rather differently than the others in th e period after December/2003.21 The possible distortions, however, should not be relevant in the context of the present study, as the comparison of efficiency indexes focuses on the larger discrepanc ies in the two methods results. 18 The above mentioned measure of productivity change does not incorporate the scale e ffect. On the basis of the computed output elasticity and the actual changes in output from year to year, we have estimated it to be equal to 3.56%, on average, from 1998 to 2003. 19 The authors report that their result is consistent with the findings of two other studies that examined the subject. 20 Mota obtains annual average productivity gains of about 5% in the period of 1994 to 2000 using data only from these two years and from 14 privatized companies. Considering that the present study provides evidence of significantly higher privatized firms productivity gains for the period of 1998 to 2000, Motas result can be taken as an upper-bound measure for average productivity gains in the period of 1994 to 1998. 21 Similar distortions may be present for the 17 companies which experienced tariff reviews along the 2003 year. We examined the issue using fo r the comparison the average SFA effici ency index for years 2002 and 2003 (SFA0203). As expected, SFA0203 and SFA2003 were highly correlated ( = .9674). Moreover, the points raised in the analysis that follows were still present when the SFA0 203 efficiency indexes were employed.

PAGE 125

125 The efficiency rankings provide d by the SFA and the engineeri ng approaches (Table 4-5) are not significantly correlated (Spearmans = 0.0682, p-value = 0.6417). The rankings show some consistency in terms of the best performers, as Enersul Coelce RGE and CAT-LEO appear in both top ten ex tracts. Nonetheless, only Eletropaulo and Jaguari appear in both models in the bottom ten. Note, in addition, that CEMIG figures in the bottom ten in one method, and in the top ten in the other. SFA efficiency estimates are significantly smaller than ANEELEFF,22 varying in the 1.045 to 1.506 interval but concentrated in the 1.045 to 1.127 (75% per centile) range, with mean 1.110. It follows that the engineering approach has cons idered firms to be, on average, more inefficient than indicated by the SFA economic benchmarki ng technique. The result is not unexpected, given that one method centers on an ideal context, while the ot her draws upon actual practice. Moreover, as the SFA method is based on strong distributional a ssumptions to disentangle the effects of inefficiency and random noise, it cannot be ruled out the possibility that some inefficiency is incorrectly at tributed to statistical noise.23 In the present study, however, the comparison of efficiency indexes is restri cted to the cases of higher divergence. According to the regulators methodology, Eletropaulo Light and CEB are considerably more inefficient than shown by the benchmarking method, as ANEELEFF exceeds SFA2003 by 0.8305, 0.7604, and 0.6312, respectively. Eletropaulo and Light however, were the two firms with the highest productivity improvements in the 1998-2003 period (Appendix C) and, according to SFA, were not distant from the averag e performance of other firms, which raises 22 The null of equality of means is rejected at the 1% significance level (p-value (H1: ANEELEFF > SFA2003) = .0007). 23 As suggested by the comparison to DEA efficiency measur es mentioned later on Section 5, it is also possible that SFA efficiency indexes are co nstrained by the half-normal distribution assumed for the inefficiency error term.

PAGE 126

126 serious concerns over their obligation to perform such profound further adjustments24 and points to the existence of flaws in the appl ication of the engineering approach.25 One possibility is underscored by the present study. In line with Peanos (2005) claim,26 the results suggest that the regulators method might have been biased ag ainst firms which operate in more densely populated areas, since five firms, from the ten which had the highest positive difference between ANEELEFF and SFA0203 belong to the top ten customer density extract. The major situations where the implem entation of the model company methodology resulted in firms being considered more effi cient than portrayed by SFA were the cases of Celesc Coelba and the seven firms which had ANEELEFF below one ( Energipe, Enersul, Coelce, Cemat, Cemig, Santa Maria and Cat-Leo ). The benefit of securing a higher rate of return over their asset base, given to these nine firms, would only be acceptable if they indeed figured in the group of best performers and had experi enced high productivity improvements in the first regulatory period, since in this case the regulator would be allowi ng them to keep part of the efficiency gains as an incentive for further productivity increments. Only Cat-Leo Enersul and Coelce however, show up in the SFA top ten segment. In the case of the other six firms, the SFA results indicate that the benefit given was probably unjustified and harmed customers through higher tari ffs. In addition, the rulings did not provide 24 As indicated by the Model Company efficiency index (1.986), Eletropaulo would have to further reduce its operating expenditures by almost 50% to be able to reap the allowed risk-adjusted rate of return (17.07%). 25 Our previous study of the regulato rs performance [Silva(2006a)] had alre ady indicated that the repositioning index announced at Eletropaulos tariff review caused a negative surprise in the market, as if it were not a conceivable number. 26 The author compared the model compan ys OPEX to actual OPEX of 12 firms submitted to tariff review in year 2003, and identified that the difference between the two meas ures was inversely related to firms customer density. The author argued that the model company method might have been efficient in avoiding reimbursements of overinvestments in more densely populated areas, and allowed an extra return to firms with less densely populated service areas, which incur higher costs to provide the service at the required quality levels. Peano noticed, however, that the positive X-factor given at the beginning of the ta riff review cycle might impact negatively the incentives for efficiency improvements.

PAGE 127

127 the appropriate incentives for efficiency impr ovements in the following regulatory term. The Cemig s case is emblematic. In spite of the cons iderable productivity increments in the 19982003 period, the firm still figured as the worst performer according to SFA2003 The fact that the engineering methodology shows the firm operating rather more efficiently ( ANEELEFF SFA2003 = -0.5522) may derive from the companys low customer density index, but suggests the possibility of a differentiated treatment to publicly owned firms, evaluated in the next section, since Celesc Celg CEEE and Copel also appear in the SFA2003 worst performers group but occupy considerably better pos itions in the Model Company ranking. It is also worth noting that f our out of the five best performers under SFA did not receive the same benefit of being able to keep part of the rents derived from accomplished productivity gains. On the contrary, the ANEELEFF indexes of two of them ( Manaus and Eletroacre ) were above 1.10, signaling that they would have to fu rther reduce their operating costs by more than 10% to be able to earn the allowed 17.07% rate of return. This apparently tough task, however, might not be difficult to achieve, as the regulator y scheme set forth for th e subsequent regulatory lag ignored the acute frontier shifts observ ed in the 1998-2003 period. The expected high productivity gains brought by tec hnological change should also al leviate the economic situation of firms which might have been unduly classified as more inefficient. Unfortunately, however, they will exacerbate the perverse effects of possible over-estimating firms efficiency. Econometric Modeling For each firm, the variable ANEELvsSFA computed as the ratio of ANEELEFF to SFA2003 expresses the divergence in the results provided by the two methods. ANEELvsSFA varies in the range of 0.633 to 1.719, with mean 1.087 (see Table 4-3). The higher the variable is (relative to unity), the more the firm was consid ered more inefficient under the regulators Model Company approach, when compared to the SFA standard, and the more the firm was harmed by

PAGE 128

128 getting a lower repositioning index (price), assuming that the SFA results are a good representation of the true values.27 Conversely, when the firms indicator is smaller (relative to unity), the firm was considered less ine fficient under the regulators methodology ( ANEELEFF< SFA2003 ). In this case, the result suggests that the regulators efficiency index was lower than it should be. Consequently, the firm, being closer to the efficient frontier than shown by the economic benchmarking method, was benefited by getting prices higher than recommended by the SFA standard.28 A ratio of .9, for example, indicates th at the firm was considered 10% closer to the efficient frontier than shown by the SFA mo del, or that its resul ting repositioning index was augmented by .1 times the operating costs (OC) participation on the estimated revenue requirement (see equation 4-2). Logically, the fir ms benefit gets higher for lower values of ANEELvsSFA At the lower bound ratio of 0.633, for instance, the multiplicative factor goes from the aforementioned .1 to nearly .37. Differences in the results provided by th e Model Company and economic benchmarking approaches are expected, as the engineering m odel does not account for substitution possibilities. Here, though, the analysis focuses on some ot her possible causes of the observed divergence between the two indicators. The investigation assumes the existence of a principal-agent relationship between the Congress (or the Government) and their de legated representatives in 27 This assumption draws upon the soundness of the parametr ic model employed, which controls for heterogeneity in operating conditions, influence of macroeconomic factors, and random shocks. The rationale, therefore, is that minor differences in efficiency assessments might be due to eventual SFA inconsistencies, but bigger divergences should be accounted to problems in the application of the engineering approach. Later in Section 5, in order to ascertain the validity of this reasoning, the results consistency is checked using performance indicators provided by an alternative methodology as th e comparison parameter. 28 The relation between the distance to the frontier and the resulting price assigned to the firm is detailed in footnote 10. Since ANEELEFF is given by the ratio of realized OPEX to the Model Companys estimated OPEX, it is important to stress, at this point, that when the firm shows up closer to the efficient frontier than it should be, it is because the Model Companys estimated operating costs are overestimated, acc ording to the SFA standard. Under the same reasoning, the engineering costs woul d be underestimated in the reverse case where ANEELEFF > 1. The analysis that follows investigates the possible causes of these overand underestimations of the Model Companys operating costs.

PAGE 129

129 regulatory agencies. In this system, interest groups can influence re gulatory outcomes. In addition, information asymmetries raise the possibi lity that the parameters employed to estimate the efficient costs do not satisfactorily capture th e effect of some cost drivers on firms actual expenditures, and affords the regulator some discretion to make choices that maximize its utility.29 The tariff review process presented two main opportunities for the regulator to exercise its judgment, possibly reflecting the influence of the industry and its customers: in the definition of the model company cost parameters and right af ter the announcement of the efficient operating cost initial estimates, when deci ding upon the acceptance or rejecti on of firms revision claims. Consequently, the investigation follows the line of previous empirical studies which focused on the determinants of regulatory outcomes, addre ssing the predictions of the economic theory of regulation associated with Stig ler (1971) and Peltzman (1976) [N elson, 1982; Primeaux, Filer, Herren, and Holas, 1984; Naughton, 1989, Nels on and Roberts, 1989; Klein and Sweeney, 1999]. The analysis, however, also accounts for the fact that the regulators decisions were taken in the context of incomplete and imperfect info rmation. As a result, the investigation not only examines the possibility of flaw s in the engineering cost parameters employed to estimate the efficient costs, but also hypothesi zes that, in the absence of the information necessary to promote 29 This framework draws upon the contributions of Stigler (1971) and Peltzman (1976), which form the basis of the economic theory of regulation. The authors posit that stakeholders face costs of organization and information, and regulators are self-interest maximizers wh ich allocate benefits across interest groups optimally, attempting to equate political support and opposition at the margin. The authors contribution helps explain the location of policy in the competitive price to the monopoly price spectrum. Generally speaking, consumers, being dispersed and having less at stake, face higher costs of organization than producers and usually do not have the required incentives to spend the necessary resources to become info rmed. In case, the prediction is that the producer interest should win the bidding for the services of a regulatory agency. However, consumers who spend a larger share of their income on a good have a higher incentive to participate in the regulatory process and should drop more votes for the politician in response to a price rise. Therefore, goods with a high shar e in the consumers budget ar e more likely to have prices close to the competitive price.

PAGE 130

130 the desired distribution of produc tivity gains among stakeholders,30 the regulator might have employed some of the available data as signals fo r firms profitability and cash flow availability. In addition, the empirical analysis examines a potential external monitoring impact on the regulators decisions. Hence, it is hypothesized that th e divergences of the regulators efficien cy measure to the SFA standard reflect two main f actors: (a) possible problems in the cost parameters employed by the engineering model, due to the difficulties pos ed by the imperfect information context or to regulators decisions; or (b) the regulators adju stments in the initial operating cost estimate, made on the basis of its utility function and the av ailable information, in a context of pressure from interest groups and direct supe rvision of the regulators job. The statistical tests are conducted through tw o complementary procedures: (a) an OLS regression of the divergence variable ( ANEELvsSFA ) on proxies for the explanatory factors mentioned above; and (b) an examination of th e possible determinants of the regulators adjustments in the OPEX estimate made during the rate setting, a more direct test for the influence of interest groups on the regulatory resu lts. In this case, the investigation employs the disclosed information regarding the initial OPEX estimated via the Model Company engineering model, the firms reported OPEX, and the final (adjusted) OPEX. These numbers are used to compute a measure of firms bargaining power, wh ich is then regressed on the political variables (and some other possible explanatory factors) These procedures are detailed below. 30 The Brazilian electricity sector regulator did not empl oy any kind of economic benchmarking procedure to estimate the productivity increments experienced by each firm during the regulatory period before the rate review.

PAGE 131

131 Specification and Data A measure of firms bargaining power can be expressed by the distance of the final OPEX to the corresponding initial estimat e or, in other words, by the we ight (W) given by the regulator to the initial OPEX estimated by the engineeri ng model. The weight is computed as follows: C E AOPEX W OPEX W OPEX ) 1 ( C E C AOPEX OPEX OPEX OPEX W where OPEXA stands for the final operating costs de fined by the regulatory agency (ANEEL), OPEXE stands for the operating costs initially esti mated with the application of the engineering model, and OPEXC is the operating costs reported by the distribution companies. These variables values, al ong with the corresponding WEIGHT computed for each firms tariff review, are portrayed in Table 4-1. Note that WEIGHT is inversely related to the bargaining effect, since the firms bargaining power can be said to be hi gher when the variable gets closer to zero. As the dependent variable is a fraction be tween zero and one, the estimation follows the procedure suggested by Papke a nd Wooldridge (1996) and employs a generalized linear model (GLM), estimated by the maximum likelihood me thod, assuming a binomial distribution for W and a logit link function. The modeling includes, as independent variables, proxies for the potential influence of interest groups, factors related to possi ble problems identified in the application of the engineering model, possible signals employed for firms profitability and cash flow availability, and proxies for a potential imp act caused by both the external monitoring of the regulators activities and a likely learning effect. These same independent variables are employe d in the OLS regression. In this case, though, an additional variable is included to control for a possible problem resulting from the labor price measure used in the SFA procedure.

PAGE 132

132 The explanatory variables are the following: income per capita ( INCOME ),31 share of electricity delivered to industrial customers ( INDSHARE ), the log of total electricity delivered, in GWh ( SIZE ), customer density ( CUSDEN ), given by the log of the number of customers divided by service area, a public company dummy ( PUBLIC ), percentage growth in residential consumption per capita from 1998 to 2003 ( CONSUMPTION ), output growth in the same period ( GROWTH ), a categorical variable indicating whether or not the Tribunal de Contas da Uniao has monitored, or not, the tariff review process ( TCU ), the number of rate reviews occurring before the firms review ( LEARNING ), and the ratio of SFA labor price to Model Company labor price ( LPDIFF ). The variables INCOME and INDSHARE are proxies for the consumers participation in the regulatory process. It is hypothesized that low-income re sidential customers should exert a higher opposition to a price increase, when comp ared to high-income customers. Since the income elasticity for electricity is less than one, poor families spend a greater share of their income on electricity and thus ha ve a greater incentive to oppos e high prices, assuming the time cost of political participation is proportional to income.32 As a result, income per capita should be negatively associated to both WEIGHT and ANEELvsSFA .33 For INDSHARE however, an opposite effect is expected, since a rise in th e share of electricity delivered to industrial customers should similarly lead to more oppositi on to high prices, as th e industry has a greater stake in lobbying for lower electri city prices than residential or commercial customers. Of 31 The state income per capita was employed for 24 companies in the sample (19 state monopolies plus five concessionaries which serve more than 90% of the consumers in the state) For the remaining 25 companies, INCOME corresponded to the weighted average income per capita of the ten biggest municipalities in the companys service area. 32 Knittel (2006), however, hypothesizes that wealth per capit a is positively correlated with the degree of residential interest group activity. 33 The lower the income, the closer should be W to one, denoting a lower firms bargaining power, and the lower should be the probability that ANEELEFF is smaller than SFA2003.

PAGE 133

133 course, the price structure (reflecting cost allo cation rules across customer categories) may be more important here than the general pri ce level, but the two are not unrelated. The SIZE variable is a proxy for the producers lobby, under the rationa le that larger companies should possess greater ab ility to influence regulatory decisions. Here, however, as suggested by Klein and Sweeney ( 1999), it is conjectured that the e xpected effect of firm size is indeterminate, as large utilities are more likely to receive careful scrutiny from the regulatory agency.34 The CUSDEN and PUBLIC variables were included in light of the points made earlier: the Model Companys results might have been biased against firms which operated in more densely populated areas, and may possibly ha ve favored publicly owned firms. It is thus expected that WEIGHT and ANEELvsSFA increases with CUSDEN and decreases with PUBLIC CONSUMPTION and GROWTH on their turn, represent pos sible signals employed by the regulator for firms profitabi lity and cash flow availability. The study tests the hypothesis that the regulator may have wanted to pass on to consumers some of the rents derived from economies of scale. Here, the expectation is that GROWTH contributes positively to both WEIGHT and ANEELvsSFA On the other hand, the residential consumption growth ( CONSUMPTION ) captures the rationing effect. Brazil e xperienced an unforeseen electricity crisis in 2001, which led to an energy rati oning in the period of June 2001 to February 2002.35 The rationing measures significan tly reduced the amount of electri city delivered and the average 34 The SIZE variable also controls for a possi ble endogeneity in the TCU monitoring, detailed below, since the closer supervision of the regulators activities during the tariff re view process concentrated in the larger firms cases. 35 The rationing aimed at a 20% reduction in energy consumption, and was implemented through a quota system where monthly energy consumption targets were established for almost all consumers (poor residential consumers were exempted). The scheme instituted penalties for non-accomplishments and bonuses for overachievements, besides allowing the trading of quotas for nonresidential consumers. The quota system met its objectives and avoided the occurrence of blackouts. Consumption levels from June to December 2001 showed a 20% load reduction, compared to the previous years consumption, and a 25% reduction if it is taken into account the new customers that entered the system in 2001 (Maurer, Pereira, and Rosenblat, 2005).

PAGE 134

134 residential consumption, bringing financial losses to concessionaires that operate in a sector where the high fixed costs cannot be adjusted (o r avoided) to compensate for the reduction in revenues. As the rationing effect differed among firms,36 the observed change in residential consumption per capita is used as a proxy for firm s losses and included in the model to test the hypothesis that the least affected firms (the ones with higher CONSUMPTION ) might have had relatively low price increases (higher WEIGHT and ANEELvsSFA ), under the rationale that the concessionaires which had lower reductions in their cash flows had lower bargaining power in the tariff review process and, c onsequently, were the ones most susceptible to rent extraction to the benefit of consumers. T hus, the expectation is that WEIGHT and ANEELvsSFA vary directly with CONSUMPTION A singular feature of the Brazili an regulatory environment consists of the existence of a governmental body (Tribunal de Contas da Uniao) which supervises the regulatory agencys performance.37 This study takes advantage of the fact that TCU closely mon itored only some of the periodic tariff review processes to examine wh ether this external monitoring has produced an 36 The differentiated effect among firms came mainly from the fact that the rationing measures varied among electric zones. The quota system was initially applied to the Northeast and Southeast/Center-West sub-markets only, with more stringent quotas being assigned to the former, when compared to the latter. Subsequently, the rationing was extended to part of the North region, encompassing softer rules than the ones applied previously. The rationing in that region was limited to the July to December 2001 period. The South regi on, in turn, was not included in the rationing. Compared to the same period of the previous year, energy consumption in the Southeast, Northeast, Center-West, North, and South sub-markets declined 31%, 28%, 25%, 10%, and 7% in the period of August to December 2001, respectively (Bardelin, 2004 ). As stated by Maurer, Pereira, and Rosenblat (200 5, p. 72), the South, despite not being forced to ration, engaged in the lo ad reduction effort as a result of appeals in the media and for fear of more drastic measures in the upcoming dry season. 37 TCU is an independent organ of the state, which assists in the external control that the Congress possesses over the whole public administration. The ag ency audits and reviews administrativ e decisions of the government to ascertain that all legal procedures and rules have been followed. TCU is not a court, but the current legislation attributes to the organ the power to order the review of some procedures undertaken and to impose sanctions and penalties in cases of strong infractions to the law. TCU ex ercises an oversight over the regulatory agencies and has examined both the procedures and substance of regulatory decisions.

PAGE 135

135 effect on regulators decisions.38 The intention is to shed light on the consequences of having an institution performing oversight of the regulatory agencys proce dures, and possibly contribute to the literature that focuses on the optimal institutional regulatory framework. The supervisions expected effect is indeterminate, though. One mi ght conjecture that the monitoring reduces the regulators discretion and leads to figures closer to the one s portrayed by SFA, assuming the SFA results are good representations of the true values. In this case, however, it is not possible to anticipate the effect on the weight measure. Noneth eless, it is also possible that the external monitoring has made the regulator exercise a higher scrutiny in the monitored cases, resulting in lower bargaining power and lo wer estimates of efficien t operational costs (higher WEIGHT and ANEELvsSFA ). The LEARNING variable is incorporated in the m odeling to avoid attributing to firms bargaining power (or lack of it) an effect due to improvements in the Model Company methods usage. This was the first time ANEEL employed the Model Company approach. It is thus reasonable to expect adjustments in the empl oyed engineering cost parameters as more rate reviews were carried out, re sulting in large changes in OPEXE at the first rate reviews, and progressively smaller changes at th e reviews conducted later on. The LEARNING variable should then be directly related to WEIGHT It is not possible, however, to anticipate the variables effect on ANEELvsSFA The continuing definition of the engineer ing cost parameters may either make the corresponding efficiency estimate converge to the economic benchmarking estimate, or not. In the computation of LPDIFF the Model Company labor price was given by total labor expenses estimated for the reference utility di vided by the corresponding number of estimated 38 In the specific cases it decided to cl osely monitor the tariff review process, TCU requested ANEEL to submit the correspondent technical notes and all other documents which supported the proposed repositioning indexes right after they were released, in order to ensure a concomitant supervision of the actions undertaken.

PAGE 136

136 employees.39 The variable is incorporated in the OL S regression only, to check whether the divergences in results provided by the two approaches are related to the fact that the SFA labor price was computed on the basis of firms actua l salaries and benefits paid, not accounting for possible inefficiencies brought by the payment of values above the market price. The higher the computed variable, the higher shou ld be the upward bias in the firms efficiency under SFA and, therefore, the higher should be the ANEELvsSFA measure. The data were assembled from the same sources employed to perform the SFA study. Summary statistics are shown in Table 4-3. The TCU monitori ng occurred in 12 out of the 49 tariff review processes examined, and 15 of the companies included in the sample are publicly owned. The average residentia l consumption growth of .3% portrays the rationing effect, whereas the computed WEIGHT varies in the interval of 0 to 0.998, with mean 0.641.40 As expected, WEIGHT is negatively and significantly (at the 1% level) correlated with the adjustment made in the initial OPEX estimate ( = -0.6371), expressed by the ratio OPEXA to OPEXE (see Table 4-1). 41 The adequacy of the variables usage as a measure of firms bargaining power is corroborated by its positive and significant correlation (at the 1% level) to the divergence measure ANEELvsSFA ( = 0.6324), a result that confirms the association between higher bargaining power (lower W ) and overvaluation of firms efficiency levels (lower ANEELvsSFA ), and vice-versa. 39 For each of the hypothesized reference utilities, total labor expenses were computed on the basis of market salaries especially gathered by a commissioned consultanc y firm, considering the service areas specificities. In addition, the estimated figure did not include some bene fits actually paid by some concessionaries (additional vacation in excess of the one-third disposed in the Constitution and profit sharing, for example). 40 In order to be able to run the GLM model, W was set equal to zero in the seven cases where the computed WEIGHT was negative, shown in Table 4-1. In these cases, OPEXA showed up higher than OPEXC essentially by virtue of revisions implemented by the regulator after the end of the standard tariff review procedure, under the claim of offering to firms which had reviews in the beginning of the periodic tariff review (year 2003) the same treatment (engineering cost parameters) given to firms which had reviews later on (years 2004 and 2005). 41 The higher the adjustment, the lower is W and the higher is the firms bargaining power.

PAGE 137

137 The variable, however, is not free of problems. There are some cases where W denotes a high bargaining power in spite of a small percentage change from OPEXE to OPEXA,42 and the data provides evidence that the mentioned adjust ments are related to the improvements in the engineering cost parameters through time as the average percentage change on OPEXE decreased steadily as more rate reviews were carri ed out, from 16.5% (for the first 12 firms in the periodic tariff review proces s) to 1.6% (for the last 13 firms). These facts endorse the inclusion of LEARNING as an additional explanatory vari able, and support a research strategy employing two investigation proced ures; the approaches employed in the present study can be viewed as complementary ways to address a complex topic. Results Analysis Regression results are remarkably consistent among the GLM and OLS procedures (Table 4-6). The evidence uncovers four main explanator y factors for the methodologies divergences in efficiency assessments. Initially, the positive and significant LEARNING coefficient on both GLM and OLS models suggests that the firms order in the periodic tariff review had implications for regulatory deci sions (and thus for the financia l well-being of companies). The first companies that went through the rate set ting experienced higher ch anges in their initial OPEX estimates and were benefite d by the adjustments in the engineering cost parameters made through time. When LEARNING decreases from the variables m ean value to a value equal to the mean minus one standard deviation, the predicted WEIGHT decreases from 0.6902 to 0.3621 (47.5%), and the divergence measure ANEELvsSFA falls 0.074 points (6.8% of the mean value). These results indicate that th e earlier-reviewed firm s ended up obtaining prices higher than 42 In these cases, OPEXE was not far from OPEXC. Thus, the final OPEX (OPEXA), despite representing only a small percentage increase on OPEXE, was very close to OPEXC, leading to a small W. See, for example, the cases of CATLEO, CELB, and CFLO in Table 4-1.

PAGE 138

138 recommended by the economic benchmarking met hod. Conversely, later-reviewed firms were considered more inefficient under the regulators approach, compared to the SFA standard, as if the regulator had become stricter after perf orming more and more rate-making processes. The results also indicate that the employed pr oxies for the consumers participation in regulatory decisions are statistica lly significant. In both cases, however, the evidence is contrary to some interpretations of interest group theory. The INDSHARE s negative coefficients indicate that firms with a higher proportion of electricit y delivered to industrial customers had a higher bargaining power in the rate review (lower W ) and were considered more efficient than shown by SFA, receiving higher prices. The estimated impact is higher for higher values of the variable.43 Moreover, when INDSHARE increases from the variables m ean value to a value equal to the mean plus one standard deviation, ANEELvsSFA falls 0.115 points (10.6% of the mean value). The findings, in case, suggest th at industrial demanders may have received higher prices than would have been approved under a SFA approach to benchmarking. The positive coefficients on the variable INCOME reveal that companies which serve more wealthy customers tend to have lower bargaining power in the tariff setting and were harmed by getting lower prices, i.e., prices seem to be lower (compared to the prices that would arise under benchmarking using SFA) when customer incomes are higher. A one-standard deviation increase in INCOME over its mean value shifts the predicted WEIGHT from 0.692 to 0.8151. Additionally, a one-unit and a one-standard deviation increase in INCOME augments ANEELvsSFA by 13.2% and 0.074 points (6.8% of the m ean value), respectively. The evidence, here, is consistent with the association between wealth per capita and th e degree of residential 43 As reported in Table 46, a one-unit increase in INDSHARE over its mean value decreases W by 7.4%.When the marginal effect is computed for a valu e one-standard deviation above the mean the result indicates that a one-unit increase in INDSHARE decreases W by 8.5%.

PAGE 139

139 interest group activity suggested by Knittel (2006), as if the variable were capturing the high discrepancy in the residential cu stomers (average) education leve ls across the different regions in Brazil, under the rationale that more educated customers face lower costs to become informed and participate in the tariff review process. The anticipated impact of customer density on the regulators results is confirmed as well. The results indicate that the more densely populat ed the service area, th e lower the companys bargaining power (here the coefficient estimate is only marginally significant) and the more harmed was the firm by receiving a repositioning index lower than the SFA benchmarking procedure would recommend. On the other hand companies operating in less densely populated areas were considered more efficient than shown by the SFA method and, consequently, benefited from higher prices. The estim ated impact is nontrivial, as the CUSDEN s change from the variables mean value to a value equal to the mean plus or minus one standard deviation shifts the ANEELvsSFA measure by 0.122 points, roughly 11. 2% of its mean. This finding corroborates Peanos (2005) claim that the regulator may have want ed to provide an extra return to firms serving less densely populated service ar eas. However, the result might also indicate a technical problem in the definition of the cost parameters employed in the engineering model, which overstated the costs incurred by firms operating under this condition. Some other results should be highlighted The aforementioned potential problem associated to the SFA labor price variable is not confirmed and neith er is the conjectured favorable treatment given to publicly owned firm s. In addition, although the regulator knew in advance which reviews would be closely monitored,44 the results indicate that supervision did not affect the types of decisi ons in a systematic way, sugge sting that ANEEL was consistent, 44 See footnote 38.

PAGE 140

140 regardless of specific oversight. In particular, the evidence does not support the hypothesis that the external monitoring may have ma de the regulatory agency be more strict in its analysis, to the detriment of the distribution companies. This finding is important, as it corroborates the view that the TCUs supervision of th e regulators activities does not increase firms regulatory risk. Finally, the positive and marginally significant coefficient found for the SIZE variable, in both GLM and OLS models, does not support the conj ectured producers in fluence on regulatory decisions, as it points to larger firms receiving prices lower than would have been approved under a SFA benchmarking. Here, the evidence s uggests that large uti lities possibly received greater scrutiny from the regulatory agency. Robustness Check: DEA Since the discussion above rests on the assumption that SFA estimates are good representations of the true effi ciency measures, the results r obustness is checked by performing the analysis with the use of efficiency es timates provided by an alternative benchmarking procedure. In case, the same dataset is employed to investigate firms efficiency levels and their evolution over time using a DE A technique. Here, the main concern was to use a specification which could control for exogenous features of th e operating environment and be comparable to the previous parametric modeling. The option was for the use of the approach proposed by Fried, Schmidt, and Yaisawarng (1999), based on a fou r-stage procedure to obtain measures of managerial inefficiency separated from the in fluence of external operating conditions. The first stage involves the calculation of an input-oriented DEA frontier under variable returns to scale (VRS), using electricity delivered ( Q ) as the output, and Opex Cap and Len as inputs. Specific DEA frontiers are computed for each year in the sample. Therefore, the procedure provides measures of the relative effi ciency of each firm in each period by reference to yearly-specific frontiers, as well as informa tion on input slacks and output surpluses of each

PAGE 141

141 observation. The efficiency scores obtained at th is stage, however, do not account for differences in the operating environment across production units. In a second stage, total input slacks are comput ed as the sum of radial plus non-radial input slacks of each observation, and expressed as percen tages of input quantities, as total slacks may depend upon external environment as well as unit size.45 The resulting total input slacks measures are then regressed on the six envi ronmental variables pr eviously mentioned ( CusDen, IndShare, ResDen, Area, Undergrd, and Income ), with the purpose of identifying the effect of external conditions on the excessi ve use of inputs. Given that i nput slacks are censored at zero by definition, three tobit regressions (one for each input) are estimated separately. More formally: ) (k j j k j j k ju Q f TIS, j = 1, N; k = 1, K where k jTIS is unit k s total radial plus no n-radial slack for input j based on the DEA results from stage 1, expressed as a percentage of actual input j quantity, k jQ is the vector of variables characterizing the operating environment for unit k that may affect the utilization of input j j is a vector of coefficients, and k ju is a disturbance term. In a third stage, the regressions estimated coe fficients are used to pr edict total input slack for each input and for each unit ba sed on its external variables. The predicted values represent the allowable slack, due to the operating environment. ) ( j k j j k jQ f IS T j = 1, N; k = 1, K 45 The definition of the total input slack measure in terms of percentage of actual input quantities is recommended by Fried et al. (1999, note 19) for the s ituation where the firm size differs significantly among firms in the sample.

PAGE 142

142 These predictions, in turn, are employed to ad just the primary input data for each unity according to the difference between maximum pr edicted slack and predicted slack, under the rationale of establishing a base equal to th e least favorable set of external conditions.46 k j k j k k j adj k jIS T IS T Max x x 1 In the final stage, the adjust ed input variables are employe d to re-run the initial inputoriented DEA VRS model, and generate efficiency scores for each firm in each period net of factors out of management control (Appendix B). In line with the procedure adopted before, the DEA efficiency measures obtained for year 2003 ( DEA2003 ) are used for comparison to the SFA and Model Company results. DEA efficiency estimates are significantly higher than SFA2003 ,47 varying in the range of 1 to 2.38, with mean 1.28. This fact, taken to gether with the evidenced similarity between DEA2003 and ANEELEFF distributions,48 suggests that either some in efficiency is attributed to statistical noise in the SFA approach, or the SFA efficiency indexes are constrained by the halfnormal distribution assumed for the inefficiency error term. On the other hand, even though DEA2003 and SFA2003 efficiency measures and rankings are not significantly correlated,49 there 46 With this procedure, a firm with external variables corresponding to this base level would not have its input vector adjusted at all, and a firm with external variables generating a lower level of predicted slack would have its input vector adjusted upward to put in on the same basis as the firm with the least favorable external environment. In other words, predicted slack below the maximum predicted slack is attributable to external conditions more favorable than the least favorable conditions prevailing in the sample for th at input. By increasing the input vector and leaving the output vector unchanged, the firms performance is purged of the external advantage (Fried et al., 1999). 47 The null of equality of means is rejected at the 1% significance level (p-value (H1: DEA2003 > SFA2003) = .0001). When the equality of DEA2003 and ANEELEFF means is tested, however, the null is not rejected. 48 Data on the respective mean and stan dard deviation are provided in Table 45. The difference in means is not statistically significant, as mentioned in the previous note, but DEA2003s distribution of effi ciency indexes is slightly more spread out than ANEELEFFs. 49 The correlation statistic and the Spearmans rank correlation amount to -0.0579 (p=.6927) and -0.0751 (p=.5966), respectively.

PAGE 143

143 is some consistency in terms of best and worst performers. Five firms appear in both top ten extracts, and four firms in both bottom ten (Table 4-5). The comparison between DEA2003 and ANEELEFF corroborates the indication that the Model Company approach understated some firms efficiency levels. Similarly to what was found in the comparison to SFA2003 ANEELEFF of firms Eletropaulo Light CEB Eletroacre Eletrocar, Piratininga Boavista and CPFL are considerably higher than DEA2003 in absolute terms. Additionally, some of the previously mentioned cases of overevaluations of firms efficiency are confirmed as well, as the model company efficiency indexes of firms Enersul Energipe Cemig Coelce Celesc Coelba and Cat-Leo are well below both SFA2003 and DEA2003 On this respect, the DEA findings provide additional support to th e indication that the benefit given to Energipe Celesc and Coelba was unjustified, since th ese firms do not belong to the DEA top ten segment either. The robustness check employs the same OLS model described in the previous section, with the difference that the depe ndent variable is now given by a new divergence measure ( ANEELvsDEA ), computed as the ratio of ANEELEFF to DEA2003 ANEELvsDEA varies in the range of 0.459 to 1.986, with mean 0.986 (Table 4-3), and is significantly (1% level) correlated with the preceding divergence measure ( = 0.7743). As reported in Table 4-6, the fitted model is not as well specified as before (smaller R2), but the Wald specification test still rejects the null that the coefficient estimates are all equal ze ro (p-value = 0.0006). The results confirm the previously noted effects of Industrial Share Income and Customer Density variables. The findings, however, do not support the Learning effect identified before, or the possible impact of the Size variable.

PAGE 144

144 Concluding Observations The present study examines the application of the Model Company approach in the Brazilian electricity distributi on sector periodic tariff review s (April/2003 to February/2006). The resulting firms efficiency measures are ev aluated with the use of efficiency estimates obtained from both a parametric and a non-parametric benchm arking model and indices of productivity changes experi enced in the six-year period before th e rate review. In the process, the study tests for possible causes of the identified divergences in efficiency assessments and checks for potential determinants of firms barg aining power in the rate setting process. Despite the criticisms made to its subj ectivity and complexity, the Model Company approach has become increasingly popular for the determination of electricity distribution tariffs in Latin America (Jadresic, 2002). It is theref ore important to verify whether the methodology has both provided an opportunity for firms to me et their break-even constraints and enabled the attainment of a welfare maximizing regulators ra te setting objectives: extracting part of the firms rents for the benefit of consumers and society, achieving allocative efficiency, and offering incentives for further productivity improvements. However, regulatory decisions are made by a regulator operating under information asymmetries, facing the influence of interest gr oups and, in the specific case examined here, subject to direct supervision of its actions. Thus, the analysis of regulatory outcomes addresses the possible impact of these factors, in addi tion to the effects of the methodology employed. The investigation reveals that the regulators objectives might not have been welfare maximizing in some situations. On the one hand, so me firms were considered to be rather more inefficient than shown by both SFA and DEA mode ls, resulting in substantially lower price increases: this result raises concerns over the co mpanies long-term financial sustainability. On the other hand, the results poi nt to the existence of firm s which the regulators method

PAGE 145

145 considered to be much more efficient than suggested by the two widely-used benchmarking methodologies. The study provides new findings on possible causes for these dive rgences in the context of a particular regulatory system. The results i ndicate that firms with a lower proportion of electricity delivered to industr ial customers, which serve wea lthier consumers and operate in more densely populated areas, had lower bargaining power in the tariff setting and were harmed by getting prices lower than recommended by the economic benchmarking methods. These results are consistent with the economic theory of regulation which posits th at political influence affects the level of prices. On the other hand, firms with opposite char acteristics had higher bargaining power and benefited from higher pr ices. The evidence is consistent with an association between per capita inco me and the effectiveness of residential interest group activity. Moreover, the findings point to a possible inaccuracy of the cost parameters employed in the engineering Model Company approach; the parame ters may inaccurately cap ture the effect of consumers dispersion (customer density) on firms operating costs, due to either the technical difficulty in defining the true parameter in a c ontext of imperfect and as ymmetric information, or a deliberate intention to avoid compensating investors in utilities opera ting in areas of higher consumer concentration, and to provide extra re turns to firms working in less densely populated areas (Peano, 2005). This benefit given at the beginning of the tariff review cycle impacted negatively the incentives for efficiency improvements provided to firms which do not appear in the top ten segments of SFA and DEA effici ency ranking. The same disincen tive was received by four of the top five firms in the SFA ranking, which c ould not keep part of the rents brought by their productivity improvements. In sum, the regu lators methodology imposed on firms a one-time

PAGE 146

146 adjustment to the virtual companys efficient ope rating costs, which in some cases were rather different than the ones estimated by the benchmarking methods. Moreover, the rulings (and associated price trajectories) i gnored the significant frontier sh ifts of almost 7% per year revealed by the parametric modeling, a point that would exacerbate the pe rverse effects of the hypothesized over-evaluations of firms efficiency. Interestingly, the findings do not provide suppor t to the hypothesis th at the monitoring of the regulators activities may lead to decisions co ntrary to firms interests and increase firms regulatory risk, one of the possibl e effects of having an institution supervise the regulators job. Regulators decisions were not affected in a systematic wa y by special oversight. Despite its specificity, the result adds to the literature on the optimal regulato ry framework design. It should be stressed that the results ou tlined above are robust to the choice of benchmarking methodology (SFA or DEA) to employ as a comparis on parameter. Moreover, the results do not support those who are concer ned with possible limitations of the SFA methodology. However, for those who are more he sitant to abandon engineering models, at a minimum the present investigation presents a way to promote greater transparency to the process and credibility for the results obtained with th e application of the Model Company method. Once the divergences in efficiency assessments are identified, and possible explanatory factors are uncovered, it remains the regulator s job to justify the choices made or demonstrate that the divergences do not come from deficiencies in the application of a particular methodology. The proposed joint use of a comparative effi ciency analysis technique benefits all stakeholders including the regulator; the agen cy could employ other benchmark techniques to alleviate potential adverse sele ction problems and consequently come up with more reliable approximations of firms break-eve n points. It would then be possi ble to better exploit the price-

PAGE 147

147 cap incentives for efficiency improvements50 and promote the desired allocation of productivity gains among stakeholders. Note, on this topic, that the redistribution of rents should ideally be based on information regarding the productivity increments of each firm during the whole regulatory period prior to the ne xt review, information that the model company approach alone cannot provide. The experience so far on setting price caps has indicated that qua ntitative benchmarking techniques may at least serve as an additiona l tool to the regulator, whose importance is underscored by information contained in comparis ons available from having a large number of companies in the regulated industry.51 Thus, there appears to be no reason for not using them in the Brazilian electricit y distribution industry. 50 The efficiency improvement incentives associated to the price-cap method are often hindered by the regulators uncertainties about firms inherent costs, which usually lead to tariffs set at a too high level, given the fear of violating firms financial sustainability constraint. Comparative efficiency analysis helps reduce the regulators informational disadvantage and enables the definition of be tter participation constraints, thereby allowing a more fruitful use of high-power ed incentive mechanisms. 51 The use of benchmarking techniques on regulatory price reviews is discussed by Rossi and Ruzzier (2000), Pollitt (2005), Stern (2005), and Dassler, Parker, and Saal (2006), among others.

PAGE 148

148 Table 4-1. Initially Estimated OPEX, Final OPEX, and Firms Reported OPEX COMPANY OPEXE (A) OPEXA (B) OPEXC (C) ANEELEFF1 WEIGHT2 ORDER3 (B) / (A) (A) / (C) AES-SUL 152,378,578 168,526,897 213,700,000 1.2680 0.7367 2 10.60% 71.30% BANDEIRANTE 178,748,147 200,857,495 254,995,034 1.2695 0.7100 6 12.37% 70.10% BOA VISTA 19,299,948 19,312,706 26,152,813 1.3542 0.9981 24 0.07% 73.80% BRAGANTINA 22,762,688 27,599,813 35,211,066 1.2758 0.6114 9 21.25% 64.65% CAIU 34,405,902 39,736,410 48,111,601 1.2108 0.6111 9 15.49% 71.51% CAT-LEO 69,812,982 73,173,281 73,047,000 0.9983 -0.0390 12 4.81% 95.57% CEAL 145,326,676 146,266,520 175,019,103 1.1966 0.9683 22 0.65% 83.03% CEB 127,222,668 145,601,583 257,359,512 1.7676 0.8588 16 14.45% 49.43% CEEE 231,085,795 235,718,790 266,328,472 1.1299 0.8685 17 2.00% 86.77% CELB 26,293,985 27,302,875 29,812,883 1.0919 0.7133 19 3.84% 88.20% CELESC 411,731,525 440,713,597 520,128,813 1.1802 0.7326 15 7.04% 79.16% CELG 455,583,137 483,121,893 580,123,409 1.2008 0.7789 23 6.04% 78.53% CELPA 229,717,734 269,031,550 310,500,443 1.1541 0.5133 5 17.11% 73.98% CELPE 374,778,267 379,210,684 476,485,325 1.2565 0.9564 20 1.18% 78.65% CELTINS 74,420,106 82,065,329 94,330,458 1.1495 0.6160 14 10.27% 78.89% CEMAR 212,939,949 217,204,197 258,866,541 1.1918 0.9072 22 2.00% 82.26% CEMAT 172,964,954 197,274,615 187,200,000 0.9489 -0.7077 1 14.05% 92.40% CEMIG 808,746,752 936,572,499 893,609,000 0.9541 -0.5063 1 15.81% 90.50% CENF 19,073,913 19,748,356 22,158,000 1.1220 0.7813 12 3.54% 86.08% CEPISA 140,241,902 141,016,014 160,151,652 1.1357 0.9611 22 0.55% 87.57% CERJ 278,164,419 297,502,578 391,983,902 1.3176 0.8301 8 6.95% 70.96% CERON 122,533,378 122,743,263 130,521,640 1.0634 0.9737 25 0.17% 93.88% CFLO 10,983,263 11,658,505 12,462,079 1.0689 0.5434 9 6.15% 88.13% COELBA 341,063,413 431,347,472 437,000,000 1.0131 0.0589 3 26.47% 78.05% COELCE 244,517,894 282,727,424 260,000,000 0.9196 -1.4680 3 15.63% 94.05% COPEL 588,545,532 606,611,885 688,548,640 1.1351 0.8193 13 3.07% 85.48% COSERN 97,792,392 113,400,305 136,100,000 1.2002 0.5926 3 15.96% 71.85% CPFL 328,589,815 421,760,792 549,100,000 1.3019 0.5775 1 28.35% 59.84% ELEKTRO 323,531,823 348,509,294 436,873,603 1.2535 0.7796 5 7.72% 74.06% ELETROPAULO 588,395,853 645,184,235 1,281,200,000 1.9858 0.9180 4 9.65% 45.93% MANAUS 87,650,951 87,948,585 102,481,654 1.1652 0.9799 24 0.34% 85.53% ENERGIPE 68,983,023 82,571,280 70,000,000 0.8478 -12.3614 3 19.70% 98.55% ENERSUL 112,343,069 130,154,623 113,300,000 0.8705 -17.6132 1 15.85% 99.16% ESCELSA 209,658,844 217,182,804 275,672,979 1.2693 0.8860 15 3.59% 76.05% LIGHT 463,351,823 516,334,111 944,760,674 1.8297 0.8899 7 11.43% 49.04% NACIONAL 19,052,515 22,337,700 29,134,454 1.3043 0.6742 9 17.24% 65.40% PIRATININGA 170,825,329 191,017,669 265,380,252 1.3893 0.7864 6 11.82% 64.37% RGE 157,117,648 170,367,818 174,089,900 1.0218 0.2193 2 8.43% 90.25% SAELPA 185,395,425 190,428,585 214,242,000 1.1251 0.8255 22 2.71% 86.54% SANTA CRUZ 40,012,199 44,081,288 48,887,827 1.1090 0.5415 9 10.17% 81.84% SANTA MARIA 16,498,783 19,771,653 19,299,425 0.9761 -0.1686 10 19.84% 85.49% V. PARANAP. 31,929,120 37,622,908 45,725,014 1.2154 0.5873 9 17.83% 69.83% CSPE 17,823,015 19,150,347 23,078,303 1.2051 0.7474 9 7.45% 77.23% DMEPC 16,750,348 17,466,270 19,635,723 1.1242 0.7519 13 4.27% 85.31% ELETROACRE 31,603,551 32,045,408 35,519,561 1.1084 0.8872 25 1.40% 88.98% ELETROCAR 11,932,292 11,958,308 16,844,993 1.4086 0.9947 21 0.22% 70.84% JAGUARI 10,677,128 11,157,355 15,783,997 1.4147 0.9060 9 4.50% 67.65% MOCOCA 12,246,733 13,167,147 16,895,193 1.2831 0.8020 9 7.52% 72.49% XANXER 10,165,730 11,223,622 12,416,298 1.1063 0.5299 15 10.41% 10.41% 1. Regulator's Efficiency Index: ratio OPEXC over OPEXA 2. The Weight shows how close the Final OPEX (OPEXA) is to the Engineering Estimated OPEX (OPEXE). 3. Firm's order in the tariff review process.

PAGE 149

149 Table 4-2. SFA Descriptive Statistics Variable 1998 1999 2000 2001 2002 2003 1998-2003 Range OPEX 98,905 85,773 84,953 74,258 70,455 70,134 80,640 [2490, 559072] (132857) (111025) (113274) (100497) (97838) (97273) (108952) Q 5,074,129 5,260,394 5,520,603 4,790,657 5,063,016 5,110,973 5,137,639 [103191, 37540051] (8442352) (8346455) (8719154) (7649132) (7569014) (7404106) (7970465) LP 38.9052 32.4873 35.9164 34.0834 39.2159 41.9181 37.1144 [6.5398, 128.4681] (18.9536) (14.7348) (18.517) (14.7472) (22.3052) (21.2112) (18.7994) MP 78.6138 72.8605 70.946 70.9966 68.4405 68.595 71.701 [60.008, 96.620] (6.7104) (4.419) (3.9627) (3.5699) (3.2352) (3.4071) (5.5173) SP 74.0168 66.802 64.4161 61.4267 53.5491 58.3822 63.022 [29.434, 98.120] (17.9825) (18.4658) (16.9462) (14.435) (11.4705) (12.7854) (16.7229) CUSDEN 25.7095 26.6959 27.9056 28.7718 30.8484 32.0821 28.6965 [6.747, 137.093] (18.6995) (19.1257) (20.0373) (20.4782) (21.8955) (22.4005) (20.4544) INDSHARE 0.2959 0.2980 0.3068 0.3132 0.3308 0.3257 0.3119 [.0333, .6438] (0.1461) (0.1434) (0.1432) (0.1413) (0.1498) (0.1568) (0.1463) RESDEN 2.1026 2.0789 2.0028 1.7162 1.6803 1.6774 1.8749 [.663, 4.572] (0.6267) (0.6282) (0.5139) (0.4687) (0.4625) (0.4167) (0.5537) AREA 129,178 129,210 129,203 131,495 126,671 126,725 128,723 [252, 1253165] (242029) (239567) (239564) (241463) (237902) (237882) (237747) NUMCUST 828,166 879,502 919,894 934,543 979,891 1,012,766 926,545 [19625, 5744178] (1099440) (1134211) (1188028) (1228822) (1255942) (1287816) (1193257) INCOME 5,769.74 5,086.45 5,160.16 4,996.71 4,386.60 4,642.68 5,001.43 [1060.012, 12747] (2804.22) (2351.86) (2379.3) (2272.11) (1880.11) (1989.73) (2317.13) CAP 3,218.57 3,269.12 3,269.12 3,142.07 3,206.25 3,206.25 3,218.73 [.1, 22728.4] (4908) (4872.48) (4872.48) (4835.87) (4751.46) (4751.46) (4792.04) LEN 41,998.10 42,957.20 42,957.20 42,959.70 42,131.10 42,131.10 42,520.10 [720.3, 379518.58] (65700.6) (65399.9) (65399.9) (66063.9) (64894.5) (64894.5) (64850.3) UNDERGRD 0.006592 0.006462 0.006462 0.005940 0.006338 0. 006338 0.006356 [0, .1391] (.0246) (.0244) (.0 244) (.0244) (. 0241) (.0241 ) (.0241) # OBSERV. 50 51 51 50 52 52 306 Mean values reported for each year and for the period 1998-2003. Standard devi ation in parentheses.

PAGE 150

150 Table 4-3. GLM and OLS Descriptive Statistics Continuous Variables Mean Std. Deviation R ange Categorical Variabl es Value Frequency WEIGHT 0.641 0.323 (0, 0.998) PUBLIC 0 34 ANEELVSSFA 1.087 0.202 (0.633, 1.719) 1 15 ANEELVSDEA 0.986 0.298 (0.459, 1.986) TCU 0 37 CUSDEN(LN) 2.825 1.511 (0.250, 7.019) 1 12 LPDIFF 0.978 0.414 (0.301, 2.474) INCOME (LN) 2.065 0.558 (0.856, 3.795) INDSHARE 0.322 0.158 (0.039, 0.644) SIZE (LN) 0.702 1.567 (-2.083, 3.491) CONSUMPTION -18.297 10.474 (-65.217, 6.820) GROWTH 12.194 13.607 (-22.525, 37.771) LEARNING 22.694 14.585 (0, 47)

PAGE 151

151 Table 4-4. Stochastic Cost Frontier Results Variable Time-trend formulations Time Fixed-Effects formulations A B C D E F Ln Opex Ln Q 0.771*** 0.781*** 0.803*** 0.708*** 0.739*** 0.756*** ( .025 ) ( .023 ) ( .035 ) ( .018 ) ( .017 ) ( .017 ) Ln LP 0.442*** 0.403*** 0.409*** 0.395*** 0.348*** 0.366*** ( .062 ) ( .064 ) ( .066 ) ( .034 ) ( .035 ) ( .034 ) ln MP 0.364*** 0.374*** 0.315*** 0.381*** 0.401*** 0.480*** ( .116 ) ( .112 ) ( .114 ) ( .068 ) ( .069 ) ( .069 ) Ca p 0.096*** 0.108*** 0.058* 0.103*** 0.102*** 0.080** ( .027 ) ( .027 ) ( .034 ) ( .028 ) ( .030 ) ( .028 ) L en 0.561*** 0.561*** 0.522*** 0.534*** 0.525*** 0.516*** ( .064 ) ( .063 ) ( .064 ) ( .063 ) ( .066 ) ( .063 ) ln IndShare -0.007 0.009 -0.013 -0.008 -0.003 0.009 ( .034 ) ( .033 ) ( .035 ) ( .033 ) ( .034 ) ( .033 ) ln R esDen 0.169* 0.157* 0.143 0.131 0.099 0.145* ( .089 ) ( .088 ) ( .092 ) ( .089 ) ( .092 ) ( .087 ) ln Income -0.179*** -0.145*** -0.186*** -0.168*** -0.139*** -0.185*** ( .035 ) ( .037 ) ( .039 ) ( .037 ) ( .039 ) ( .040 ) ln A rea 0.072*** 0.074*** 0.066*** 0.074*** 0.073*** 0.073*** ( .012 ) ( .011 ) ( .013 ) ( .012 ) ( .013 ) ( .012 ) ln CusDen 0.500*** 0.496*** 0.466*** 0.469*** 0.462*** 0.465*** ( .061 ) ( .061 ) ( .064 ) ( .062 ) ( 0.064 ) ( .061 ) Under g rd 4.765*** 4.480*** 4.486*** 4.830*** 4.465*** 4.601*** ( .589 ) ( .582 ) ( .563 ) ( .602 ) ( .611 ) ( .583 ) T -0.054** -0.052* -0.051** ( .027 ) ( 0.026 ) ( .025 ) ln Q *t -0.015*** -0.014*** -0.007 ( .005 ) ( .005 ) ( .005 ) ln LP *t -0.012 -0.010 -0.014 ( .017 ) ( .017 ) ( .017 ) ln MP *t 0.001 0.001 0.047 ( .029 ) ( .027 ) ( .030 ) Ts q -0.004 -0.004 -0.005 ( .007 ) ( .007 ) ( .007 ) Private -0.110** ( .047 ) Private *t 0.006 ( .012 ) Privtzed -0.094* ( .052 ) A lws p riv -0.081 ( .087 ) D1999 -0.039 0.049 -0.044 ( .033 ) ( .183 ) ( .032 ) D2000 -0.106*** 0.048 -0.102*** ( .033 ) ( .294 ) ( .033 ) D2001 -0.150*** 0.029 -0.147*** ( .034 ) ( .346 ) ( .033 ) D2002 -0.268*** -0.075 -0.273*** ( .035 ) ( .355 ) ( .036 ) D2003 -0.303*** -0.137 -0.308*** ( .036 ) ( .337 ) ( .035 ) Cons -0.150** 0.001 -0.142 -0.200*** -0.581 -0.437)*** ( .070 ) ( .052 ) ( .089 ) ( .052 ) ( .367 ) ( .105 ) lnsi g 2v Cons -3.964*** -3.734*** -4.242*** -3.898*** ( .307 ) ( .080 ) ( .486 ) ( .242 ) lnsi g 2u Q 0.109* 0.606 -0.063 0.131** ( .064 ) ( .761 ) ( .102 ) ( .051 ) Cons -4.423*** -15.104 -3.590*** -4.502*** (1.193) (14.702) (1.029) (.956) Statistics N 306 306306306 306 306Ll 128.429 136.29144.157120.077 124.98343 134.31818Chi2 21833.984 23964.36817857.73621535.4928332.588 17974.398 Legend: p<0.10; ** p<0.05; *** p<0.01 Sta ndard deviation in parenthe sis. Coefficients on translog squared and interaction terms are omitted.

PAGE 152

152 Table 4-5. Efficiency Rankings and Indexes Ranking ANEELEFF SFA2003 DEA2003 Company Eff. Index Company Eff. Index Company Eff. Index 1 ENERGIPE 0.848 RGE 1.045 AES-SUL 1.000 2 ENERSUL 0.871 CAT-LEO 1.057 CEMIG 1.000 3 COELCE 0.920 CELB 1.059 COELCE 1.000 4 CEMAT 0.949 ELETROACRE 1.060 CPFL 1.000 5 CEMIG 0.954 ELN/AM (MANAUS) 1.062 ELETROPAULO 1.000 6 SANTA MARIA 0.976 ENERSUL 1.065 ELN/AM (MANAUS) 1.000 7 CAT-LEO 0.998 COELCE 1.068 LIGHT 1.000 8 COELBA 1.013 COSERN 1.069 PIRATININGA 1.000 9 RGE 1.022 LIGHT 1.069 RGE 1.000 10 CERON 1.063 CEAL 1.070 ELETROACRE 1.000 11 CFLO 1.069 CEMAR 1.072 ELETROCAR 1.000 12 CELB 1.092 CENF 1.072 JAGUARI 1.000 13 XANXER 1.106 ESCELSA 1.074 DMEPC 1.024 14 ELETROACRE 1.108 BANDEIRANTE 2 1.074 MOCOCA 1.073 15 SANTA CRUZ 1.109 CSPE 1.077 BANDEIRANTE 2 1.096 16 CENF 1.122 BOA VISTA 1.080 XANXER 1.116 17 DMEPC 1.124 PIRATININGA 1.081 CELB 1.120 18 SAELPA 1.125 ELEKTRO 1.082 CFLO 1.138 19 CEEE 1.130 MOCOCA 1.083 COSERN 1.144 20 COPEL 1.135 COELBA 1.084 CELPA 1.153 21 CEPISA 1.136 SANTA CRUZ 1.088 CELESC 1.170 22 CELTINS 1.149 ELETROCAR 1.090 NACIONAL 1.171 23 CELPA 1.154 CELPE 1.090 ELEKTRO 1.183 24 ELN/AM (MANAUS) 1.165 CPFL 1.092 CSPE 1.206 25 CELESC 1.180 ENERGIPE 1.093 BOA VISTA 1.217 26 CEMAR 1.192 AES-SUL 1.093 ENERGIPE 1.221 27 CEAL 1.197 CEPISA 1.093 CELPE 1.233 28 COSERN 1.200 CERON 1.094 SANTA MARIA 1.233 29 CELG 1.201 CELTINS 1.095 CERON 1.258 30 CSPE 1.205 DMEPC 1.096 SAELPA 1.259 31 CAIU 1.211 SAELPA 1.097 CENF 1.266 32 V. PARANAPANEMA 1.215 CFLO 1.101 ESCELSA 1.274 33 ELEKTRO 1.254 SANTA MARIA 1.101 COELBA 1.284 34 CELPE 1.257 CEB 1.106 CEEE 1.337 35 AES-SUL 1.268 NACIONAL 1.113 CEMAR 1.340 36 ESCELSA 1.269 CERJ 1.115 CEB 1.379 37 BANDEIRANTE 2 1.270 CEMAT 1.127 CEMAT 1.437 38 BRAGANTINA 1.276 CELPA 1.129 CEAL 1.449 39 MOCOCA 1.283 COPEL 1.132 COPEL 1.499 40 CPFL 1.302 CEEE 1.133 CERJ 1.522 41 NACIONAL 1.304 CELG 1.140 CEPISA 1.577 42 CERJ 1.318 JAGUARI 1.147 BRAGANTINA 1.600 43 BOA VISTA 1.354 CAIU 1.148 ENERSUL 1.618 44 PIRATININGA 1.389 BRAGANTINA 1.151 V. PARANAPANEMA 1.647 45 ELETROCAR 1.409 ELETROPAULO 1.155 SANTA CRUZ 1.656 46 JAGUARI 1.415 V. PARANAPANEMA 1.179 CELG 1.692 47 CEB 1.768 XANXER 1.196 CAIU 1.733 48 LIGHT 1.830 CELESC 1.283 CAT-LEO 2.174 49 ELETROPAULO 1.986 CEMIG 1.506 CELTINS 2.381 Mean 1.202 Mean 1.110 Mean 1.283 Std. Deviation 0.217 Std. Deviation 0.072 Std. Deviation 0.302 25% Percentile 1.106 25% Percentile 1.074 25% Percentile 1.024 75% Percentile 1.270 75% Percentile 1.127 75% Percentile 1.437

PAGE 153

153 Table 4-6. Regression Results GLM OLS Variable Coefficient Marginal Effect ANEELvsSFA ANEELvsDEA Industrial share -3.466** -0.741** -0.728*** -0.561** (1.560) (0.339) (0.156) (0.271) Income (ln) 1.223*** 0.262** 0.132** 0.151* (0.468) (0.103) (0.050) (0.075) Size (ln) 0.313* 0.067 0.048* 0.039 (0.190) (0.041) (0.027) (0.028) Consumption 0.011 0.002 0.001 0.003 (0.018) (0.004) (0.002) (0.002) Growth 0.003 0.001 -0.001 0.001 (0.016) (0.003) (0.002) (0.003) Customer density (ln) 0.303* 0.065* 0.081*** 0.124*** (0.176) (0.038) (0.023) (0.040) TCU monitoring -0.694 -0.156 -0.058 -0.078 (0.560) (0.132) (0.067) (0.104) Public company 0.144 0.030 -0.075 0.078 (0.488) (0.102) (0.063) (0.085) Learning 0.094*** 0.020*** 0.005** -0.001 (0.020) (0.005) (0.002) (0.003) Labor price effect -0.029 (0.097) Intercept -3.515*** 0.756*** 0.533*** (1.232) (0.143) (0.172) Statistics N 49 49 49 Log pseudolikelihood -18.376 R2 0.634 0.522 Robust standard errors in parentheses Legend: p<.1; ** p<.05; *** p<.01

PAGE 154

154 CHAPTER 5 SUMMARY AND CONCLUSIONS This dissertation is composed of three empirical essays (chapters) on the reforms accomplished in the Brazilian electricity s ector. Initially, an event-study methodology is employed to examine the actual behavior of th e created autonomous regulatory body, vis a vis the predictions of the theories regarding the pattern of the govern ment intervention in business. In sequence, a parametric benchmarking approach is used to investigate whether the price-cap incentive mechanism effectively lead to performa nce improvements in the electricity distribution sector, checking for difference in performance between public and private firms, and looking at the possibility of efficiency catch-up. Then, the SFA efficiency estimates and indexes of productivity change, along with efficiency measures provided by a non-parametric benchmarking technique, are utilized to eval uate whether the use of the model company approach in the distribution comp anies first periodic tariff review enabled the attainment of the welfare maximizer regulators rate setting objectives. The event-study examines the regulators perfor mance in the Brazilian electricity sector using a methodology specially designed to a regulatory context, which explicitly accounts for the possibility of event anticipation. Despite th e more pronounced interests that characterize a developing countrys regulatory en vironment, the results indicate that the regulator has acted relatively independently, w ith its decisions not favoring a sing le interest group. The findings are similar to the ones obtained in previous studies that focused in the United Kingdoms context, and do not support the claim that Brazilian regula tory agencies are captured by the industry. On the contrary, the evidenced unpredictability of regulatory agencys decisions suggests that the regulatory agency has favored different interest groups at diffe rent times, supporting the claim that the utility maximizing regulator will not exclusively serve a single economic interest.

PAGE 155

155 The observed unpredictability, on the other hand, reinforces the need of improvements in the regulatory discussion process, with the adoption of measures to in crease the transparency and to promote more substantive public hearings. The study suggests that the need to provide incentives for new investments has had a significant role in the regulator y process. In addition, electricity companies have been compensated for the regulatory risk they face. The estimates indicate that regulatory decisions led to an increase in firms market value over the period examined and a ccount for part of the difference of the sample securities pe rformance with respect to the market. Some specific findings are also worth noting. The adoption of the a sset base repositioning cost methodology was not that harmful to distri bution companies, as one would anticipate in light of the press coverage rela ted to the issue, and the eviden ce raises the concern over the objectivity and transparency of the methodol ogy employed in the distribution companies periodic tariff review, suggesting the need of improvements. Moreove r, the results indicate that the Governments proposal to review the regulat ory agencies responsibi lities and performance was seen as a step back in th e electricity sector regulation and increased the regulatory risk. The second study confirms the th eoretical predictions regard ing the impact of incentive regulation on firms performance. Brazilian elec tricity distribution comp anies have experienced high productivity growth ra tes after the sector reforms, above what was found in a previous study for the period before the reforms. The producti vity increment relates to the closing of the efficiency gap present in 1998, and is driven by the performance of privatized and public companies. Privatized firms responded more aggressively than public firms to the new incentives brought by price-cap regulation, de noting that incentives were highe r to profit-oriented managers

PAGE 156

156 operating under a shareholders pressure to qu ickly recoup the invest ments made. The studys estimate of privatized firms in cremental annual productivity grow th rate (3.36%), on the other hand, brings about the need to tailor specific efficiency improveme nt incentives to public firms, since it represents not implemen ted-but-achievable productivity ga ins, which could have resulted in lower tariffs to customers. The subset of firms privately owned before the reforms shows up as more efficient, on average, than other firms in the beginning of th e period examined. Its pr oductivity growth rate evidenced in the present study, therefore, is consistent with a limited space for efficiency improvements on operating and maintenance expenses on the more efficien t firms subject to a rate of return regulation scheme. Given the sens ibly higher productivity growth rates experienced by other firms, always private fi rms face a decline in their efficiency levels over the period. This research provides another possible explanation. It shows that the observed decline in these firms mean efficiency level derives, fundament ally, from their low productivity growth in 2003, which might therefore indicate a possi ble strategic behavior of some of these firms, associated to the periodic aspect of the price-cap incentive regulation scheme. The results suggest a possible occurrence of stra tegic behavior of anot her sort as well. In spite of plausible economi es of scope, vertically integrated distribution companies show up as more inefficient than other firms, raising the possi bility of cost shifting. Stricter rules regarding cost allocation and/or a closer look at these companies accoun ting numbers may be appropriate. Interestingly, the study reveals that the high performance improvement experienced by privatized firms in the period does not come from mere reductions in costs brought by deterioration in the quality of service provided, a result that also indicates the effectiveness of the quality regulation instruments implemented by the regulator.

PAGE 157

157 All these findings ultimately provide a better understanding of the cost opportunities faced by each firm, and consequently enable the establis hment of prices conducive to a greater social welfare. The regulator has had the opportunity to define new electricity di stribution prices in the periodic tariff review that star ted in 2003. On that opportunity, the choice was for the use of the model company approach to estimate each firms efficient operational costs. This papers findings provide the basis not only for eval uating the regulators decisions in those circumstances, notably with respect to their consequences in terms of both distribution of productivity gains among stakeholders and incentiv es for further efficiency improvements, but also for examining the model company approach itsel f. The approachs usage is not pacific in the theory and its implementation in the Brazilian context has been dis puted among the parties involved. The performed evaluation of the results obtained with th e use of the model company approach reveals that the regulat ors objectives might not have been welfare maximizing in some situations. On the one hand, some firms were considered to be rather more inefficient than shown by both SFA and DEA models, resulting in substantia lly lower price increases: this result raises concerns over the companies l ong-term financial sustainability. On the other hand, the results point to the existence of firms which the regu lators method considered to be much more efficient than suggested by the two wi dely-used benchmarking methodologies. The study provides new findings on possible causes for these dive rgences in the context of a particular regulatory system. The results i ndicate that firms with a lower proportion of electricity delivered to industr ial customers, which serve wea lthier consumers and operate in more densely populated areas, had lower bargaining power in the tariff setting and were harmed

PAGE 158

158 by getting prices lower than recommended by the economic benchmarking methods. These results are consistent with the economic theory of regulation which posits th at political influence affects the level of prices. On the other hand, firms with opposite char acteristics had higher bargaining power and benefited from higher pr ices. The evidence is consistent with an association between per capita inco me and the effectiveness of residential interest group activity. Moreover, the findings point to a possible inaccuracy of the cost parameters employed in the engineering Model Company approach; the parame ters may inaccurately cap ture the effect of consumers dispersion (customer density) on firms operating costs, due to either the technical difficulty in defining the true parameter in a c ontext of imperfect and as ymmetric information, or a deliberate intention to avoid compensating investors in utilities opera ting in areas of higher consumer concentration, and to provide extra re turns to firms working in less densely populated areas. This benefit given at the beginning of the tariff review cycle impacted negatively the incentives for efficiency improvements provided to firms which do not appear in the top ten segments of SFA and DEA effici ency ranking. The same disincen tive was received by four of the top five firms in the SFA ranking, which c ould not keep part of the rents brought by their productivity improvements. In sum, the regu lators methodology imposed on firms a one-time adjustment to the virtual companys efficient ope rating costs, which in some cases were rather different than the ones estimated by the benchmarking methods. Moreover, the rulings (and associated price trajectories) i gnored the significant frontier sh ifts of almost 7% per year revealed by the parametric modeling, a point that would exacerbate the pe rverse effects of the hypothesized over-evaluations of firms efficiency.

PAGE 159

159 Interestingly, the findings do not provide suppor t to the hypothesis th at the monitoring of the regulators activities may lead to decisions co ntrary to firms interests and increase firms regulatory risk, one of the possibl e effects of having an institution supervise the regulators job. Regulators decisions were not affected in a systematic wa y by special oversight. Despite its specificity, the result adds to the literature on the optimal regulato ry framework design. It should be stressed that the results ou tlined above are robust to the choice of benchmarking methodology (SFA or DEA) to employ as a comparis on parameter. Moreover, the results do not support those who are concer ned with possible limitations of the SFA methodology. However, for those who are more he sitant to abandon engineering models, at a minimum the present investigation presents a way to promote greater transparency to the process and credibility for the results obtained with th e application of the Model Company method. Once the divergences in efficiency assessments are identified, and possible explanatory factors are uncovered, it remains the regulator s job to justify the choices made or demonstrate that the divergences do not come from deficiencies in the application of a particular methodology. The proposed joint use of a comparative effi ciency analysis technique benefits all stakeholders including the regulator; the agen cy could employ other benchmark techniques to alleviate potential adverse sele ction problems and consequently come up with more reliable approximations of firms break-eve n points. It would then be possi ble to better exploit the pricecap incentives for efficiency improvements and promote the desired allo cation of productivity gains among stakeholders. Note, on this topic, that the redistribution of rents should ideally be based on information regarding the productivity increments of each firm during the whole regulatory period prior to the ne xt review, information that the model company approach alone cannot provide.

PAGE 160

160 The experience so far on setting price caps has indicated that qua ntitative benchmarking techniques may at least serve as an additiona l tool to the regulator, whose importance is underscored by information contained in comparis ons available from having a large number of companies in the regulated industry. Thus, there appears to be no reason for not using them in the Brazilian electricit y distribution industry.

PAGE 161

APPENDIX A DETAILS ON EVENT STUDYS SAMPLE AND DATA

PAGE 162

162 Table A-1. Missing Observation Problem (N umber of computed stock returns missing) COPEL ELETR ELETBCEMIGCESP LIGHTCELESTRAC T EMAE FCATACOELCETRANPCERJ PAULFL Total missing 0 0 0 0 2 1 4 52 58 119 266 334 382 552 % missing 0.00 0.00 0.00 0.00 0.15 0.07 0.29 3.79 4.23 8.67 19.39 24.34 27.84 40.23 Missing 1998 0 0 0 0 0 1 2 52 13 38 46 196 6 2 Missing 1999 0 0 0 0 0 0 2 0 4 67 51 138 0 8 Missing 2000 0 0 0 0 2 0 0 0 10 8 46 0 55 34 Missing 2001 0 0 0 0 0 0 0 0 2 2 85 0 104 124 Missing 2002 0 0 0 0 0 0 0 0 27 0 24 0 157 226 Missing 2003 0 0 0 0 0 0 0 0 2 4 14 0 60 158 *1: COELCE was not included in the sample because it has a high nu mber of missing observations in all years examined. It was ch ecked the possibility of its inclusion in the regressions for year 2003, but the fact that this security has 38 observations missing in the 2002-2003 period was reducing significantly the total number of observations and preventing the analysis of the significance of some announcements.

PAGE 163

163Table A-2. Companies in the Sample COMPANY STOCK SECTOROWNERSHIP ENERGY DELIVERED TO FINAL CONSUMERS (GWh) PARTIC. IN DISTRIB. MARKET ENERGY GENER. (GWh) PARTIC. IN TOTAL ENERGY GENER. TOTAL ENERGY SOLD (GWH) ENERGY BILLED (R$1,000) COPEL PNB M Public 17,629.1 5.8% 16,825 4.6% 22,648.1 2,028,705 ELETROPAULO PN D Private 37,424.0 12.2% 37,424.0 4,732,541 ELETROBRAS PNB G*1 Public 22,911.7*2 7.5% 165,022*3 45.1% 211,889.4*4 12,222,302 CEMIG PN M Public 37,542.1 12.3% 32,561 8.9% 42,479.1 3,668,206 CESP PN G Public 2,122.7 0.7% 32,505 8.9% 31,526.7 1,416,175 LIGHT ON D Private 23,783.9 7.8% 4,144 1.1% 23,802.3 3,087,772 CELESC PNB D Public 12,006.3 3.9% 374 0.1% 12,203.6 1,233,600 TRACTEBEL ON G Private 21.9 18,605 5.1% 22.178,7 826,093 EMAE PN G Public 2,614 0.7% 3,689.3 167,985 FCATAGUAZES PNA D Private 1,005.9 0.3% 233 0.1% 1,036.7 132,745 TRANSM. PAULISTA PN T Public 650,148 CERJ ON D Private 7,325.9 2.4% 241 0.1% 7.656.5 1,085,935 Total 52.9% 74.7% Source: Setor Eletrico Ranking 2001 Cadernos de Infra-Estrutura BNDES. D stands for Distribution; G for Ge neration; T for Transmission; and M for mixed companie s (vertically integrated). *1: Although the energy delivered to final customers is very re presentative, it accounts for only 10.8% of the total energy sol d by the company. *2: Energy delivered by Chesf (7,546.3), Eletronorte (14,963.8) and Eletronuclear (401,6). *3: It does not include the energy from Eletronuclear and CGTEE. *4: It does not include the energy from Eletronuclear and CGTEE. It incorporates, however, the energy from Itaipu.

PAGE 164

164 APPENDIX B SFA AND DEA EFFICIENCY INDEXES SFA DEA COMPANY 1998 1999 2000200120022003199819992000 2001 20022003AES-SUL 1.062 1.056 1.076 1.078 1.101 1.093 1.000 1.000 1.000 1.000 1.000 1.000 BANDEIRANTE 1.378 1.205 1.146 1.000 1.000 1.000 BANDEIRANTE 2 1.077 1.074 1.000 1.096 BOA VISTA 1.071 1.101 1.057 1.046 1.085 1.080 1.238 1.381 1.333 1.196 1.205 1.217 BRAGANTINA 1.079 1.091 1.121 1.120 1.128 1.151 1.325 1.294 1.575 1.623 1.629 1.600 CAIU 1.101 1.107 1.106 1.149 1.132 1.148 1.397 1.364 1.517 1.639 1.692 1.733 CAT-LEO 1.073 1.066 1.078 1.075 1.066 1.057 1.869 1.582 1.873 2.024 1.946 2.174 CEAL 1.090 1.079 1.084 1.073 1.071 1.070 1.575 1.464 1.555 1.656 1.664 1.449 CEB 1.113 1.096 1.103 1.105 1.097 1.106 1.443 1.376 1.359 1.295 1.393 1.379 CEEE 1.154 1.121 1.113 1.161 1.174 1.133 1.520 1.447 1.451 1.372 1.397 1.337 CELB 1.110 1.094 1.083 1.065 1.063 1.059 1.460 1.357 1.361 1.481 1.418 1.120 CELESC 1.424 1.358 1.353 1.304 1.512 1.283 1.572 1.447 1.420 1.218 1.261 1.170 CELG 1.065 1.135 1.171 1.181 1.126 1.140 1.536 1.515 1.828 1.984 1.761 1.692 CELPA 1.092 1.165 1.176 1.188 1.151 1.129 1.412 1.333 1.239 1.136 1.217 1.153 CELPE 1.227 1.202 1.124 1.098 1.102 1.090 1.381 1.304 1.307 1.387 1.399 1.233 CELTINS 1.089 1.096 1.085 1.118 1.100 1.095 2.410 2.058 2.257 2.801 2.525 2.381 CEMAR 1.083 1.108 1.123 1.118 1.081 1.072 1.377 1.460 1.451 1.441 1.433 1.340 CEMAT 1.063 1.114 1.135 1.145 1.112 1.127 1.818 1.445 1.618 1.678 1.575 1.437 CEMIG 1.422 1.620 1.593 1.493 1.498 1.506 1.000 1.000 1.000 1.000 1.000 1.000 CENF 1.157 1.124 1.129 1.135 1.090 1.072 1.309 1.309 1.316 1.403 1.441 1.266 CEPISA 1.111 1.139 1.138 1.094 1.129 1.093 1.748 1.706 1.642 1.692 1.724 1.577 CERJ 1.144 1.126 1.116 1.125 1.126 1.115 1.437 1.342 1.534 1.490 1.570 1.522 CERON 1.111 1.118 1.130 1.089 1.095 1.094 1.486 1.406 1.451 1.178 1.277 1.258 CFLO 1.048 1.059 1.079 1.071 1.084 1.101 1.047 1.227 1.107 1.075 1.124 1.138 COELBA 1.104 1.075 1.091 1.121 1.080 1.084 1.486 1.109 1.383 1.709 1.439 1.284 COELCE 1.114 1.097 1.103 1.081 1.068 1.068 1.115 1.060 1.100 1.120 1.096 1.000 COPEL 1.144 1.131 1.148 1.057 1.142 1.132 1.453 1.192 1.330 1.064 1.302 1.499 COSERN 1.081 1.056 1.062 1.064 1.072 1.069 1.300 1.009 1.235 1.326 1.318 1.144 CPFL 1.216 1.185 1.132 1.127 1.129 1.092 1.073 1.000 1.127 1.000 1.004 1.000 ELEKTRO 1.122 1.109 1.081 1.070 1.082 1.174 1.164 1. 193 1.004 1.183 ELETROPAULO 1.489 1.289 1.390 1.210 1.131 1.155 1.000 1.000 1.000 1.000 1.000 1.000 MANAUS 1.051 1.098 1.058 1.074 1.061 1.062 1.000 1.160 1.136 1.000 1.000 1.000 ENERGIPE 1.091 1.070 1.098 1.111 1.112 1.093 1.342 1.062 1.316 1.404 1.387 1.221 ENERSUL 1.074 1.080 1.085 1.068 1.066 1.065 1.773 1.475 1.748 1.783 1.712 1.618 ESCELSA 1.108 1.135 1.114 1.095 1.078 1.074 1.208 1.074 1.215 1.295 1.171 1.274 LIGHT 1.233 1.179 1.148 1.131 1.105 1.069 1.000 1.000 1.055 1.000 1.000 1.000 NACIONAL 1.065 1.075 1.086 1.063 1.088 1.113 1.000 1.088 1.112 1.116 1.189 1.171 PIRATININGA 1.127 1.081 1.000 1.000 RGE 1.049 1.059 1.064 1.055 1.057 1.045 1.499 1.000 1.142 1.133 1.092 1.000 SAELPA 1.088 1.083 1.099 1.100 1.103 1.097 1.546 1.264 1.502 1.616 1.553 1.259 SANTA CRUZ 1.080 1.071 1.078 1.091 1.077 1.088 1.328 1.222 1.565 1.799 1.773 1.656 SANTA MARIA 1.091 1.081 1.080 1.103 1.083 1.101 1.182 1.067 1.182 1.282 1.224 1.233 V. PARANAP. 1.090 1.116 1.118 1.147 1.138 1.179 1.295 1.321 1.493 1.567 1.647 1.647 COCEL 1.115 1.087 1.083 1.092 1.100 1.132 1.000 1.000 1.000 1.000 1.000 1.000 CSPE 1.093 1.063 1.052 1.050 1.053 1.077 1.473 1.323 1.222 1.112 1.151 1.206 DMEPC 1.146 1.112 1.115 1.119 1.106 1.096 1.217 1.285 1.292 1.299 1.233 1.024 ELETROACRE 1.038 1.062 1.071 1.065 1.088 1.060 1.000 1.000 1.000 1.000 1.000 1.000 ELETROCAR 1.076 1.054 1.058 1.075 1.087 1.090 1.000 1.000 1.000 1.000 1.000 1.000 JAGUARI 1.135 1.081 1.068 1.091 1.099 1.147 1.000 1.000 1.000 1.000 1.000 1.000 MOCOCA 1.056 1.044 1.045 1.053 1.060 1.083 1.000 1.000 1.000 1.000 1.093 1.073 SULGIPE 1.047 1.066 1.071 1.096 1.103 1.146 1.000 1.181 1.235 1.292 1.321 1.471 CPEE 1.148 1.100 1.055 1.066 1.071 1.101 1.686 1.479 1.064 1.014 1.041 1.145 XANXER 1.103 1.098 1.148 1.143 1.141 1.196 1.000 1.052 1.202 1.172 1.131 1.116 M ean 1.128 1.121 1.121 1.113 1.113 1.111 1.327 1.243 1.314 1.341 1.318 1.279 Std Deviation 0.100 0.092 0.092 0.072 0.084 0.070 0.299 0.224 0.269 0.354 0.314 0.298

PAGE 165

165 APPENDIX C MALMQUIST TFP INDEXES COMPANY 1999/1998 2000/1999 2001/2000 2002/2001 2003/2002 2003/1998 A E S -SUL 8.48% 6.74% 8.45% 6.64% 10.27% 47.67% B ANDEIRANTE 23.83% 15.79% 43.39% B ANDEIRANTE 2 10.78% 10.78% B OA VISTA -0.44% 7.60% 5.24% 0.73% 5.80% 20.16% B RAGANTINA 2.32% 1.18% 4.77% 4.55% 3.71% 17.61% CAIU 3.30% 4.37% 0.63% 6.68% 4.36% 20.78% CA T L EO 5.25% 3.89% 5.64% 6.83% 7.46% 32.62% CEAL 6.88% 5.65% 7.59% 7.27% 7.97% 40.72% CEB 9.12% 7.23% 8.28% 9.83% 8.73% 51.31% CEEE 11.98% 10.12% 5.06% 8.85% 14.32% 61.21% CELB 5.51% 5.95% 7.00% 5.65% 6.27% 34.29% CELES C 13.76% 9.52% 13.50% -6.77% 28.56% 69.48% CELG 0.90% 5.12% 8.18% 14.80% 9.06% 43.65% CELPA 0.00% 6.25% 6.69% 11.63% 10.90% 40.32% CELPE 10.24% 15.50% 11.40% 8.97% 11.25% 71.96% CELTIN S 2.61% 4.84% 1.26% 6.67% 6.05% 23.23% CEMAR 3.60% 5.31% 7.83% 11.25% 9.02% 42.70% CEMA T 1.68% 4.97% 6.52% 11.00% 7.04% 35.08% CEMIG -4.75% 12.86% 18.56% 11.59% 11.91% 59.16% CEN F 5.56% 2.47% 2.73% 7.92% 6.17% 27.31% CEPISA 3.00% 6.23% 10.55% 3.60% 11.25% 39.42% CERJ 9.74% 9.64% 8.16% 9.28% 11.00% 57.86% CERO N 5.19% 4.97% 10.43% 6.60% 7.83% 40.16% CFLO 1.00% 0.76% 3.90% 2.44% 2.64% 11.17% COELBA 11.03% 7.22% 6.18% 13.60% 9.89% 57.77% COELCE 9.61% 7.74% 11.11% 10.62% 9.98% 59.63% COPEL 10.56% 8.11% 19.19% 2.18% 12.59% 63.90% COSER N 9.47% 6.72% 7.51% 7.48% 9.19% 47.40% CPFL 12.26% 14.95% 11.01% 10.75% 15.22% 82.79% ELEKTRO 10.16% 12.03% 10.81% 8.90% 48.91% ELETROPAULO 25.34% 2.29% 26.20% 19.81% 10.63% 114.45% ELN/AM ( MANAUS ) 1.37% 11.13% 6.16% 9.46% 8.89% 42.55% ENERGIPE 8.40% 3.84% 5.76% 7.43% 9.84% 40.48% ENERSUL 6.10% 6.87% 9.36% 8.47% 9.01% 46.63% ESCELSA 5.43% 10.50% 10.62% 11.05% 10.24% 57.77% L IGH T 14.44% 13.09% 12.34% 13.60% 15.04% 90.02% N ACIONAL 1.98% 2.46% 6.13% 1.93% 2.55% 15.92% PIRATININGA 15.38% 15.38% R GE 6.43% 7.15% 9.23% 8.76% 10.82% 50.14% SAELPA 6.36% 4.74% 6.79% 7.18% 8.34% 38.17% SANTA CRUZ 4.20% 3.26% 3.38% 6.68% 4.80% 24.36% SANTA MARIA 3.05% 2.53% 0.59% 5.23% 2.26% 14.36% V. PARANAPANEMA 0.94% 3.77% 1.74% 5.89% 1.79% 14.86% COCEL 4.13% 2.47% 1.62% 2.22% 0.32% 11.20% CSPE 5.49% 4.12% 3.57% 3.35% 1.86% 19.76% DMEP C 5.55% 2.73% 2.65% 4.60% 4.68% 21.86% ELETROACRE 0.94% 2.98% 4.75% 2.26% 7.44% 19.63% ELETROCAR 2.89% 0.93% 0.20% 1.46% 2.87% 8.61% J AGUAR I 7.74% 4.61% 1.53% 3.54% 0.16% 18.67% M OCOCA 2.54% 1.80% 1.65% 2.59% 1.47% 10.45% SULGIPE -0.91% 0.90% -0.62% 1.47% -1.21% -0.40% CPEE 6.48% 6.64% 1.59% 2.55% 0.71% 19.13% X ANXER 1.64% -3.16% 2.58% 2.83% -1.85% 1.90% Mean 6.04% 6.03% 6.94% 6.88% 7.77% Cumulative Index 6.04% 12.44% 20.25% 28.52% 38.50%

PAGE 166

166 REFERENCE LIST Ai, C., Martinez S., & Sappington, D. (2004), I ncentive Regulation and Telecommunications Service Quality, Journal of Regulatory Economics 26, 263-285. Amihud, Y., & Mendelson, H. (1986), Asse t Pricing and the Bid-Ask Spread, Journal of Financial Economics 17, 223-249. ANEEL (2003), Metodologia e Clculo da Emp resa de Referncia relativa rea de concesso da AES SUL, Nota Tcn ica N 047/2003/SRE/ANEEL, Anexo I. Antoniou, A., & Pescetto, G. (1977), The Eff ect of Regulatory Announcements on the Cost of Equity Capital of British Telecom, Journal of Business Finance and Accounting 24, 125. Armstrong, M., & Sappington, D. (2003), Recent De velopments in the Theory of Regulation, Handbook of Industrial Organization (Vol. III) M. Armstrong and R. Porter, eds., Elsevier Science Publishers, forthcoming. Atkinson, S. E., Cornwell, C., & Honerkam p, O. (2003), Measuring and Decomposing Productivity Change: Stochastic Distance Func tion Estimation Versus Data Envelopment Analysis, Journal of Business & Economic Statistics 21, 284-294. Bagdadioglu, N., Waddams Price, C. M., & We yman-Jones, T. G. (1996), Efficiency and Ownership in Electricity Distribution: A Non-Parametric model of the Turkish Experience, Energy Economics 18, 1-23. Bardelin, C.E.A. (2004), Os efe itos do Racionamento de Energia El trica ocorrido no Brasil em 2001 e 2002 com nfase no Consumo de Energi a Eltrica, Dissertao de Mestrado, Universidade de So Paulo. Beneish, M. (1991), The Effect of Regulatory Changes in the Airline Industry on Shareholders Wealth, Journal of Law and Economics 98, 395-430. Beneish, M., & Whaley, R. (1996), An anatomy of the S&P game: The effects of changing the rules, Journal of Finance 51, 1909-1930. Berg, S., Lin, C., & Tsaplin, V. (2005), Regulat ion of State-Owned and Privatized Utilities: Ukraine Electricity Distribution Company Performance, Journal of Regulatory Economics 28, 259-287. Binder, J.J. (1985a), On the use of the Mult ivariate Regression Model in Event Studies, Journal of Accounting Research 23(1), 370-383. Binder, J.J. (1985b), Measuring the effects of regulation with stock price data, Rand Journal of Economics 16(2), 167-221.

PAGE 167

167 Binder J.J. (1998), The Event Study Methodology Since 1969, Review of Quantitative Finance and Accounting, 11, 111-137. BNDES (2001), Cadernos de In fra-Estrutura Setor Eletrico Ranking 2001, Vol. 1. Botasso, A., & Conti, M. (2003), Cost Inefficien cy in the English and Welsh Water Industry: An Heteroskedastic Stochastic Cost Frontier Approach, Working Paper. Brown, A., & de Paula, E. (2002) Strengthening of the Institu tional and Regulatory Structure of the Brazilian Power Sector, Report, The World Bank. Brown, A., & de Paula, E. ( 2004), PPIAF Phase 2 Project for Brazil Power Sector, Task 4: Strengthening of the Institutio nal and Regulatory Structure of the Brazilian Power Sector, Report, The World Bank. Brown, S.J., & Warner, J.B. (1985), Using Daily Stock Returns: The Case of Event Studies, Journal of Financial Economics 14, 3-31. Buckland, R., & Fraser, P. (2001), Political and Regulatory Risk: Be ta Sensitivity in UK Electricity Distribution, Journal of Regulatory Economics 19(1), 5-25. Burns, P., & Weyman-Jones, T. G. (1996), Cost Functions and Cost Efficiency in Electricity Distribution: A Stochast ic Frontier Approach, Bulletin of Economic Research 48, 41-64. Byatt, I. (2004), Regulation and Asse t Valuation, Report, The World Bank. Carroll, K.A., & Landim, D.J. (1993), Measuring ma rket response to regulation of the cable tv industry, Journal of Regulatory Economics 5, 385-399. Chordia, T., Shivakumar, L., & Subrahmanyam, A. (2004), Liquidity Dynamics Across Small and Large Firms, Economic Notes by Banca Mont e dei Paschi di Siena SpA 33(1), 111 143. Coelli, T., Perelman, S., & Romano, E. (1999), Accounting for Environmental Influences in Stochastic Frontier Models: With Application to International Airlines, Journal of Productivity Analysis 11, 251-273. Coelli, T., Prasada Rao, D., & Battese, G. (1998), An Introduction to E fficiency and Productivity Analysis Boston, MA: Kluwer Academic Publishers. Cubbin, J., & Tzanidakis, G. (1998), Regressi on versus data envelopment analysis for efficiency measurement: an application to the England and Wales regulated water industry, Utilities Policy 7, 75-85. Dassler, T., Parker, D., & Saal, D.S. (2006), M ethods and trends of performance benchmarking in UK utility regulation, Utilities Policy 14, 166-174.

PAGE 168

168 de Oliveira, A. (2003), The Political Econom y of the Brazilian Power Industry Reform, Working paper. Di Tella, R., & Dyck, A. (2002), Cost Reduction s, Cost Padding and Stock Market Prices: the Chilean Experience with Price Cap Regulat ion, Working Paper No. 03-050, Harvard Business School. Dnes, A.W., Kodwani, D.G., Seaton, J.S., & W ood, D. (1998), The Regulation of the United Kingdom Electricity Industry: An Even t Study of Price-Capping Measures, Journal of Regulatory Economics 13, 207-225. Dnes, A.W., & Seaton, J.S. (1999), The regulati on of electricity: results from an event study, Applied Economics 31, 609-618. Elyasiani, E., Hauser, S., & Lauterbach, B. ( 2000), Market response to liquidity improvements: Evidence from exchange listings, Financial Review 41, 1-14. Estache, A., Perelman, S., & Trujillo, L. (2005) Infrastructure Performance and Reform in Developing and Transition Economies: Evidence from a Survey of Productive Measures, World Bank Policy Research Working Paper n. 3514 Estache, A., Rossi, M. A., & Ruzzier, C. A. (2004) The Case for Intern ational Coordination of Electricity Regulation: Evidence from the Meas urement of Efficiency in South America, Journal of Regulatory Economics 25(3), 271-295. Fama, E. F. (1976), Foundations of Finance (Basic Books, New York). Ferreira, C.K.L. (2000), Privatizacao do Setor Elet rico no Brasil, in Pinheiro, A. C. and K. Fukisaku (eds.), A privatizacao no Brasil: o caso dos services de utilidade publica Rio de Janeiro: BNDES / OCDE. Filippini, M., & Wild, J. (2001), Regional differen ces in electricity dist ribution costs and their consequences for yardstick regulation of access prices, Energy Economics 23, 477-488. Frsund, F., & Kittelsen, S. (1998), Productiv ity Development of No rwegian Electricity Distribution Utilities, Resource and Energy Economics 20(3), 207-224. Foster, V., & Antmann, P. (2004), The Regulator y Challenge of Asset Valuation: A Case Study from the Brazilian Electric ity Distribution Sector, ENERGY Working Notes 2, World Bank. Galetovic, A., & Bustos, A. ( 2002), Regulacin pro empresa efic iente: quin es realmente usted?, Estudios Pblicos, 86, 145-182. Goldstein, A. (1999), Brazilian privatization in international perspective: the rocky path from state capitalism to regulatory capitalism, Industrial and Corporate Change 8, 673-711.

PAGE 169

169 Gomes, V., Pessa, S. A., & Veloso, F. A. ( 2003), Evoluo da Produtividade Total dos Fatores na Economia Brasileira: Uma Anlise Comparativa, mimeo. Gomez-Lobo, A., & Vargas, M. (2001), La regul acin de las empresas sanitarias en Chile: una revisin del caso de EMOS y una propuesta de reforma regulatoria, Documento de Trabajo N 177, Departamento de Economa, Universidad de Chile. Greene, W. (2005a), Reconsidering heterogeneity in panel data estimators of the stochastic frontier model, Journal of Econometrics 16, 269-303. Greene, W. (2005b), Fixed and Random Eff ects in Stochastic Frontier Models, Journal of Productivity Analysis 23, 7-32. Grifell-Tatj, E., & Lovell, C.A.K. (2003), The Managers versus the Consultants, Scandinavian Journal of Economics, 105, 119-138. Hattori, T. (2002), Relative Performance of U.S. and Japanese Electric ity Distribution: An Application of Stochast ic Frontier Analysis, Journal of Pro ductivity Analysis 18, 269284. Hattori, T., Jamasb, T., & Pollitt, M. G. (2003), A Comparison of UK and Japanese electricity distribution performance 1985-1998: le ssons for incentive regulation, DAE Working Paper WP 0212 University of Cambridge. Irastorza, V. (2003), Benchmarking for Distribu tion Utilities: A Problematic Approach to Defining Efficiency, The Electricity Journal 16(10), 30-38. Jadresic, A. (2002), The Model Company Appr oach for Tariff Regulation in Electricity Distribution, Report, The World Bank. Jamasb, T., & Pollitt, M. G. (2001), Benchmark ing and Regulation: International Electricity Experience, Utilities Policy 9, 107-130. Jamasb, T., Nillesen, P., & Pollitt, M. G. (2003), Gaming the Regulator: A Survey, The Electricity Journal XX, 68-80. Joskow, P. L. (2005), Incentive Regulation in Th eory and Practice: Elec tricity Distribution and Transmission Networks, mimeo. Klein, C.C., & Sweeney, G. H. (1999), Regulator preferences and utilit y prices: evidence from natural gas distribution utilities, Energy Economics 21, 1-15. Knittel, C.R. (2006), The Adoption of State El ectricity Regulation: The Role of Interest Groups, The Journal of Industrial Economics 54(2), 201-222. Kridel, D., Sappington, D., & Weis man, D. (1996), The Effects of Incentive Regulation in the Telecommunications Industry: A Survey, Journal of Regulatory Economics 9, 269-306.

PAGE 170

170 Kumbhakar, S. C., & Hjalmarsson, L. (1998), Relative Performance of Public and Private Ownership under Yardstick Competition: Electricity Retail Distribution, European Economic Review 42, 97-122. Kumbhakar, S. C., & Lovell, C. A. K. (2000), Stochastic Frontier Analysis Cambridge University Press. Laffont, J-J., & Tirole, J. (1993), A Theory of Incentives in Procurement and Regulation Cambridge, MA: MIT Press. Lamdin, D.J. (1999), Event studie s of regulation and new results on the effect of the cigarette advertising ban, Journal of Regulatory Economics 16, 187-201. Lowry, M. N., & Kaufmann, L. (1999), The Cost Structure of Power Distribution Pacific Economics Group, Madison, WI. Lynch, A., & Mendenhall, R. (1997). New ev idence on stock price effects associated with changes in the S&P 500 index, Journal of Business 70, 351-383. Maurer, L., Pereira, M., & Rosenblatt, J. (2005) Implementing Power Rationing in a Sensible Way: Lessons Learned and International Best Practices, Report 305/05, The World Bank. Morana, C., & Sawkins, J.W. (2002), Stock Ma rket Reaction to Regulatory Price Reviews in the English and Welsh Water Industry, Journal of Regulatory Economics 22(2), 185-204. Mota, R. L. (2003), The Restructuring and Priv atisation of Electricity Distribution and Supply Businesses in Brazil: A Social Cost-Benefit Analysis. Working Paper, Churchill College, University of Cambridge. Mota, R. L. (2004), Comparing Brazil and U.S.A. Electricity Distributio n Performance: What was the Impact of Privatisation?, Worki ng Paper, Churchill College, University of Cambridge. Naughton, M.C. (1989), Regulatory preferences a nd two-part tariffs: the case of electricity, Southern Economic Journal, 55, 743-758. Nelson, R.A. (1982), An empirical test of the Ramsey theory and Stigler-Peltzman theory of public utility regulation, Economic Inquiry, 20, 277-290. Nelson, J.P., & Roberts, M.J. (1989), Ramsey numbers and the role of competing interest groups in electric utility regul ation, Quarterly Review of Economics and Business, 29, 2142. Neuberg, L. (1977), Two Issues in the Munici pal Ownership of Elect ric Power Distribution Systems, Bell Journal of Economics 8, 303-323. Oliveira, A., & Pires, J.C.L. ( 1994), Setor Eltrico Br asileiro: diagnstico e perspectivas. Rio de Janeiro: IE/UFRJ, mimeo.

PAGE 171

171 Papke, L.E., & Wooldrige, J. (1996), Economet ric methods for fractional response variables with an application to 401(k) plan participation rates, Journal of Applied Econometrics 11, 619-632. Peano, C.R. (2005), Regulaco Tarifria do Setor de Distribuio de Ener gia Eltrica no Brasil: Uma Anlise da Metodologia de Reviso Tari fria Adotada pela Aneel, Dissertao de Mestrado, Universidade de So Paulo. Peltzman, S. (1976), Toward a More General Theory of Regulation, Journal of Law and Economics 19, 211-40. Pires, J.C.L., & Goldstein, A. (2001), Agncias Reguladoras Brasileiras: Avaliao e Desafios, Revista do BNDES 8 (16), 3-42. Pollitt, M. G. (2005), The Role of Efficiency Estimates in Regulatory Price Reviews: Ofgems Approach to Benchmarking Electricity Networks, Utilities Policy 13, 279-288. Prager, R.A. (1992), The effects of deregulatin g cable television: evidence from the financial markets, Journal of Regulatory Economics 4, 347-363. Primeaux, Jr. W.J, Filer, J.E., Herren, R.S., & Hollas, D.R. (1984), Determinants of regulatory policies toward competition in the electric utility industry, Public Choice, 43, 173-186. Roberts, M. J. (1986), Economies of Density and Size in the Produc tion and Delivery of Electric Power, Land Economics 62, 378-387. Rose, N.L. (1985), The incidence of regulat ory rents in the moto r carrier industry, Rand Journal of Economics 16, 299-318. Rossi, M.A., & Ruzzier, C.A. (2000), On the regula tory application of efficiency measures, Utilities Policy 9, 81-92. Salinger, M. (1992), Standard Errors in Event Studies, Journal of Financial and Quantitative Analysis 27(1), 39-53. Salvanes, K. G., & Tjtta, S. (1994), Productivity Differences in Multiple Output Industries: An Application to Elec tric Distribution, Journal of Productivity Analysis 5, 23-43. Sappington, D. (2002), P rice Regulation, in The Handbook of Telecommunications Economics. Volume I: Structure, Regulation, and Competition edited by M. Cave, S. Majumdar, and I. Vogelsang. Amsterdam: Elsevi er Science Publishers, 225. Sarafidis, V. (2002), An Assessment of Comp arative Efficiency Measurement Techniques, Europe Economics Occasional Paper 2, London. Sawkins, J.W. (1996), Balancing multiple interest s in regulation: an event study of the English and Welsh water industry, Journal of Regulatory Economics 9, 249-268.

PAGE 172

172 Schipper, K., & Thompson, R. (1983), The Imp act of Merger-Related Regulations on the Shareholders of Acquiring Firms, Journal of Accounting Research 21, 184-221. Schwert, G.W. (1981), Measuring the Effect s of Regulation: Evidence from the Capital Markets, Journal of Law and Economics 24, 121-145. Serra, P. (2002), Regulacin del sector el ctrico chileno, Perspectivas, 6, 11-43. Shuttleworth, G. (2003), Firm-Specific Productive Efficiency: A Response, The Electricity Journal 16(3), 42-50. Silva, H.C.D. (2006a), An Empirical Assessm ent of the Regulators Performance in the Brazilian Electricity Sector, Un iversity of Florida, mimeo. Silva, H.C.D. (2006b), Privatization, Incentive Regulation, and Efficiency Improvements in the Brazilian Electricity Distribution Indust ry, University of Florida, mimeo. Simar, L., & Wilson, P. (1998). Sensitivity Analys is of Efficiency Scores: How to Bootstrap in Nonparametric Frontier Models. Management Science 44: 49. Stern, J. (2005), UK regulat ory price reviews ant the role of efficiency estimates, Utilities Policy 13, 273-278. Stigler, G.J. (1971), The Theory of Economic Regulation, Bell Journal of Economics and Management Science 2, 3-21. Wang, H. (2002), Heteroskedasticity and Non-M onotonic Efficiency Effects of a Stochastic Frontier Model, Journal of Productivity Analysis 18, 241-53. Weisman, D.L. (2000), The (in)efficiency of th e efficient-firm cost standard, The Antitrust Bulletin, Spring, 195-211.

PAGE 173

173 BIOGRAPHICAL SKETCH Hamilton Caputo Delfino Silva is a Senior Anal yst at Tribunal de Contas da Uniao TCU (Federal Audit Office), an indepe ndent organ of the state, which assists in the external control that the Congress possesses over the whole public administration. Mr. Silva was awarded the Bachelor in Busi ness Administration degr ee in 1986, from the Universidade de BrasiliaBrazil. In 1994, he was awarded a Master in Business Administration degree from Fundacao Getulio Vargas in Sao Paulo, Brazil. Mr. Silva joined the University of Florida as a doctoral student in 2001. After completing the course requirements and being admitted to candidacy, Mr. Silva moved back to Brazil in December of 2004, where he split his time betw een working at TCU and performing the job needed to finish his dissertation. Mr. Silva was awarded the Doctor of Philosophy degree in Economics in August of 2007.