<%BANNER%>

Theoretical and Empirical Analyses of Incentives and Public Ownership

Permanent Link: http://ufdc.ufl.edu/UFE0021071/00001

Material Information

Title: Theoretical and Empirical Analyses of Incentives and Public Ownership
Physical Description: 1 online resource (97 p.)
Language: english
Creator: Sabbioni, Guillermo Sebastian
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: adverse, efficiency, incentives, multiagent, procurement, sanitation, selection
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: This dissertation includes both theoretical and empirical research in economic efficiency. The first chapter theoretically evaluates how to design procurement models when two agents may have better information than the principal about their production costs and about the similarity (i.e., correlation) of those costs between agents. The results indicate that if the uncertainty about the correlation is not severe, the principal should design a contract as if she knew that correlation is the lowest possible. The drawback with such mechanism is that the high-cost agent earns rent if correlation is high. In contrast, if the uncertainty about the correlation is more severe, the principal should not tailor the solution to any of the potential correlations. This contract reduces the rent of the high-cost agent if correlation is high by giving rent to the low-cost agent if correlation is low. Although the welfare level may be different, the characterization of the results holds when both agents observe the correlation and do not report it, as well as in the case where only one agent observes and reports the correlation to the principal. Finally, when the probability of the low-correlation setting is very small, the principal may find optimal to exclude the high-cost agent from the contract if correlation is low, allowing her to achieve the socially optimal situation if correlation is high. The second chapter empirically evaluates the relative efficiency of different type of water and sewerage operators in Brazil. The analysis consists of a cost comparison between public and private operators, and between state-level and municipal-level operators. In a first stage, a cost function is estimated utilizing a fixed-effects panel data model. In a second-stage, the firm-specific costs from the first stage are explained by means of firm-type indicator variables. The results illustrate that water and sewerage provision in Brazil is characterized by substantial economies of scale, indicating that state-level provision is economically more efficient. The results also show that there is not an economically significant difference between the costs of private and public firms.
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 Guillermo Sebastian Sabbioni.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Sappington, David.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021071:00001

Permanent Link: http://ufdc.ufl.edu/UFE0021071/00001

Material Information

Title: Theoretical and Empirical Analyses of Incentives and Public Ownership
Physical Description: 1 online resource (97 p.)
Language: english
Creator: Sabbioni, Guillermo Sebastian
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: adverse, efficiency, incentives, multiagent, procurement, sanitation, selection
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: This dissertation includes both theoretical and empirical research in economic efficiency. The first chapter theoretically evaluates how to design procurement models when two agents may have better information than the principal about their production costs and about the similarity (i.e., correlation) of those costs between agents. The results indicate that if the uncertainty about the correlation is not severe, the principal should design a contract as if she knew that correlation is the lowest possible. The drawback with such mechanism is that the high-cost agent earns rent if correlation is high. In contrast, if the uncertainty about the correlation is more severe, the principal should not tailor the solution to any of the potential correlations. This contract reduces the rent of the high-cost agent if correlation is high by giving rent to the low-cost agent if correlation is low. Although the welfare level may be different, the characterization of the results holds when both agents observe the correlation and do not report it, as well as in the case where only one agent observes and reports the correlation to the principal. Finally, when the probability of the low-correlation setting is very small, the principal may find optimal to exclude the high-cost agent from the contract if correlation is low, allowing her to achieve the socially optimal situation if correlation is high. The second chapter empirically evaluates the relative efficiency of different type of water and sewerage operators in Brazil. The analysis consists of a cost comparison between public and private operators, and between state-level and municipal-level operators. In a first stage, a cost function is estimated utilizing a fixed-effects panel data model. In a second-stage, the firm-specific costs from the first stage are explained by means of firm-type indicator variables. The results illustrate that water and sewerage provision in Brazil is characterized by substantial economies of scale, indicating that state-level provision is economically more efficient. The results also show that there is not an economically significant difference between the costs of private and public firms.
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 Guillermo Sebastian Sabbioni.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Sappington, David.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021071:00001


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_AAAAMM INGEST_TIME 2010-11-13T15:49:43Z PACKAGE UFE0021071_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 56717 DFID F20101113_AACKRI ORIGIN DEPOSITOR PATH sabbioni_g_Page_15.pro GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
20c5f49555f33122d22e04a52222329c
SHA-1
8eedd41e5d9e5d3c8e1bf8ffba38dd8d31d2de28
1053954 F20101113_AACKQT sabbioni_g_Page_95.tif
7d8cef2a1edc8740f30bfc5d61834b9d
d753e5769b45e56b6e5ad76901b4969ab793bc3d
46499 F20101113_AACKRJ sabbioni_g_Page_18.pro
3cdbae425703eab1533a375081026bee
30349811ef78b982e629ce66c5a09b7b43ad694a
F20101113_AACKQU sabbioni_g_Page_96.tif
355b135243e74abb9c0c070adddf583d
64a9f9dec92e562ad99616aa21fe63730f9e568f
60502 F20101113_AACKRK sabbioni_g_Page_19.pro
d218d478478fe987e507516cc2c9484e
de8b026b1b7ac97fb97cac04c7eac803ddf232ab
8247 F20101113_AACKQV sabbioni_g_Page_01.pro
aa03413650a2d71e45c13450c72d987d
845c4de0c49029ab4445aa1ce909a9bff031b6b0
50452 F20101113_AACKRL sabbioni_g_Page_20.pro
d69c6cc53a856d13e40cba4d50333e1c
8c7d78a69b0a92ca8df1a6c88ef38aa9bf50223f
1355 F20101113_AACKQW sabbioni_g_Page_02.pro
42d54817a8f545ffc29e1d03a7b530c5
f9274fae580ca35e3f20bbeaa9b231759b85ff93
42730 F20101113_AACKSA sabbioni_g_Page_42.pro
a051e40bb6c981ae5363164e53b5dd47
461f0e75a9955b8eb28a906bc12529dd2a071675
61303 F20101113_AACKRM sabbioni_g_Page_21.pro
8de6cda311e0bc5de22083612ccc99dc
ebfc43eb25f3118e61c1976ae8d2dd8508aac02e
4685 F20101113_AACKQX sabbioni_g_Page_04.pro
597fa0b3e5a78d72da07b7f253a7bbf4
7d85fa5d5d4838a46cd49cd1af58309fb6cda08c
57218 F20101113_AACKSB sabbioni_g_Page_43.pro
950c67e58ccb34aabbd8e7cd2965f4b4
1ecf6924a70b0ab28cb0ffb0686be48b624d71d0
50723 F20101113_AACKRN sabbioni_g_Page_22.pro
057d47c583aaa396ad0f055e8c99ca59
54c86579e4c2ec683b7c085569d58e4105fb294c
90840 F20101113_AACKQY sabbioni_g_Page_05.pro
815e73527b2aac29e430ad5a1e567cb0
92787a967f9f347bb1e1e607042c3471bb32bdc1
54556 F20101113_AACKSC sabbioni_g_Page_44.pro
a17d6045d5ab6db79a1025fa7bf0098f
2c21f9cfb089b4d34d2c326a7cd7fd1628b0ddac
46286 F20101113_AACKRO sabbioni_g_Page_23.pro
e6383262fbf00dbf12e619e830ec054d
2291df1000bcafc70efdf050ab55ca37516fe18a
3877 F20101113_AACKQZ sabbioni_g_Page_06.pro
5ac4a71f9c9174e9c61fd38331352ca8
e8239f561c36087d45c0027d44acc5b1f4f294b2
27322 F20101113_AACKSD sabbioni_g_Page_45.pro
c04f5016de38e574057253a5ffdc9867
194f38769c118f697ae2a8d905adef58786fc16c
13460 F20101113_AACKSE sabbioni_g_Page_46.pro
3a7a9e3620e6999e9ae3d17bf545339c
2a28e3f43ab01243bed313764065a15d23d99604
50757 F20101113_AACKRP sabbioni_g_Page_26.pro
308f33d5b46742ccff83d457ae608d1c
3b89f4bd8bc55c840f05e8aa03779fea9094fa7f
11729 F20101113_AACKSF sabbioni_g_Page_50.pro
9e6659b6280bb682c9ca780ae7512172
075a5fdf5d4c1dee72ac835dfcb83a82aebfcb2b
48333 F20101113_AACKRQ sabbioni_g_Page_27.pro
e53f756a514fe6ee79ad441b1f6a04a6
e74af8d9705bafc3f3a37ecc6c86028424680607
9894 F20101113_AACKSG sabbioni_g_Page_52.pro
bb5bd8dd37f9ef5690f2e0a33c254824
ab78878f181aaf2ea20f0b90ba8902eee1bd0818
44113 F20101113_AACKRR sabbioni_g_Page_28.pro
c55bc5b9dfff8f1f1993990dfc4035a7
ef080c97d077c96ed9b31b1b2cc6ca6faff29ee9
56057 F20101113_AACKSH sabbioni_g_Page_54.pro
2d26ccbc202fd90707f37dde8148f195
eee75467561d6ae27cfa4de1b6992889d33fd30e
44182 F20101113_AACKRS sabbioni_g_Page_29.pro
07f79c95869f158c5edfd2765095478d
42f5f5b61804416cd00d9332df71837d4a5a39aa
55728 F20101113_AACKSI sabbioni_g_Page_55.pro
c21ab58c25c3922a94b592de2028ab92
45f7dbaec2e5203ccffa4a50ecf78c1670239246
53806 F20101113_AACKRT sabbioni_g_Page_30.pro
5ac2a06fca2aadd103f1263987816480
fcca9f2b5d9812c74804afb6049bb8fcc091d1ef
63636 F20101113_AACKSJ sabbioni_g_Page_56.pro
190dbb0fc3e86be30392a03461928670
e9383a4b25de68f5c55f929e8b4606a9d3b5e1d9
55125 F20101113_AACKRU sabbioni_g_Page_32.pro
f11229d48ab0b8cc230f174080135c04
a544d1044889e72ed9de8db1c662505c8e4799ae
45264 F20101113_AACKSK sabbioni_g_Page_57.pro
ed73fec037107973e74c0f31e89794aa
bf2b380ac91c2a59035ee1bdad5da8ee90558265
37668 F20101113_AACKRV sabbioni_g_Page_33.pro
ecbc01d8b81c08d27d951ea2c003f78f
0eb378cdd43bbc9827c2282b7e3227ecb1d6fb40
68174 F20101113_AACKSL sabbioni_g_Page_58.pro
0a3ff13c3f3f07607c887bf6a3d18314
a868cc8eb9ce70798fef2d3e6ac22098b349af31
57065 F20101113_AACKRW sabbioni_g_Page_34.pro
ff9a8b8f25d2d397c8617902d972bc8c
c6b0cdb00f6efd870fd5f03037312d784eae12fe
51828 F20101113_AACKSM sabbioni_g_Page_59.pro
edcddc6d8d23cfdd8fb81f4880dc223a
dd93ada7c86d52f5b458fa8c927e8491f41973f0
33565 F20101113_AACKRX sabbioni_g_Page_35.pro
54adb58bf6156819d2b73169a7ca2bc7
cde72d58653e11b6a1c026b1d371e10742726f67
19592 F20101113_AACKTA sabbioni_g_Page_75.pro
86e12221726a35ef7d5712f8232db752
f0944787963f3c93d10801482992d3557a6ac926
63538 F20101113_AACKSN sabbioni_g_Page_60.pro
5eec7223368c5d099bb8fbbbecb9f571
c8e1663ae778d493c773501c4bcbdd10d486994e
58902 F20101113_AACKRY sabbioni_g_Page_37.pro
61ff081862229eeca22513f407e0cce2
cafb52680868f7fb17065117897268e027af5ee2
15760 F20101113_AACKTB sabbioni_g_Page_76.pro
5cab3be533b5ddfffc8ab63aad341c8c
e4df6626a5cbe9860b039e4adb7a6a3aa0217c36
102055 F20101113_AACKSO sabbioni_g_Page_62.pro
13b6a30c92e821e2b03e8d43fa90fa5f
e4fe4be417f1711aaf23b74229e2cb639a30cce7
40962 F20101113_AACKRZ sabbioni_g_Page_41.pro
8cdbc06c288ff584baa551486d86c571
9ee4e89ab18b410f8de0e1cb3488357672f2718c
16962 F20101113_AACKTC sabbioni_g_Page_78.pro
a530f1554598b74ecb4c7681aba644a0
111b924b8d57ebe107a50806c23f16595c1f3018
79861 F20101113_AACKSP sabbioni_g_Page_63.pro
10a28fe280b5a1ea8e125403ffc75977
d94182d657a7f61219b47e01adab1b100123f0b5
31804 F20101113_AACKTD sabbioni_g_Page_79.pro
ba335b9836fe42a37439e2d426098dfc
21858b4bd7efed53e2fec78f32d37f91e0085cdc
33832 F20101113_AACKTE sabbioni_g_Page_81.pro
ec185bb77ab3753772824d6553efdcc4
7c1fc0c02a91e6ac03257f0eb14528e958540612
65969 F20101113_AACKSQ sabbioni_g_Page_64.pro
600d30c7f56e8a678b10c92811c0cfa2
1e640f077fb9a36d6a52f4282b3a805394b12e25
25649 F20101113_AACKTF sabbioni_g_Page_82.pro
efdf8da041e7ecf92b99ed6f4dd7369d
87870bd6105b79a922ca35d34a039fd6556aec98
55437 F20101113_AACKSR sabbioni_g_Page_65.pro
31a6e78235daf6e30d4e169d75c7407d
196fc3d1828b99e76bc9959c5036f10d51de3ab8
33578 F20101113_AACKTG sabbioni_g_Page_83.pro
52fc3dd254a149fb6331199dc4fd7085
89f6d25d184c319033dd7a80d28229cf6f903b95
57112 F20101113_AACKSS sabbioni_g_Page_66.pro
9356c5a6b9110d2856126aa93d709a27
3fc258eb4ff226aa86c342d081764d80942589fd
26699 F20101113_AACKTH sabbioni_g_Page_84.pro
97fdc91b6e1c9cb1002099b5804d4b32
29197c19ac204b2f79fe0187bb25cf64bf71e501
54300 F20101113_AACKST sabbioni_g_Page_67.pro
f160198c85e5df3b0adac37398072944
c8be3d9e5bb822bafd781b1a366b0af42d1ea581
25436 F20101113_AACKTI sabbioni_g_Page_85.pro
f83670f5201456dc2ad71480f2540ff4
6040a31dfe41b4304fd25b9a287bd819727c6f39
55633 F20101113_AACKSU sabbioni_g_Page_68.pro
c34b8fdfa862d3eaf38acc7550880069
b290a43e19b5a1e9ea0fe60d03abea36465e161b
25662 F20101113_AACKTJ sabbioni_g_Page_86.pro
687d371a985be1ca8d17387d1f96fab2
9289a36e16ac20e18f22b8796ef9f858d8368bfd
27389 F20101113_AACKSV sabbioni_g_Page_69.pro
20d4b717cb09416457d1965778e852bb
de1e5cc80760680b437f11c9e881103c317f7c61
32394 F20101113_AACKTK sabbioni_g_Page_87.pro
20f4e613b86afa6ee8c155cf2752a22c
79c914e0a84f008651361d470415ac8b09e6f3d3
33204 F20101113_AACKSW sabbioni_g_Page_71.pro
fb23d1c68cfa16882aeec3a85a976543
1f1f01c6889788340b293c45c2ce1ec485e0e908
31301 F20101113_AACKTL sabbioni_g_Page_88.pro
75267c7bbafb82bd01ec5b171be0fee5
7045662558fe4e0341e0966cb92bc70d5f315b43
25200 F20101113_AACKSX sabbioni_g_Page_72.pro
3b31d464db94238e8d3d520ce7e565db
9fd1b31aa900e0ca8f70064e67e170dcfe0fcedf
2202 F20101113_AACKUA sabbioni_g_Page_11.txt
505c0a02b8b8911a48b9c2f4f1d1ebdd
36dbe3d3533501a78aedd6e574268428c772eb36
32024 F20101113_AACKTM sabbioni_g_Page_89.pro
8d066dbf6a57cbbb66ea737c374892cd
c4e42bf8cbd64100fed841be2acc919da5790394
15462 F20101113_AACKSY sabbioni_g_Page_73.pro
8b0fcc8253f85f51fb135c1350dd71a2
6baa3759a354a667619105f934e1273a9bd994a6
2133 F20101113_AACKUB sabbioni_g_Page_12.txt
1f35113c70d14670c29c4946375db7a9
416b16289f65d6d69071b208301baa539aee7e92
9947 F20101113_AACKTN sabbioni_g_Page_91.pro
63cb050fc27c073c6bfc5524858384f9
61423a8192b0030878c0916a5e38631aa88220f5
10142 F20101113_AACKSZ sabbioni_g_Page_74.pro
75626e5c43a0e337855cd61207663130
acea5f27593af2502e9131445be83ad5f8021d22
2218 F20101113_AACKUC sabbioni_g_Page_13.txt
f9bd1675b10cb5726a7781ff9aa20680
dee7cd3b508260d67818fac4f73a8c968f6085bd
58958 F20101113_AACKTO sabbioni_g_Page_94.pro
64a6fdb4a9cc6adb34a5bdf322d6d004
d3425a094f635c0d4ecdf76f93fffd6e76196177
2219 F20101113_AACKUD sabbioni_g_Page_15.txt
8550121758457f122f50b31860c47087
52670daaae725a170863fecfc7929eb3c2da8212
54698 F20101113_AACKTP sabbioni_g_Page_95.pro
32d2b2340aaa7fa7e25e544056865092
b060aec57f715b02986683b47b73724d399ec330
2110 F20101113_AACKUE sabbioni_g_Page_16.txt
ce32b24f819102f623f2990c2d093ff8
2e27d31b16365aafbc6ba9a411f15d4280e7fbc0
5257 F20101113_AACKTQ sabbioni_g_Page_96.pro
3a2de4beb37748626dd48ae74f36e813
5be3006f42ee9b493ecc07cd2c65b6b981214c59
2349 F20101113_AACKUF sabbioni_g_Page_17.txt
81ca349e6a4c163578cd6be059d10ced
f4b152fd747d5ad6e5dc5d2ae696a8acec8fb9b4
2074 F20101113_AACKUG sabbioni_g_Page_18.txt
eca9f035c90a45499fd2a9bc64bc9c80
0accfe63631eabcc75d65ebcd10e5de3659014c5
43906 F20101113_AACKTR sabbioni_g_Page_97.pro
a708d1df4eb9afb7ed8f639aa8ac1fba
7d5a4f91ccbdea7cdc88427917db5eff9aeb7a96
7073 F20101113_AACLAA sabbioni_g_Page_56thm.jpg
911d36560f1b583af8b5bf8bec36802e
a315d37876e540f142065d3814e41a48198aa7cb
1997 F20101113_AACKUH sabbioni_g_Page_20.txt
dd58b546d0d5c1145543dcda7c52ec1a
73341ae035d5cc65afec6871e4a77c0442c0bb63
108 F20101113_AACKTS sabbioni_g_Page_02.txt
468ecf60bbde5036888a715f429349d4
2aed7090ccbff762137ec47727e192029909c9ec
21666 F20101113_AACLAB sabbioni_g_Page_57.QC.jpg
7e7d8c322293ac28753ae5c18290229d
34ec9895e58cb4f10a13e4173a0461e1d03b48bb
2420 F20101113_AACKUI sabbioni_g_Page_21.txt
a5559eb81fbaa5638c99241b0339b8fa
f6f9d04b99054ad6b3ea343020c7aa03ae2ae916
165 F20101113_AACKTT sabbioni_g_Page_03.txt
af1f0faf52df1b79c71516a3b655d0ee
716dd2126ebb7d5c4e79a79297ad7b98d15f9539
5796 F20101113_AACLAC sabbioni_g_Page_57thm.jpg
35f855a64202dd2458a702284079c72c
72e71264eac4b5fc59590b07f5a641402599a132
2012 F20101113_AACKUJ sabbioni_g_Page_22.txt
c77ad9c28017447893873b7ea82f4e47
4d127ffea3c9a2a2e9752e1d3c13d1de35267be3
226 F20101113_AACKTU sabbioni_g_Page_04.txt
f94da16febbb81af506cb5f6f10273af
c9bb738c4f3fc07a98ce8b8f29707779e4536769
28146 F20101113_AACLAD sabbioni_g_Page_58.QC.jpg
5a855d9acc5555cf91153d91f2d53e9f
ab5fbec44ade523696606c15fa2720f9afc8e54e
1982 F20101113_AACKUK sabbioni_g_Page_23.txt
09df5f022e554632023bca56e3749ffa
5e43acdd7cff63d55bc5e29236aa81b8d38e31f7
3705 F20101113_AACKTV sabbioni_g_Page_05.txt
b0b59f9a1bbf2d11c43df5b381387dce
e6135f1787a35108f949e7ed8328073e728b3889
7404 F20101113_AACLAE sabbioni_g_Page_58thm.jpg
c0f6124ee4ad57770127c85317acc506
3dca46075699052676e45defcacfb974287f4742
2044 F20101113_AACKUL sabbioni_g_Page_25.txt
c8157d63ffb11ee568467287d3939444
46b1a2ddf21e7d5a5641f0aa6b79a14f77e912eb
156 F20101113_AACKTW sabbioni_g_Page_06.txt
db72332331587984b0643124fc2a2668
8c60e0a5e4f9300c6ff7c84cc1c666345abbb8d1
22377 F20101113_AACLAF sabbioni_g_Page_59.QC.jpg
1a009f181c8a2a991f28ed0c7febc46c
1ae191d165a650a43350e4f9a83282554cbfbed0
2286 F20101113_AACKVA sabbioni_g_Page_43.txt
6fffd2247efad1add1d8a9d710b7c279
d74a7b235d90a3cfc1cb6b6b0b6cf2f7588e875d
2043 F20101113_AACKUM sabbioni_g_Page_26.txt
c6d12c26e0f247e9eb5af83bc35122dd
5bfea0fe3fc6680be3ff58f8c458497af048aea7
1196 F20101113_AACKTX sabbioni_g_Page_07.txt
0bd9fba87f868b5e9b461d680bcb410f
40e263e4714659867cd3e1b9e0dd90293115ddfe
6261 F20101113_AACLAG sabbioni_g_Page_59thm.jpg
e18d6c3abbeacfa3511440aafbd63626
6db392c1108a1f30b209f1127b09a3dcb8f637e6
835 F20101113_AACKVB sabbioni_g_Page_46.txt
25837a53ad8d14a9015391f47a6ffc2c
c9bac6cc5f76fe50c4de9c9f9e9c6739e718dcc0
1968 F20101113_AACKUN sabbioni_g_Page_27.txt
77fb0542070ca12ae79ad54ac23498e4
63326dfe86ef502046002944b9196f107f860960
870 F20101113_AACKTY sabbioni_g_Page_08.txt
83e3a496dcf51cf3d711a89ca8b547e4
15bfbf5abe220478252b6882f9310e205688c8ab
6571 F20101113_AACLAH sabbioni_g_Page_60thm.jpg
c8225c3adf2d1d353e04cb159a2cb7cc
3933b8884da602043ba20e0afdae3738a62da74c
604 F20101113_AACKVC sabbioni_g_Page_47.txt
8ea44af780b5f5b1a43ab812b03e97a5
7915aadc48092e85e0760bc730a3be8593fce521
1858 F20101113_AACKUO sabbioni_g_Page_29.txt
fa7d7b73d4b6b43aff4a55963b564293
d3ef32f9638251ec7a9bec2daaf6c3e2e0a18548
2035 F20101113_AACKTZ sabbioni_g_Page_09.txt
15edbe0b1427c2a45e670317953511ad
7088d3a6ecec779cb6e7f821913b813569da874b
23451 F20101113_AACLAI sabbioni_g_Page_61.QC.jpg
bce475791068e8ab9abf90dc81bc7be7
a23fff8c22850bd883c268825d9370c89a9ab298
780 F20101113_AACKVD sabbioni_g_Page_49.txt
5e3cc9242afd3d0938200d10358a74d7
54e6a0554487406b9dd08f7b16b4165aa94c8486
2167 F20101113_AACKUP sabbioni_g_Page_30.txt
d357223b39471dee0011bf64c6434d4d
5eb773db734c7d38176a967b089a8d66bf527fad
6654 F20101113_AACLAJ sabbioni_g_Page_61thm.jpg
d94addbff33cb0714c4e05b9397aec69
fbc377d761f2072c1435341d1acec474a4c09e4d
624 F20101113_AACKVE sabbioni_g_Page_50.txt
8517cd1027af192119c49e40912b0c9b
2b53599c19e4c9a7c7341075f69ae6e02ec9c94c
2115 F20101113_AACKUQ sabbioni_g_Page_31.txt
815beabbd8deebfe37387fee50cc7b9c
5426321787c984de8bb6e0d1097412593a474fbe
30061 F20101113_AACLAK sabbioni_g_Page_62.QC.jpg
08a60728320014741ead07473615046d
cc507b988448684b9fbf50517b55cefc81597437
658 F20101113_AACKVF sabbioni_g_Page_51.txt
4c2f1f8c8544d849481d89c21584f756
1778a10a4c111de3185d7e0ddc5c39a4f452238f
2176 F20101113_AACKUR sabbioni_g_Page_32.txt
9f51ddf5ada336b414c8951a9d1bbbd3
7b61f24e78dbb5dfb968f2631bf6c6c0804ba9ff
7711 F20101113_AACLAL sabbioni_g_Page_62thm.jpg
c28ca11c6874e925704640e9b8619a6a
fe2004bacc3119e6e6a930955c04f67cdba68193
521 F20101113_AACKVG sabbioni_g_Page_52.txt
f84d28e0b7bb3ef848cffedbdfa46cb0
cb41b0391ec78c8d8aca415020b1db794e15f1ae
14464 F20101113_AACLBA sabbioni_g_Page_71.QC.jpg
046311d3065922f97dab94bfba01b489
f7da809588c7b1327af6023d6c38627fa6265f11
30818 F20101113_AACLAM sabbioni_g_Page_63.QC.jpg
f8e36030f572b61c7765265ee43e7989
e885c9fc98c0cd4925a10b476c911315d2e72ac0
2204 F20101113_AACKVH sabbioni_g_Page_54.txt
794d880a795df9924c1347c07e704dd4
93d731d9e6deac33d75c52879bf9ae4bcbe3adec
1640 F20101113_AACKUS sabbioni_g_Page_33.txt
5eebf2fa4591460e16f6d0241c954717
fd187434e41336efbd20a27c36f586d4becc1a93
4690 F20101113_AACLBB sabbioni_g_Page_71thm.jpg
7d3fb6935bad4768e2690741802598f1
51959fe202d5483ebbd64f70d654c1d5e6924f77
8005 F20101113_AACLAN sabbioni_g_Page_63thm.jpg
43ee4430ae81d7ca6fd010829148533b
635dbccedf73addacaabb3846f180ad19ffcf6e5
2198 F20101113_AACKVI sabbioni_g_Page_55.txt
b97a01fae19dfb9a26c0c5a7733dc004
4fc2320524b808bdf3c3d2658b4440878abc39ca
2237 F20101113_AACKUT sabbioni_g_Page_34.txt
2d9f90ef5b43fcf21250a2ace0e15792
2ada4fb31434e7662fc3f1c0b08d7c4a537083ba
11568 F20101113_AACLBC sabbioni_g_Page_72.QC.jpg
e2ba7040a6763041fded0dbac936739b
a0cc4bb28ecf4ac1ccb730bc32a5bc8839294059
27278 F20101113_AACLAO sabbioni_g_Page_64.QC.jpg
5bb8c6849a5acc2902a24f4edd9481bb
b62624fd6b230a8678fe2cb9d364b5c367945a5b
2504 F20101113_AACKVJ sabbioni_g_Page_56.txt
7a4798eaddca30f62e169d2b6e5b7475
c4f5d42cea01a057f87aca0e193248faa8010f2d
1607 F20101113_AACKUU sabbioni_g_Page_35.txt
439bdf2d78ce3838cdd4d3ac88fa60d0
380dbbc02a7f919dc996db09f5f92a35ae672e56
3972 F20101113_AACLBD sabbioni_g_Page_72thm.jpg
b78cd6b5ffd736457f8171581839fbf2
e52f4f6f569d6cd22457e2ed7cc34593f788b4a2
7580 F20101113_AACLAP sabbioni_g_Page_64thm.jpg
3d32970a08050561ee53c00a47862cf9
518beb9b0fb0eea860c81631b6ee79d0eb6428d3
1802 F20101113_AACKVK sabbioni_g_Page_57.txt
67be6b92f69b07b592f1ef24b88c5971
4b81a81ebcf3b8d8254d06c69ffacb3832535555
2312 F20101113_AACKUV sabbioni_g_Page_37.txt
da54e84b21a9b808379a35a4936b26d6
71467a2ab772c2bd4407acd755fa49231f18e067
8608 F20101113_AACLBE sabbioni_g_Page_73.QC.jpg
543a8b4bb1fc82c40af2198dda369d47
497c77392e312f911e21c7c7d38ef8278f43e9ad
24642 F20101113_AACLAQ sabbioni_g_Page_65.QC.jpg
a6110e5be00631bb6f3dfd9da58aafb7
564aeca5052a0f16e128c56f5d436e27d4430727
2164 F20101113_AACKVL sabbioni_g_Page_59.txt
9ac47de7f2ef47bd763e77efe9d9bb9c
ff9755b12cbf4d1884731da73288fcc9beef4c01
2113 F20101113_AACKUW sabbioni_g_Page_39.txt
8ac53e77891bcd1cf4fe741683100604
78b99051630d86cc2a49227d06c799e02b8425ab
2381 F20101113_AACLBF sabbioni_g_Page_74thm.jpg
0813e7c5a9018f849c06e7b47f06e5ce
70448c73d470d706718320cffcda238a43fcd3c2
6919 F20101113_AACLAR sabbioni_g_Page_65thm.jpg
0ca3c3d9653b64f93a81cf27d813032c
68d86e5d187f07eac5cbc400ee197a23519d52c3
2013 F20101113_AACKUX sabbioni_g_Page_40.txt
2fc65e1de984ef25e38a4b40e9b1e28b
09d2ee2d3522c4a4dfd57d91d6e934891f389403
3588 F20101113_AACLBG sabbioni_g_Page_75thm.jpg
137492f186da3ae31e90860cb97412f8
15ba8b9b21f3e3e8a1e5716681922793644f4e15
1537 F20101113_AACKWA sabbioni_g_Page_81.txt
3cc825372011fe866571771ca25b94b7
5719fab4c4008478ab78931ed55b2bc71b77941a
26580 F20101113_AACLAS sabbioni_g_Page_66.QC.jpg
8318b33325371ca8e95bf3edbe2d1451
5f7c20da5a635c9ce83d1f5e0bc20f9481f11b33
2510 F20101113_AACKVM sabbioni_g_Page_60.txt
bc62bdeac4c6a57dc585a2536878c4ee
e81354b10ce03cc3bc20a4555721f178610ae448
2069 F20101113_AACKUY sabbioni_g_Page_41.txt
fbb75793e39e616e171a38f42a1808ac
95f3cab28b145ea2f58476206981078dad0bdea0
8733 F20101113_AACLBH sabbioni_g_Page_76.QC.jpg
a943c8a5ef5e67f76294930013e189b3
8df2ca6a11b2d0473d20f060e50f3020d57758d0
1269 F20101113_AACKWB sabbioni_g_Page_82.txt
4ca99ef5f83d342e9c47218617b2d43f
a3cfae51468e607562efe0f8ae8cae1ab8683e44
6902 F20101113_AACLAT sabbioni_g_Page_66thm.jpg
d4b4bef56def8cc700e034177f1b77a0
9e72c35b5ce33d8f1e5ab5526879c02d44181b0a
2287 F20101113_AACKVN sabbioni_g_Page_61.txt
702a01e0baff9fa25924ac653b288ebb
e1a8f2b741fe36fa3e125cc1d437ce5636db9f0e
1746 F20101113_AACKUZ sabbioni_g_Page_42.txt
08a067a5c1c750fd786362975e287dc2
73d8adffd1b84bd114ff5f0f01a56a8c4b0c54f5
2954 F20101113_AACLBI sabbioni_g_Page_76thm.jpg
9d03e903fc7ba002d7b3ff83557e9554
bebb03baf15fb0ca0ad99042f0bb9d28fcc30f45
1568 F20101113_AACKWC sabbioni_g_Page_83.txt
15972f48c897d5305b6156414c26e0c9
8110e5749f19e7c3dc4c178a53c6a630c6071118
6657 F20101113_AACLAU sabbioni_g_Page_67thm.jpg
ec79ac2e681222744b281a80756a3794
347e066abf739ff02fa7f0b2ed06ffd2fe44df97
3113 F20101113_AACKVO sabbioni_g_Page_63.txt
6d26b84d40924356290fa3d13c545185
efa944e4de7c23e97ee9506593548b8454653809
10290 F20101113_AACLBJ sabbioni_g_Page_77.QC.jpg
e166c40da3c325a433769e3b39d03c0f
e29853d7183b22976f1999c956a90a240e28c29c
1445 F20101113_AACKWD sabbioni_g_Page_85.txt
89d7e442ed915f2af73a4bce972ccd61
9f9641ce3feb4f4604f25c35fb1aff899fdcd7b0
25602 F20101113_AACLAV sabbioni_g_Page_68.QC.jpg
912e49b4e9fe17ec55f2ea8f016ea72d
4b545021b1714b7edbcbffe03056a8acdd8489eb
2615 F20101113_AACKVP sabbioni_g_Page_64.txt
81d9820d2d30f500ef69bec37bf86e11
1a8e1548f829dc13a117635e940ec9cd6faa7dba
3603 F20101113_AACLBK sabbioni_g_Page_77thm.jpg
03daa93752f06625e7d8d96112aed37b
2eb0792d9e59e66e57460c30e3e5e47efd9a9c2f
1366 F20101113_AACKWE sabbioni_g_Page_86.txt
277bd9b66d44d66691427dd428abf9a2
39fabaa8bd3ab92af0210122b20f01fe3d4a9f44
6982 F20101113_AACLAW sabbioni_g_Page_68thm.jpg
d053712fd5ed6e0b354a0e01a5a04923
269891a9f50196f59a8529fe8cb1c81dbbb6dda1
2273 F20101113_AACKVQ sabbioni_g_Page_66.txt
a527b4a4a22536e0b7ee21aec4003670
29ba4315d410998e14c4cf289a9bb5afafa6c905
8516 F20101113_AACLBL sabbioni_g_Page_78.QC.jpg
93ada8332864ad21423f5bb92b5bfc9b
8c7352e95b8b89c7b3b08a64e69f4d9e8f9b9f14
1617 F20101113_AACKWF sabbioni_g_Page_88.txt
5c3742652ddfe570aedc9d145df4a9c2
11ad55ea00e03a232e9aac303a15d0d711ec285c
3988 F20101113_AACLAX sabbioni_g_Page_69thm.jpg
aa642f771c9c568a8fc0b4eb9055eb5f
58705c03e36c6967e7c0acac46b14bdfae8d3ec1
2188 F20101113_AACKVR sabbioni_g_Page_68.txt
54ae7cd7fc3359f4bf651bced7792063
3e6455fb7cb6d693f1ca63ba27afe0ca2842a141
15450 F20101113_AACLCA sabbioni_g_Page_86.QC.jpg
68243ccad35e4e4f3ad0daf9456a09bd
eda816c641284a8af435858e805fcf2ab6088484
2949 F20101113_AACLBM sabbioni_g_Page_78thm.jpg
f9160030032192200033418c60473611
de15499363ffe874ddbad5a62388953a2df8b04a
1618 F20101113_AACKWG sabbioni_g_Page_89.txt
fa1129f93a843379186e78e797369e23
52c0acbf442dc8335a9a400a432119455edca784
8795 F20101113_AACLAY sabbioni_g_Page_70.QC.jpg
4febd48f53191e672a8b4e9b43973443
5ca6ca5052c3712dfc880d630509f72bdd6dc6de
945 F20101113_AACKVS sabbioni_g_Page_70.txt
dd4d072f0d06608997835dd4aaa96f8c
41268ab8a33c774394c4f18f5842f91a5b9ceada
4646 F20101113_AACLCB sabbioni_g_Page_86thm.jpg
c5629c4fd5ae100b5de040cf5a36c217
3f2f864c2a092da59e1bb9b54179e2018a8abff7
17768 F20101113_AACLBN sabbioni_g_Page_79.QC.jpg
d63945e35ad35b2dc0197e422ce30a11
a24f44022f347ec11a260cfd4549646b660a5ad2
1465 F20101113_AACKWH sabbioni_g_Page_90.txt
0a5531d27eaaeda7581fb897ef3d4b9a
2e51dc1576dd7e0356deb0a2838c7190e9272b32
2761 F20101113_AACLAZ sabbioni_g_Page_70thm.jpg
217916a069147da7aa1a8bf7760f0bbe
f96d9a9adec7de68afd42b75bb39ce554a36a6eb
15832 F20101113_AACLCC sabbioni_g_Page_87.QC.jpg
fce0a9a30ca1c0b7f0c45c39847bc725
bc7acc3ff1359f8da38b3a77cbfc441df7f676b0
5226 F20101113_AACLBO sabbioni_g_Page_79thm.jpg
e2924355fe0880801052cbfba8abe422
9e6a1f23647099014f0e1a65ac23cabd29c96cf7
2281 F20101113_AACKWI sabbioni_g_Page_92.txt
45ebdfeb2b3a5e4ee2f6c90c01de648e
bf8b11ff812f9ebc6f050ba4ee2e0df72e31705d
1496 F20101113_AACKVT sabbioni_g_Page_71.txt
c2777c93f9c5a5d72d684599415be5bc
0ae9439e01738a82952468dfb53dbe675be3f524
4802 F20101113_AACLCD sabbioni_g_Page_87thm.jpg
387643b0129d6152cf1fd706af8cacb5
dc4b53810503cadb5f5c0d0c16fea7f84c14b35a
13932 F20101113_AACLBP sabbioni_g_Page_80.QC.jpg
5f9852e69041af771f572b1ba61775ee
d438c27306d5710281a90dcf0fa5eceb797f1143
2410 F20101113_AACKWJ sabbioni_g_Page_93.txt
524ca058e0dd664630ccb60e98295ab1
d490f4397dc37a81d67b85cf697bb3cb85457bfd
649 F20101113_AACKVU sabbioni_g_Page_73.txt
03ab394f4b29c287b864fae80d4d1125
e0c1033d59b187b623bd440ffe5599f7a552031d
15591 F20101113_AACLCE sabbioni_g_Page_89.QC.jpg
58cd4a74554ecea0c4df91af722afbd6
e3124de177a01e4d0e537516d77233e7c900ba58
4522 F20101113_AACLBQ sabbioni_g_Page_80thm.jpg
9fa467d17144b228acc7d13cca7be4e5
0a5b2311ea929149b2dde171106848f6a2d80b03
2343 F20101113_AACKWK sabbioni_g_Page_94.txt
b36aebbee063eee52a870062ab5399fa
30031176d8476c0a835ab34f32d45af1944f0a37
549 F20101113_AACKVV sabbioni_g_Page_74.txt
47c54a234631a996446706629a6d179e
b9fdafd1f69d11252d0a05ee3106e03c45d4967f
4692 F20101113_AACLCF sabbioni_g_Page_89thm.jpg
ab1ebfbdf6b1606c11051fe86f6c0885
4e4f94af4ba098c6e67d84e2d453c7c35276422c
16550 F20101113_AACLBR sabbioni_g_Page_81.QC.jpg
fa737838a3b8ad097b2706c35851dbb6
78eef19eadcc4cb7f0267508561792871d19f20c
2196 F20101113_AACKWL sabbioni_g_Page_95.txt
dbba7a587809c054b6f628dc0a1937c8
43bca7fd2c731b9789eb0d59042a5aa28273cd64
843 F20101113_AACKVW sabbioni_g_Page_75.txt
3c6fff0cadab33456321152ead62d1d6
7132d611799cd99a12c1b31d619da4220ebf8b15
2419 F20101113_AACLCG sabbioni_g_Page_91thm.jpg
86459cdd4cdc3fb10dc47c955954e0a1
5eada3040a2effae356d1a0a2603a36915973c35
10283 F20101113_AACKXA sabbioni_g_Page_08.QC.jpg
1e5eb1f30628ec1e51d7e4154bee66c0
bff789c50b91614c47134cf69203e5de9213460d
13545 F20101113_AACLBS sabbioni_g_Page_82.QC.jpg
076afed135d6aba4b397ea0d6b271296
698e2c08db03ad0835133d144cc52e4012b0c7e1
1788 F20101113_AACKWM sabbioni_g_Page_97.txt
63b808dd001ab20728df90e083de4021
1480d3ed64c7a6ab83ce942f0ba576d18bc5c92b
657 F20101113_AACKVX sabbioni_g_Page_76.txt
9ed10f53ee25941d7791ad5a4ed977ff
31f93a52fe1667cc8071c20e2e73c684497a455a
6703 F20101113_AACLCH sabbioni_g_Page_92thm.jpg
948078173f4eadde57af33d24b50690e
e59ef3ca8b3a4df6e25e993426bdf7743d37d524
21119 F20101113_AACKXB sabbioni_g_Page_09.QC.jpg
7310d881ea61116c2b5fca541080cadf
321c223eb0ff75fa1f5346c8d69e0c90f7462023
4238 F20101113_AACLBT sabbioni_g_Page_82thm.jpg
cb2ae7759162fd3e66cca910e77f1438
f1c82919e7d80546462d4da3658829269decf98a
520434 F20101113_AACKWN sabbioni_g.pdf
14f39a0d359789303b637d4991326a64
8b5b603c6bf71181a4e448d034cd42db793be116
850 F20101113_AACKVY sabbioni_g_Page_77.txt
8a1da532b091438335c6b41445855296
f58692b425437040caf9bea8f0f0b3bd2b298c7b
25126 F20101113_AACLCI sabbioni_g_Page_93.QC.jpg
c2836c7b4eb8b7dfd051effdf335679e
69ba2b29b12f6b56e77af6d05cf257060d55e632
5992 F20101113_AACKXC sabbioni_g_Page_09thm.jpg
afe25dd7a9a631b6d32361c6e7c66d1c
6488290c194db2d049883180025d22178ee10a7b
17898 F20101113_AACLBU sabbioni_g_Page_83.QC.jpg
f4db2c8eb9996f24a3856a33d687ec26
d170bab87a8ef939b3b27591b3e985388b8367b4
6930 F20101113_AACKWO sabbioni_g_Page_01.QC.jpg
872b067960be05a518e33717c66d24e9
0703b56254f07dcbba8d13af49694291f5475a8a
1473 F20101113_AACKVZ sabbioni_g_Page_80.txt
247397a769a52ce34ec3e66ab3fb4017
2d31017eeb714088c1152277bf76be9623a34e23
3369 F20101113_AACKXD sabbioni_g_Page_10thm.jpg
bf1fb63b288fcb1898a8a7633f0ac183
5e1118aa66908c0cecc2decdd9653affe34b689a
5080 F20101113_AACLBV sabbioni_g_Page_83thm.jpg
94099072371fa705e1579bc45ae719ef
732ac922cb56e851c6b86d4986e2c913d51e3413
3246 F20101113_AACKWP sabbioni_g_Page_02.QC.jpg
d51b9a3d84675fd7c2a5e24a111bc678
5a087853ccf39d83a7c7f546ea7a9e7dd695fb6f
26041 F20101113_AACLCJ sabbioni_g_Page_94.QC.jpg
e5fd1fa371dbc4dbca8ac4478619fdd6
d0ecf5e1a1be9de484771be92ca5aaeca5d8cb5f
6811 F20101113_AACKXE sabbioni_g_Page_11thm.jpg
2bec0ebf1dd86f5f51a922b33cbbd916
44d9ec0acf59ed8843cd882d275dffea5710f9a6
12846 F20101113_AACLBW sabbioni_g_Page_84.QC.jpg
e96aefdf408bf5ca5bedcd6aa452d055
2db5e6839285a1c34b43a71069aaa148e8700b60
1375 F20101113_AACKWQ sabbioni_g_Page_02thm.jpg
dd7f35d19e5d2c21a91c02b4381091bd
89b1897ce82fd371026da978d9c5d6ff8297a21a
7135 F20101113_AACLCK sabbioni_g_Page_94thm.jpg
f65a4109447062f8e08b495a7e85f45a
8bed3e2dcafc7013a80e2ddd9046196bcdb56c0f
24950 F20101113_AACKXF sabbioni_g_Page_12.QC.jpg
1bf56dc8d7c366d25d4171e3ef506786
8f83a469b0e789d1557f2b6556290a0f53333d55
4278 F20101113_AACLBX sabbioni_g_Page_84thm.jpg
6326d3329dee0f521178fed984a6de63
1b6502bf2bcab73cbfabb7b2b04ff10f818bb090
1589 F20101113_AACKWR sabbioni_g_Page_03thm.jpg
bc56a64898f15623134a974126dd9827
f11d90f8eb2ea9555f054905ffa1dadceb563fe8
6577 F20101113_AACLCL sabbioni_g_Page_95thm.jpg
b0348edbda6de3379c959862b39741c2
231b10124e54c0fd4ca44692501ab4982f091aed
6712 F20101113_AACKXG sabbioni_g_Page_12thm.jpg
1c1cdb0eb8650382cf3548d3f5a2ceab
4d5488d413bdd71320e4a321db3a1ff0b9de64e0
13730 F20101113_AACLBY sabbioni_g_Page_85.QC.jpg
8855a1f86c5ddaaf0d79b27105bcac52
943119e69edf7a63c4be93f4dc81e70adb3da250
4953 F20101113_AACKWS sabbioni_g_Page_04.QC.jpg
bf1edae75a13d6bae67188ee74b2333e
825b40d9c4ba0b43dfb0428c621db1301a6e9e96
1642 F20101113_AACLCM sabbioni_g_Page_96thm.jpg
59b787e14e830028ed5e45230f7ad077
793ce07fd7012cd8086cc994f9f3ceab6bd14fa4
7040 F20101113_AACKXH sabbioni_g_Page_13thm.jpg
8ac46c8ab500b639426359d37fc0dfa3
6a01256881590675d0cdb3bad21683205ff99370
4191 F20101113_AACLBZ sabbioni_g_Page_85thm.jpg
7e727355bff9583a6a55b8227c1ea03a
da353020d4d5c43680bddd67edb749a92ab823b0
1844 F20101113_AACKWT sabbioni_g_Page_04thm.jpg
c583f012f4f36f6907596c4b29069c3e
9230ced36a2e4659d7fadcf33c3cafa90b006656
22266 F20101113_AACLCN sabbioni_g_Page_97.QC.jpg
2e06480dae39d4bf6dc1aa0ed3995158
00dec3b0a05e076160c3a933cc7d0dfc3caa0e98
23817 F20101113_AACKXI sabbioni_g_Page_14.QC.jpg
b7f09ae46b9c2664cc360051242a5986
647287d89a90d335adcdde026a04502097eca3a5
6010 F20101113_AACLCO sabbioni_g_Page_97thm.jpg
a8df73c1cd3b3435d21d177999083ccf
241c4e3d7726d61373e869dd89c62b7c91f905bf
6821 F20101113_AACKXJ sabbioni_g_Page_14thm.jpg
2510643064bbb8ee747bda19357d0a87
4720eed72a4f324ecdb21b6d7ffd7a649bb73dd6
22329 F20101113_AACKWU sabbioni_g_Page_05.QC.jpg
27330334932b4c87ab51b34e0a6c9f52
7ef768a6f1d9e5abe3c18dd57193efcfc0ad9263
113345 F20101113_AACLCP UFE0021071_00001.mets FULL
af262d6ec5ead10f4de68324121b20b2
628ba977968c39f141b1db1e7a50a12ebe4b6599
25101 F20101113_AACKXK sabbioni_g_Page_15.QC.jpg
4aa9e65f835f946bbd93b7a5363904b2
241cdcd136c71fc0b7d189a91d17de65b9f8e20b
5947 F20101113_AACKWV sabbioni_g_Page_05thm.jpg
f386b0d6dedfa8bf27b2c0cc34a72846
ea69b7d3bdfaf792322bb87a1fb2ede42fa0282c
23398 F20101113_AACKXL sabbioni_g_Page_16.QC.jpg
5feadd54f584eba88be68803cb8844ab
f3c4ff7c28d4ecb6b9c871b25232ac131e598ac2
3555 F20101113_AACKWW sabbioni_g_Page_06.QC.jpg
803fc04f539428704f5fccaf19bcd363
4ce972330869c359cd3b9ed7d6e9911dc934bda8
6761 F20101113_AACKXM sabbioni_g_Page_16thm.jpg
d6662fea18f89b98574309a74877ba2a
b97f50ae1d32ee57e1657f19cd10b469bdc30ce5
1427 F20101113_AACKWX sabbioni_g_Page_06thm.jpg
2161666dc08ab29c806595f7b40656b3
52bf925fe7b81bfe916253995ebdc73d504c5cb6
24007 F20101113_AACKYA sabbioni_g_Page_24.QC.jpg
129dc1e069de718edb6ef2412af8d2c3
1d24bcd2ea4d97d839d8001da837ed005f5cb38b
23621 F20101113_AACKXN sabbioni_g_Page_17.QC.jpg
5cb05c121d43d50f420da24d6574ed1a
45365c7ebda5139db7d6ffc52a8410c8503c3f5a
11093 F20101113_AACKWY sabbioni_g_Page_07.QC.jpg
0b26ea65a70c89e447d1140f9740f90f
f1e30d79fe8b68358810ac74ae6fb54cd50c292c
7063 F20101113_AACKYB sabbioni_g_Page_24thm.jpg
a5a0f74dd183ea8d8a5946054f873904
9652ea939a3e5bb24548ea9560650a92dffe0438
6896 F20101113_AACKXO sabbioni_g_Page_17thm.jpg
5d8f81273ed765195b229cba14cbe16d
acf3b1771de4b21a8347ef871b6b20fc7cf1b2ae
3051 F20101113_AACKWZ sabbioni_g_Page_07thm.jpg
dd6471a0e752771f8a8409fcceedab6e
3884b79e5dd9979c6caeedadfa82117d6ae829b6
6794 F20101113_AACKYC sabbioni_g_Page_25thm.jpg
e850881c28c891c2c02e2c21f8bed591
332beef1138805f0d87b13fa63ae55efdb34940c
18928 F20101113_AACKXP sabbioni_g_Page_18.QC.jpg
1a9875fd2bef4866b9f552f763c6b6df
f629e38e34989b0a10b5337a9e1df8eacbe7ff76
23463 F20101113_AACKYD sabbioni_g_Page_26.QC.jpg
41551952f891750c81b297a34524ec28
cb08e9f2d8acce6b4011b0bf5a7fac5a2000e199
5513 F20101113_AACKXQ sabbioni_g_Page_18thm.jpg
751a08fcb21088863b4e0fc2aa8c1d39
9543db43412a5129b6bd238d02d456ee4eb950ed
6478 F20101113_AACKYE sabbioni_g_Page_26thm.jpg
8684c632a915275d05ab18e634cd0ce3
6211fdcc7b98256ac20fd64d1227b4844e6bea6a
23403 F20101113_AACKXR sabbioni_g_Page_19.QC.jpg
47f2bf7d2d730996cccd0e4aa0360b24
1a166e3506fba59acac4277c924e623e0aad232c
22847 F20101113_AACKYF sabbioni_g_Page_27.QC.jpg
1653627e818e0d1d40ac342d2c6b7faa
2b4c45ec07ba8fdd1a25421ca8690470549f14aa
6634 F20101113_AACKXS sabbioni_g_Page_19thm.jpg
2f9315fa4166f1b160decbe7f828de91
27d5b1f6b6840b4e0831091d626df8007791880d
6532 F20101113_AACKYG sabbioni_g_Page_27thm.jpg
49d7748950aa66ce98ae71acb7b518b7
49d42bfbc4500eb3c570b0e9f7794a8473c34cf1
23128 F20101113_AACKXT sabbioni_g_Page_20.QC.jpg
7942b32ee38ff53be3519ce2b3d85740
af5a1f2d5a6cb99ff3f7930ce1a891676eb044db
19396 F20101113_AACKYH sabbioni_g_Page_28.QC.jpg
98c119daf5f17f46f1e2b0ffb120913b
2bc677756b6e0d1f030d1367d014571d8da2b26e
6318 F20101113_AACKXU sabbioni_g_Page_20thm.jpg
0674a918fbcfaed1de015d5a5192980b
0556bc19cccdb75706edadfd4974cc6cca1b3775
19129 F20101113_AACKYI sabbioni_g_Page_29.QC.jpg
fe762fad588307fc325d56340bb7b63a
c16822a2b9f84691ce8899640e8bd2462f467d57
23715 F20101113_AACKYJ sabbioni_g_Page_30.QC.jpg
18c115541e2c22d26f1d0ccb1ab640a4
cc4addd62b2636f734c23158f4eae34d4ac1021b
6969 F20101113_AACKXV sabbioni_g_Page_21thm.jpg
12fd2703a157d34c16449a58dbaf2ed1
eab74a73f5bd2199d2b844fbbd18a5e0acbb58f3
6664 F20101113_AACKYK sabbioni_g_Page_30thm.jpg
b3276fd5dcfa4fa543575a315349c9b6
61cf57bed6de1a4e1328b953410091b024ffccb8
23622 F20101113_AACKXW sabbioni_g_Page_22.QC.jpg
3d138bcd6ec9aeff3e61edb3772008aa
a6fac16469d42a58ac0fba6a67f6f8d09a277208
25561 F20101113_AACKYL sabbioni_g_Page_31.QC.jpg
cfca6ecd356b84883f5d1b619506eabd
c74827b027336f1d9629d7ce6b826ab41fa772b4
6549 F20101113_AACKXX sabbioni_g_Page_22thm.jpg
c2b50371f2bcd39993afb7725587d4cc
0146e0e7b7d07e18a3ea6151f9a6059edc1468b8
21292 F20101113_AACKZA sabbioni_g_Page_39.QC.jpg
6b8ca0b447269d3d43ac76440e70d42d
5e5830ccffe99e703e39410bae7eebec8ca1ee20
6842 F20101113_AACKYM sabbioni_g_Page_31thm.jpg
0203f8f796a223db3589afd459403134
3e01b9d9df841007e6d9bd4fef4072117e306ce3
21874 F20101113_AACKXY sabbioni_g_Page_23.QC.jpg
1849af1e364af5d5d1dcd9b9ece17518
ab5e080db2a9b2086247e94fd60ef493f78dea5c
6137 F20101113_AACKZB sabbioni_g_Page_39thm.jpg
4515ebb71e1d2083ef79f50816b23c90
83888708eb3ee46b741d1fdcaa4b5f6441e6fd49
22691 F20101113_AACKYN sabbioni_g_Page_32.QC.jpg
a2b4abd0a8a8d20dd106475f7e141ecb
f23e0e988a5263b040df7000bc7cd11bdba9e01c
6150 F20101113_AACKXZ sabbioni_g_Page_23thm.jpg
440625d93fe5db496780e63f6896ad28
6639d6cfca39a460aa418ea7b972c2985aece860
24609 F20101113_AACKZC sabbioni_g_Page_40.QC.jpg
94af35109e74aeb22f5b31a611afe773
a34d0ea76cc3342fc61cc6754a0face444f755c7
6326 F20101113_AACKYO sabbioni_g_Page_32thm.jpg
19e3ef2585c5d98d69811104503f84b2
9b4573c32793ed717901608247e441a42a86bff6
6605 F20101113_AACKZD sabbioni_g_Page_40thm.jpg
c59f9a7a920aed7a7a8d0e79989b640c
551affee71ad21ea4462eef79cc3c41138491cdc
18992 F20101113_AACKYP sabbioni_g_Page_33.QC.jpg
54c94fd2dfd199b11a16d83792edf06c
cf13b3921f97963a1347d78d5e57089ad595130a
17007 F20101113_AACKZE sabbioni_g_Page_41.QC.jpg
fac2473622ec485391279ae19f406ce2
f2251e177a8d2deba7a09c10d96f8f0bc02ad044
5554 F20101113_AACKYQ sabbioni_g_Page_33thm.jpg
9b6bee5ef78b141304075e4a81b63e76
3ca23390d47d9f59a677a99bf71d7bdbce7b217b
5209 F20101113_AACKZF sabbioni_g_Page_41thm.jpg
cb8ddcf8c19cea34d58711d825110864
ac70a3b5c2009d7032c951802d9af6d7ab2aa6ea
22814 F20101113_AACKYR sabbioni_g_Page_34.QC.jpg
d11c433390ad6fa03b5a42f06b7c098d
4bea67f69ec53bda89af7c35be246b5a92e1b563
19387 F20101113_AACKZG sabbioni_g_Page_42.QC.jpg
0ecf496da7a9edc97a662fa14b1a42fc
091474cb7aa1933667514133a3baec424c55d391
6585 F20101113_AACKYS sabbioni_g_Page_34thm.jpg
07fcd5c263d3a94ee5436ee6b7f8eb6c
95c38f61f506d40deee575d403adb55c9dfa5ede
6105 F20101113_AACKZH sabbioni_g_Page_42thm.jpg
9ac85362cdf29d321fc4d42ccc65868c
a8e06308a601d8c0a3f6c47e47febdaf0750551e
15840 F20101113_AACKYT sabbioni_g_Page_35.QC.jpg
b778d7fc4c0a5b02b844f918e21a0c94
4add0d4b62b97a477a88939f9963b826cddc879d
24963 F20101113_AACKZI sabbioni_g_Page_43.QC.jpg
668e7bc30573fe1ce3de5d2e4b1a4d49
139b586910f16e32e541e2a8f7621b3d5c6d542a
4796 F20101113_AACKYU sabbioni_g_Page_35thm.jpg
50fcfd1a331f024104711b6249e7c0c3
ea67a7900e99c61af22cf76d2ee35dc544db9f45
6958 F20101113_AACKZJ sabbioni_g_Page_43thm.jpg
801b250291257db639cdbc6444820e9d
6323b9eed677d0bb76ce10c0d000a6520d49041f
6999 F20101113_AACKYV sabbioni_g_Page_36thm.jpg
81efcd7845cf9db6a81bc66742d92786
1f580fb8c923f623f2fd2c4af1a3053ced10249e
25566 F20101113_AACKZK sabbioni_g_Page_44.QC.jpg
4d4d721a785d4b21ac28e8fd71390e96
920f3c099556eda84563301aa48d44f7434ecaa3
14272 F20101113_AACKZL sabbioni_g_Page_45.QC.jpg
2c51205af90767980fd21e453b6f44b7
4118d6c17e8538a186541af0564dcaf204b2eb3d
24063 F20101113_AACKYW sabbioni_g_Page_37.QC.jpg
9c7ed36ccf616321bd824a0c469ea9ac
5c359b8f9884d5a0e99f4691b929fbdca967e0c2
4016 F20101113_AACKZM sabbioni_g_Page_45thm.jpg
8316aea522b935e2d65d1229fe72f9bb
2c71fb90d289e7af1dcad35ace32ce1c39d1710d
6823 F20101113_AACKYX sabbioni_g_Page_37thm.jpg
b76865349845aefe8dd971c5e2a0aec4
cd20aabc16102dd8dc42868eb264ba3260b99a11
7071 F20101113_AACKZN sabbioni_g_Page_46.QC.jpg
471e1f40d7f1aad1432d3c8876433ebc
31bcd54187353eff4dc9ad25d4450b08cb1c8f3a
20774 F20101113_AACKYY sabbioni_g_Page_38.QC.jpg
e8756aeb0755c4a26b63644dc5f2fa49
d4019228d4e00e8a582c11d01deb96f4bec34ea4
2292 F20101113_AACKZO sabbioni_g_Page_46thm.jpg
83ad14ad779137dd73901d7cbc5800f0
8281ab9282d3b0c0f817cb3ef0319500ee4c9424
6196 F20101113_AACKYZ sabbioni_g_Page_38thm.jpg
e25541dd1903731d7989cb7c2380e511
c26ef596f7d6ab1fa6dbe8583f62f04161d5fd23
8846 F20101113_AACKZP sabbioni_g_Page_47.QC.jpg
8553419f1afaf604616dbf1e8898f9fd
50dc40d821f2f6c118a6ba31f94d05fe305c6a05
2962 F20101113_AACKZQ sabbioni_g_Page_47thm.jpg
0fe1487d08a6160d9d55abae385631b8
d43e8b2889c1de34345fca9d8074dd186f201de7
21308 F20101113_AACKCI sabbioni_g_Page_46.jpg
00024349b962015bfa3df87964274570
fba96195cd6eecde3ac0fa1a4149eed6b688e56a
6922 F20101113_AACKZR sabbioni_g_Page_49.QC.jpg
873fbfe3dca658a4f4f745dff8e48d93
80004d5199f6a0d55e10edb148c4392aa37be7fe
53248 F20101113_AACKCJ sabbioni_g_Page_41.jpg
c89ab0c7bd6ae200f8126a6e438b5034
832714fafbc609200f426a305c16a90146451ad7
2302 F20101113_AACKZS sabbioni_g_Page_49thm.jpg
c8886316419d5123d1307c3f16b0305f
26a011bc70a0fb2d8ea6a968c9972b53dc303292
439 F20101113_AACKCK sabbioni_g_Page_01.txt
afcf2bef67b3dec36a19cdfde3b457b6
4eaa9dcd445684bfba653f5674320eb9f24ce0ff
8318 F20101113_AACKZT sabbioni_g_Page_51.QC.jpg
119e046a68e09d77ecd66bb6e94812a5
a6853f7bd92565b097f2c6d1bb06fa878beb704f
31848 F20101113_AACKDA sabbioni_g_Page_77.jpg
70ab351a309aa001da3970520ddb2c09
dff0133e2cf6cb8b55efb259eb682e04070c21dc
111612 F20101113_AACKCL sabbioni_g_Page_61.jp2
390e6c2566e65e0c54291b7fb200c1ce
71db4f6c067122b1ac679dc81319c84379b65b9c
2841 F20101113_AACKZU sabbioni_g_Page_51thm.jpg
18db7f02eda30b393e999203d7908228
5bafa53ddaaa55634dd30d83c19db5c2b23e1ea3
8371 F20101113_AACKCM sabbioni_g_Page_50.QC.jpg
5060094e9c7783b9d115e4c69bc1c70a
87769a3aa1c8e3957ea16623a853e2a4b709f471
2876 F20101113_AACKZV sabbioni_g_Page_52thm.jpg
d0ec9c280d250fd84fdac71fc9d4fe47
6971923e4635318b0802e466a9bbd078822eb0a6
25398 F20101113_AACKDB sabbioni_g_Page_11.QC.jpg
9c203a4eb3303d8c6152bf8210b46675
d54b5976cd3a3e1505045e62abe9c9e7b1b459d4
F20101113_AACKCN sabbioni_g_Page_84.tif
04f5adffefb2c27f1eb95747ce51f272
dd1e510064643e704c56def6290b5ccc0ac9bed9
6451 F20101113_AACKZW sabbioni_g_Page_53thm.jpg
b495cccd9634442b4f93c551cf9562cc
69b426938448fdad4d5b543a0e1bbe6d9e787f26
67469 F20101113_AACKDC sabbioni_g_Page_81.jp2
eef1712067e87e389ac652d06cd1d6ed
ee0e2b2bb2235c548ea5cc2df0e6f2b32a3edd8c
1092 F20101113_AACKCO sabbioni_g_Page_69.txt
5204e44fe0f957d30e46683d93598117
fa192cea1ec42ba31e1849f14abaa354092b626a
108555 F20101113_AACKDD sabbioni_g_Page_30.jp2
568412070cef2f66fcb5c1c16e665aa9
77f146ae512d5cc037a92f4812c0b9fdbe1bbfc4
36293 F20101113_AACKCP sabbioni_g_Page_72.jpg
09f1b00f2d099aeed42e18e1e4d4a8d1
d691c81843eab3448198f12166c507c6e2f1c867
25913 F20101113_AACKZX sabbioni_g_Page_55.QC.jpg
d58118e36c867d1cb3ff3ae809e26658
7e1ea8e3f26fdce6b06364bd28d81593dbfb858a
43847 F20101113_AACKDE sabbioni_g_Page_80.jpg
20796e3e607a9a784b3149af330ce049
cec409b844fc259d4484457abc0acdf7c39818f3
112240 F20101113_AACKCQ sabbioni_g_Page_36.jp2
efd6f2a9fa5b7af41b5454d3c1c5b06f
c53c1771aa6456a65400e3f1dcf14aa8e7dbde98
6861 F20101113_AACKZY sabbioni_g_Page_55thm.jpg
2a67cc00a6c83309e3c3d8aaa67bf2e9
ea74dda1e1761aef07e6cc29b1fe05c37fa72d6d
F20101113_AACKDF sabbioni_g_Page_78.txt
a2dde6171a6298b6b15339430000e1c6
b81d95ee3ec732cc94519826e5def11fbd533aa1
2180 F20101113_AACKCR sabbioni_g_Page_36.txt
b224f4a827251c0c18e4b976a2a727a7
7114954ec3a74c233fffd6ee7d2c9262176260b1
26430 F20101113_AACKZZ sabbioni_g_Page_56.QC.jpg
57670b34e0510b9eccf2c5ff58fa1ef1
52d32f3d2f43df9c7feceabc082fdb49a847fd58
F20101113_AACKDG sabbioni_g_Page_68.tif
75843c313251bab3f326397ae7c4813f
c86c77743bf665a92faee0252cfa2dc3f232d263
F20101113_AACKCS sabbioni_g_Page_57.tif
d9e16124addd3a2f5e4e5e7a411f09ac
87194b549c47d633e91fdeeb43ace7c302ab7baf
56130 F20101113_AACKDH sabbioni_g_Page_92.pro
822a42096d2b029c59971385c5570c87
3ae8aa124a199aaf695f373b15d93c3421311814
113343 F20101113_AACKCT sabbioni_g_Page_11.jp2
62fd0384c09d9dbc957a76d705287933
38cba131c890f4fb8e5fdf49af47b44235074a07
27481 F20101113_AACKDI sabbioni_g_Page_54.QC.jpg
2f9b59c94fb9f6a31ba04ef1bdf10a01
1221181f6fa346e3864f1f0ed87d30c27eb5e8a5
19818 F20101113_AACKCU sabbioni_g_Page_77.pro
faa4994a38bc01d5d3fd3e381b40a67b
3b33b43fa6fe7987bc408c196bc8dbff2b50a8dc
55178 F20101113_AACKDJ sabbioni_g_Page_36.pro
0f03de793f51f707a436436b075ac8fa
7dd7df7220f519491e424b2c931a208376c53fe1
5878 F20101113_AACKCV sabbioni_g_Page_28thm.jpg
4fdf3803156ae78929c3f752d0670277
b0a1a3a2fbc9195bee72b711208e6d6b591a92da
F20101113_AACKDK sabbioni_g_Page_37.tif
2a5a9152df16be4a2fab30e560b81913
fb7d47526e0df7e289f5815a9235713d012bbbbe
289098 F20101113_AACKCW sabbioni_g_Page_48.jp2
eec5cd8ba41a2b2bc63cbac0a4360d05
078f2e29c2d9678c1c2e51b460d13b0b1441dcd0
25949 F20101113_AACKDL sabbioni_g_Page_13.QC.jpg
ffd069287637c8f7ba9fa524b2b8ec7a
448a3bb47ec7e9ac69c6945bf2a1355db3be327d
2072 F20101113_AACKCX sabbioni_g_Page_24.txt
7c87b767071f7d0cc8ef4062d2fae4b0
6823d296872f0482f7592f5a7ebd85f666218938
19123 F20101113_AACKEA sabbioni_g_Page_70.pro
c36e45b17e0dbe94ceacb5a8ff14aabb
25461020a4d00eab97ed55ab7f55140aa206d026
28444 F20101113_AACKDM sabbioni_g_Page_80.pro
3900ecb11a4d4b15a51555c5baedee49
3b8ba4fea303aa4bf6f694a3e0a649ba57392360
74828 F20101113_AACKCY sabbioni_g_Page_30.jpg
4eeaa458d78729ac3298655b995bb2bd
0198db0970e3cfb0f986c7798054929213be4f74
24486 F20101113_AACKEB sabbioni_g_Page_36.QC.jpg
cc16c25028dd0b33a5fb592babf27050
4618cdeb750e0a4d1dc5e2ae687e217a25a52471
59439 F20101113_AACKDN sabbioni_g_Page_17.pro
e5a1563b5ac4dcbb98320b5c38bc39a0
8122a83167abcda60f62e7f5f0a9cdd8be93fd54
3970 F20101113_AACKCZ sabbioni_g_Page_62.txt
79d6165257bbc1cd47c38414983516be
1da0d3ea327ce7ca4f616356a56b8a52d291ecdd
6802 F20101113_AACKDO sabbioni_g_Page_74.QC.jpg
ee9d474335737d68006b1d86ee2ce9e5
d0b2396c012651949512471f6e28f537170dff5d
10926 F20101113_AACKEC sabbioni_g_Page_51.pro
5fe6868cb362cda5318b997f3c7c299a
b205b425b4569f9152d2bde1a4d70a6b3612364f
11977 F20101113_AACKDP sabbioni_g_Page_47.pro
cff7ab95d8c73273e655c0e0fa753632
ce65938f285300ee3cfe8a925fbc57960eb27dc9
6848 F20101113_AACKED sabbioni_g_Page_93thm.jpg
974be2c48f24e1fd3d2473c9934da52b
9f1a7dd43ed39a51fa9b7c2b4ddc7f7d3702bfd4
25271604 F20101113_AACKDQ sabbioni_g_Page_08.tif
e5ddf339dc1bd1f15a9ecc4addce5919
6abf2c305682bffb5b13e3af1e4d51a68626261c
F20101113_AACKEE sabbioni_g_Page_93.tif
aad58f573f20254678b824ed2970da19
8cad63ab018c043fd85574edfd80927e3fdf3676
57182 F20101113_AACKDR sabbioni_g_Page_61.pro
1edc284ac0d12c8e975f9dceac6d6416
35a38482e5cd167f284a8b79aa6970f0db48ea26
2058 F20101113_AACKEF sabbioni_g_Page_14.txt
21395d12130444c8aa249e6d9ec12326
82cf19f1ac1e5fa8fb24bbcfc18cdc4687a52c42
11086 F20101113_AACKDS sabbioni_g_Page_10.QC.jpg
d452a1e553c304772de25e9b530fb191
19be72dbc7d1a4129e5d677aaa03d00bb5e03fcb
F20101113_AACKEG sabbioni_g_Page_14.tif
2ce23d9c631be7572dcf1bfdbec41b39
d271aea65443381dcd9627443c3f86012ea01fe9
54658 F20101113_AACKDT sabbioni_g_Page_38.pro
b461987800411f840d27ddd203d73c08
9198c1a384ec4e6c65f33780663705afcb32db2a
F20101113_AACKEH sabbioni_g_Page_73.tif
cca89db4cbdd9437027119035d56d6cf
a3e5b458f98d1c3c8cc51fb0d9603a8548dc83bb
42469 F20101113_AACKDU sabbioni_g_Page_69.jpg
cd292c045c24554a6f7048b856fca896
e88a4282d3b0e41661c939cbd101d9a9af838411
19735 F20101113_AACKEI sabbioni_g_Page_74.jpg
16b14deabc53cbcaa6259589c5c0e726
0e8ef76db7ce0abea6ad0175693deda1139d89c1
10169 F20101113_AACKDV sabbioni_g_Page_75.QC.jpg
9f9af8331dc1924099b6e06ed4c4be54
1c0590f3ea26e312dfb4a59ad9c30cec5c353f5d
84416 F20101113_AACKEJ sabbioni_g_Page_92.jpg
00128363ca4a68c2aaf95f779e940302
c68caf30bea327bd6e4b88e3d9cc6e97f75893ea
F20101113_AACKDW sabbioni_g_Page_39.tif
ee7496ca1ee8d637147c01bf5d0e9e8f
52424f55206f0ccafeafc1a8b457b8d93fb41ffb
56605 F20101113_AACKEK sabbioni_g_Page_84.jp2
7f8c531520015ce8d68f626510d71299
a0040bb367442fb26e212afd73f69b6857aaf12b
71355 F20101113_AACKDX sabbioni_g_Page_22.jpg
04b493ba4c00367a6cddf84b7890eceb
5394e5c29a2be7fe67ac2578d1507c3e267c0f10
28380 F20101113_AACKFA sabbioni_g_Page_90.pro
169b6eecfef10a203a6e6c38586c78aa
2e8bbc3e6f89f26329c337cd80537446b7e373cb
F20101113_AACKEL sabbioni_g_Page_50thm.jpg
5ef5edaea02c763a4f4471cfbabe00be
d6edca0a205b1e9a24c60ca4ceadaa507688661e
217 F20101113_AACKDY sabbioni_g_Page_96.txt
03f15b5d06a3e069e91c49d5c45ef57b
a6e47f864596dd0a59efb3735f6d16f692ffb426
2670 F20101113_AACKFB sabbioni_g_Page_58.txt
878b80408783f88c87fb8caa3a1017e9
ce5374017362e9762b01eb771338177c2547918a
F20101113_AACKEM sabbioni_g_Page_91.tif
f999f6f3832b964b08c799ab1f626b15
b507704277f8096882a85ca4258099e27b39b665
25070 F20101113_AACKDZ sabbioni_g_Page_48.jpg
68cd08d684497e07912e32b3ababfb43
d40c08dd6876bfd7834d37414e70d2d95d900629
1130 F20101113_AACKFC sabbioni_g_Page_84.txt
29c44fe391372d6492b2261d3126720d
dbde9e08d202f3ee4634479d7f4a97b5c13e7826
74412 F20101113_AACKEN sabbioni_g_Page_16.jpg
2ab931d3bfda6b2fd66d10b41bce117d
5121c7b9bd8f7354f650f7b44c9c7498b669b793
7392 F20101113_AACKEO sabbioni_g_Page_91.QC.jpg
eebda8fa3dff654c1f361720137634ac
07fa9aec8b3230b2aa6b2e11f52628cc6714da89
116345 F20101113_AACKFD sabbioni_g_Page_43.jp2
8f1114909d539ffe465daa593c28b4e0
75308c14dc791b9c5ffd4f19f9b92b7ab822f0ab
F20101113_AACKEP sabbioni_g_Page_71.tif
7d96ca474567ca08f706693e3e96cfab
54cb80e8cd3f3eb53ef8a8ec6385a5302bf44bdc
1601 F20101113_AACKFE sabbioni_g_Page_87.txt
21628825ef2564f15e92992fe86754ad
236f660be2e06ef9a04398b51a58dcc50fcebbee
2371 F20101113_AACKEQ sabbioni_g_Page_01thm.jpg
1391896970e43f47bdd1e0e8955974e8
b7ceb9fc4fea970abfbab31706d10a0211a59136
52045 F20101113_AACKFF sabbioni_g_Page_79.jpg
0129556c07f44945e67efb5a1e6eade1
ebd5d320d5f5e05a0ec868195315d09ced9c1a69
899289 F20101113_AACKER sabbioni_g_Page_07.jp2
efbd3f68faf8d080a08278a8f23e5a1f
4a60de407a0e5ab0a4e4849832e3cd36a8185323
10731 F20101113_AACKFG sabbioni_g_Page_48.pro
1b172b4c27ba8c0f7712ecf7da0a109c
856a83046e5e83987f048700b463da4191a0bdd0
4738 F20101113_AACKES sabbioni_g_Page_96.QC.jpg
d9357f945bce2b6ae265ee3d79b94b5b
8d164cae261eb021703c66339e2dc349296b331a
75576 F20101113_AACKFH sabbioni_g_Page_34.jpg
6c0346b41baa9287fb6d8f968aaff09e
060e3560b02dcf7a8eab8ae97e8494f88e5f8b8d
23397 F20101113_AACKET sabbioni_g_Page_95.QC.jpg
916c3e84967bfe7ab3c586ba9747d00d
deda2684a6b49444108c89805a88713466f58d07
F20101113_AACKFI sabbioni_g_Page_60.tif
75a41102e93516e8d58d62b7e836a1de
b5bbb5570e72b1ab2d832442d5a61e2f2f014323
85878 F20101113_AACKEU sabbioni_g_Page_18.jp2
dc7137a2c4acbd425f15c7952f946987
5733d0d1e04df2fb15bb5059a80f8a21481f10b3
1051973 F20101113_AACKFJ sabbioni_g_Page_58.jp2
cbe628f86b9e63ef286bd7969d6837ab
1bb8303a17bad97099db2757f2417c940a682ecf
F20101113_AACKEV sabbioni_g_Page_15thm.jpg
5faa6cee268af0c4970cf0fe5fd25cca
64c89557e5d4efbbced0068a5dc4dc950d61ce88
53486 F20101113_AACKFK sabbioni_g_Page_16.pro
c3dfce61c0b9a0329d45b206e39540e1
1401ddf76d2fc078ad435205e1de55048ea21294
7926 F20101113_AACKEW sabbioni_g_Page_48.QC.jpg
7d6a1a3a5c5d50f645c2ef2bc02decfa
3bca63159f50527296e4e4e1fe2666b7cff9345b
38176 F20101113_AACKGA sabbioni_g_Page_70.jp2
437060d6ff68498e87b9c5542313ea61
9dae0f87c3215e3b954913bb794434a5e0befbdc
85417 F20101113_AACKFL sabbioni_g_Page_54.jpg
9007ab923f1ed9de86f398f64996cf1b
e0a9bae1400b3157a21420eae7ff5aeca1f226d7
599 F20101113_AACKEX sabbioni_g_Page_48.txt
fb6c82298e81c99b87a572cb98875c60
23e9edccd2ae01925270f4233bad178ebf247fa0
3847 F20101113_AACKGB sabbioni_g_Page_03.QC.jpg
088d0993c5599b8afc9977927cd625f3
cd86626cc34855e016eb6a5e20faacd74d0e679c
24179 F20101113_AACKFM sabbioni_g_Page_53.QC.jpg
2841576761848d9440d3367ba3467af3
a5d0dc0a70cd1d0bdbb69179bac02c5b9ba13c81
4434 F20101113_AACKEY sabbioni_g_Page_88thm.jpg
802fba3796c830faabc685bcc11e95dd
5c94c54b086c8bc8bb881d28ab27a4f32e994b0c
1807 F20101113_AACKGC sabbioni_g_Page_28.txt
362849482ba0b22ae4d1d9775ee6381b
7f2c5f9eecda3423b9b7795768e3bb2fce66cd81
2177 F20101113_AACKFN sabbioni_g_Page_67.txt
e5d8ddf3f206a0da7d98f3620cb5a0ab
ae73bf5f79710b122279d11ab6d99e7bc1b56a63
49892 F20101113_AACKEZ sabbioni_g_Page_53.pro
789d1d8c57d3469570ab459ed08cf975
0254a09617cb4c9be8f319c21db042622df2c9c3
F20101113_AACKGD sabbioni_g_Page_08thm.jpg
c9f13d13afee8173aa12dc1bf339176d
e5dc04fc3d947eb87c50eb7eac99be9ba0338723
F20101113_AACKFO sabbioni_g_Page_10.tif
ad599e3827995decdbdf1b5ffde1aab9
d1cb8c7b6cd7ddf2a870af7cbeacc0637e76b2d5
97657 F20101113_AACKFP sabbioni_g_Page_57.jp2
df5356e19003b41b058a7303c38a1499
dbacc980d7b9fc2fdacaae7636a53cfa191bf2e6
61583 F20101113_AACKGE sabbioni_g_Page_88.jp2
7e66302134d8ef4f52435013c7652c44
4a88cb44bcfe4deb301d7d425b052ab7b25a12a7
F20101113_AACKFQ sabbioni_g_Page_86.tif
5a052aede7d1fa631da13ba7fff66f36
ac2c8538a1a5ba5e73fbe43596c213441facf30b
107570 F20101113_AACKGF sabbioni_g_Page_62.jpg
e2a910c93ea3731e5f89fcff7f068a74
77bc7dc9c9f354422e60040ef8698fc3f87cf89d
50135 F20101113_AACKFR sabbioni_g_Page_40.pro
9b3ad74c605a5423760cc2045c723386
0afc9cc2013a8b3b41409bf27de2275749298118
25183 F20101113_AACKGG sabbioni_g_Page_67.QC.jpg
06aa5c324f4501091780286154f03775
afe0eceeef0bb41fce28823ba500c0a24fec043e
23652 F20101113_AACKFS sabbioni_g_Page_25.QC.jpg
56da1b60a479daad4d227d8e891c36d6
f0fba0a760663da3aa73fb126f679cc663507dd7
2447 F20101113_AACKGH sabbioni_g_Page_19.txt
f1adeb60aeb8e472210b3f200da64fd1
51aca795e802e43e9d04483081f85cd27e1e0374
F20101113_AACKFT sabbioni_g_Page_77.tif
718b5aa8e2d271c14c332a2b82b59c33
e016f73f8e0eef786396d9a2c81bc51c5a74e2ea
F20101113_AACKGI sabbioni_g_Page_46.tif
7fd95dc28b5836092ab63cf9367f86eb
3a6d42d6658f07e098ba757a54a362c7eb560186
F20101113_AACKFU sabbioni_g_Page_47.tif
f6f7e071e8a210bf5114aff1b128e62d
a7cf7fda464fcf23eec0c3ed28b424bb01d92d7e
117471 F20101113_AACKGJ sabbioni_g_Page_60.jp2
587e1e0b87fa85e447458640647bef0a
72cd18746ec1e5405d59fea0dfe846bf939dd896
1087 F20101113_AACKFV sabbioni_g_Page_45.txt
f75ea32a2abff3aa6839c14339a2bfed
54c04255a61035a489f1048c1da2024db37a7f44
25239 F20101113_AACKGK sabbioni_g_Page_21.QC.jpg
742131c308041b5e0766c4fe95c12f38
252b00c2368dc79a911b7589a602e8eff64b3f7a
F20101113_AACKFW sabbioni_g_Page_85.tif
c77e91ff5707fd7b99bee09e71104023
d0dee5983594b0c8b65d82319a1e9ad0a595571e
13704 F20101113_AACKGL sabbioni_g_Page_49.pro
0394ba17db4f186c022afe988b4788f3
83cdef7b4e72f4f27dd87fc10cff171a5604af28
F20101113_AACKFX sabbioni_g_Page_49.tif
8a276ee485f6f1aac7be0ed568aa6bfa
7d0026d3bd1f091ceca4a88ff6790cc059953ed7
60326 F20101113_AACKHA sabbioni_g_Page_93.pro
b9f62746cd6404fc6c5ecb11fd047a80
858a3691d34c51236e9b3cbe78bb32caae61f366
1078 F20101113_AACKGM sabbioni_g_Page_72.txt
7520772593ed77a09d2c9e8386478ecb
491087fface87f2f8325103c4994b429a1e4cdb5
F20101113_AACKFY sabbioni_g_Page_92.tif
0249cdbcabe693b54060812eaa8407ae
b218f84eadd17d5822b5ce41fff7dc7958e4472b
F20101113_AACKHB sabbioni_g_Page_26.tif
278096d587fa8c5d3d6644818078cf48
5409e6c7940b23c77aaf3516d68fce7b9fec3516
27767 F20101113_AACKGN sabbioni_g_Page_73.jpg
023e7d7a4e9f074a09f88523237bcc12
eb921eb832ad49811661e2c7f021de70798f0963
1511 F20101113_AACKFZ sabbioni_g_Page_79.txt
0ba656adc4a6048785ea3174ee8a800d
c66b3418a459b12c6cf6bea4d07a4ebb4626f9de
25251 F20101113_AACKHC sabbioni_g_Page_50.jpg
a6a171d935f00c5893e857a0fb7c7d73
cd6b003d4dd41380462f4333f2bccfd94a4c1d75
836 F20101113_AACKGO sabbioni_g_Page_10.txt
a7aa01e1dd38f62cc9e3b32972f6c91f
5a5f986bb50b319b8633c8f460c64932ef4aa6f8
93363 F20101113_AACKHD sabbioni_g_Page_58.jpg
f7a97429a9cd9591d9064deeb308d481
195c247ce8d37eddb7af9d5372a660e3042bf9b2
3032 F20101113_AACKGP sabbioni_g_Page_03.pro
ceaed6ab23a06699ddbab91079b6409e
c5d5198bf294fbb9c90f55fdaf617c2431a31cbd
2078 F20101113_AACKHE sabbioni_g_Page_53.txt
e6b5fc7a322f966a6b35949ba34a8218
d9e0800f80484e95cb4b04b8bed8fa67b8468d5d
2959 F20101113_AACKGQ sabbioni_g_Page_73thm.jpg
59a5aab447cf2ddc7bf2d256a4e8de26
d158219dadd798b7b7cea5f34c13a6209333555f
17684 F20101113_AACKGR sabbioni_g_Page_90.QC.jpg
5bf2d1f778ab197c78530328d544ee94
3a3793a0fcfbfc608b8dc6cfaaeb93022c609ee1
27608 F20101113_AACKHF sabbioni_g_Page_47.jpg
f96891a71eaedc1eeb2dea549997c81f
47a903cdff1db262c112a90bd42704733b5cc7eb
F20101113_AACKGS sabbioni_g_Page_80.tif
d93ebad68960a0c42246efd917619453
603fdb3b06a407ecfb025cc83c0dfb9011ffac3e
126840 F20101113_AACKHG sabbioni_g_Page_94.jp2
bb096c3b54854e9bb98eb79d802ab4a7
cc0f8a9dcb1c918a34728d3948a36abcfd45255a
25087 F20101113_AACKGT sabbioni_g_Page_46.jp2
d85ed67ca16c1b092fc2a943edbed582
dbbcc9d1bb4b7b7b77d345df69a471da7b3e45eb
13607 F20101113_AACKHH sabbioni_g_Page_88.QC.jpg
7bcd816567d4cdb7d7708e1357ea37fc
2796110bcfd3bcf2fb33e9a47bedc78f4266dd7b
565 F20101113_AACKHI sabbioni_g_Page_91.txt
3a3812a9506841af086dfba4277a04d2
37ffe512c6a9fd7319674b9bda0fd587cd44ed97
52684 F20101113_AACKGU sabbioni_g_Page_24.pro
f21c56030d4403924f4224796c0672bf
8e5e8d056537f52d0e89f3cdcb0fcaead94b6d4e
F20101113_AACKHJ sabbioni_g_Page_83.tif
ec7f00849fc3e8f890e051c18ce8e66f
c0e94056c88488cc17ac772b205b8e4f3632ecef
5330 F20101113_AACKGV sabbioni_g_Page_90thm.jpg
ff6d780fe39e5197f84cada588820d7b
e6ff14c2d4853e9353da5785b98a92f25ba24ee0
50890 F20101113_AACKHK sabbioni_g_Page_25.pro
22e1df5f4737273789fea5cb6dc6cc65
3282b539db6680cf89de9bf8e841b48589cdd6df
24746 F20101113_AACKGW sabbioni_g_Page_92.QC.jpg
f1586b951e09e0b795712f851a9f3c3b
83ec355705c5d4dee0185c46e1924d35db729aa1
5038 F20101113_AACKIA sabbioni_g_Page_81thm.jpg
f002669f6ec32b0ea3aba37d120cb8c8
cac2ef95342f238ac26b85798064bf3b0a82292e
108985 F20101113_AACKHL sabbioni_g_Page_14.jp2
6ff5d4b0dac8f84ca234e270e40a1ee1
20971ea540e222d5dc3e574acba601e599759511
F20101113_AACKGX sabbioni_g_Page_87.tif
f8b88ee2c3b5fa986d49a0335594dc36
71737242262392a51d3c00471038ca0b56ef7ae7
F20101113_AACKIB sabbioni_g_Page_23.tif
2d36f5cc37b9bd7afcb4a0d0a58e962a
5c856beb9bd24fd6da4839c58e5c55c3bcdf6b8f
5730 F20101113_AACKHM sabbioni_g_Page_29thm.jpg
1387140411896c11c2cb1ab1ae34918d
ca03042b53501a1ec4f04fdc45f0a2d4cde40717
F20101113_AACKGY sabbioni_g_Page_30.tif
25377c27489e849f22b3210c0523f169
08115a621e8805a7ac039261a2f60cad7abe9919
6968 F20101113_AACKIC sabbioni_g_Page_44thm.jpg
7af5e604a4bc9c64eb9fb897aa2fe43c
653840ba14f65c9065b0293a6582ea8b4d9927f9
95184 F20101113_AACKHN sabbioni_g_Page_39.jp2
c8a0eb725fb1a5b2efcf30e617816718
15d7b538c67a2cd0e6daaafbaacd14e4d853025f
2147 F20101113_AACKGZ sabbioni_g_Page_44.txt
8270c3e908bc44d9b35dc5396a6605bc
2c9a26bd65fbe46a2aadb452d365e9fd30051ab0
23479 F20101113_AACKID sabbioni_g_Page_60.QC.jpg
e8173e677f57b80a9a82b8f21960067f
809e9ad5c6909aa61af9875f298651b3513b0b47
8609 F20101113_AACKHO sabbioni_g_Page_52.QC.jpg
7075c4440395c56e96c8050827b7478a
84c5600ebbe1f0087d399b20a59ae029da21ac01
F20101113_AACKIE sabbioni_g_Page_97.tif
643f749f40384387daa1b8caa961601e
f293e55effed3e2c46e75d5f216034baf16e59d3
53499 F20101113_AACKHP sabbioni_g_Page_31.pro
daa1786a09775a34f3efb166db4fcc63
6c11ed9d0f1038765147ded92e3a0c5af15aa757
59371 F20101113_AACKIF sabbioni_g_Page_18.jpg
0bcd53e36bec4b94bee676d6f65844be
7e4cf14d27dd3020199c1da3a01a19c74f1c6dd1
2706 F20101113_AACKHQ sabbioni_g_Page_48thm.jpg
d6453237e4163b180eb83bf8039eec83
4ac787a8a4802aad2aa538f0fdc9b0d908c4cd9e
75372 F20101113_AACKHR sabbioni_g_Page_12.jpg
ace32ef407d8f0d1ba3a0cce78a4043b
dc6aa4975a30f3219dd06dbb98f0f9f678f3a42a
2278 F20101113_AACKIG sabbioni_g_Page_38.txt
26c714e629ae5abc4d24abf9f2e7a6be
a9cd1d4175f3729b78cadffd5da1227002e75374
48303 F20101113_AACKHS sabbioni_g_Page_39.pro
e0af9e198fc78a064e8b17e52d565c24
37520e6284518d547e411908635f202f1ec4d152
7498 F20101113_AACKIH sabbioni_g_Page_54thm.jpg
fab73f5a54f7a7dfc094e7184393391d
de61c256b4ddcdc8d5208c042b04f2a905ebf19e
F20101113_AACKHT sabbioni_g_Page_31.tif
b844e2b5be753b060b719e24db2b501f
5bfb5195e8d63da33df71023f7abec393efd55fe
97908 F20101113_AACKII sabbioni_g_Page_09.jp2
bced6d3a372bd99007b43ba0d089a45a
382687e5beef713595b12c9fc27e23d5f256327c
13991 F20101113_AACKHU sabbioni_g_Page_69.QC.jpg
4fe2f50ea9e9a0f76a711f29c42f1370
72ae4bbfa1089b18044e00939db77eff41c89533
F20101113_AACKIJ sabbioni_g_Page_65.txt
6e81756723a3018de25105e5ec3c951e
b71fee7abc8362340ffe97a29c226fd7d01cbce4
F20101113_AACKHV sabbioni_g_Page_04.tif
d254de189b76eab16c01f1b151cfddac
e44fb8e553389facf89a7e3d3698550c420f4501
146910 F20101113_AACKIK UFE0021071_00001.xml
716e10b4105842d5c2ea94ede02f9bdd
7825d81ffc8cacc18c4d76e01a417d32bf2db473
67588 F20101113_AACKHW sabbioni_g_Page_38.jpg
a8beeb8a88e7edcad1596f4bff0a06b5
379e94663ecb5f51fe7349d97064c86b0a406756
23780 F20101113_AACKHX sabbioni_g_Page_01.jpg
f9cab9f3ac17f3b73bcab7a53e0961c2
2a8f491db4f0d1f6058cb06428862ad080aebad0
73862 F20101113_AACKJA sabbioni_g_Page_17.jpg
2ebc18f3ec93f412ede8996c521fb568
77a4a80fe65fcd7c9a87bf8a598d9c1ab70333fa
67172 F20101113_AACKHY sabbioni_g_Page_97.jpg
d5b3e17cbc34a73c3886ad7a6755f704
bf33636ed56d2782ed33324f067e7dea27fcc77d
75408 F20101113_AACKJB sabbioni_g_Page_19.jpg
fae0b3c0b62f5e94a6e6df7cd1ac3cb3
cd2f49209d14f762547ba51b7db0342dbf643486
10619 F20101113_AACKIN sabbioni_g_Page_02.jpg
8901106e665bd1fd2353c350ecad5e31
3077f4089b7ff2a5ba1c484dcfe3ecfe4e9c3685
F20101113_AACKHZ sabbioni_g_Page_32.tif
900bb9ef67c4c3624dae9cc861ff1a8d
d32ec5def9a981837b5851d4601af7685e77e28c
71718 F20101113_AACKJC sabbioni_g_Page_20.jpg
cf789cd85f5f3795bc8eae2f814d23e6
f0a387430ece0c11f1e441be1779872d71bdb578
11893 F20101113_AACKIO sabbioni_g_Page_03.jpg
6b4a688547a083b5db4409fb0d831109
a0de2bf78df7fced2d77091cb87c3e2aa0f2f602
80642 F20101113_AACKJD sabbioni_g_Page_21.jpg
6751666b84fdee0fb28a02b3f6bee4f3
3044e28d7d14a1b1f509c0455817461e723532dc
15291 F20101113_AACKIP sabbioni_g_Page_04.jpg
5fb2499dd26f5803a26a9206b103bf8a
1667eb993ec7aa80628e36dd1aca7d5e023a3046
63819 F20101113_AACKJE sabbioni_g_Page_23.jpg
8bf29becf52ec985319c7f7fdb8d3947
6c6e7d1fad5e87f15b6fa2bdf9f7f8e418081c78
83874 F20101113_AACKIQ sabbioni_g_Page_05.jpg
d74d889ac33c435bca51436a72abc20d
48591738a31655056da3fa059d061a595930b9fe
72279 F20101113_AACKJF sabbioni_g_Page_24.jpg
55acafdb864fd1acc96d6a8b51b362dd
7c6472694ba7d203d95b6a0796dfe66742f3ed80
11722 F20101113_AACKIR sabbioni_g_Page_06.jpg
75d6ad462cef985a82ce0b0ab56d7c99
952c5141a1ebcf8ffa1a4012eb41305cfdd58e9b
71936 F20101113_AACKJG sabbioni_g_Page_25.jpg
15682e67c2fc9c0f1a7e8419acee6fd7
c6b7d717a9769f6bbef35afd6e88decdfefa1c0e
37025 F20101113_AACKIS sabbioni_g_Page_07.jpg
017ced91431f54c4d86b0cc43e63bc35
46318faf9e1fbd4aba6acfd517297dd7236fc184
35143 F20101113_AACKIT sabbioni_g_Page_08.jpg
7e613479aa2be4eb994e816c186b6431
bb1d0cf1cb66d974858bb3258682ccafb7ab7c7f
68954 F20101113_AACKJH sabbioni_g_Page_26.jpg
641af8713ad5d7941536a4a8e3d88933
029c5e10aa2c8a93c82a26cd29ed21b08db44167
67600 F20101113_AACKIU sabbioni_g_Page_09.jpg
22a80f96be568a62982abfedb2c6140f
67ca8700480fc3b5a3333924d5d07eb60b572517
70735 F20101113_AACKJI sabbioni_g_Page_27.jpg
15ec38009778f7da97054e8035f1aaee
4cd068f90252b1ac770241cb79068ba5382b9d7a
34196 F20101113_AACKIV sabbioni_g_Page_10.jpg
5c93e14c14746d8468935d20bf6f293d
199aee1d706babb008b7221f5e8bb46b23b57e17
60728 F20101113_AACKJJ sabbioni_g_Page_28.jpg
167828ad2ca2068e799f9c48a54132cc
3d9f06540a3052e64e52fc1fc82ee5b795efc468
76674 F20101113_AACKIW sabbioni_g_Page_11.jpg
bd75b6b92e4cc500a97f905291eaaf9a
0caed806a70c4dd72b3e8ea70b99052127a5a7fc
61406 F20101113_AACKJK sabbioni_g_Page_29.jpg
0b47086da3bf335219a23308a1434d30
ff9d2baf7ec761e7d768b51dbaaa631f539171b8
79940 F20101113_AACKIX sabbioni_g_Page_13.jpg
f3f71c5b4b3bc19294514ec8fe208396
fa8553cb74ba2c5f2a5e25b123351664435ab193
71638 F20101113_AACKKA sabbioni_g_Page_53.jpg
9836fd45a4576ebc2d87713a7ec8d6f9
72f831dca487700caa01e247998d216ad4047f7c
75494 F20101113_AACKJL sabbioni_g_Page_31.jpg
f2bb1cb7a3f66ba4f0607947364176df
35af046760ef63e8d81df9528173f6993707c2ad
72524 F20101113_AACKIY sabbioni_g_Page_14.jpg
c459c39e87fcf020446836b59dcf3768
699de1b05895b8bf829be5490fac3e31d44583e3
78415 F20101113_AACKKB sabbioni_g_Page_55.jpg
3bc8841d69f60d7723de5e69d3f4f7f1
15442124825a2d8fe1a0039580a67b9397d244b7
71998 F20101113_AACKJM sabbioni_g_Page_32.jpg
565bfaabaf2b15e4f188b374ab5742b3
1d664fc8880fc4651261c580528787c14b3eadf0
74444 F20101113_AACKIZ sabbioni_g_Page_15.jpg
f5b745825d6fa4c916006dd55246d1a1
d63d56a39cf8069686ffa4663a4c09d757bfc45a
81943 F20101113_AACKKC sabbioni_g_Page_56.jpg
c9ab65c5b2927ef8c0ddf57fbf9073bc
37c1ce27466ceb568007a21444923499db20be2a
57347 F20101113_AACKJN sabbioni_g_Page_33.jpg
193374d0713823205ff30e973c52cbaf
acc77dadeb0e90abbbb44b80f24735e16e07639f
64467 F20101113_AACKKD sabbioni_g_Page_57.jpg
308446159961e82656cfdc31a70fbce0
9daeb2dfe88e83118b2733ca023f08ae321e0ebf
47776 F20101113_AACKJO sabbioni_g_Page_35.jpg
d45e3f999defc51ea7f5fbf53b63c1f6
0daa702820214e58f8ef9f2c33dc31434d8fe6d5
66312 F20101113_AACKKE sabbioni_g_Page_59.jpg
70852e2c42ff379080e177d0cf8cdecb
020220d0643de0b07292e890796c4fc90552f7b6
74470 F20101113_AACKJP sabbioni_g_Page_36.jpg
57b2938b23ca8ea30d220fb3272a48eb
e5a71697f9c0d40ee1bae647cfb61e3b4652c50e
78591 F20101113_AACKKF sabbioni_g_Page_60.jpg
25836895f54391452e13f7645bca068a
9a94ab25ba8a21cb728232e8ef6a4fd501578113
77179 F20101113_AACKJQ sabbioni_g_Page_37.jpg
4929f8f49074c6f7925d507cc04381cd
4f89b4eddd7e29b1462d43400c40992b457e6d55
72689 F20101113_AACKKG sabbioni_g_Page_61.jpg
6736ef86a56a89f450c2161c6322f1ac
bbefed10b08095d8b0fb7615d8ddad50bef47e6b
65825 F20101113_AACKJR sabbioni_g_Page_39.jpg
66001c88eecb7c92ba5bf3aa34490a32
b3e2278afa77ddb13fd9bc9eb8ae02f594a0293b
106112 F20101113_AACKKH sabbioni_g_Page_63.jpg
9a12881dc16889b4b80d43116cb78dab
696cc2b62720a13e6e3fbecb6183ea7fd6584e5f
71373 F20101113_AACKJS sabbioni_g_Page_40.jpg
d5988ad3f48ad8beef3e65356602041d
4c4e86e713e181a1a68ac35ee39b4b55d5b0c05f
58388 F20101113_AACKJT sabbioni_g_Page_42.jpg
78ec9005777d1287ddf787bad0c52912
2429e0e80d79530cd79634725d33fb28db51df93
91352 F20101113_AACKKI sabbioni_g_Page_64.jpg
7d2a6b189a052af7dcc622ebfc1227d2
5cf7498db7ab20f073e9fe03d2df627893d47780
78742 F20101113_AACKJU sabbioni_g_Page_43.jpg
2a6d0bad443f3328af5f863e96232830
02e2e23650fe059d3a80a10badeacf60f90c4ba5
74391 F20101113_AACKKJ sabbioni_g_Page_65.jpg
dbb3b955c07c26b09b92570f037ac318
426b2eea5fee7abdc130603708a04bc08f413bfd
75726 F20101113_AACKJV sabbioni_g_Page_44.jpg
96075e3e9c29ddbd7de241297c7386f6
b9cb00c0066e886cf0866645e12fe872d7ec1260
79734 F20101113_AACKKK sabbioni_g_Page_66.jpg
fc2808701045b0ff22d7b2219626f08b
d591b9f000b9f35358a40dc8fb4a057f2f6748d3
42959 F20101113_AACKJW sabbioni_g_Page_45.jpg
c314da439b02ac00a0348ab7f97b3385
615ab5cacd3b68b1a95e2fe0720ab9e85973273c
45873 F20101113_AACKLA sabbioni_g_Page_89.jpg
6b123b8e475d4d7ed4adc19f70694551
e4016b1e5fe73eee39f3c336a24fb682bea4e9a5
76220 F20101113_AACKKL sabbioni_g_Page_67.jpg
9c38997e47ea17d3bbe849f6b44f2217
9e59bc662678d08dc01cb13bbeac89f06710d1b3
21139 F20101113_AACKJX sabbioni_g_Page_49.jpg
74bc71abc852489dc731cbe1c2fff6d4
e10979fb3f623a9208932b76d45c1c86908dafdd
54465 F20101113_AACKLB sabbioni_g_Page_90.jpg
c6e5a60c9e046fbaddd94842b870b85f
e25e3846aa5d67a1e4d50716c33d99376b84261d
77102 F20101113_AACKKM sabbioni_g_Page_68.jpg
177ec835a7132c1596d9264ccc7f769d
3b2a6138903ac6082a1d3c8438896354747c8126
26298 F20101113_AACKJY sabbioni_g_Page_51.jpg
45cd93b25bf63dc363df2255df4d70b7
97d5852fd49347df3642f722e2788f9923dcb4e1
21514 F20101113_AACKLC sabbioni_g_Page_91.jpg
e96b7156adde6adbb6305c7092ced839
ff50c98703a9d4c6a47f2ec39e79217e12597d11
29000 F20101113_AACKKN sabbioni_g_Page_70.jpg
2fee157214150ca850df9d73ca038ebb
c1e8c91dbf51d3e8ebc92da2bcc8158cbada4ca9
27073 F20101113_AACKJZ sabbioni_g_Page_52.jpg
87eb281bdd7d2df89f591b99bdde26b2
c71219cac2a3c88031237fae360f9635658cb5e2
82857 F20101113_AACKLD sabbioni_g_Page_93.jpg
3e8e1309e9e6c88c138658be5c61493a
e5150f731d1c9a060b42458a03ea3d42fe1da15c
46766 F20101113_AACKKO sabbioni_g_Page_71.jpg
51d6a66d538f4387ad9ac11dfdc6916c
5eb3925913d24516730b3c4d204745c1f38dad65
84173 F20101113_AACKLE sabbioni_g_Page_94.jpg
ebe4485dd15168ab6bf7e4d35717ade9
aaee118f7268dfa81f1d206fb0bab54131640681
31622 F20101113_AACKKP sabbioni_g_Page_75.jpg
e7e10ff8986fc7d30f6a0476d74526fe
d746b4607abc69d07c3232bd0d58959494d87aca
75720 F20101113_AACKLF sabbioni_g_Page_95.jpg
cf98c824c3a41e77ceab9d669a62299b
0b1864a191918db0039b1f8c2ebaf8bd2f82a1a6
28132 F20101113_AACKKQ sabbioni_g_Page_76.jpg
640fbcecf8ef76898c0aca7f88128fe9
89fc5d163486465796057fd1f9c964567c334b75
15100 F20101113_AACKLG sabbioni_g_Page_96.jpg
12c2f4451f68ffd5e0505864dd5f7752
a19258be3672c1a139e7e11c684142ea8ecc8ac1
27469 F20101113_AACKKR sabbioni_g_Page_78.jpg
0d53d98ef6c232eb81a8eb51f086634b
91dd2f139ea4eeefc585d67dd5f1e1c592ca6f3e
24957 F20101113_AACKLH sabbioni_g_Page_01.jp2
6678a0c43fa4000227c35ddbfd3af53a
6f82b3114c3f8631f17327c1ff5b6bec5b8ec9e8
49036 F20101113_AACKKS sabbioni_g_Page_81.jpg
a146c1d81e5558fc113dd650980a6d37
366a2320632ac9b379192be733f4182d628145c6
6540 F20101113_AACKLI sabbioni_g_Page_02.jp2
c03c1e3f2fa14493426cf62514c261c8
bbf94484ed988bcc4f105c97b4113f5d8c55f469
38898 F20101113_AACKKT sabbioni_g_Page_82.jpg
c4b9ce9225124291432d1952349b4546
52ead6e66445a90d398681e7bb75eba77d98dfd0
49473 F20101113_AACKKU sabbioni_g_Page_83.jpg
8fff622f3b9c1e733c8458f662dd9610
d814fcffaf3ba3f119c0950a03c8c7bda1e40878
9649 F20101113_AACKLJ sabbioni_g_Page_03.jp2
de8b2675ebc1b9cabff098c9537667dd
fe59c4c4069e0a0c2c2ba33b8b1fa30d0b1327cc
40810 F20101113_AACKKV sabbioni_g_Page_84.jpg
103768f45b10a21a919d59b1b783baab
ec5751ec20688f0e3111eec6dd8bec2be2ccf1ee
14335 F20101113_AACKLK sabbioni_g_Page_04.jp2
e1d61ed1df81b4239d163a2d6d7f939d
d5433e890961c994a24208c0d552a5d1d60520f6
40393 F20101113_AACKKW sabbioni_g_Page_85.jpg
fc9bc83209d96aa65622c731171de938
5c2019dc4b5972fc4fe72b30b651159078532c39
1051978 F20101113_AACKLL sabbioni_g_Page_05.jp2
4dbd22c3f55f1a09fd6096aed1ea96cd
039576d1cb155af009b76e10c03e48a5ffe14165
44678 F20101113_AACKKX sabbioni_g_Page_86.jpg
99f9f3f1e236ad4aa874d7e5f7927b5a
df0b11e1ce5d5e08cb34e40e68b901adb1ce6440
105885 F20101113_AACKMA sabbioni_g_Page_25.jp2
41c59aedc74c718f4ea2a0d028d7f513
1f6d495967e23aa486f357b17f7213af31b13b24
90453 F20101113_AACKLM sabbioni_g_Page_06.jp2
7025ed12499b001fa797d5160e511673
2ec583350eed63694546f0ff010734c0c54b8949
46621 F20101113_AACKKY sabbioni_g_Page_87.jpg
a832ee5317374d4b1435e7ed10f00298
3321be87f3b897256cdc545f78a733d5aea732b6
101752 F20101113_AACKMB sabbioni_g_Page_26.jp2
08c6b523b11a4d02eae6951853bd62e2
df04ae196f289d1084b9e3d6eb1223b8655c9485
774173 F20101113_AACKLN sabbioni_g_Page_08.jp2
d4bd17a93e53535636a3ba9120ad3859
d2f8ead353cd1394be4f0f17cc5e553d7200b445
42781 F20101113_AACKKZ sabbioni_g_Page_88.jpg
d4fb8c9537a8cdbabb962f4a177cfda1
9dd98322e831f1eac389fd5e4631883d5bc8f096
103279 F20101113_AACKMC sabbioni_g_Page_27.jp2
9367785c9acfe9e3bbb858c2240f53a7
e93c2862d3ea6af460b06d0bf590d0619f061d01
47506 F20101113_AACKLO sabbioni_g_Page_10.jp2
ba99e650ad29c9df1a20d9120d84b067
1e818a48a08f9d72878def03115f5e50844525a5
90372 F20101113_AACKMD sabbioni_g_Page_28.jp2
b4c6bb7977cf478d40c0107a4c2a66a0
8ab0f192bb0907dec1a545b3f9f35c8557d97094
114939 F20101113_AACKLP sabbioni_g_Page_12.jp2
c1b1591dce2ee60bfb6eae72bf72a67e
b5fd262ec3388a385655686e7ee7d8c2332590b8
92139 F20101113_AACKME sabbioni_g_Page_29.jp2
9b6f7be002a94918d6de8484e16f4793
a2405e4541f4413de3fa02e1f3e84b0d897814b1
120304 F20101113_AACKLQ sabbioni_g_Page_13.jp2
91d7af25b068b3c43bb71a1ec6a27bd3
d5e3f74ec4598f6e6887226de2e2daa7daf64554
108723 F20101113_AACKMF sabbioni_g_Page_31.jp2
168bebd69192806bbd4c6fa5b9170d7a
b28943a9a651acc05cc84070c09e0f3756852d8a
107645 F20101113_AACKMG sabbioni_g_Page_32.jp2
b29c251cab6cb4f803ba8861c9ffa6d1
f475ed221556c49a5d6afc8d1ef9c428c0a5141c
110635 F20101113_AACKLR sabbioni_g_Page_15.jp2
e71f95fe8185529f5963745f3457895b
3bd00bc622b816aa18374e15517b6b58c0b31785
83660 F20101113_AACKMH sabbioni_g_Page_33.jp2
9c81bb878706cf3f6fa0478d2023758e
7eeb7b52ceab5e73a2565fe6da72d12c67446ee3
110793 F20101113_AACKLS sabbioni_g_Page_16.jp2
5bc5448c182801ff9a895c40dd463e03
0c29b209cce94b7e0d243ba39bc4e8e091bca360
110247 F20101113_AACKMI sabbioni_g_Page_34.jp2
4e350573d8354a63f585629b9cfec8d9
f8be1f182b48b002dfba270bc40bfff49c86560a
115119 F20101113_AACKLT sabbioni_g_Page_17.jp2
0db6cd4bb7713a5f959e1e413e3eb772
7b32bc5bd68a520ab218562274bd46adfffc94b8
69666 F20101113_AACKMJ sabbioni_g_Page_35.jp2
fe5f0610f1c19cfb78ed832c185728f1
0b2aaf7c02d25bb93ca3527ca5ee2228d18e7f11
113382 F20101113_AACKLU sabbioni_g_Page_19.jp2
1548ea188e27c0ac687311d401f00f3b
251b80e5f99d8535c7c911f3e34bee6c2bfc8cbc
105498 F20101113_AACKLV sabbioni_g_Page_20.jp2
a71091ebbb4add86932d30299088a0d3
2da011a31a8615690f1fac5bc87fe9ec15f7cf25
116057 F20101113_AACKMK sabbioni_g_Page_37.jp2
46534b3404813ee325958530895e4557
126544eaee6ae697579738463a0c5d2cace816cc
121628 F20101113_AACKLW sabbioni_g_Page_21.jp2
5399cd60159002a00a06a0cc27173f28
d33d7575cd404a941f28e72615368fe112185b54
103804 F20101113_AACKNA sabbioni_g_Page_59.jp2
874151caa062cd26f7f4c3758d6eb231
072c199b906a5114a0f144b454c9932379574491
101580 F20101113_AACKML sabbioni_g_Page_38.jp2
85b57fb8b72cb032ff2b34526a912277
67270121211ab36fbb27c629898a3a4ea4043f8a
107228 F20101113_AACKLX sabbioni_g_Page_22.jp2
f89b47354df1faa6783c5f3408578700
5cd70c3f5d44383e87cca74f026772c4e273ee63
176217 F20101113_AACKNB sabbioni_g_Page_62.jp2
9f7877445738b05060b58b2b36b02950
2f9940ce1d6f7f01b7b86a0fb3eaffd79127544e
105302 F20101113_AACKMM sabbioni_g_Page_40.jp2
15ff93086a10e40c973fce0da4a931c3
cfbb7f3334ef1c5260f7e8ca41c3e41e9b6832c3
93859 F20101113_AACKLY sabbioni_g_Page_23.jp2
e9803f25398bd9505fd1e585370c1200
197ac3a75a862c3fe087a7f8a7ab2be25be7303c
1051964 F20101113_AACKNC sabbioni_g_Page_63.jp2
481aec7d459afff304361721d0dd9c27
0f02c5188762934155895c4916298bb4d9ef515f
76219 F20101113_AACKMN sabbioni_g_Page_41.jp2
3bef9c56b9fef96028e9ec60996b70c4
50c54b9fdf1f056ef7755372a17894eae3ff3993
107405 F20101113_AACKLZ sabbioni_g_Page_24.jp2
3896050c7467a44701ed19d0898c5622
87a31304af75a22c28b39b0d0e967ca333e0125a
1051966 F20101113_AACKND sabbioni_g_Page_64.jp2
f5406d1a410c5b7c2a430e9b169580b8
59e5c8ed79c4794f2c44113088615ecca48133a6
90946 F20101113_AACKMO sabbioni_g_Page_42.jp2
baefba976717b56f8d0bf6387e5eb2a1
b59ac4a15734b37ec7c79582efe11aae6fd9f09d
116091 F20101113_AACKNE sabbioni_g_Page_65.jp2
f088c22003bd604aea5771e574e2ed12
67f48020d338b6cb43d485a712865ec4b12c0536
114425 F20101113_AACKMP sabbioni_g_Page_44.jp2
ed589c67aecc059e47a253b74db61960
3d10bcdbf7553d037090bc6b4b4dca70d443524d
120700 F20101113_AACKNF sabbioni_g_Page_66.jp2
d781db9fa7a2f6d31635b866c7451915
204eebc83a9e07737171622af5c73651c4f707e3
59954 F20101113_AACKMQ sabbioni_g_Page_45.jp2
b238282082852be24541e493979ffdba
ae47ca953e0b012aa449e9392461fc73e4873bb3
117000 F20101113_AACKNG sabbioni_g_Page_67.jp2
e235b95458de09fb71554d63a690264b
cc0ac2d04604e544ee77fddcf09fed5838b4a48b
296080 F20101113_AACKMR sabbioni_g_Page_47.jp2
3b8ffee126344d42cbb524bae450e914
26a31b83d2e13e8d755c6695d0380d086e7c182d
118628 F20101113_AACKNH sabbioni_g_Page_68.jp2
63b7cb2876f4332eb4794122ae7ee35c
3723a754e1a7772838230dff5debe97ee4004064
25439 F20101113_AACKMS sabbioni_g_Page_49.jp2
bb423294e1a949bf8849878ca6b92f31
58370752b2411f46d9d876263e7a4852b5136d8e
62389 F20101113_AACKNI sabbioni_g_Page_69.jp2
431f4f6ce7831876b9c4da6f105a3de0
5aa83b9eb140e77c365454161a1edc418a592fde
271843 F20101113_AACKMT sabbioni_g_Page_50.jp2
1b4606200abac40d98a71a24f7657c58
a5f572483920b0d5543b075b17f2dd636852e058
68093 F20101113_AACKNJ sabbioni_g_Page_71.jp2
f7c02300c5c646038d6f323e8df9add6
4bf9c8807d184cc51f19669824de35b01cd9f479
282442 F20101113_AACKMU sabbioni_g_Page_51.jp2
3539f51a59570c5dbca5b6da99bdeea2
aeef5aa253e635af61a137061c722ca47566c855
52882 F20101113_AACKNK sabbioni_g_Page_72.jp2
f9b5899b72dade5e0eb461205acb7e88
0fb1e91f1bc981da0f8129a72362674b9de40b42
287702 F20101113_AACKMV sabbioni_g_Page_52.jp2
520eb554265893e160da2c6bc7c8df9a
9763e62120153d7dcf59843fae2a56d15191f440
108144 F20101113_AACKMW sabbioni_g_Page_53.jp2
198ced5b4f2c25474439f2f46bff2af6
6fa75e26f720840b542ea2bf22d3178b68feabc5
35269 F20101113_AACKNL sabbioni_g_Page_73.jp2
288217739e967b2c3c349d05032d74ff
c76d395c89f7f50c916d71f5534688332f2aa2b3
1051951 F20101113_AACKMX sabbioni_g_Page_54.jp2
06588b079a6cebd6441e87291522cee3
3aaf40be55e786090ffc3a7a309a402173601ad2
24999 F20101113_AACKOA sabbioni_g_Page_91.jp2
384aa29122cf88bf5a55f0ea5a3142a2
9816343c40232bcb3024040b578d1861fe59a8a4
20281 F20101113_AACKNM sabbioni_g_Page_74.jp2
388ac03c9564ee4aef8f7a33bcef6144
b710b9af24ac684ee416c62f7bdb5c292f125506
119400 F20101113_AACKMY sabbioni_g_Page_55.jp2
31e89992ba7fa6bc40c7929b3c333122
fa7f76f24637dd1094115753fe887fc161ccf88d
122122 F20101113_AACKOB sabbioni_g_Page_92.jp2
29ae69b812d5dd2aa49d5ac02e231b40
c469c0e84d6426b9f9b21c8c87232bd62a98a64a
44500 F20101113_AACKNN sabbioni_g_Page_75.jp2
59a2ae753cc3794bb2dba73e24b328f3
681ef8bd9e159efaf29bd18dc21b267561ccf18d
125882 F20101113_AACKMZ sabbioni_g_Page_56.jp2
7f4e70f9072e6a4fee7b0157f0d5056f
05da16baba67da28d6fc403f176ade4b42339e16
130074 F20101113_AACKOC sabbioni_g_Page_93.jp2
a9223bc401dddd2f077149fcbcba863e
2272edc22e39d39b45041d77bae3d95a067aa4fd
36197 F20101113_AACKNO sabbioni_g_Page_76.jp2
4c181b70be230aa374d77652b1abbcea
e098bbbf9f4f526eafcc503212385884c7cfb512
120549 F20101113_AACKOD sabbioni_g_Page_95.jp2
a454a32f5de6541c0b10dc95efccf54c
e006e8bf09d7c875856fdf6f4ff2e28b495a2c72
45147 F20101113_AACKNP sabbioni_g_Page_77.jp2
58c8a0439b1208e611d9a5b4f61534cf
4ff16597ceebe90468ef0209dd5bc655a5fa31ec
15156 F20101113_AACKOE sabbioni_g_Page_96.jp2
c74ecc9870fa2e8562e085a431ec35ab
c4052a60c42e1fddfbf234d1a707c439ac1cfb43
33529 F20101113_AACKNQ sabbioni_g_Page_78.jp2
32113df846aea7d2a9d56d638558394a
bcb913f4616d9b57424487b16c85d525ae953877
98834 F20101113_AACKOF sabbioni_g_Page_97.jp2
82df881f91ae7b7ada1cad31d84f9306
363b1d1f6084650b9509007ae686d9f7207cdc8f
69582 F20101113_AACKNR sabbioni_g_Page_79.jp2
184907787e9ce5efbdb6e23ca6f8fb7a
d35fddf5ec500fddbc82427e22f3e04cf3ec2c6e
F20101113_AACKOG sabbioni_g_Page_01.tif
65b144546e867e90216f20df019ee102
c7d284bce6c12dc4a645be80b6da9921f3a7cffd
58838 F20101113_AACKNS sabbioni_g_Page_80.jp2
1e7b3c70a4a2312565f01e996037de4e
4f468065a34884579811333b05f9b0ee1eb543f6
F20101113_AACKOH sabbioni_g_Page_02.tif
03cb74b1cc2b78885998a5b4bcf1eb61
3703981bd3abe6813996ad0fe92a16ef674ba04f
51119 F20101113_AACKNT sabbioni_g_Page_82.jp2
94f65480c2e4d53a3c01b54b87f4fc56
fa0c0f1989a3128f1956450cfd72d8e2993a2153
F20101113_AACKOI sabbioni_g_Page_03.tif
0cfc9351605a9f465c30c776a20c983c
cead48308a9c1b74488edb9af8655d35733267a8
69533 F20101113_AACKNU sabbioni_g_Page_83.jp2
0ad12f1f05c43145bc8d49262d27f35a
572719979456a961765d780fe5023b1af3c731a7
F20101113_AACKOJ sabbioni_g_Page_05.tif
e5e965949fb084bd69ff4cb88fea80f1
3b388d60d46e736c8d791cccd1f45f14d0912c67
56827 F20101113_AACKNV sabbioni_g_Page_85.jp2
06b5bf3089b44d442321e90cbe6a04ea
00e0663a72f5b5729abf505d2b72c652b961be83
F20101113_AACKOK sabbioni_g_Page_06.tif
28ee0214611a7f5258af0fdc133ece79
735c15b5da0c2ec23016f2de89906147d7f21517
60217 F20101113_AACKNW sabbioni_g_Page_86.jp2
1ebe658d44e0fe2397fbb5e0419eafa2
0b8b3be5fedc9c315792eba5486453bbbfdd1f8e
F20101113_AACKOL sabbioni_g_Page_07.tif
c9601b7dd4ac091caaec574b98c074e2
43613cc2ff6d8125d68d954abaaae97a3b5a6a56
68267 F20101113_AACKNX sabbioni_g_Page_87.jp2
283e61777757d13855b56dfc713084d4
21a879f89a2b3ab95996f975493b5e321ca5ecb7
F20101113_AACKPA sabbioni_g_Page_27.tif
518b571370b5e67108b739aaa4975c67
20408929cf8092507bea89359f9cd42a2b0c135b
62272 F20101113_AACKNY sabbioni_g_Page_89.jp2
348b1d8da94fcc0adfcffe4ca9201d3e
4eaa7d31bfb1b3e25f6fd8403a9ee9f8ffc99b5c
F20101113_AACKPB sabbioni_g_Page_28.tif
5cbb0da4a8712549369fdfd94d7cdaf7
3387c386da84109e32992c75def358375b42aee5
F20101113_AACKOM sabbioni_g_Page_09.tif
e1ae39ed5734209849f35c0ae20b9916
f47c405ef5a1e8d60ae1f5843dbae922141e477c
74155 F20101113_AACKNZ sabbioni_g_Page_90.jp2
06ff864a326f6451287c858ff0c37bc4
1d83e92677a97a7025e4adf7db6a15db4bf76919
F20101113_AACKPC sabbioni_g_Page_29.tif
d99d6366835d087116b0909d97bab6a4
16f3c893812e094f2c59fa8c24dbf59d9391766b
F20101113_AACKON sabbioni_g_Page_11.tif
37585e91e381e168397eee13f4500dcb
59035a7fa8468396976ec5f4daf3a24379f5713f
F20101113_AACKPD sabbioni_g_Page_33.tif
4c95153af63bcf51fc0b7edbbd5ca9c6
98526c65c6b3f93c5cf74c83a770e2afa5088faf
F20101113_AACKOO sabbioni_g_Page_12.tif
a7b0b0be33f9ac9cf5ec091cb688a694
5dc7377bc4cc1fd93e0484c04486750a8a442cf0
F20101113_AACKPE sabbioni_g_Page_34.tif
87630a14507b192eb148995d88166aeb
05b4d3eb88c29b0952e54e1380a6e42f5b806479
F20101113_AACKOP sabbioni_g_Page_13.tif
423f932bd23eb97cff088002a2e12532
bbb9e74c03b574156b14cfd3e8d0853b3eb451c6
F20101113_AACKPF sabbioni_g_Page_35.tif
e85715b6b4716fc15dbfe78936a5f2cf
60ef48e3890f10677351b4ee86ad93a9764260bc
F20101113_AACKOQ sabbioni_g_Page_15.tif
24173454d7cd650b33218592a68870db
da259f18459dc96791c062f7bf85ebe0291674f5
F20101113_AACKPG sabbioni_g_Page_36.tif
48f27eee56a9141e046ef5ff8338e8ef
37d41107b93cd3d2f9ada3ebe445a5be1b2f7e0e
F20101113_AACKOR sabbioni_g_Page_16.tif
427292ae4d612e42f3a8b99b8fab234f
2467b8393823af4aa8f8aedf9ad01868b3d8fd78
F20101113_AACKPH sabbioni_g_Page_38.tif
98134c2eee8a9f021a9a0d05da665343
ef67f5366a329b986576093a0520877bcb520e0e
F20101113_AACKOS sabbioni_g_Page_17.tif
ca5bf01b3437767e877e3ff44c13ed6f
3941682cc969f3b2884840b01351b84826b4f531
F20101113_AACKPI sabbioni_g_Page_40.tif
5f0fad498e3f2d804d14a366738252ef
2117692a141e706d2d28b0ebc26eabda0507b411
F20101113_AACKOT sabbioni_g_Page_18.tif
deda8004b400cf974de4fdda4e784c26
524ef6609e4388e9447b9f19ff6cda09ca4b4f40
F20101113_AACKPJ sabbioni_g_Page_41.tif
8743b61abd8102e6a7c5544305612714
9a40c53cfd8dcb660ee6090dfd9f1ff1755a2324
F20101113_AACKOU sabbioni_g_Page_19.tif
59fe674aa897e2e51c93de4d61c26be6
891259f218da0748bbad5aff332c8bada5b4b725
F20101113_AACKPK sabbioni_g_Page_42.tif
204bd525d0632cd23de716fa2146211b
d5b133eb3e6674b5120b463617cadb10bea750cb
F20101113_AACKOV sabbioni_g_Page_20.tif
92323d354482765959fbfe4db11c5b65
ae62bacbf5bcc1689c958d3ad9589604bc91a0db
F20101113_AACKPL sabbioni_g_Page_43.tif
dee0c26117bc48fb4fe4aa17026bb5d7
a76e24d937f9fb76477d83faef6896985aa5b8a3
F20101113_AACKOW sabbioni_g_Page_21.tif
0467ba57cf04a94148d2e89fc69312a1
8f459d0cd304ef4f8985e1d151869dd4e8980602
F20101113_AACKQA sabbioni_g_Page_63.tif
75d0905a81f0e04a359daa2562f2d09d
b4ffc1e29fd5c84648984efd642d2f9c366bd0bd
F20101113_AACKPM sabbioni_g_Page_44.tif
2461bf73de8ae1666547f9ceddcdd474
623dd3001ea6104555c84308201920f8dc7ddff5
F20101113_AACKOX sabbioni_g_Page_22.tif
94565820e556fce248cf654f901470a2
aeaaeefa7632ae208d13c0710bb35d2b01326c30
F20101113_AACKQB sabbioni_g_Page_64.tif
70e0a7a2b2f7f93a6705358548cf5b4c
db217d25d56aae4af0a303fe1ff1270120a95b69
F20101113_AACKOY sabbioni_g_Page_24.tif
e058246a267099278ce1793f07a79357
c278a3e9ace6afd9f314a827d9cc6a49c8e0d3c5
F20101113_AACKQC sabbioni_g_Page_65.tif
ebe802d3f4bca7cf8e929eea53d8b1e7
dad98b8fd8b99fa9cb8cb20fa67430781129d195
F20101113_AACKPN sabbioni_g_Page_45.tif
91b064cdab875648de104b45540b7ae9
0ba6f07dda98733be94ad9015e9e8f310331a072
F20101113_AACKOZ sabbioni_g_Page_25.tif
cdd8aff26cb1d55b0650671e6ee096bf
6e2329ef89fe551f029939f647cb50a6f12a260a
F20101113_AACKQD sabbioni_g_Page_66.tif
5afa09e3e6741afd15ead8f3415af8b6
58a636ed85760bdcaea994c5b2c78f8ff32e2aac
F20101113_AACKPO sabbioni_g_Page_48.tif
6f0a2fb0803f588cc1dc25b26b23dc14
4ef8514cb26a4632cf0d8c12a78d142a46852c70
F20101113_AACKQE sabbioni_g_Page_67.tif
489fbebddd78f06e5608d2fffc233fa2
92741e6ee72a9b80ea5173f175815393ad823443
F20101113_AACKPP sabbioni_g_Page_50.tif
90fb2940fa12e2846488208e49636ec0
555eb9c36232b40d7c293cd09c324e6ac351314e
F20101113_AACKQF sabbioni_g_Page_69.tif
1a2d3b6793d900ccd97688c5ac4fef34
86b1f5cc4ab845911891e678ac0c66ba00d2a3ce
F20101113_AACKPQ sabbioni_g_Page_51.tif
ad64dfd8371320ab1471c4592d86f55f
433857fee506a5836474e4c030799c0adb0905e3
F20101113_AACKQG sabbioni_g_Page_70.tif
855dcfce676ef9ebf21b1f1cc34eb0cb
7cc574ceb62cc3a9400dc4f133de07b3297da2c7
F20101113_AACKPR sabbioni_g_Page_52.tif
d79528146f413d69a3d6c2fb09856ac6
3e57d8656f67819617c51891af5f95bdeec4f5a9
F20101113_AACKQH sabbioni_g_Page_72.tif
250304c224408576c6b4a74ca721470a
1fe3497ccbaa987bb4da839f4d4108838aa72a4f
F20101113_AACKPS sabbioni_g_Page_53.tif
3ec14503a50cf57b78b1859c337692c8
8c5820317443ab133061a70624db1e8b6b7c3ecb
F20101113_AACKQI sabbioni_g_Page_74.tif
04448114ddba0d31ad8cd8aca631cc43
595e428e58b657bcb7b311a1b1460ea4de43773f
F20101113_AACKPT sabbioni_g_Page_54.tif
13b79e6cc60e3ff52ee51dd1dbe53f42
82fe2bf96d7ff5eb7d57ac6429cdaffabf15d602
F20101113_AACKQJ sabbioni_g_Page_75.tif
52e63db145dfbb62343ba68f90fcc76c
48d79c8f9d3983802aa9aeb186144071fdc838f2
F20101113_AACKPU sabbioni_g_Page_55.tif
7c907d7901c42e2e3099fc8074cc7555
12a99c69858bd44d7495c0910780ef42e4d75d65
F20101113_AACKQK sabbioni_g_Page_76.tif
3baae4caec6af8500b7dac30bc50ff61
00c9053c221f279c4ac58ae862caef81d172434c
F20101113_AACKPV sabbioni_g_Page_56.tif
919c157b2ff988f75a6c6b8400d25ebf
82805a1dc4eb24184dbaa7a5d2686759aca86a66
F20101113_AACKQL sabbioni_g_Page_78.tif
dee1b57257dd3def425d7ebc4841d3f0
ce9e7b36ec6febfff140482333ff3402f19ee8a7
F20101113_AACKPW sabbioni_g_Page_58.tif
47a9db9c3a0d7b737a4546359b6c5dd7
9afec2d19b36ee3ddfa701cf955582d2b2594f19
F20101113_AACKQM sabbioni_g_Page_79.tif
8aede41f71128af678a5036f55273c98
7d6bb72ce22b9197dba656ad39cff425f98611d5
F20101113_AACKPX sabbioni_g_Page_59.tif
220f79365ac225ef0a47e185969b6505
8c67801185c399ca7fd52d2d209a9190148e490b
28340 F20101113_AACKRA sabbioni_g_Page_07.pro
7452ef9f5aa8a05463d03ca4e7f19fc3
321b31de7d23521a7dbea4ea822d95438bc8a986
F20101113_AACKQN sabbioni_g_Page_81.tif
18b4073af4bbcb83156e842a3e609caf
b9b5395aacac8f3dff1d4c89a8b4dbed9f21abd3
F20101113_AACKPY sabbioni_g_Page_61.tif
52cb8314f787cf3824b600a7d784db58
c69df442a454fec32b56ae790179df790232e077
21128 F20101113_AACKRB sabbioni_g_Page_08.pro
a404de451406000c879cb66b6cb6d3f7
5eb6b2a2afe6beacd0ec8f9a2fc014c422f7b764
F20101113_AACKPZ sabbioni_g_Page_62.tif
bdf1f16eacf7d45c1ec79fbe2540109e
189a241270f60d40a47008b8c4365509322e3d09
46256 F20101113_AACKRC sabbioni_g_Page_09.pro
ad7a8e9060b01d3c6fd6f0575ceaa0c3
89f172db05080b5dd5c375c17479c272ee413577
20921 F20101113_AACKRD sabbioni_g_Page_10.pro
42c55f2da8a1e44c88c0cdca226c32db
25991dd047fe165c25c7c5120aaa662c9b62d3da
F20101113_AACKQO sabbioni_g_Page_82.tif
96a9e5085f78e4061f9bc6b679599bf8
4a63efe2076ed3e798b569aae42b8e3b5a68f741
53686 F20101113_AACKRE sabbioni_g_Page_11.pro
4a77b10105d81c66f206b2aab9c945db
cc913d16a5972b1fd9120c4180829a7dee3395be
F20101113_AACKQP sabbioni_g_Page_88.tif
a169c8a6ee2be552165a304ce5bd3698
53d7a1fed78a48041600e883386290b60c7770ff
54327 F20101113_AACKRF sabbioni_g_Page_12.pro
4b30b30a4971d1e654edf314ea3e533c
b77848ee4fa98de8b0a247d4b499621a39ec77a1
F20101113_AACKQQ sabbioni_g_Page_89.tif
4a2cf9590c1aa83cc475a3ce4574d62c
b396b61c264f97d42a2763c5695819291b10ac5e
56683 F20101113_AACKRG sabbioni_g_Page_13.pro
24cea6aca1bc37df21c0638ea50124e5
ccadc4dd2a68b804a1f64929678ce1e4808acfe1
F20101113_AACKQR sabbioni_g_Page_90.tif
e3a72d556db8a72cbc12fb290785403c
9d7d043af1081afc2cf26a22a91c075e8825831c
51146 F20101113_AACKRH sabbioni_g_Page_14.pro
5c63d64454041a7b2ab8ee20150b6800
3096e116a2074cbbde94baa5e884d0733486880d
F20101113_AACKQS sabbioni_g_Page_94.tif
d98cd8685f0768734f9afaaafa4743cf
f4f94f82937e2ecd0b06221d660f880fa6034c38







THEORETICAL AND EMPIRICAL ANALYSES
OF INCENTIVES AND PUBLIC OWNERSHIP





















By

GUILLERMO SEBASTIAN SABBIONI PEREZ


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 Guillermo Sebastian Sabbioni Perez



























To my wife Carolina, to my daughter Francisca,
to my parents Jorge and Nelyta and to Inda, Maru and Vicky









ACKNOWLEDGMENTS

I thank David Sappington for his invaluable help and advice. I also thank Sanford Berg,

Steven Slutsky, Jon Hamilton and Chunrong Ai for their comments.









TABLE OF CONTENTS

page

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

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

LIST OF FIGURES .................................. .. ..... ..... ................. .8

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

CHAPTER

1 MULTI-AGENT CONTRACTS WITH UNKNOWN COST CORRELATION ..................11

Introdu action ................... .......................................................... ................. 11
T h e M o d el ................. .......... .......... ............................. ................ 14
Uninformed Principal: Limits on Communication.............................................................. 21
If B oth A gents Observe the Correlation................................... .................................... 22
If Only One Agent Observes the Correlation ............... ..........................................32
Uninformed Principal: No Limits on Communication.........................................................33
If B oth A gents Observe the Correlation................................... .................................... 34
If Only One Agent Observes the Correlation............................................................... 34
Lim its on Com munication and Exclusion ........................................ ......................... 40
C onclu sions.......... ............................... ................................................43

2 EFFICIENCY IN THE BRAZILIAN SANITATION SECTOR................ ..................53

Introdu action .............................................................................. ................. 53
Overview of Brazil's Water and Sewerage Industry ...........................................................56
M e th o d o lo g y ..................................................................................................................... 5 9
R e su lts ................... ...................6...................4..........
S en sitiv ity C h eck s ................................................................66
C onclu sions.......... ..........................................................67

APPENDIX DERIVATION OF THE SOLUTIONS TO CHAPTER 1 ..............................79

L im its on C om m unication ................................................................................................ 79
W hen Correlations are Relatively Sim ilar............................................... 79
W hen Correlations are R relatively D different .................................................................. 81
N o Lim its on C om m unication ............................................................83
W hen Correlations are Relatively Sim ilar............................................... 83
W hen Correlations are Relatively Different ................................................................ 87
Limits on Communication and Exclusion ................................. ...............90

L IST O F R E F E R E N C E S ....................................................................................................92









B IO G R A PH IC A L SK E T C H ............................................................................... .....................97









LIST OF TABLES

Table page

2-1 Average statistics by operator-type for 2004.................... .......... ............ .............. 70

2-2 Summary statistics for first-stage regressions....................................... 71

2-3 First-stage LSD V regression results ............................................................................ 72

2-4 Second-stage regression results .............................................................. .....................73

2-5 Ranking of firm-specific costs across firm-types. .................. ...........................74

2-6 First-stage LSDV regression results using a balanced panel..........................................75

2-7 Second-stage regression results using a balanced panel ....................................... 76

2-8 First-stage LSDV regression results excluding the Regional type ..................................77

2-9 Second-stage regression results excluding the Regional type .......................................78









LIST OF FIGURES


Figure page

1-1 Timing at [P-NOC], when both agents observe but do not report the correlation.............46

1-2 A solution to [P-CM] that can also be a solution to [P-NOC] ............... ............. .....47

1-3 Welfare under the two alternative mechanisms at [P-NOC]. ..........................................48

1-4 Timing at [P-CO], when only agent A observes and reports the correlation...................49

1-5 Welfare at [P-EXCL] below welfare at [P-NOC] .......................................................50

1-6 Welfare at [P-EXCL] sometimes larger than welfare at [P-NOC] .................................51

1-7 Welfare at [P-EXCL] almost always larger than welfare at [P-NOC].............................52









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

THEORETICAL AND EMPIRICAL ANALYSES
OF INCENTIVES AND PUBLIC OWNERSHIP

By

Guillermo Sebastian Sabbioni Perez

August 2007

Chair: David Sappington
Major: Economics

This dissertation includes both theoretical and empirical research in economic efficiency.

The first chapter theoretically evaluates how to design procurement models when two agents

may have better information than the principal about their production costs and about the

similarity (i.e., correlation) of those costs between agents. The results indicate that if the

uncertainty about the correlation is not severe, the principal should design a contract as if she

knew that correlation is the lowest possible. The drawback with such mechanism is that the high-

cost agent earns rent if correlation is high. In contrast, if the uncertainty about the correlation is

more severe, the principal should not tailor the solution to any of the potential correlations. This

contract reduces the rent of the high-cost agent if correlation is high by giving rent to the low-

cost agent if correlation is low. Although the welfare level may be different, the characterization

of the results holds when both agents observe the correlation and do not report it, as well as in

the case where only one agent observes and reports the correlation to the principal. Finally, when

the probability of the low-correlation setting is very small, the principal may find optimal to

exclude the high-cost agent from the contract if correlation is low, allowing her to achieve the

socially optimal situation if correlation is high.









The second chapter empirically evaluates the relative efficiency of different type of water

and sewerage operators in Brazil. The analysis consists of a cost comparison between public and

private operators, and between state-level and municipal-level operators. In a first stage, a cost

function is estimated utilizing a fixed-effects panel data model. In a second-stage, the firm-

specific costs from the first stage are explained by means of firm-type indicator variables. The

results illustrate that water and sewerage provision in Brazil is characterized by substantial

economies of scale, indicating that state-level provision is economically more efficient. The

results also show that there is not an economically significant difference between the costs of

private and public firms.









CHAPTER 1
MULTI-AGENT CONTRACTS WITH UNKNOWN COST CORRELATION

Introduction

This chapter analyzes a multi-agent principal-agent model where the agents are privately

informed not only about their production costs but also about the extent to which those costs are

correlated. The analysis is performed with and without limits on communication, and considers

the possibility that either one or both agents may be perfectly informed about the correlation. The

results show that full rent extraction is impossible if the informed agents cannot communicate the

realization of the correlation to the principal. Additionally, the results also illustrate that even

with communication on the correlation the principal may not be able to fully extract rent from all

agents if only one of them is informed about the exact correlation.

The most important finding of the analysis under limited communication is that the high-

cost agent earns positive rent if correlation is high. When limits on communication on the

correlation prevail, the principal cannot tailor the payments to the exact correlation coefficient.

As a result, the extreme lottery of payments for the high-cost agent should ensure that non-

negative expected rent is obtained under both potential correlations. Extreme payments for the

high-cost agent are necessary to prevent the low-cost agent from exaggerating cost. Since

expected rent is calculated using two different sets of conditional probabilities (one for each

correlation), and since the high payment is obtained with higher probability (and the low

payment with lower probability) under the high-correlation environment, the principal can

eliminate rents for the high-cost agent only if correlation is low.

Similarly, the most important finding of the analysis without limits on communication is

that if only one of the agents is perfectly informed about the extent of cost correlation, again the

high-cost agent enjoys positive rent when correlation is high. Although the rent for the high-cost









agent if correlation is high parallels the results with blocked communication, the source of this

rent is slightly different. Instead of being explained by the inability of the principal to tailor

payments to each correlation, when communication on the correlation is available the

mechanism designer needs to prevent the high-cost agent from untruthfully underreporting the

correlation. If correlation is high, the high-cost agent would claim that correlation is low because

he would then be assigned a high payment more frequently (and a low payment less frequently)

than what he should if the principal designed two independent mechanisms, one for each

correlation report.

When limits on communication prevail, the rent for the high-cost agent if correlation is

high is explained by his required participation if correlation is low combined with the fact that

his rent is always higher under the high-correlation scenario than under the low-correlation

setting. A natural question is then whether the principal should always require non-negative rent

for the high-cost agent if correlation is low. Numerical examples illustrate that if the probability

of the low correlation scenario is low enough, it could be optimal to exclude the high-cost agent

from the contract if correlation is low.

This work is related to two streams of research in the mechanism design literature. First,

since the agents) have private information on more than one dimension, the results contribute to

the literature on multidimensional mechanism design. Dasgupta and Maskin (2000) show that if

bidders' types are multidimensional and independently distributed there may be no efficient

auction (i.e., one that assigns the object to the bidder that values it the most). Jehiel and

Moldovanu (2001) illustrate that efficient mechanism design may be impossible in a social

choice setting, where each agent can receive multidimensional signals from all other agents and

where the signals are independently drawn for each agent. The difference between the former









two analyses and the present study is that here the two elements of the two-dimensional type are

not independently distributed. Trying to rank the revenue between a First Price and a Second

Price auction, Fang and Morris (2006) consider an auction where the bidders, besides knowing

their own type, also receive a signal about the other bidders' types. In Fang and Morris' model

there is a separation between each player's valuation and his signal about his opponent's

valuation, since both realizations are drawn from two independent probability distributions. In

their model, the beliefs of the first bidder about the second bidder depend on the type of the

second bidder rather than on the type of the first bidder. In this study, on the other hand, the

agent's privately observed cost at the same time provides information about the other agent's

cost. Miller et al. (2007) consider a problem where agents' types are multidimensional and

continuous, but they relax Jehiel and Moldovanu's assumption that the agents' private

information is independently distributed. Miller et al. show that incentive-compatible

implementation of any decision rule is possible provided that the agents' types satisfy one of

their correlation conditions. The present work is different in that the principal is not required to

utilize budget-balanced transfer payments and in the discrete support of the agents' types.

Second, when addressing the case with limited communication on the correlation, the

present chapter also makes a contribution to the literature on robust mechanism design. This

stream of research attempts to relax the assumption of too much common knowledge at the time

of contract design. Too much common knowledge refers to perfect information about all the

relevant parameters of the problem at hand, as explained by Bergemann and Morris (2005),

Bergemann and Valimaki (2006) and Heifetz and Neeman (2006). The influential statement by

Wilson (1987) has been the starting point of this emerging literature: "Game Theory is deficient

to the extent it assumes features to be common knowledge, such as one player's probability









assessment about another's preferences or information." One way of relaxing the assumption of

too much common knowledge is by focusing on contracts that are not fine-tuned to the exact

environment faced (e.g., to the exact correlation between the agents' costs).

The origins of both streams of research can be traced back to the seminal work by Demski

and Sappington (1984) and Cremer and McLean (1985, 1988), who showed that if agents are

sufficiently risk-neutral and if they face no limited liability constraints, mechanisms can be

designed to ensure costless truthful revelation of private information provided the agents' types

are correlated. McAfee and Reny (1992) extended the analysis to the case where agents may

have infinitely many possible types.

This chapter starts with a description of the model employed. First, the case with limited

communication on the correlation is analyzed. Second, the situation with no limits on

communication is presented. A following section evaluates the possibility of excluding some

agents from the game if their participation is too costly in terms of welfare. Conclusions are

presented in the final section.

The Model

A principal contracts with two agents, A and B, each producing one unit of output. For

each agent, the cost of producing one unit of output is given by C = e. The parameter

p e {/f, JH )} is the agent's initial cost, with A -/,H -/ L > 0. Meanwhile, e denotes the cost-

reducing effort the agent exerts. Effort e generates disutility yr(e) > 0 for the agent, where /y(e)

is increasing and convex. The final cost C is perfectly observable by the principal, but she

cannot distinguish between the cost parameter / and the effort level e. Each agent knows his

own initial cost, but not the cost of his counterpart.1 The principal reimburses each agent for his


1 The principal is subsequently referred as "she", while the agents are referred as "he."









observed final cost C and may provide additional payment (t). The utility of each (risk-neutral)

agent is given by the difference between the transfer payment t and the disutility yr(e) of the

cost-reducing effort exerted: U = t y/(e).

The agents' initial costs are assumed to be positively and imperfectly correlated. The

prevailing probability structure consists of p, = pHH = a / 2 and pLH = p = (1 a)/ 2, where

p, denotes the joint probability that agent A (B) has cost t, and agent B (A) has cost /j, with

i,j e {L,H}. This formulation presumes that the probability that either agent faces an identical

counterpart is invariant to the agent's cost realization. For example, the probability that agent A

(B) has low cost is a and the probability that he has high cost is 1- a when agent B (A) has low

cost.2 The positive and imperfect correlation between the agents' costs is captured by a e

(2, 1) .3, 4 This ensures that the probability that an agent has low cost is higher when the other

agent also has low cost than when the other agent has high cost: a > 1 a if a e (Y, 1).

The parameter a takes on the value a, with probability ,l and the value a, with

probability u0, where a, > a, and u + u0 = 1. Thus, when alpha is a, the agents' costs are

highly correlated. This setting is referred to as the high-correlation scenario. When alpha is a0,

the agents' costs are less correlated (although still positively). This setting is referred to as the

low-correlation scenario. This information is common knowledge at the start of the game.




2 From Bayes' rule.
3 The correlation coefficient is given by 2ca 1 > 0. See Es6 (2005).
4 The values a = 0.5 and a = 1 are excluded from the analysis. These values would constitute no correlation and
perfect (positive) correlation, respectively.
5 The prior presented satisfies the Beliefs-Determine-Preferences (BDP) property from Neeman 211114). In short, a
prior has the BDP property if each agent has different beliefs about the other agents' types depending on his own
type. Neeman shows that full rent extraction results hinge on the BDP assumption.









The principal never observes the exact realization of the correlation. In contrast, at least

one of the agents acquires precise information about the correlation before contracting. This

means that besides her information disadvantage about the agents' initial costs, the principal is

also less informed than (at least one of) the agents about the exact correlation. In addition, two

alternatives are considered regarding limits on communication. In the first alternative setting, the

agents) who are informed about the correlation are unable to communicate its realization. The

second case consists of a framework with no limits on communication, in which the informed

agents) report the actual correlation observed.

The principal designs a mechanism where simultaneously and independently both agents

report (some or all of) their private information.6 Each agent always reports his privately

observed cost ,/, as it is standard in the literature. Additionally, the principal can also request the

more informed agents) to report the actual correlation. The decision of whether to ask for a

correlation report depends on the particular circumstances faced by the principal. As explained

later in more detail, contracts that do not depend on the exact correlation between the agents'

types constitute one important example of the issues addressed by the emerging literature on

robust mechanism design.

Agent A submits the report rA and agent B submits the report rB Based on the joint

report r = (rA, rB), the principal assigns a pair [t(r), C(r)] to each agent. A pair entails a transfer

t(r) from the principal to the agent and a final cost C(r) that the agent must meet.7 Both the

transfer (t) and the final cost (C) depend on all information (r) reported to the principal.

Therefore, each agent knows that the [t, C] pair received can vary with the report of his


6 Collusion possibilities are not considered here.
7 From the cost function C = e, a final-cost target is equivalent to an effort target.









counterpart. In the cases with asymmetric agents (i.e., when only one agent observes the

correlation), the analysis assumes that the principal offers the same menu of options to both

agents. The reason behind this assumption is that the agents may refuse to accept different menus

of contracts, on the grounds that the principal would be treating them unfairly.

The agents will only contract with the principal if they anticipate non-negative expected

rent from doing so. However, the computation of the expected rent depends on the agent's

knowledge about the correlation. For example, if a low-cost agent observes a low correlation, he

knows that the other agent has low cost with probability a, and high cost with probability

1 a,. On the other hand, if an agent is uninformed about the correlation, he only knows the

expected probability of facing each type of counterpart.9 An agent who is uninformed about the

correlation only knows that the other agent has similar cost with probability a and different cost

with probability 1 &, where &c /o0ao + ala .

The principal maximizes the sum of the value derived from both units produced, plus the

agents' profits, minus the cost of social funds. 10 The principal values each unit produced at S,

where S is assumed constant and sufficiently large. The social funds are the transfers and the

reimbursed costs received by the agents, which the principal delivers at a cost of 1+ A per unit.11

If the principal observed the agents' costs:12

W = 2S + Y[t' (' C')] (1+ A) -[t + C'].
IG{A,B} IG{A,B}

8 Opportunity rents are normalized at zero and no ex-post limited liability constraints are imposed.
9 Every time an agent is described as "informed" or "uninformed", it is meant from the point of view of whether
the agent has perfect information on the correlation. As a remainder, all agents are always perfectly informed
about their own production costs.
10 See Laffont and Tirole (1986).
1 The parameter > 0 measures the distortion introduced by the excise-tax system utilized to raise social funds.
12 qf(P-C) is used when specific reference to the cost parameter / needs to be made. Otherwise, y(e) constitutes
more compact notation.









Standard manipulation yields:13

W = -(+ A) -[-e' + y(e')]- A [t' (e')] (1-1)
It{A,B} It{A,B}

Equation 1-1 shows that if the principal observed the agents' costs, the solution would be

characterized by t = Iy(e) and e = e for both agents, where y' (e*) = 1.14 The socially efficient

level of effort e would be delivered without any rent being afforded to any agent.

Demski and Sappington (1984) and Cremer and McLean (1985) (CM) illustrated how this

result can be replicated if the principal does not observe the agents' costs but she knows exactly

how these costs are correlated.15 Let t,, y(B, C,) denote the ex-post rent for an agent (A or

B) that reports cost 8, when he has cost /, and when the other agent reports cost 8/k, for i, j, k

e {L,H}. This problem [P-CM] constitutes the benchmark situation of this chapter:

Maximize

W= -(1+A) I2p, [-ej +,(ej)]- A 2p,[tj -, (ej)] (1-2)
i,j~{L,H} ij~ L,H}

subject to:

u, c[t r(P, C,, )]
( a[ 8 i, je {L,H},i j (1-3)-(1-4)
+ (1- a)[t,j V(/, C, )] > 0

u, > a[t, r,(P, C ,)]
i, je L,H},i j (1-5)-(1-6)
+ (1 a)[t, V(A, C, )]






13 The constants 2S and the 's are dropped for the sake of exposition.
14 From C = e, an extra unit of effort reduces the final cost also by one unit. Therefore, q '(e) = 1 denotes the
equality of the marginal cost and the marginal benefit of exerting cost-reducing effort.
15 It is well documented that risk aversion and/or limited liability prevent the principal from achieving the first-
best. See Sappington (1983) for the single-agent case. See also Cremer and McLean (1988), Demski, Sappington
and Spiller (1988) and Robert (1991)









Equation 1-2 incorporates the four possible cost combinations that can arise and their

respective probabilities. Equations 1-3 and 1-4 ensure non-negative expected rent for the agents.

Equations 1-5 and 1-6 ensure that the agents report their costs truthfully in equilibrium. 16

The solution to [P-CM] permits payments that fully extract (expected) rent while ensuring

e, = e *, for i,j e {L,H} To accomplish this, the principal sets relatively high payments for

cost reports that match and relatively low payments when cost reports do not match. These two

degrees of freedom allow for an infinite number of solutions to [P-CM]. After selecting

tHH > tH and tLL> tm the principal finds tH and tLH such that no rent is obtained in

expectation by either agent, and so UH = UL = 0. In particular, consider the bound tH > tH ,

where:17

l-a
t" r((eHH) + [-- [ HL + (1- a HH > V(eH), (1-7)
2a -1

with H, V(e) (eH, A) > 0, for i e {L,H} .18


The lower bound tHH > tH prevents the low-cost agent from exaggerating his cost while

ensuring that a high-cost agent earns non-negative rent in expectation. This is achieved with a

high payment tHH > V(eHH) if both agents report high cost and a low payment tHL < (eHL) if a

high-cost report is not matched with the same report from the other agent. If the low-cost agent

lies, he is relatively likely to receive the low tHL payment and relatively unlikely to receive the


16 This is known as Bayes-Nash implementation. Stronger solutions would be obtained if Ex-Post or Dominant
Strategy implementation was required. Ex-Post implementation consists on the requirement that the incentive
constraints are satisfied even after the other agents make their (truthful) report. Dominant Strategy
implementation consists on the requirement that an agent is always better off by reporting the truth, even when
the other agents may lie.
17 The lower bound tLLM" guarantees that a high-cost agent does not claim low cost, which is usually not
constraining for the principal.
18 (H is the profit earned by a low-cost agent in the single-agent case. This profit is explained by the savings in
disutility of effort that the low-cost agent would enjoy if he exaggerates his cost.









high tHH payment because, due to the positive correlation among costs, the other agent is

relatively likely to report low cost.

It is important to notice that tm" is increasing in H, which means that t"n is increasing

in effort eH, for i e {L,H} .19 The effort exerted by the high cost agent has a positive impact on

the "reward" (i.e., payment above disutility of effort) that an agent receives if a high-cost report

is matched by his counterpart. Similarly, the effort exerted by the high cost agent positively

affects (the absolute value of) the "penalty" (i.e., payment below disutility of effort) that an agent

receives if a high-cost report is not matched by the other agent. In other words, the larger is e,,

the higher tHH has to be (and the lower tHL has to be) to become unattractive for a low-cost

agent. This feature of the payments has important implications for the solution to the principal's

problem when she is imperfectly informed about the correlation.

The subsequent analysis focuses on settings where the principal, unlike the agentss, does

not know exactly the extent of the correlation between the agents' costs. In such settings, one or

both agents have better information than the principal about both their costs and the exact

correlation. As it will be seen, the principal cannot achieve the first-best outcome in those

settings.

The ensuing analysis is performed with and without limits on communication about the

correlation. The situation with full communication of all private information follows the standard

optimal contracting approach. The case with blocked communication on the correlation deserves

additional explanation. It is important to understand why the principal may not utilize all

available communication channels to gather as much information as possible.



19 It also means that tHH.n is increasing in the cost advantage of the low-cost agent, A.









Uninformed Principal: Limits on Communication

Consider, first, the setting where the informed agents) are unable to report the correlation

to the uninformed principal. In this context, it is impossible for the principal to fine-tune the

contract to the exact correlation observed by the agentss. As explained below, specific

circumstances prevent the principal from making a contract conditional on the actual correlation.

Consequently, the principal designs a mechanism where both agents report only their cost:

rA =/ and rB =/ with i,j{L,H}.

For example, this could be the situation faced by a regulator when publicly-owned firms

are privatized quickly.20 In the pre-privatization stage, the regulator may not know the identity of

the operators that will take control of the soon-to-be privatized firms. Alternatively, the regulator

may not know exactly which production technologies the new firms will utilize. As a result of

this uncertainty, the regulator may ignore the exact correlation between the firms' costs when she

designs the regulatory contract.21 The inability of the agents) to subsequently report the realized

correlation can be a stylized means for capturing prohibitive costs of acquiring more precise

information about the correlation at a later stage. For example, it may be too costly or it may take

a long time for an expert auditor to provide accurate information about the correlation.22

Limited communication on the correlation is a key element of the new literature on robust

mechanism design. This strand of research attempts to relax the implicit assumption of too much

common knowledge at the time of contract design.23 As such, this literature focuses on contracts


20 When a government makes a tender for the construction of highways may serve as another example.
21 In the auction setting presented by Arya et al. (2005), for example, the authors suggest that the contract designed
can be utilized with different pairs of bidders (i.e., with pairs of bidders that may have different correlations
among their valuations for the object auctioned).
22 In the context of a fast privatization process, political pressure sometimes requires that a regulatory framework
be set up early enough.
23 See Bergemann and Morris (2005) and Heifetz and Neeman (2006).









that are not fine-tuned to the exact environment faced (e.g., on contracts that do not rely on the

exact correlation between the agents' types). According to researchers, the main motivation is

that real-world mechanisms seem to be simpler than what it would be required by a theoretically

optimal contracting approach. For example, Arya et al. (2005, p. 15) suggest that the robustness

problem might "help us better understand observed institutions." Additionally, Bergemann and

Valimaki (2006, p. 3) say that "practitioners have often been led to argue in favor of using

simpler but apparently sup-optimal mechanisms." Also, Bergemann and Morris (2005, p. 1)

emphasize that "the optimal mechanisms solving the well-defined planner's problem seem

unreasonably complicated."

Two cases are examined below: one in which both agents are informed about the

correlation and one in which only a single agent is so informed. In the context of the

privatization process introduced before, the situation with only one agent informed about the

correlation could arise when only one of the firms has been operating in the industry for a

sufficiently long time (perhaps in other geographical areas). Therefore, only one of the firms

would be endowed with sufficient expertise to permit an accurate assessment of the extent of

cost correlation in the environment. This experience would allow the firm to be well informed

about the correlation between its own cost and the cost of the (less experienced) counterpart.24

On the other hand, if both firms have been operating in the industry for a similar length of time,

they could both know exactly the extent to which their costs are correlated.

If Both Agents Observe the Correlation

The principal computes expected welfare:




24 Although plausible, this chapter does not consider the possibility that a more experienced agent could
presumably be more likely to have low cost.









W=-(1+2) -2a,[-eJ+w(e,)]-A -2pr[t,- W(e,)], (1-8)
I,jG{L,H} i,j]{L,H}

with pJ -= lP,\ +oPj, and where p,' denotes the probability of joint costs /, and fLJ

when a = a,, for i,j e {L,H} and s e {0,1}. The difference between Eq. 1-8 and Eq. 1-2 is

explained by the fact that p, is perfectly known by the principal at [P-CM], whereas it can only

be estimated (PJ ) when correlation is uncertain and its communication is blocked.

To ensure that both agents participate and report their costs truthfully for all cost and

correlation realizations, the next eight constraints have to be satisfied:25

t 'i, e {L,H},i j se {0,1} (1-9)-(1-12)
+ ([t )[ VC, ) > 0

S- i, jJ {L,H},i j se {0,1} (1-13)-(1-16)
+(1- a)[t, -v(, -C )]

With no communication the principal solves problem [P-NOC]:

Maximize Eq. 1-8 subject to Eq. 1-9 to Eq. 1-16.

The only difference between the constraints at [P-NOC] and the constraints at [P-CM] is

that the constraints at [P-NOC] consider the possibility of two potential correlations (alpha could

be either a1 or ao). The timing is depicted in Figure 1-1.

At the solution to [P-NOC], the two degrees of freedom that the principal had available at

[P-CM] are utilized to satisfy the four additional constraints. There are two alternative ways in

which the principal can perform this task. Which of the two mechanisms is optimal depends on

the specific values of the parameters, as is later illustrated in detail.26


25 Before concluding, the last section discusses whether it is always optimal to require participation and truthful
revelation by all agents in all scenarios.
26 See the Appendix for a formal derivation of all the results.









In the first potential solution to [P-NOC], one degree of freedom is utilized to set

t, = L y(e, ), while the other degree of freedom is used to set tHH equal to the lower bound that

[P-CM] would require if correlation is known to be low (i.e., to set tHH equal to t"n from Eq. 1-

7 if a = a ). This payment structure is subsequently referred to as the CM-Low mechanism,

because it is identical to one of the solutions to [P-CM] if the principal knows that correlation is

low.

The fact that t, = Vy(eL) implies that tLH = I(eLH), regardless of which correlation

occurs. Since t, = yI(e, ), the principal can do no better than to set tLH = I(eLH). Setting

tLH < I(eLH) would imply a violation of the participation constraints of the low-cost agent, while

setting tLH > I(eH ) would imply unnecessary positive rent for him. This is no different than

what the principal can accomplish at [P-CM], when fully extracting expected (and ex-post) rent

from the low-cost agent by giving him non-stochastic transfers. These payments automatically

guarantee that the participation constraints for the low-cost agent are satisfied under both

potential correlations because they are also satisfied ex-post (i.e., for any cost report of the other

agent).

At the same time, tHH = tH (a0) ensures that the incentive compatibility constraints for

the low-cost agent are satisfied for both potential correlation coefficients. This happens because

the lottery of payments for the high-cost agent (tHH, tHL ) tailored to the low-correlation setting is

sufficiently "extreme" if correlation is high.27 In other words, if the payments for the high-cost

agent are not attractive for the low-cost agent if correlation is low, they are even less attractive

under the high-correlation setting. The reason for this result is simple. If the low-cost agent lies

27 A lottery of payments for the high-cost agent is subsequently called more "extreme" when the difference
between tHH and tL gets larger.









under the high-correlation setting, he is "extremely" likely to receive the low tL payment (i.e.,

even more likely than in the low-correlation scenario). This happens because the high-correlation

framework imposes on the low-cost agent a very high chance of facing a similar agent, from

al > ao. Figure 1-2 depicts this solution, showing the CM well-known result of increasingly

extreme payments for the high-cost agent when correlation becomes arbitrarily small.28

The problem with the CM-Low mechanism comes from the fact that t"m > yr(eHH), as

shown by Eq. 1-7. This is a key feature of the solution to [P-CM]: in order to prevent untruthful

reporting from a low-cost agent, a high-cost agent obtains positive (ex-post) rent if he faces an

identical agent. This rent is not a problem at [P-CM], because the principal offsets the high

payment tHH > I(eH ) with a low payment tL < r(eH ), such that the high-cost agent earns no

expected rent. However, the positive and negative (ex-post) rents for the high-cost agent come at

a cost at [P-NOC]: they prevent the principal from fully extracting rent from the high-cost agent

under both correlations, as is now illustrated.

At [P-NOC], the principal does not know which correlation is realized. Therefore, she has

two sets of potential weights to use when calculating the expected rent of the high-cost agent. As

she does at [P-CM], the principal would like to average positive and negative (ex-post) rents

such that (ex-ante) rent UH is reduced to zero. However, achieving this goal under two sets of

conditional probabilities is impossible. This is apparent after a reexamination of the participation

constraints for the high-cost agent:

U aU HH+ ( )UHL > 0, s E{0,1

with UHH tHH (eHH ) > 0 and UHL tHL (eHL ) < 0, from Eq. 1-7.



28 This result can also be obtained analytically by taking the derivative of Eq. 1-7 with respect to alpha.









If the principal extracts all rent for the high-cost agent at the high-correlation scenario, she

would not provide him with at least zero rent in the low-correlation setting. That is, U 0 = 0 is

impossible without violating Uo > 0. This result is explained by the necessary positive and

negative ex-post rents for the high-cost agent (UHH > 0 and UH < 0), combined with a, > ao.

Therefore, the expected rent for the high-cost agent is always larger if correlation is high:

aUHH + (1- a )UHL > OUHH + (1 0)UHL,

As a result, the principal can only extract rent from the high-cost agent under the low-

correlation setting, leaving him with positive rent if correlation is high:29


= a- o [aD + (1 0 )HH ] >0
S2a -1 H


Recalling that H, '(eH,) -(e A) > 0, it can be seen that rent U, increases with

effort level e,, for i e {L,H}. As it happens at [P-CM], the payments for the high-cost agent

include rewards and penalties which discourage untruthful reporting by the low-cost agent. The

problem is that these rewards and penalties increase (in absolute value) with effort eH,. That is,

the higher the effort required from a high-cost agent, the more extreme his lottery of payments

needs to be (i.e., higher tHH and lower tL) to become unattractive for a low-cost agent. That is

not a problem at [P-CM], where no rent is provided to either agent. At [P-NOC], however, there

is rent for the high-cost agent if correlation is high. This rent depends on the effort he delivers,

because his effort has a positive impact on the rewards and penalties included in his payments.





29 The principal provides rent to the high-cost agent if correlation is high because she requires his participation if
correlation is low. This assumption is relaxed in the last section of this chapter, where the principal no longer
requires that the high-cost agent earns non-negative rent if correlation is low.









Consequently, the principal asks for suboptimal effort levels eH < e and eHH < e in order to

reduce the rent U, of the high-cost agent if correlation is high:30


V' (e=H) HH < 1
(1 + A) (2ao 1) ia


(1 + 2) (2a0 -1) (1-a)

The CM-Low mechanism works well if correlation is low, where no rent is commanded by

either agent. The problem arises if correlation is high, in which case rent is obtained by the high-

cost agent. The welfare loss generated by rent UV is severe if the potential correlations are very

different. The reason is that under the CM-Low mechanism, the principal tailors one of the CM

solutions to the low-correlation setting (recall tHH = t (a0)). If this scenario does not occur,

however, the welfare loss depends on how "wrong" the principal is with respect to the exact

degree of correlation (i.e., how different the actual correlation is high with respect to what the

principal thought low). Depending on this difference, therefore, the welfare loss under the CM-

Low mechanism calls for an alternative solution to [P-NOC]. Intuitively, this second contract

should work better than the CM-Low contract when the potential correlations are substantially

distinct with respect to the correlation at the low-correlation setting.

In the second potential solution to [P-NOC], the two degrees of freedom available at [P-

CM] are utilized in a different way than under the CM-Low mechanism. This alternative

mechanism does not consist of a CM solution tailored to any of the potential correlation

coefficients. In this alternative contract, the two degrees of freedom available at [P-CM] are

employed to guarantee ex-post truthful reporting by the low-cost agent. That is, one degree of


30 The efficient level of effort is delivered after a low cost report: eL = eH = e*.









freedom is used to set tLL y(eLL) = tHL -(eHL A) while the other degree of freedom is used

to set tLH I(eH ) tHH -y(e A). As a result, both incentive compatibility constraints for

the low-cost agent are satisfied because they are satisfied ex-post (i.e., for any cost report of the

other agent):

a, [tLL y(eLL)] [tHL (eHL A)]
+ (1- a, ){[tLH (e )] [tHH (eHH A)]} = 0

This mechanism is subsequently referred to as the Ex-Post mechanism. The disadvantage

of the Ex-Post contract is that tLL # y(eL) which means that the principal cannot fully extract

rent from the low-cost agent under both correlations. The reason for this result is, again, the

impossibility of setting equal to zero the average of positive and negative (ex-post) rents under

two possible sets of weights. At most, expected rent can be zero under only one of the potential

correlations. As a consequence, under the Ex-Post mechanism the low-cost agent obtains rent

when correlation is low:


U o0 a'-a0 [a HH +(1 )OHL]>0
a, + a, -1

If correlation is high, however, the rent for the high-cost agent is lower than under the CM-

Low mechanism:31


U, = l- a [aL HL +(IaLHH]>O
a, +a, -1

The principal cannot completely eliminate rent U' because positive and negative (ex-

post) rents for the high-cost agent are still required. Recall that tHH > I(eH ) and tHL < I(eHL)



31 The fact that rent UH' is lower under the Ex-Post mechanism than under the CM-Low mechanism is easily
verified from the effect of al > ao in the denominator of both expressions: a, + ao 1 > 2ao 1. The term in
brackets [ ] has only a second-order effect on rent.









prevent untruthful reporting from the advantaged low-cost agent: he would probably receive a

low payment if he claims high cost.

It can be seen that rents under the Ex-Post mechanism depend on the difference between

the potential correlations (a1 a ) measured with respect to the "middle" correlation coefficient

(a, + ao -1). Recalling that 2a -1 denotes the correlation coefficient between costs, the

"middle" correlation coefficient al + ao -1 is defined as the correlation at the middle point of

the distance between al and ao. For example, if a, =0.8 and ao =0.6, the middle correlation

would be the correlation coefficient if alpha was 0.7. It is important to notice that the middle

correlation coefficient is not the expected correlation 2d -1, which would incorporate the

probabilities of each scenario in the calculation. This confirms that the Ex-Post mechanism does

not consist of a CM mechanism tailored to any of the potential correlations.

Like under the CM-Low mechanism, rents under the Ex-Post mechanism also increase

with the effort e., delivered by the high-cost agent. The explanation resides again on the

positive impact of his effort on the rewards and penalties that his payments include to discourage

untruthful reporting by the low-cost agent. Therefore, the principal reduces rents under the Ex-

Post mechanism through the suboptimal effort levels eHL < e and e < e* :32

'(e = (a, a) pu, (1-a,) + pO <
(1+ A) (ao + a, -1) a


'(eHL) 1 A (a1 -ao) pj1a, +p o (1 -ao)
=' (eHL L<1
(1+ A) (ao +a -1) 1 a





32 Again, the efficient level of effort is delivered after a low cost report: eL = eLH = e









As it was mentioned before, which of the two alternative mechanisms yields a higher

welfare level depends on the parameters of the problem. The CM-Low mechanism is optimal for

[P-NOC] if:

a- < (1-17)
2ao -1 P,i

Equation 1-17 can be evaluated in terms of how different the two potential correlations are,

measured with respect to the correlation at the low-correlation setting.33 For similar correlations

(small (a, a ) /(2ao 1)), the CM-Low mechanism yields a higher welfare level. This happens

because the welfare loss that the CM-Low mechanism entails is directly proportional to how

different the potential correlations are. As explained above, the rent that the CM-Low

mechanism provides to the high-cost agent under the high-correlation scenario is too large if the

potential correlations are very distinct. Remember that this rent if correlation is high depends on

how "wrong" the principal is with respect to the exact degree of correlation (i.e., how different

the actual correlation is high with respect to what the principal thought low). Then, the

principal should utilize the CM-Low mechanism only if the potential correlation coefficients are

relatively similar. In contrast, if the potential correlations become too distinct, the principal

should employ the Ex-Post mechanism. This contract does not consist of a CM contract tailored

to any of the potential correlation coefficients. Figure 1-3 plots the behavior of welfare under

both alternative contracts as a function of how different the potential correlations are.34

Equivalently, Eq. 1-17 can also be analyzed in terms of the relative likelihood of each

correlation coefficient. When /u is so large that Eq. 1-17 holds, the principal ensures that no rent

is afforded under the (relatively likely) low-correlation scenario. She does so by implementing

33 Remember that 2a 1 denotes the correlation coefficient.
34 The graph assumes a constant value of a2 while al is shown in the horizontal axis.









the CM-Low mechanism, which provides no rent to either agent if correlation is low. Although

this payment structure provides rent UH1 to the high-cost agent if correlation is high, the

principal still applies the CM-Low mechanism if the high-correlation setting is relatively

unlikely (i.e., if u, is so low that Eq. 1-17 holds).

On the other hand, if /u is so large that Eq. 1-17 does not hold, the principal reduces the

(relatively likely) rent U. as much as possible by means of the Ex-Post mechanism. She cannot

completely eliminate this rent because tHH > I(eHH) and tL < VI(eHL) still prevent untruthful

reporting from the low-cost agent. The drawback to the Ex-Post mechanism is that the low-cost

agent obtains rent UL if correlation is low. Yet, the principal is willing to accept rent U if the

low-correlation scenario is relatively unlikely (i.e., if /u is so low that Eq. 1-17 does not hold).

The results obtained are similar to those of Arya et al. (2005). In an auction environment,

they also find a condition that determines which of two mechanisms the principal should apply.

Furthermore, the proposed mechanisms and the circumstances under which they should be

employed are similar to the ones presented in this study. If the uncertainty about the correlation

is not severe, the principal should utilize a CM-Low type of mechanism.35 On the other hand, the

principal should employ an Ex-Post type of mechanism if the uncertainty about the correlation is

36
more severe.

Additionally, after considering the parameter values assumed by Arya et al., the condition

that determines optimality in their paper is the same as Eq. 1-17 in this study. Instead of a

discrete support for the parameter alpha, they assume a uniform distribution between a lower

bound a > 0.5 and 1. Given such a probability structure, they find that the principal should

35 Bayesian-Nash is the name given by Arya et al. to such a contract.
36 Dominant-Strategy is the name given by Arya et al. to such a contract.









employ the CM-Low type of mechanism if a > 2/3. Now notice the result of substituting

U0 = ,1 = 0.5 and a, = 1 into Eq. 1-17:

1 -ao 0.5
2ao -1 0.5

Manipulation yields a0 > 2/3, which means that their condition can be considered a

special case ofEq. 1-17 in this chapter.37

If Only One Agent Observes the Correlation

Consider now a situation where communication of the correlation is still blocked, but

where only agent A observes the exact correlation before contracting. This is a framework where

only agent A is endowed with a strong knowledge and expertise about the industry. Meanwhile,

agent B and the principal remain ignorant about the correlation. They never acquire such

privileged information, knowing only the probabilities of facing either a, or a Returning to

the previous privatization example, this asymmetry of agents could arise when firm A has been

operating in the industry for a long time (perhaps in other markets), while firm B is relatively

new in the industry. In such a context, it is plausible that not all the agents are perfectly informed

about the correlation between their costs.

Since it is assumed that the principal does not design a different menu of options for each

agent even when she distinguishes the experienced agent A from the inexperienced agent B, the

solution to this problem is identical to the solution to [P-NOC]. The explanation for this result is

simple. To ensure A's participation, the principal must guarantee that participation and incentive



37 There is only one difference between their results and the ones in the present study. Unlike here, the Ex-Post
type of mechanism from Arya et al. does not provide any rent to the low-valuation bidder (which is the
equivalent of the high-cost agent). (See their Corollary 2, on page 11). The linearity of the agents' utility in their
model is the conjectured explanation for such a difference (i.e. there is no convexity like the one coming from
the disutility of effort in the present model). A formal proof is beyond the scope of this study.









compatibility constraints are satisfied under both potential correlations. Therefore, the constraints

are automatically satisfied for the inexperienced agent B, who computes expected rent not only

across both potential types of counterpart, but also across both potential correlations. Formally, if

the constraints from [P-NOC] are satisfied, then the equivalent constraints for an agent that only

observes a are also satisfied:

i, j.E {L,H},i j
+ (1 -()[tj y( C, )] > 0


i, j f{L,H},i j
+ (1 a)[t, 6(#, C, )]

As a consequence, the solution to [P-NOC] also applies if only one of the agents observes

the correlation but the principal offers the same menu of options to both of them.

Uninformed Principal: No Limits on Communication

In the second case of interest, nothing prevents the principal from asking for a correlation

report from the informed agentss. Unlike in the case with limits on communication, the

Revelation Principle applies in this context. Therefore, the principal can restrict attention to

truthful and direct mechanisms where the agents report all their private information. This

situation is interesting because an agent who is informed about the correlation becomes a two-

dimensional type of agent. The (simple) situation where both agents observe the correlation is

presented first. Next, the (more interesting) setting with only one agent informed about the

correlation is investigated.









If Both Agents Observe the Correlation

It is well known that if some element of private information is common knowledge for

both agents, the principal can elicit that information at no cost.38 She can do so by designing a

mechanism where both agents report their individual cost and the realized correlation: rA =

(/,, a,) and rB = (/6 ah), with i,j {L,H} and s,h e {1,0}. Since it is impossible to have

different (and truthful) correlation reports, the principal can design a "forcing" contract that

threatens the agents to very large negative rents if the correlation reports differ.39 Therefore, even

when the principal has imperfect information about the correlation, the first-best situation can

still be replicated if both agents can communicate the actual correlation observed. Hence, the

solution to this problem is identical to the solution to [P-CM].40

If Only One Agent Observes the Correlation

If the correlation is known only by agent A, the principal designs a mechanism where the

inexperienced agent B only reports his cost, rB = (/), while the experienced agent A reports his

cost and the realized correlation, rA = (/j as), with i, j e L, H) and s e {0,1}. The timing of

the game is depicted in Figure 1-4.

As explained above, both agents are offered the same menu of contracts even when the

principal knows about their asymmetry in terms of their information on the exact correlation. Let

ts y(/8 C ) denote the ex-post rent for an agent (A or B) that reports cost /, when he has




38 See Fudenberg and Tirole (1991).
39 In the work of Tangeras (2002), for example, the element of private information that is common knowledge for
both agents is the "industry" portion of their costs, which has to be added to each agent's idiosyncratic cost.
Their cost structure follows the model proposed by Auriol and Laffont (1992).
40 There also exists a Nash equilibrium in which both agents lie about the correlation. As pointed out by Fudenberg
and Tirole, the possibility of multiple equilibria in "shoot-them-all" mechanisms "gave rise to a large literature
on unique Nash implementation" (1991, p. 293).









cost /,, when the other agent reports cost /k and when the experienced agent A reports that

alphais as, for i, j,k e {L,H} and s {0,1}.

The principal computes expected welfare:


W= /-((1+A) I2pi[-e + s o{0,1i} i,J{L,H} i,JG{L,H}

The difference between Eq. 1-18 and Eq. 1-8 is that the principal now assigns different

contracts for different correlations. To ensure that the experienced agent participates and reports

his cost truthfully for all cost and correlation realizations, conditional on a truthful correlation

report, the next constraints have to be satisfied:

u, -: n,[tl ',( CC -)]
i, je {L,H},i # j s e {0,1} (1-19)-(1-22)
+ (1- a,)[ti y(, C;)] > 0

> a,[t (?,-c )]
i, jeL,H},i j s {0,1} (1-23)-(1-26)
+ (1- a)[t; (A -C )]

Additionally, conditional on a truthful cost report, the experienced agent A should not lie

about the correlation:

Us as[th Vf'8- C,,)]
Sath i, j{e L,H},i # j s, h {0,1} (1-27)-(1-30)
+ (1- )[t, (# C,)]

Finally, agent A should not simultaneously misreport his cost and the correlation:

U'" 2a [th Vf(,8 _Ch)]
i, je {L,H},i j s,h {0,1} (1-3)-(1-34)
+ (1-_a,)[th _-( _Ch)]
'\ U/II ii~\r









With only one agent informed and with communication of the correlation, the principal

solves problem [P-CO]:41

Maximize Eq. 1-18 subject to Eq. 1-19 to Eq. 1-34.

Intuition could suggest that once a correlation report becomes available, the principal

should again be able to achieve the first-best by means of two different CM mechanisms, one for

each correlation. That is not the case, however. The problem is that at [P-CO], the agent who is

informed about the correlation may not have the incentive of truthfully reporting its realization.

In particular, a high-cost agent would gain by claiming that correlation is low when it is actually

high. An intuitive explanation follows.

Consider a payment structure that consists of two independent CM mechanisms (i.e., one

for each correlation). Based on the correlation report, the principal could offer non-stochastic

payments to the low-cost agent and an extreme enough lottery (i.e., low tL and high t' ) to the

high-cost agent, for s e {0,1}. Under each correlation, the high-cost agent's lottery would extract

all his rent and it would also prevent a low-cost agent from exaggerating his cost. The problem of

these two independent CM mechanisms is that if the experienced agent has high cost, he would

enjoy rent under the high-correlation setting if he claims that correlation is low. The explanation

behind this incentive follows.

After the high-cost agent underreports the correlation, he would be assigned the high

payment toH with probability a, (instead of with probability ao) and he would be assigned the

low payment to with probability 1 a (instead of with probability 1 a ). Since a, > a, this

average of positive and negative ex-post rents yields rent for the high-cost agent if correlation is

41 As in the case with no communication and only one agent informed, it is not necessary to impose participation
and incentive compatibility constraints for the inexperienced agent B. Those constraints are automatically
satisfied when the principal offers the same menu of contracts to both agents.









high. Therefore, a high-cost agent would want to report that correlation is low when the high-

correlation environment arises. Although they prevent untruthful cost reporting by the low-cost

agent, extreme payments to the high-cost agent generate a new incentive if there is asymmetric

information about the correlation. A high-cost agent would earn positive rent if correlation is

high by reporting that correlation is low. As a result of this incentive, full rent extraction is not

possible at [P-CO]. The two-dimensional type of the informed agent does not allow the principal

to consider the problem as if they were two independent CM problems.

The rent for the high-cost agent if correlation is high is unavoidable.42 This result is similar

to what happens at [P-NOC], although the source of the rent at [P-CO] is different. At [P-NOC],

the rent for the high-cost agent at the high-correlation setting occurs because the principal cannot

tailor his payments to the actual correlation. This comes as a result of designing only one set of

payments that has to work well regardless of the correlation. At [P-CO], however, the principal

can design different payments for different correlations. Yet, the rent for the high-cost agent if

correlation is high still prevails. At [P-CO], this rent exists to discourage him from claiming that

correlation is low when it is actually high.43

The solution to [P-CO] is similar to the solution to [P-NOC] in that Eq. 1-17 again

determines the optimal mechanism. If the low-correlation scenario is relatively likely (i.e., if u0

is so large that 1-17 holds), the principal only affords rent to the high-cost agent in the (relatively

unlikely) event that correlation is high:




42 The incentive of the high-cost agent to claim that correlation is low when it is actually high is always binding at
[P-CO].
43 If [P-CO] could be considered as two independent CM problems, there would be four degrees of freedom (recall
that a standard CM problem has two degrees of freedom). Since one degree of freedom is utilized to prevent
untruthful correlation reporting from the high-cost agent if correlation is high, problem [P-CO] has three degrees
of freedom instead of four.










1 = a [ao0o ]+(I-ao)o j>0
S 2a0 1


The principal provides rent to the high-cost agent if correlation is high by increasing the

payment t' (without reducing t' accordingly) above the t, that an independent CM

mechanism would require (i.e., one designed specifically for the high-correlation setting).44 As a

result, the lottery (t ,, ti ) under the low-correlation scenario is relatively less attractive than

the lottery (t' tG ) obtained if truthfully revealing that correlation is high. Consequently, the

rent provided by the lottery (t ,, tE ) prevents the high-cost agent from underreporting the

correlation.45

To reduce the rent of the high-cost agent if correlation is high, the principal also requests

less than the efficient effort from the high-cost agent if correlation is low. This suboptimal effort

e < e reduces how extreme his payments need to be if correlation is low, because it reduces

the rewards and penalties included in those payments as a means to avoid untruthful cost

reporting by the low-cost agent:46


V'(eHH) = 1 '< ( )HH 1; and
(1 + A) (2a -1) /oua,


v"(eZL) =1 A- (a a\ ma \L 1.
(1 + A) (2a 1) ( ao)



44 Equation 1-7 would determine this value, for a = al.
45 Since the lottery for the high-cost agent if correlation is high is even more extreme than what an independent CM
mechanism would require, a low-cost agent does not have the incentive of claiming high-cost under the high-
correlation environment. He also does not have the incentive of claiming high cost and low correlation. The
reason is as follows. If the high-cost/low-correlation lottery is unattractive for a low-cost agent if correlation is
low, it is even more unattractive for a low-cost agent if correlation is high (the low payment tm0 would arise
even more frequently, from al > ao).
46 Since the effort exerted by the high-cost agent in the high-correlation scenario does not have any impact on rents,
its socially efficient level is required: eHH' em e*. Also, eLL eLH eLL eLHO e*









As a consequence of this effort distortion, the less extreme payments if correlation is low

become less attractive for a high-cost agent who would untruthfully underreport the correlation.

Making the lottery (tH ti ) under the low-correlation scenario less extreme contributes to the

rent reduction for the high cost agent if correlation is high, because the required increase in tH

is smaller than what an independent CM mechanism would dictate.

The contract just described is similar to the CM-Low mechanism at [P-NOC].47 As such,

welfare decreases substantially if the principal applies this mechanism when the correlation

coefficients are very different (with respect to the correlation at the low-correlation setting).48

The reason comes again from the rent for the high-cost agent, which increases with the

difference between both potential correlations. Therefore, if Eq. 1-17 does not hold, the high-cost

agent commands smaller rent in the (relatively likely) event that correlation is high:

S1 al -ao (1- a + L]>
a, +a, -1

Like at [P-NOC], the principal reduces rent U1 at the cost of affording rent to the low-

cost agent in the (relatively unlikely) event that correlation is low:

U,=o [a 0 +] ->0
20- k'o 1D [aHOH + ((1 aO0)D]HL
a + aO -1

The suboptimal effort levels in this case are as follows

,,o z (l ao) [1oo + (1- ) a o0 ,1
y'(eHH) =1 HH < 1; and
(1+ A)(ao +a -1) / oa

/', A (a, ao) [PO(1-- ao)+/]i] O' ,' (eL) = 1-
(1 + A) (ao + a -1) /0( (1 -ao)


47 Although it is not required due to the degrees of freedom available, the principal could offer non-stochastic
payments to the low-cost agent under both correlations: t1 = tLH t = tLHO v(e*)
48 Again, recall that 2a -1 denotes the correlation coefficient.









This contract resembles the Ex-Post mechanism at [P-NOC]. Rents again depend on the

difference between the potential correlations (a, a ) with respect to the middle correlation

coefficient (a, + a, 1), which refers to the correlation at the middle point of the distance

between al and a Another similarity with the Ex-Post mechanism from [P-NOC] resides on

the fact that the incentive constraints that prevent the low-cost agent from exaggerating cost are

binding under both correlations instead of only at the low-correlation setting.49

Limits on Communication and Exclusion

The analysis so far has assumed that the principal never excludes any agent from the game.

That is, participation and truthful revelation of private information has been required from all

agents under both potential correlations. The results from the optimization problems performed

have illustrated the welfare loss imposed by this assumption. In particular, problem [P-NOC]

showed that rent for the high-cost agent is always higher under the high-correlation scenario than

under the low-correlation setting (U1 > U ). Recall that the explanation resides on the

implications of the link between UHH > 0 and UHL < 0 (which prevent untruthful reporting by

the low-cost agent) and a, > ao. As a consequence, the best the principal could do is reduce rent

U to zero and ad allow positive rent U A natural question then arises: should the principal

always require non-negative rent UO > 0 for the high-cost agent if correlation is low?

The answer depends on the welfare achieved with and without the participation of the

high-cost agent if correlation is low. Intuitively, if the probability of the low-correlation scenario

is low enough, it may be optimal to exclude the high-cost agent from the contract. Then, consider





49 Recall that this is the reason why this mechanism was given the Ex-Post name.









the following problem, where the principal designs a mechanism that would provide negative

rent for the high-cost agent if correlation is low. The principal computes expected welfare:


W = -(1+ 2) 2T,, [-eL, + Y(eL,)] + L 2pH [-eH +
ze{L,H} JE{L,H}
S(1-35)

-A, 2 L, [t, I(e, )] + ,L2p' [tH -- )] j
z{L,H} {jL,H}

Equation 1-35 arbitrarily excludes the high-cost agent from the game if correlation is low

(i.e., if alpha is ao). To ensure that the high-cost agent does not participate (whether truthfully or

lying) if correlation is low, the next two constraints have to be satisfied:

U a [tHH -/H CHH ) (1-36)
+(1- ao)[tHL V( CHL)] <0

ao [tLH (H CLH)] (1-37)
+ (1- ao)[t, -y(,H CLL)] <0

Like at [P-NOC], to ensure that the low-cost agent participates and reports his cost

truthfully for both correlations, the next four constraints have to be satisfied:

Us a,[t Vf(68 C, )A
U [C sE {0,1} (1-38)-(1-39)
+ (1- a)[tLH (fL -CL)] > 0

Us > a,[t Vf(68 CH )A
U[ HL ( CL)] e {0,1} (1-40)-(1-41)
+ (1- as )[tHH -(8L CHH)]

Like at [P-NOC], to ensure that the high-cost agent participates and reports his cost

truthfully if correlation is high, the next two constraints should be verified:

UH = at [HH 6(H CHH) (1-42)
+ (1- a, )[tHL 68H CHL )] >0

U', > a [tL H LH
H I ILH H -CLH)] (1-43)
+ (1- a )[t, -y(H CLL)]









With no communication on the correlation and excluding the high-cost agent from the

game if correlation is low, the principal solves problem [P-EXCL]:

Maximize Eq. 1-35 subject to Eq. 1-36 to Eq. 1-43.

At the solution to [P-EXCL], the socially efficient effort level is restored: e = e *, for

i, j e {L, H}. No rent is provided to the low-cost agent under any correlation (U1 = U = 0) and

no rent is earned by the high-cost agent if correlation is high (U1 = 0). As expected, the high-

cost agent would obtain negative rent if correlation is low (U < 0), so he does not participate in

that scenario. The principal need only ensure that tL is low enough:


t, ao + a -1

Like in the previous problems, by setting tL small enough the principal guarantees that a

low-cost agent does not exaggerate his cost. Additionally, tL small enough ensures that the

high-cost agent earns negative rent if correlation is low (U < 0) because all rent is extracted if

correlation is high (U' = 0). Recall from problem [P-NOC] that U1 > U implies that U < 0

if U, = 0.

The welfare loss at the solution to [P-EXCL] is attributed to the zero effort delivered by

the high-cost agent if correlation is low. Therefore, if the probability of the low-correlation

setting is sufficiently low, the solution to [P-EXCL] may be an alternative to the Ex-Post

mechanism obtained at [P-NOC]. Recall that the Ex-Post contract is optimal at [P-NOC] when

the low-correlation setting is relatively unlikely (i.e., when po is so low that Eq. 1-17 does not

hold). Some conclusions can be drawn from numerical examples.









Figure 1-5 shows that welfare at the solution to [P-EXCL] may be still below the welfare

achieved under the Ex-Post mechanism (which constitutes the solution to [P-NOC]), even for a

relatively small probability of the low-correlation setting (5% in the example). Figure 1-6 shows

that a slight decrease of that probability (to 3%) can raise welfare at the solution to [P-EXCL]

above the welfare achieved at the solution to [P-NOC]. Finally, Figure 1-7 illustrates that when

the probability of the low-correlation scenario drops substantially (to 1%), the welfare under the

Ex-Post mechanism is almost always below the welfare attained at the solution to [P-EXCL].

Therefore, depending on the parameters of the problem, the figures show that it may not be

optimal to require participation and truthful revelation of private information from all agents

under all scenarios.50

Conclusions

This chapter contributes to the recent and growing literature on both multidimensional and

robust mechanism design. The results from [P-NOC] show that full rent extraction is impossible

if two informed agents cannot communicate the realization of the correlation to the principal.

This finding coincides with the results obtained by other researchers on robust mechanism

design, like Arya et al. (2005), Bergemann and Morris (2005), Heifetz and Neeman (2006) and

Neeman (2004). Additionally, the solution to problem [P-CO] illustrates that even with

communication on the correlation the principal may not be able to fully extract rents from the

more informed agents. These results coincide with the findings by authors investigating

multidimensional mechanism design, like Miller et al. (2007).

When limits on communication on the correlation prevail, the principal cannot tailor the

payments to the exact correlation coefficient. As a result, the high-cost agent earns positive rent

50 Although the analysis was not performed, it is conjectured that the Ex-Post type of mechanism that constitutes a
solution to [P-CO] if Eq. 1-7 does not hold could also be replaced by a contract that excludes the high-cost agent
from the game if correlation is low.









if correlation is high. This result is the most important finding of the analysis under limited

communication. The reason behind the rent for the high-cost agent under the high-correlation

setting is that his extreme lottery of payments should ensure that non-negative expected rent is

obtained under both potential correlations. Since expected rent is calculated using two different

sets of weights (one for each correlation), and since the high payment is obtained with higher

probability (and the low payment with lower probability) under the high-correlation

environment, the principal can reduce rents for the high-cost agent to zero only if correlation is

low. As a remainder, extreme payments for the high-cost agent prevent the low-cost agent from

exaggerating cost.

When full communication is restored, and if only one agent is perfectly informed about the

extent of cost correlation, the principal is still unable to fully extract rent from both agents under

both correlations. Even when the mechanism designer can tailor payments to each correlation

coefficient, the high-cost agent still obtains positive rent if correlation is high. This is the most

important result of the analysis without limits on communication. Although the rent for the high-

cost agent if correlation is high parallels the results with blocked communication, the source of

this rent is slightly different. Instead of being explained by the inability of the principal to tailor

payments to each correlation, when communication on the correlation is available the principal

has to worry about a high-cost agent untruthfully underreporting the correlation. He would do so

because he would obtain a high payment more frequently (and a low payment less frequently)

than what he should if the principal designed two independent CM mechanisms, one for each

correlation report.

The rent for the high-cost agent if correlation is high is driven by the fact that non-negative

rent is required for him under the low-correlation setting. If the principal could exclude the high-









cost agent from the game if correlation is low, the socially efficient effort level can be restored

without affording any rents to the (participating) agents. The cost of such a contract is that no

effort is delivered by the high-cost agent when correlation is low, because he does not participate

due to the negative rent he would obtain. If the probability of the low-correlation scenario is

small enough, numerical examples show that the principal can achieve higher welfare by

designing a contract that excludes the high-cost agent from the game when correlation is low.

Finally, the usual criticism of the risk neutrality assumption also applies to this study. The

solutions obtained make use of that assumption, since agents here voluntarily participate in the

mechanism proposed by the principal even when they could potentially obtain large negative ex-

post rent. Therefore, the impossibility of the full-rent-extraction result would probably be more

pronounced if risk aversion or limited liability constraints were imposed.













All Each agent The principal Both agents Each A specific The agents
parties learns his offers two observe the agent (t,C) pair is produce,
learn al, own cost contracts correlation reports assigned to transfers are
ao and parameter (3 but do not his cost each agent made and
p1, po report it depending costs are
on both reimbursed
reports



Figure 1-1. Timing at [P-NOC], when both agents observe but do not report the correlation.















Payments under the CM-Low contract


- tLL(CM-Low)
----- tHH(CM-Low)
- tHL(CM-Low)
- tLH(CM-Low)


- -- -----


0.56 0.60 0.63 0.67 0.71 0.74 0.78 0.82 0.85 0.89 0.93 0.96 1.(




.. -


-0.20


-0.40


-0.60


-0.80


Parameters:


ao= 0.55
P-o= 0.01
1t = 0.99

X= 0.60
A= 0.20


1.00


0.80


0.60


0.40


0.20


0.00


Figure 1-2. A solution to [P-CM] that can also be a solution to [P-NOC].


l(e)=

1 e
4


''
''
''
''













Welfare at P-NOC





\ -



--- ....

"- ...


2.42

2.40

2.38

2.36


2.34

2.32


2.30

2.28


2.26

2.24
0


Figure 1-3. Welfare under the two alternative mechanisms at [P-NOC].


- - CM-Low P-NOC

- Ex-Post P-NOC


(X1


Parameters:

ao= 0.55
-o = 0.70
pi = 0.30

S= 0.60
A= 0.20


l(e)=
1 e
4


.56 0.60 0.63 0.67 0.71 0.74 0.78 0.82 0.85 0.89 0.93 0.96 1.(













All Each agent Agent A The principal Each agent A specific The agents
parties learns his observes offers four reports his (t,C) pair is produce,
learn a,, own cost the contracts cost and assigned to transfers are
ao and parameter (3 correlation agent A each agent made and
p1, po reports the depending costs are
correlation on both reimbursed
reports


Figure 1-4. Timing at [P-CO], when only agent A observes and reports the correlation.













Welfare at P-NOC vs P-EXCL


2.40

2.39 -
\
2.38

2.37

2.36

2.35

2.34


- -


Ex-Post P-NOC

Altern P-EXCL




0.56 0.60 0.63 0.67 0.71 0.74 0.78 0.82 0.85 0.89 0.93 0.96 1.
0.56 0.60 0.63 0.67 0.71 0.74 0.78 0.82 0.85 0.89 0.93 0.96 1.(


Figure 1-5. Welfare at [P-EXCL] below welfare at [P-NOC]


Parameters:

ao= 0.55
lo = 0.05
pi = 0.95

S= 0.60
A= 0.20


2.33

2.32

2.31


l(e)=
1 e
4













Welfare at P-NOC vs P-EXCL


2.40


2.39


2.38 \


2.37


/
/


2.36 -


2.35 -

Ex-Post P-NOC
2.34
Altern P-EXCL

2.33

ao
2.32 .
0.56 0.60 0.63 0.67 0.71 0.74 0.78 0.82 0.85 0.89 0.93 0.96 1.CI


Figure 1-6. Welfare at [P-EXCL] sometimes larger than welfare at [P-NOC]


Parameters:

ao= 0.55
to = 0.03
t1 = 0.97

S= 0.60
A= 0.20


Tl(e) =
1 e
4















Parameters:
I

S ao= 0.55
to = 0.01
t1 = 0.99

S= 0.60
A= 0.20


l(e) =
1 e 4
4


2.40


2.39


2.38


2.37


2.36


2.35


2.34


2.33


2.32


0.56 0.60 0.63 0.67 0.71 0.74 0.78 0.82 0.85 0.89 0.93 0.96 1.AC



Figure 1-7. Welfare at [P-EXCL] almost always larger than welfare at [P-NOC]


Welfare at P-NOC vs P-EXCL


-- Ex-Post P-NOC

-- Altern P-EXCL


L









CHAPTER 2
EFFICIENCY IN THE BRAZILIAN SANITATION SECTOR

Introduction

Early work on the relative performance of water and sewerage (WS) utilities by Crain and

Zardkoohi (1978) tried to determine whether private U.S. water utilities attained a more efficient

level of operation than public ones. Since then, a number of papers have been published on the

efficiency of WS utilities. Some authors have also focused on the private vs. public issue, while

others have tested other hypotheses, like the existence of economies of scale, economies of

scope, or the possible homogeneity or homotheticity in the production technology. Data

availability limited the types of studies: early papers focused mostly on utilities in the U.K. and

U.S. because these countries pioneered the collection and publication of data on WS firms.

Until a decade ago, little research was conducted into the efficiency of WS utilities in

developing countries. Then, studies began to address the performance of water systems using

quantitative techniques. A number of papers focused on Asian and African water utilities, most

of them supported by the World Bank.1 These studies provided insights for countries

implementing World Bank policies to increase coverage and quality of WS services in their

regions.

Politically potent and economically important, WS utilities generate public concern over

efficiency. Using data from Brazil, this study extends the standard approach by comparing not

only the performance of public and private WS firms, but also the performance of different types

of public WS operators.

There are four different types of WS providers in Brazil. The first type consists of regional

public operators, which provide services at the state level. The other three types of WS operators


1 For example, Estache and Rossi (1999), Estache and Rossi (2002), and Estache and Kouassi (2002).









provide services at the local (municipal) level. The first local type consists of private firms that

have signed concession contracts with the municipalities where they operate. The other two local

types both consist of publicly-owned operators, but they differ in their legal status. One type

consists of local public providers that are organized similarly to a corporate business. They are

called public-corporative operators throughout this chapter. The other type consists of local

public providers that are run like not-for-profit organizations. They are called public-non-

corporative providers throughout this study.

Besides the public vs. private discussion, controversy exists in Brazil about whether

municipalities or states should be responsible for WS provision. The Association of State-Owned

Sanitation Firms (AESBE), for example, argues that WS services should be provided at the state

level for two reasons: one, the larger scale of operation permits scale economies; two, there is a

possibility of cross-subsidization between poorer and richer municipalities.2 In contrast, the

National Association of Municipal Sanitation Services (ASSEMAE) favors municipal provision

on the grounds that WS services are an essential necessity for the population.3 ASSEMAE does

not explain in detail why the essential nature ofWS services calls for municipal and not state

provision, but standard arguments are based on local control, responsiveness to citizen concerns

and awareness of local conditions.

This study finds that WS provision in Brazil is characterized by economies of scale.

Therefore, since an increase in output generates a less-than-proportional increase in (operating)

costs, WS provision at the state level should be preferred. The potential efficiency gains are not

trivial when one recalls that Brazil has a population of 187-million people.


2 Associaqio das Empresas de Saneamento Basico Estaduais. See Ihip \ \ .aesbe.org.br (last visit: March 26,
2007).
3 Associaqao Nacional dos Serviqos Municipais de Saneamento. See http://www.assemae.org.br (last visit: March
26, 2007).









In a first stage, a fixed-effects panel data model with data for 2000-2004 is employed. A

cost function is proposed, identifying firm-specific (operating) costs which account for

inefficiency and other unobserved heterogeneity. In a second stage, those firm-specific costs are

explained by means of firm-type and other time-invariant indicator variables. The results show

that regional operators have lower firm-specific costs than local-private and local-public

providers. This finding indicates that the efficiency gains of the state providers from their scale

of operation are augmented by their lower firm-specific costs.

This work also shows that the local-public vs. local-private comparison depends on how

the local public provider is organized. Private and public-corporative providers have lower firm-

specific costs than public-non-corporative providers. This finding indicates that the WS operators

organized as not-for-profit organizations have the highest firm-specific costs in Brazil.

In spite of the firm-specific cost differences found, it is worth mentioning that these

differences represent a small portion of operating costs. The first-stage regressions illustrate that

the output produced, input prices and other technological factors explain most of the variation of

operating cost, regardless of the firm-type. As a result, the firm-specific cost differences found

are not substantial from an economic point of view.

Quantifying the relative efficiency of the Brazilian regional operators, Tupper and Resende

(2004) use Data Envelopment Analysis (DEA) with data for 1996-2000. The efficiency scores

obtained are considered in the construction of a proposed linear reimbursement rule that

constitutes a yardstick mechanism. However, the authors acknowledge that its implementation is

constrained by the weak current regulatory framework. Utilizing DEA with data for 1998-2002,

Seroa da Motta and Moreira (2006) argue that the government level at which conceding authority

resides is not a crucial barrier to the Brazilian sanitation sector's development when looking at









the operators' performance. Unlike this study, they find that ownership does not matter for

productivity gains for municipal services. Like this chapter, they find that regional operators

benefit from larger scale economies.

Evidence on the beneficial effects of private sector participation (PSP) in the Latin

American sanitation sector is not conclusive. In Argentina, Bolivia and Brazil, for example,

Clarke et al. (2004) find that even when connection rates to piped water improved following the

introduction of PSP, connection rates similarly improved in the control regions that never

privatized. In contrast, Galiani et al. (2005) find that child mortality in Argentina fell 8% in the

areas that privatized their water services. They conclude that while privatization is associated

with reductions in deaths from infectious diseases, it is uncorrelated with deaths from causes

unrelated to water conditions. The ambiguity on the beneficial effects of PSP in the Latin

American sanitation sector coincides with the results from other regions.4

This chapter first presents an overview of Brazil's water and sewerage industry. The study

then illustrates the two-stage methodology utilized. A following section present the results

obtained. After performing some sensitivity checks, conclusions are provided in a final section.

Overview of Brazil's Water and Sewerage Industry

In 1971, Brazil created a national plan for WS provision (PLANASA).5 This plan

delegated authority for the provision of WS services to twenty seven newly born state-owned

companies. According to PLANASA, these public companies were the only sanitation entities

authorized to obtain financing from the National Housing Bank (Banco Nacional de Habitacao -

4 For the US, Bhattacharyya, Parker and Raffiee (1994) found evidence of greater efficiency in public utilities,
Crain and Zardkoohi (1978) found evidence in favor of private operators, and Byrnes et al. (1986), Feigenbaum
and Teeples (1984) and Fox and Hofler (1986) found no difference between public and private operators. For
Asia, Estache and Rossi (1999) found evidence in favor of private operators while in a later study Estache and
Rossi (2002) found no difference. For Africa, Estache and Kouassi (2002) found evidence in favor of private
firms.
5 Plano Nacional de Saneamento. See Soares (2001) for a detailed description.









BNH). This feature made PLANASA attractive for the municipalities that were interested in

expanding their sanitation systems.6 About 3,200 municipalities joined the new plan, awarding

concessions to the state-owned companies for 20 to 30 years.7 On the other hand, about 1,800

municipalities never adhered to PLANASA, providing WS services on their own ever since. The

three types of local operators (private, public-corporative and public-non-corporative) provide

WS services within the municipalities that never joined the system.

The PLANASA model started to decline in the 80's. After 1986, the BNH was unable to

finance the required expansion of the WS sector due to a weak fiscal situation of the federal

government.8 Anti-inflationary policies may have also played a role, since the government at that

time pressed for low water tariffs to keep inflation under control. The 80's decade was also

characterized by an emphasis on decentralization, best illustrated in the constitutional reform of

1988. The centralizing concept of PLANASA, on the other hand, was more in accordance to the

military regimes of earlier decades.

There have recently been some attempts to define a new framework for the WS sector. Bill

4147/2001, for example, intended to allow for more private participation. This bill defined the

states to be the conceding authority in metropolitan areas. However, the constitutional reform of

1988 granted to municipalities the right to make concessions for public services of "local"







6 See Faria (2005).
SIn many cases, however, there was never any formal contract between the municipality and the state-owned
companies.
SPLANASA formally extinguished in 1992. See Parlatore (1999).










interest.9 Due to that controversy about the interpretation of the Constitution, Bill 4147 never

became law. 10 11

A new Bill 5296/2005 also attempted to redefine the rules for the WS sector. A

Parliamentary Commission approved it on July 2006 after many modifications and it recently

became Law 11445 in January 2007. This new bill specifies that municipalities have the

conceding authority over services of local interest. Nevertheless, the implications of the new

Law 11445 are yet to be seen.

Due to the heated debate about where the conceding authority resides, only some

municipalities that never adhered to PLANASA have made concessions to private operators.12

These private companies provide WS services to less than 4% of the population. 13 The Brazilian

Association of Private Water and Sewage Operators (ABCON) suggests that only through a

more active private participation will the WS sector meet the high investment levels required. 14

According to 2004 data, approximately 112,000 people are directly employed in the WS

sector, almost 90% by the state-owned companies. The national coverage for water services is

roughly 85%, although the sewage coverage is below 60%. On average, almost 30% of the

treated water produced is unaccounted for, due to leaking through broken pipes and illegal


9 Water distribution and sewerage collection are defined as services of "local" interest. On the other hand, water
catchment and water and sewerage treatment are defined to be of local interest only in case of exclusive use by
the municipality. See Minist6rio das Cidades, http://www.cidades.gov.br (last visit: March 26, 2007).
10 There was also a strong opposition from the public and representative institutions, like the Brazilian Association
of Sanitary and Environmental Engineering (Associaqio Brasileira de Engenharia Sanitaria e Ambiental -
ABES). See http://www.abes-dn.org.br (last visit: March 26, 2007).
11 A Nacional Water Agency (Agencia Nacional de Aguas ANA) was also created in 2000. However, the main
function of ANA is to monitor the utilization of water resources. Its role as a regulator is yet to be defined.
12 See Vargas and De Lima (i 2' 4).
13 This figure contrasts with the situation in other infrastructure sectors like telecommunications, railroads and
electricity, where private participation is much more active. See Oliveira and Fujiwara (2005) and Pinheiro
(2003).
14 Associaq~o Brasileira de Concessiondrias de Serviqos Puiblicos de Agua e Esgoto. See http://www.abcon.com.br
(last visit: March 26, 2007).









connections. Furthermore, only 50% of the sewage volume collected receives some type of

treatment. Table 2-1 shows these statistics by operator-type.

Methodology

Duality theory implies that the production technology of a firm can be modeled with a cost

function, where the firm's cost depends on its output level and the prices of the inputs employed

in production. Other factors may also affect the firm's technology and hence the firm's costs.

Specifically:

c =c(q,w,z) (2-1)

where c denotes cost, q denotes output level, w denotes input prices and z includes other

control variables.15 To empirically estimate Eq. 2-1, a panel data framework is adopted:

Y, = Xtfl + u, + E, (2-2)

where Y,, denotes the dependent variable for individual i at time t, X,, denotes the vector

of explanatory variables, u, accounts for time-invariant heterogeneity at the individual level and

E,, denotes random statistical noise. Heterogeneity is the denomination of the observed and

unobserved unique individual characteristics.

Fixed or random effects models can be adopted for panel data. The fixed effects model

allows unobserved heterogeneity to be correlated with the explanatory variables. In contrast, the

random effects model assumes that any unobserved heterogeneity is distributed independently of

the covariates. In the context of this study, correlation between unobserved heterogeneity and the

explanatory variables is hard to rule out. Such a correlation would exist, for example, if the firm



15 Control variables have been sometimes denominated hedonic measures, referring to the approach introduced by
Spady and Friedlaender (1978) for the trucking industry. They emphasize that the service dimensions provided
by the firm should enter the cost function as explanatory variables. Feigenbaum and Teeples (1983) first
introduced the hedonic approach to the water sector.









can modify its output level based on private information about its unobserved inefficiency.

Therefore, a fixed effects model for panel data is employed. The fixed effects formulation allows

unobserved inefficiency to be captured by the firm-specific coefficients. 16 In a panel data

framework, the cost function from Eq. 2-1 takes the following form:

c,, = /o +, fq,, +8 zI, + Pz,, + E, +u,, (2-3)

where the f 's are parameters to be estimated and u, denotes cost inefficiency and any

additional unobserved heterogeneity. It is assumed that the ,, are i.i.d and uncorrelated with the

covariates. In contrast, the u are allowed to be potentially correlated with the explanatory

variables. 17 The least squares dummy variables (LSDV) estimator is utilized, including also a

year-specific effect:18

c,, = a, + P q,t + Pf. i",, + l z,, + Yt + ,, (2-4)

where the firm-specific intercepts a, f, + u, account for inefficiency and any other

unobserved heterogeneity. Utilizing this formulation, Schmidt and Sickles (1984) proposed the

measure a,* = a, min(a) to construct a ranking of relative inefficiency. Their approach

permits the computation of individual inefficiency terms relative to the most efficient firm in the

sample. 19 That calculation might be appropriate when one is concerned about efficiency at the

individual-firm level. For example, a regulator could set the tariff of firm "A" based partly on its


16 According to Greene (2005), assuming that inefficiency is time-invariant is not a problem in short panels. This is
especially true in the water industry, which is characterized by low technological change.
17 It is unnecessary to make any distributional assumption on the inefficiency term p If one is willing to make
distributional assumptions on the [i, Maximum Likelihood would theoretically allow for more efficient estimates
than both fixed and random effects models. Nevertheless, Kumbhakar and Lovell (2000) and Murillo-Zamorano
(21 114) mention several papers that after performing empirical comparisons of the three approaches generate
similar efficiency rankings, especially at the top and bottom of the distribution.
18 The LSDV estimator is equivalent to the within-groups estimator.
19 Ashton (2000) constructs an efficiency ranking of British water firms utilizing that measure.









relative efficiency with respect to firm "B", as suggested by Shleifer (1985). However, the focus

of this study is not yardstick comparison between individual operators but rather between

different types of firms. The goal of this chapter is to identify whether regional, local-public or

local-private firms are relatively more efficient in providing WS services in Brazil. Thus, an

alternative analysis is pursued.

After estimating Eq. 2-4, the predicted firm-specific costs a, are computed. A high &,

indicates an inherently high cost for firm i, even when controlling for output, input prices and

other exogenous variables. Once the predicted firm-specific costs are obtained, an additional

regression is performed. In this second-stage regression, the dependent variable is the predicted

firm-specific cost (per unit of output), while firm-type indicators and regional dummies are the

explanatory variables:20


-= +A 3*Type, + *Region, + (2-5)
q,

The vectors 3 and contain the coefficients for each of the Type and Region indicator

variables. Meanwhile, q, represents the average output of firm i for the period under analysis. If

the coefficients 3 for the firm-type dummy variables are statistically significant in Eq. 2-5, there

will be evidence of relatively distinct firm-specific costs between the different operator-types.

Following the extant literature, Operating Cost is utilized to represent the dependent

variable c on Eq. 2-4.21 Wage is employed to represent input prices w, since they account for




20 Although output is present in the first stage regression, the firm-specific costs are going to be correlated with
output by the nature of the fixed-effects model. That is the explanation behind the utilization of firm-specific
costs per unit of output.
21 Although it is also conceivable to use total cost as the dependent variable, that would require data on the price of
capital, since depreciation charges constitute a large share of total costs. Since reliable data on the price of
capital are unavailable, this study focuses only on operating cost, which excludes depreciation.










more than 40% of the operating cost. 22 Wages were calculated as the ratio of total labor

expenses divided by the number of employees, as it is standard in the literature.23

Although the volume of water produced seems like the most appealing output variable, the

number of connections has also been widely used by researchers.24 Thus, both Volume and

Connections are employed as two alternative measures of output q. As control variables, this

study includes Network Length,25 the Percentage of Urban Population,26 a Metering Index, 27 a

Fluorination Index28 and a Sewerage Dummy that equals 1 if the firm also provides sewerage

collection (not all operators provide both services).29


22 Data on other input prices are limited. Some data on energy consumption suggests that energy is the second most
important input, representing around 20% of operating cost.
23 More detailed data on input prices would theoretically allow for more efficient estimates utilizing the Seemingly
Unrelated Regressions (SUR) model proposed by Zellner (1962). This model consists of a multivariate
regression system. Besides the cost function, the input-demand share-equations are utilized, enhancing the
efficiency of the estimation because the same coefficients participate not only in the cost function but also in the
input-demand share-equations.
24 The number of connections is employed by Ashton (2000), Estache and Rossi (1999), Estache and Rossi (2002)
and Teeples and Glyer (1987). The volume of water produced is utilized by Antonioli and Filippini (2001),
Aubert and Reynaud (2005), Bhattacharyya, Harris et al. (1995), Bhattacharyya, Parker and Raffiee (1994),
Bottaso and Conti (2003), Corton (2003), Crain and Zardkoohi (1978), Cubbin and Tzanidakis (1998), Estache
and Rossi (1999), Estache and Rossi (2002), Fabbri and Fraquelli (2000), Fox and Hofler (1986), Kim (1987),
Stewart (1993) and Teeples and Glyer (1987). The number of customers is used by Antonioli and Filippini
(2001), Aubert and Reynaud (2005), Fabbri and Fraquelli (2000) and Saal and Parker (2000).
25 The length of pipes is utilized by Antonioli and Filippini (2001), Bottaso and Conti (2003), Corton (2003),
Cubbin and Tzanidakis (1998), Fox and Hofler (1986), Kim (1987) and Stewart (1993).
26 A proxy of density is used by Bottaso and Conti (2003), Fabbri and Fraquelli (2000) (ratio between population
served and the length of pipelines) and Teeples and Glyer (1987) (connections per mile of line). The percentage
of non-domestic consumers is employed by Bottaso and Conti (2003), Estache and Rossi (1999), Estache and
Rossi (2002), Fox and Hofler (1986), Kim (1987) and Stewart (1993).
27 The percentage of metered connections is used by Cubbin and Tzanidakis (1998), Estache and Rossi (2002),
Feigenbaum and Teeples (1983) and Teeples and Glyer (1987).
28 A proxy for quality is used by Antonioli and Filippini (2001) (dummy indicating if water has to be chemically
treated before distribution), Estache and Rossi (1999) (continuity), Estache and Rossi (2002) (continuity),
Feigenbaum, and Teeples (1983) (water treatment index), Fox and Hofler (1986) (tests of water quality and tests
of organic contamination), Saal and Parker (2000) (percentage of water that is compliant with key parameters
relative to the compliance percentage for England and Wales) and Teeples and Glyer (1987) (water treatment
index).
29 Other control variables have been also employed in previous literature. The percentage of water losses is used by
Antonioli and Filippini (2001) and Bhattacharyya, Harris et al. (1995). The storage capacity is used by
Feigenbaum and Teeples (1983), Fox and Hofler (1986) and Teeples and Glyer (1987). A dummy indicating if
the utility has to purchase water from other utility is employed by Aubert and Reynaud (2005), Feigenbaum and










Earlier literature suggests that a longer network should be associated with higher costs due

to its maintenance (fixing leaks, for example). Previous research also suggests that a higher

metering index should be associated with higher costs due to the reading and maintenance of the

meters. The fluorination index should also be associated with higher costs due to a more intense

chemical treatment of water before delivery. The sewerage dummy is also expected to show a

positive sign, capturing the higher operating cost of providing both water and sewerage services.

Finally, the effect of a higher proportion of urban population in the area served is difficult to

predict. On the one hand, many researchers argue that having customers densely located in a

small area reduces costs. On the other hand, Feigenbaum and Teeples (1983) argue that "we

should expect that it is more costly to supply more densely developed service areas, which

requires more hydrants, higher water pressure and greater peak capabilities for fire protection."30

The main source of data is the National System of Sanitation Information (SNIS) of

Brazil.31 Operators voluntarily join the SNIS, which started collecting data in 1995. The number

of firms providing data has increased each year ever since. This study utilizes an unbalanced

panel for 2000-2004. There are approximately 180 observations for 2000 and 340 observations

for 2004, with almost 1200 observations in total.32 The SNIS is part of the Modernization


Teeples (1983), Fox and Hofler (1986) and Teeples and Glyer (1987) (they use the proportion of water that is
purchased). The number of districts served is used by Corton (2003). A dummy indicating if the utility obtains
water from surface sources is used by Aubert and Reynaud (2005), Bhattacharyya, Harris et al. (1995), Bottaso
and Conti (2003) (they use river sources), Estache and Rossi (1999) and Fox and Hofler (1986). I ignore the
reasons that support the inclusion of a proxy for capital stock, as done by Antonioli and Filippini (2001) (number
of water wells), Aubert and Reynaud (2005) (average net base rate divided by the estimated price of capital),
Bhattacharyya, Harris et al. (1995) (residual of the revenue less variable costs) and Bottaso and Conti (2003)
(replacement costs of net tangible assets). Such a variable would be an appropriate covariate in a production
function, but not in a cost function
30 Feigenbaum and Teeples (1983, p.674). They confirm that result in their paper.
31 Sistema Nacional de Informacoes sobre Saneamento. See www.snis.gov.br (last visit: May 7, 2007).
32 Missing data on some variables explain the variation in the number of observations across the two models
presented in the next section. The online data are split into several files, each containing only a certain group of
variables (financial, descriptive, operational, etc) and a certain type of firm (regional, local, etc). These
spreadsheets were pooled together for this work.









Program of the Sanitation Sector (PMSS), which Brazil started in 1992 with financial support

from the World Bank.33 To adjust monetary figures for inflation, data on the Brazilian Consumer

Price Index (IPCA) are utilized.34

Results

Summary statistics of the variables used in the first-stage LSDV regressions are presented

in Table 2-2, discriminated by operator type. The size difference between the regional and local

operators is evident when observing the average output, network length and operating cost of

each type of firm. For example, from the point of view of the number of connections, the average

regional operator is 14 times bigger than the average public-corporative provider.

Table 2-3 presents the results from both LSDV regressions, according to Eq. 2-4.

Alternatively, both Connections and Volume are positive and statistically significant. The

hypotheses that their coefficients are equal to 1 is rejected, which provides evidence of

increasing returns to scale regardless of the output variable chosen. For example, a 10% increase

in the volume of water produced generates only a 0.98% increase in operating cost.35 This

confirms the argument made by AESBE, which favors state-level provision due to economies of

scale. As expected, Wage also has a positive and statistically significant effect on cost.

From the set of control variables, the Metering Index shows the expected positive sign and

statistical significance. This means that when the fraction of metered connections is higher,



33 Program de Modemizacao do Setor de Saneamento. The PMSS resides on the sphere of the National Secretariat
of Environmental Sanitation (Secretaria Nacional de Saneamento Ambiental), which depends on the Ministry of
Cities (Minist6rio das Cidades). See www.cidades.gov.br (last visit: May 7, 2007).
34 The Indice Nacional de Precos ao Consumidor Amplo (IPCA) is constructed by the Brazilian Institute of
Geography and Statistics (Instituto Brasileiro de Geografia e Estatistica IBGE). See www.ibge.gov.br (last
visit: May 7, 2007).
35 To test for economies of scale that may vary with output, specifications including the square of the output
variable were run. The results were not satisfactory. The Volume2 variable was not statistically significant, while
Volume remained statistically significant with little change in the value of its coefficient. Meanwhile, both
Connections and Connections2 became statistically insignificant.









operating cost increases. This is in line with earlier research suggesting that reading and

maintaining the meters has a positive impact on cost. The Sewerage Dummy coefficient is

positive and statistically significant. This means that collecting and treating sewage increases the

operating cost of a water provider. Finally, Network Length, the Percentage of Urban Population

and the Fluorination Index are statistically insignificant.

The first-stage LSDV regressions in Table 2-3 also illustrate that output, input prices and

other technological factors explain most of the variation of operating cost.36 As a result, the firm-

specific costs provided by the first-stage regressions represent a small portion of the operating

cost (less than 1%).

After running the first-stage LSDV regressions, the predicted firm-specific costs &, are

obtained. Following Eq. 2-5, a second-stage regression is then performed, where the firm-

specific cost (per unit of output) plays the role of the dependent variable. Firm-type and five

region-indicators are the time-invariant explanatory variables in the second-stage regression.

Dummies for the five different regions in which Brazil is divided are included because

geographic heterogeneity may differently affect the cost of access to raw water.37 Even when the

specific effect of regional heterogeneity is not the focus of this study, it is important to control

for that time-invariant characteristic. If the firm-type dummy variables are statistically significant

in the second-stage regression, there will be evidence of relatively distinct firm-specific costs

between the different types of operators.

Table 2-4 presents the results of the second-stage regression. The negative and statistically

significant coefficient of the Regional dummy indicates that regional providers have lower firm-


36 The high explanatory power of both models remains even if the firm fixed-effects (not reported) are excluded. In
that case, the R2 is still above 0.95.
7 These regional dummies could partially capture different energy prices as well.









specific costs than all other operators. At the local level, meanwhile, the firm-specific cost

comparison between private and public operators depends on how the public provider is

organized. The positive and statistically significant coefficient of the Public-Non-Corp dummy

variable indicates that the local-public-non-corporative operators have higher firm-specific costs

than both the local-private and the local-public-corporative providers. Table 2-4 also shows that

there is not a statistically significant difference between the firm-specific cost of local-private

and local-public-corporative operators.

Table 2-5 illustrates the value of the firm-specific costs, obtained from the results of the

second-stage regression. The table shows that the regional public operators are the lowest-cost

WS providers, while the local-public-non-corporative operators are the highest-cost WS

providers in Brazil. Although the differences are substantial and statistically significant, it is

worth recalling that the firm-specific costs are not a significant portion of the operating costs.

Sensitivity Checks

As a first sensitivity check, a balanced panel was used. Since the sample size increases

over the years, it is important to check that incorporating new firms does not affect the analysis.

The results are presented in Table 2-6. The positive and significant effect of both output

variables (Connections and Volume) remains. The same is true for the Wage variable across both

specifications. Table 2-7 contains the results of the second stage regressions. It can be verified

that the lower firm-specific cost for the Regional type is confirmed when using a balanced panel,

regardless of the output variable chosen. However, the higher firm-specific cost for the public-

non-corporative type is not obtained when a balanced panel is employed. The reason behind this

statistically imprecise result could be the loss of observations when utilizing a balanced panel.

That is, most of the observations dropped when using a balanced panel correspond to local firms

(which include the public-non-corporative type) rather than Regional ones.









As a second sensitivity check, and utilizing again the larger unbalanced panel, the Regional

type was excluded from the sample. The reason for this sensitivity check is that the size

differential between the Regional and Local operators could affect the conclusions. Table 2-8

shows the results of the first stage LSDV regression and Table 2-9 contains the results of the

second stage regression. The first stage regressions show that the positive and statistically

significant effect of output, wage and the metering variable are again verified, along with the

sewerage dummy. Furthermore, the coefficients obtained are similar to those in Table 2-3 and

Table 2-4. For example, the measure of economies of scale indicates that a 10% increase in the

volume of water produced generates a 0.94% increase in operating cost. The second stage

regressions show that the higher firm-specific cost for the public-non-corporative type is again

confirmed, providing confidence about the conclusions drawn earlier. The statistically

insignificant difference between the private and public-corporative operators is also verified.

Conclusions

Brazil is a country that lacks perfect access to WS services. Efficiency improvements could

free up funds for network expansion, which would constitute a step towards a desired full service

situation. Therefore, greater attention for cost-containment is needed, regardless of the

jurisdictional and ownership/organizational status of the WS operators. Improving our

understanding of relative performance can help policy makers focus on the sources of differential

cost patterns.

The results of this study suggest that, at least for Brazil, evidence of economies of scale is

enough to claim that WS provision at the state level is more efficient than WS provision at the

municipal level. Economies of scale generate substantial cost savings, which far outweigh any

potential differential in firm-specific costs. As such, the argument made by AESBE seems more

compelling than the argument made by ASSEMAE.









In addition, this chapter finds evidence of inherently lower firm-specific costs (per unit of

output) for regional WS firms than for all other types of WS operators in Brazil. These lower

costs reinforce the efficiency gains the regional firms achieve through actual economies of scale.

Finally, this study shows evidence of higher firm-specific costs (per unit of output) for local-

public-non-corporative providers than for local-private and local-public-corporative providers.

Future research could examine what features generate the intrinsic cost differences among

operator types. In particular, the higher firm-specific cost for the public-non-corporative type

deserves further attention. It may be important to check whether the not-for-profit motive of

those organizations actually drives their higher firm-specific costs.

Even when cost differences between the different types of WS operators were found, it is

worth noting that these differences represent a small portion of operating costs. The first-stage

regressions illustrated that output, input prices and other technological factors explain most of

the variation of operating cost, regardless of the firm-type. As a result, the firm-specific cost

differences presented are significant from a statistical point of view, but less significant from an

economic perspective.

Clearly, much work remains. For the purpose of rewarding good performance and

penalizing weak performance, scholars and practitioners need to develop efficiency-measuring

procedures that can pass legal challenges. The process must continue to build on the pioneering

research of those whose work is cited in the references. In particular, the publication of league

tables is one way to put pressure on the weakest performing WS utilities. Similarly, the managers

of WS utilities in the top 20 percent might be awarded some share of the cost savings that can be

attributed to their efforts. Those promoting improvements in WS sector performance can take

steps to reduce production costs and free up cash flows for network rehabilitation and expansion.









Identifying, implementing, and evaluating good incentive systems represent a challenge for

regulators.

A final issue that also deserves future research follows. The analysis in this chapter only

considered relative measures of efficiency. The goal was to identify sources of cost differences

between the different types of operators. However, cost savings for the entire industry could also

be estimated utilizing the results obtained. For example, reducing water losses would also free up

funds for network expansion. Table 2-1 shows that water losses in 2004 stood at almost 30% on

average, while they were 49% for the state operators. Thus, a reduction of water losses by 10%

should not be hard to achieve. Yet, it could represent an almost 1% lower operating cost. Some

researchers suggest that the only explanation behind the lack of incentives for reducing water

losses is that it may be cheaper to produce more water instead. Although geography might

validate this statement in some cases, the issue deserves further exploration.










Table 2-1. Average statistics by operator-type for 200438
Type of firm # Connect. Empl. Water Sewer. Water Sewer.
cover, cover, losses Treatm.
Private 31 30470 88.9 80.1% 50.4% 29.1% 54.4%
Public Non-Corp 296 18851 99.2 86.3% 63.0% 26.6% 46.5%
Public Corp 11 75180 400.3 98.8% 68.5% 41.3% 36.6%
Regional 25 1104748 2978.6 71.3% 33.8% 48.7% 74.5%
Total 369 91660 300.7 85.1% 58.5% 28.9% 51.3%


38 The sample also includes six Microregional operators, which are not the focus of the analysis. These are public
operators that are neither regional nor local, since they provide services to just a few municipalities.










Table 2-2. Summary statistics for first-stage regressions39
Variable Private Public Non Corp. Public Corp. Regional
N=62 N=913 N=42 N=121


Operating Cost


Connections


Volume


Wage


Network Length


Dummy sewerage


Urban %


Metering %


Fluorination %


9,102,144
(12,500,360)


50,539
(67,902)

22,239
(41,240)

15,959
(6,953)

642
(840)


0.82
(0.39)

0.88
(0.09)

0.88
(0.19)

0.33
(0.46)


5,088,097
(13,124,270)


24,251
(39,282)

8,552
(17,315)

11,512
(6,256)

294
(457)


0.56
(0.50)

0.77
(0.21)

0.77
(0.32)

0.30
(0.44)


25,661,470
(34,681,050)


74,573
(67,960)

34,899
(33,649)

18,082
(9,349)

1,005
(1,044)


0.98
(0.15)

0.95
(0.05)

0.86
(0.21)

0.35
(0.47)


239,346,200
(338,851,600)

1,021,909
(1,234,135)

396,122
(583,563)

31,147
(10,613)

11,398
(12,930)


0.94
(0.23)

0.77
(0.12)

0.74
(0.28)

0.20
(0.36)


N=1163
(Standard deviations in parenthesis)
















39 Volume is in 1000m3/year and Network length is in Km. Operating cost and Wage are in Reais/year deflated
using 2000 as the base year. For the Volume variable, summary statistics are for 1172 observations. The 1163
observations include 25 observations for the Microregional category, which are not reported.










Table 2-3. First-stage LSDV regression results40
Dependent Variable: Operating Cost C


Connections


Volume


Wage


Network Length


Dummy sewerage


Urban %


Metering %


Fluorination %


Constant


Observations
R-squared


Year and firm fixed-effect not reported
Standard errors clustered at the state-level
* significant at 10% ** significant at 5% *** significant at 1%













40 Operating Cost, Volume, Connections, Wage and Network length are in In form. The statistical significance of
all coefficients is very similar when the standard errors are clustered at the Region level.


connections


Volume


0.427
(0.112)***


0.150
(0.043)***


0.024
(0.073)


0.122
(0.058)**


0.070
(0.246)


0.382
(0.177)**


0.001
(0.035)


8.496
(1.133)***


1163
0.99730


0.098
(0.040)**

0.157
(0.040)***

0.089
(0.088)

0.138
(0.067)**

0.082
(0.258)

0.409
(0.165)**

0.005
(0.035)

11.274
(0.713)***

1172
0.99729










Table 2-4. Second-stage regression results
Dependent Variable: (In) Firm-Specific Cost
per Unit of Output (from LSDV regressions)


Public Non Corp


Public Corp


Regional


Constant


Observations
R-squared


Omitted type: Private
Region fixed-effect not reported
Robust standard errors in parenthesis
* significant at 10% ** significant at 5% *** significant at 1%


Connections


0.331
(0.127)***


-0.009
(0.217)


-1.154
(0.190)***

-0.960
(0.124)***


380
0.31726


Volume


0.294
(0.124)**

-0.073
(0.176)

-0.270
(0.158)*

2.640
(0.116)***

380
0.04428









Table 2-5. Ranking of firm-specific costs across firm-types.
Firm type Firm-Specific cost per unit of output
($/Connection) ($/1000m3)


Index (Regional= 100)


Regional 0.12 10.70 100 100

Public Corp 0.38 13.03 314 122

Private 0.38 14.01 317 131

Public Non Corp 0.53 18.80 441 176










Table 2-6. First-stage LSDV regression results using a balanced panel
Dependent Variable: Operating Cost Connections Volume


Connections


0.631
(0.152)***


Volume


Wage


Network Length


Dummy sewerage


Urban %


Metering %


Fluorination %


Constant


Observations
R-squared


0.185
(0.022)***


0.014
(0.074)

0.034
(0.048)

-0.158
(0.135)

-0.002
(0.203)

0.023
(0.028)


0.127
(0.043)***

0.186
(0.026)***

0.082
(0.104)

0.046
(0.045)

-0.150
(0.139)

-0.009
(0.188)

0.023
(0.030)

13.287
(1.053)***

766
0.99711


7.894
(1.155)***


758
0.99713


Year and firm fixed-effect not reported
Standard errors clustered at the state-level
* significant at 10% ** significant at 5% *** significant at 1%










Table 2-7. Second-stage regression results using a balanced panel
Dependent Variable: (In) Firm-Specific Cost Connections Volume
per Unit of Output (from LSDV regressions)


Public Non Corp


0.176
(0.219)

-0.005
(0.301)


Public Corp


Regional


Constant


Observations
R-squared


-1.966
(0.259)***

-2.702
(0.328)***


170
0.54194


0.265
(0.179)

0.056
(0.166)

-0.339
(0.182)*

2.795
(0.249)***

170
0.11137


Omitted type: Private
Region fixed-effect not reported
Robust standard errors in parenthesis
* significant at 10% ** significant at 5% *** significant at 1%










Table 2-8. First-stage LSDV regression results excluding the Regional type
Dependent Variable: Operating Cost Connections Volume


Connections


0.439
(0.130)***


Volume


Wage


Network Length


Dummy sewerage


Urban %


Metering %


Fluorination %


Constant


Observations
R-squared


0.145
(0.045)***

0.022
(0.072)

0.119
(0.058)*

0.062
(0.252)

0.424
(0.194)**


0.094
(0.040)**

0.152
(0.041)***

0.083
(0.087)

0.135
(0.066)*

0.077
(0.264)

0.460
(0.179)**

0.005
(0.039)

11.37
(0.721)***

1047
0.99575


0.004
(0.039)


8.433
(1.326)***


1042
0.99579


Year and firm fixed-effect not reported
Standard errors clustered at the state-level
* significant at 10% ** significant at 5% *** significant at 1%









Table 2-9. Second-stage regression results excluding the Regional type
Dependent Variable: (In) Firm-Specific Cost Connections Volume
per Unit of Output (from LSDV regressions)

Public Non Corp 0.331 0.286
(0.127)*** (0.120)**

Public Corp -0.037 -0.077
(0.220) (0.170)

Constant -1.005 2.711
(0.124)*** (0.113)***

Observations 354 354
R-squared 0.1322 0.01567

Omitted type: Private
Region fixed-effect not reported
Robust standard errors in parenthesis
* significant at 10% ** significant at 5% *** significant at 1%









APPENDIX
DERIVATION OF THE SOLUTIONS TO CHAPTER 1

Limits on Communication

When Correlations are Relatively Similar

IfEq. 1-17 holds, solve [P-NOC] imposing only Eq. 1-9 (multiplier denoted with PL), Eq.

1-10 (multiplier denoted with PLO), Eq. 1-12 (multiplier denoted with PH0) and Eq. 1-14

(multiplier denoted with IL ). Solving from the first order conditions with respect to payments

yields:


PL1 =u1 >0, PLO =A O- a- >a0,
S2ao -I)


Io = a-a >0 and P, =A L +) a+ao >0.
2ao -1 2ao-I )

Setting equal to zero the first order conditions with respect to the effort vector:

aL
= -(1+ A)[p/la + p/oao][-1+ y'(eLL)]+ /2[pla + uoao ]0'(eLL)
BeLL
PL llj (eLL) PLOaO (eLL) LO aO (eLL) = 0

Substituting the value of the multipliers and simplifying: (eL) = 1

DL
= -(1 + ,)[p,(1 -a) +o(1 -ao)][-1 +'(eL)]
aeLH
+ A[p/ (1 a) + / (1 ao)]y' (eLH) PL(1 a )y'(eLH)- PLo(1- ao)y'(eLH)
-ILO(1 -ao)'(eLH) =0

Substituting the value of the multipliers and simplifying: W' (eH) = 1


aL
= -(1 + A)[1ua, + Oao[ ][-1+ V' (eH)] + A[/ a, + ,0ao ]W' (eHH)
SeHH
- PHOaOV' (e,) + LO (1 ao )' (eH, A) = 0

Substituting the value of the multipliers and simplifying:









A al -aO 1(1-ao) <
y' (eHH 1 A a HH<1
(1+ 2) (2a, 1) ( u a, + 0ao)

QL
= -(1 + A)[p~(1 a) + o (1 a)][-1 + (eHL)] + [p1 (1 ) + o(l -o)]I' (eL)

-PHo (1- ao )y(eHL)+ IOa (eHL A) = 0

Substituting the value of the multipliers and simplifying:

A a1 a0 11a0o D
(1 + A) (2ao -1) (1 p/1a /o0)

Solve from the binding constraints to obtain the payments. From Eq. 1-9 and Eq. 1-10:

tLL = y(e*) and tLH = V(e)

Substituting in Eq. 1-14 and simplifying:


tHH =-- I(eHL A) + (eHH A) HL
1 a 1 a

Rearranging terms in Eq. 1-12:


tHH = (eHH )+ HL ) L
ao ao

Equating the right-hand-sides of the last two equations:


tHL = y(eHL) a- [(1 l )HH + aoDHL]
2ao -1

Substituting back in Eq. 1-12:

1 a
tHH = (eHH )+ a- [(1 a- 0)HH +a oHL
2a, -1

The excluded constraints (Eq. 1-11, Eq. 1-13, Eq. 1-15 and Eq. 1-16) are satisfied (not

binding) by the payments above. In particular, substituting the payments in Eq. 1-11 yields the

rent for the high-cost agent if correlation is high:










UH -[(1_ a HH + aHL] > 0
2a, -1

When Correlations are Relatively Different

IfEq. 1-17 does not hold, solve [P-NOC] imposing only Eq. 1-9 (multiplier denoted with

PL1), Eq. 1-12 (multiplier denoted with PH ), Eq. 1-13 (multiplier denoted with ILI) and Eq. 1-

14 (multiplier denoted with IL0 ). Solving from the first order conditions with respect to

payments:

ILI 'U (a, ao) 0 (2a 1)
S 1> 0, > ILo -/o >0,
a, +a o -1

2a' -1 2a -1
PLi =A 2a o >0 and PH,,A 2l1- >0
a0' + a -1 a1 + a( -1

Setting equal to zero the first order conditions with respect to the effort vector:

dL
=- -(1+ A)[/a, + oao + ][-1 + (eLL)] + A[p/a, + pOao ] '(eLL)
OeLL
PLj/af' (eLL) IL,,ay' (eLL) ILoao (eLL) = 0

Substituting the value of the multipliers and simplifying: Vy'(eLL) = 1

-- = -(1 + A)[/p/ (1 a) +p /o(1 ao)][-1+ (eLH)] + A[//i (1 1) + //o (1 ao )] (eLH)
SLH

-PLI1 al '(eLH)- L (1 al)'(eLH)- I (1 c o)i(eLH) = 0


Substituting the value of the multipliers and simplifying: y' (eLH) = 1

aL
= -(1 + +)[pla, + p0ao ][-1+ q' (eHH)]+ A[p/a1 + /pOa ]y' (eHH)
BeHH
- PHOo' (eHH ) + IL1(1 a l)' (e A) + ILo (1 Ca )'(eHH A) = 0

Substituting the value of the multipliers and simplifying:










(eH) =1 a -ao u, (1 a1 )+ Oo H <
w' (eHH HH
(1+2A) (a, +ao -1) (P, + aoo)

aL
= -(1 + )[/i (1 a) + a o(1- ao)[-1 + '(eHL)]
,eHL

+ Z[/l (1 a) + /0 (1 -ao )]t' (eHL) PHO (1 a0 )' (eHL) + Ila1' (eHL -A)
+ ILOa (eHL A) = 0

Substituting the value of the multipliers and simplifying:

Ay a1 -ao0 ja,1+ + ( -o (1 o),H <
'( )= ( ^1 <1
(1+ A) (a1 + ao -1) (1 p, poao)

Solve from the binding constraints to obtain the payments. From Eq. 1-9:


1-a,
tLL = ry(eLL ) [tLH ry(eLH )]


From Eq. 1-12:


tHH = y(eH ) [tHL (eHL )
ao
50

Substituting the last two expressions into Eq. 1-13 and Eq. 1-14 yields:


tHL = y(eHL) [(1 _-1)aHH +1 a HL]
a, +ao -1


tLH = y (eLH )+ a [aoHH +( 0 o)HL]
a, +a, -1

Substituting tL in Eq. 1-9 yields:


tLL = I(eLL) --1 [a0 HH +(1 0 )HL].
a1 +ao -1

Substituting t, in Eq. 1-12 yields:


tHH = y( )+ a [(1 a1)(D + a HL
al +aO -1









The excluded constraints (Eq. 1-10, Eq. 1-11, Eq. 1-15 and Eq. 1-16) are satisfied (not

binding) by the payments above. In particular, substituting the payments in Eq. 1-10 and Eq. 1-

11 yields the rent for the high-cost agent if correlation is high and the rent for the low-cost agent

if correlation is low:


UH1 = 0 [ 1 HH 1 -H) > 0
a, +oao -1


ULo = 1- [a HH++ (1- ) HL> 0
a + ao -1

No Limits on Communication

When Correlations are Relatively Similar

IfEq. 1-17 holds, solve [P-CO] imposing only Eq. 1-19 (multiplier denoted with y19 ), Eq.

1-20 (multiplier denoted with 720), Eq. 1-21 (multiplier denoted with 21 ), Eq. 1-23 (multiplier

denoted with 23 ) and Eq. 1-30 (multiplier denoted with 30 ). Solving from the first order

conditions with respect to payments yields:

a23 -A1 ao > 0(2a 1) + /1 (a, + ao 1)>
Y23 = jl/d1 > 0, y20 = 2 > 0,
2a -1 2a -1


719 A Uo -U1 -- > 0, 21 = A > 0 and 30 = Ai > 0.
2ao -1

Setting equal to zero the first order conditions with respect to the effort vector:

8L
o- = .',,"i'-'(eo)- (1 + A)oao + y19ao~"'(e) + y c',,i''(eL) =0
LL

Substituting the value of the multipliers and simplifying: y'(e ) = 1

8L
o- = poaoV'(eHH) (1 + ))/ 0 + y.,_, ,HH)
eHH









- 723(1- ao) (eH -A) y,30ol '(eH) =0

Substituting the value of the multipliers and simplifying:


(l+ A) po ao (2ao 1)H

aL
-- = o0(1 -ao)'(eH )-(1+A)o0(1 -ao)


+ 79 (1- a, )' (e0 ) + 23 (1- o )I'(eH) =0

Substituting the value of the multipliers and simplifying: y' (e ) = 1

aL
oL= 0 0(1 o)yI'(eL)- (1 + A)Po(1- ao) + 720 (1 Lo) '(eL)
SeHL

-7. '-,,'(eo -A) 30 (1 )' (eL)=0

Substituting the value of the multipliers and simplifying:

Vy'(eo A 1 A2 ( a, (' (o o , (1 (e) += A) ( a) (2ao- 1)HL

aL
= p, a, '(e1) (1+ A),la, + y2aly'(eV )=o
ae (la,)'(eLL 11 =0211 L)=0
eLL

Substituting the value of the multipliers and simplifying: ey'(e) = 1

aL
- =/la, y'(e, )-(1+ A)j)ia + y3oal (e,) =0
elHH

Substituting the value of the multipliers and simplifying: y'(e) = 1

aL
e (LH =1,( l)'( )(11a)yl+)(1 a) + Y2(1 1) L'(e) =0
Substituting the value of the multipliers and simplifying: (e
Substituting the value of the multipliers and simplifying: V'(eLH ) =









aL
L 1 (e)V)=o
ae=H ,(1 c-t )y'(eL) (1 +A)M1 1- ) +/3o (1 a L) =0
OeHL

Substituting the value of the multipliers and simplifying: /' (e) 1

Solve from the binding constraints to obtain the payments. The five equations are as

follows:

a [t (e )]+ (1 )[t e)] 0 (1-19)

a0 [to H (eO)] + (1- a )[toL (eL)] 0 (1-20)

a1 [tL (eL)] +(1- )[tH r(el )] 0 (1-21)

a [t y(eL,) t + (eL A)]
(1 0[ o o(1-23)
+ (1 a)[t y(eH )-t + y(eo, -A)]=

a, [t- y(e, ) t H + y(e H, )](1-
(1-30)
+ (1- a, )[tL y(e, ) t + y(e)] 0

The system has an infinite number of solutions. Selecting arbitrary values for tH tHL and

tL a solution can be characterized as follows:

S(-0 o) o
tLL (eLL) [LH LH+(eL)



2ao -


2a0 -1
1i = Vy(el) -(I-- [LH- V(el
tLL L 1 H H
a,


t, =V((e,)- [t -( (el +)] o +1(I- o)LO
a1 a, (2a0 1)








Some of the excluded constraints are automatically satisfied by the payments above. For

the rest of the excluded constraints to be satisfied, the following bounds on the arbitrary

payments should hold:


I(eH )< tH

y(e1 ) a o ( 0) o
)2a+ 2--1
2a -1 1+ [ + (1 a )LL i

Y/(el H)+ 01 [a H + (1 a aO)WL
a, + a, -1


where YL r(e, + A) (e' )> 0, for i {L,H} and se {1,0}.


1 ao Ho L 1
V(eHL ) 0~ t L < min
HL 2ao -1 +(1- )0 HH
I )ao IH


Vf(elHL--
2al -11+(1-a )HH1

(1-a )(ag -ao) oaoHL
(2 1)(2o 1) +(1-


1 1 L I0 IOHL
HLao +a +1ao) HH

(1- ao)(a- ao) oaoHL
(ao + a, -1)(2ao -1) + (1 a0 )0H


I(eH ) 2 1(0, H L
2ao -1
,(eH )<+H [ H I

y(eH)+ 0 2ao -1
al +ao -1 L + )
+ + [alo +(1_- 1 0 a l )

In particular, substituting the payments into Eq. 1-22 yields the rent for the high-cost agent

if correlation is high:


U1 = a( -a [( +(1-ao)(o> 0
H(2a 1)









When Correlations are Relatively Different

IfEq. 1-17 does not hold, solve [P-CO] imposing only Eq. 1-20 (multiplier denoted with

y20), Eq. 1-21 (multiplier denoted with y21), Eq. 1-23 (multiplier denoted with Y23), Eq. 1-30

(multiplier denoted with y30) and Eq. 1-33 (multiplier denoted with y33). Solving from the first

order conditions with respect to payments yields:

2a, -1 2ac -1
720 = 2a'-1 >0, y21 = 2a 1>0,
O +aI --1 ao +a -1


723 0= 'UO>0, 730 = 1 >0 and 7/ (,33 = )- (2a- 1)>0
aO + a -1

Setting equal to zero the first order conditions with respect to the effort vector:

aL
oL = oaoVy'(eO ) (1+ A)p/ao +y. y, ei''( )=0
eLL

Substituting the value of the multipliers and simplifying: y'(e = 1

8L
o-I = ,,,,i '(eo ) (1 + A)p/oa + y ,,.,,i '' (e H)
SeHH

23(1 ao0)Vy(eH A) y30aV y'(eHH) y33(1 ,e -A)

Substituting the value of the multipliers and simplifying:

/' (eH) [/0a + /1 (1- a )] (a, a) ( ,<
w '(eHH HH
(1 + ) /ao0a ( + a1 1)

aL
o-L = 0(1 -ao)W'(eH)-(1+A)/o (1 -a) +7 23(1 -o)'(eH )=0
eLH

Substituting the value of the multipliers and simplifying: "y'(eH ) 1

8L
-a = e (1 ao)y'(eL) -(1 + A)/o(1 -ao) + 7o(1 -o)'/(eL)
SeHL









y- c( '.'(e'L A)- 3o0( l- )y'(eoL) 33c/' (eL A) =0

Substituting the value of the multipliers and simplifying:

/,(eL )=I A [/uo(1- ao)+/,u ] (a, -o ) ,
w' (e /1 'i HL
(1+A) ,o(1 -o) (ao + a-1)

L I
a = la, '(eL )- (1 + A),Ia + y/21 a '(e ) YL 33l(eLL) =0
eLL

Substituting the value of the multipliers and simplifying: y'(el) 1

aL
/ = ula '(eH1) (1 + A) a, + y,30 a (e,) =0
eHH

Substituting the value of the multipliers and simplifying: y'(e) = 1

8L
-- /1(1- aI)y'(e (H (1 +)/I(1- a) + 21(1 -a)y'(e )
OeLH

+ 33(1- a )'(el )=0

Substituting the value of the multipliers and simplifying: y' (e ) = 1

aL
L /(1- )'(eL) (1 (1 -a + /30(1 -a)'(e})=0
OeHL

Substituting the value of the multipliers and simplifying: y'(e) =1

Solve from the binding constraints to obtain the payments. The five equations are as

follows:

a, [to (eO )]+ (1- ao )[to -(eL )] 0 (1-20)

a [t~L L(e )] +(1- )[tH -y(e H)] 0 (1-21)

a [to y(e) tL +(e A)] (1-23)
+ (1 )[t (1-23)
+(1-a )[to -(eH )- t, + V(e, -A)]= -









a1 [t'H (eH)-toHH + y(eH )] (1-30)
+ (1- a )[tL (eL) t + y (e,)]= 0
1[ [tLL L L

[L -(eLL) HL +(eL (1-33)
+(1 a)[tH I(e )- tHH + H(e -A)]=

The system has an infinite number of solutions. Selecting arbitrary values for tH, tHL and

tLH, a solution can be characterized as follows:

tLL -- r(el) [tLH V/(el)]
a1
to =V(eo o (1 ) +
tHL = y(eHL) a0 -1 1 al)HH + aDHL]
a, + 1

o = o 1-ao 1[(1 ) + 0]H
a1 + a0 -1

o /(a1 ) [0 o H ]o 1 a0 [to H /(eHo ]
tLL =o )+ HH + (1 o0 )IL LH
a, (a, + a, 1) a,

tHH =V (e1H)+ al ao 1[(- al)( + a00 (HL1- ) -(e1 )
a, (a1 + a 1) a

Some of the excluded constraints are automatically satisfied by the payments above. For

the rest of the excluded constraints to be satisfied, the following bounds on the arbitrary

payments should hold:

a 0 a0 OH 0 HH
Vf(e )+ ao 0 L (1ao)(a- ao) a HH
y(e0L)(Do
2ao -1 + (1- ao)YTL (a +a -1)(2o 1) _+(1 L

tLH < min< (eH) + a HL + [(1 -i a, ) L + a, LH
a1 +ao -1 al + ao -1

)+(e--) + a [(I)oH +(1- O)DL]
(a, + a, 1)










L (ef1 ) HH1 ( 1- a, -H 0 H1-L
(L 2D-1_+ 0 (a,+ 0-1)(2a,-1) + a 0


t+' 1
(a' + ao 1)



YV(el ) + a (I i )T' + al(a- a) (HH
2a, 1 L+a (a +ao-1)(2a -1) + a, I

t < min< (e ) + + ( a)
ao + ao -1 a)
)7(el)+( +(H -o)0L
(a, + ao 1)



low-cost agent if correlation is low and the rent for the high-cost agent if correlation is high:

U (a,+ a, -
L (a= + ao >0, [, ( yz9 + (/ I > and 4 D 0 > 0.

(a1 + 1)



Limits on Communication and Exclusion

Solve [P-EXCL] imposing only Eq. 1-38 (multiplier denoted with 73,), Eq. 1-39

(multiplier denoted with y3g) and Eq. 1-42 (multiplier denoted with 742). Solving from the first

order conditions with respect to payments yields:

/73 = A0 >0, /39 A1 >0 and 42 Ai1 >0.

Setting equal to zero the first order conditions with respect to the effort vector and

substituting the value of the multipliers yields '(e, ) = 1, for i,j e {L,H}.

Solving from the binding constraints yields the payments. The solution is given by:











tH =(eLH), t =(eLL) and tHH = y(eH -) 1 HL- y(eHL)], with tiL arbitrary.
a1

Some of the excluded constraints are automatically satisfied by the payments above. For

the rest of the excluded constraints to be satisfied, the following upper bound on tHL should hold:



y,(e*);

tHL < min y(e*) l [y(e*)- y(e -A)];
2a, -1

y(e*) a- [-y(e*) y(e -A)]
ao +a, -1









LIST OF REFERENCES

Antonioli, B. and Filipini, M. (2001) 'The use of a variable cost function in the regulation of the
Italian water industry', Utilities Policy, 10, 181-187.

Arya, A., Demski, J., Glover, J. and Liang, P. (2005) 'Quasi-robust multiagent contracts',
Carnegie Mellon Working Paper.

Ashton, J. (2000) 'Cost efficiency in the UK's water and sewerage industry', AppliedEconomics
Letters, 7, 455-458.

Aubert, C. and Reynaud, A. (2005) 'The impact of regulation on cost efficiency: An empirical
analysis of Wisconsin water utilities', Journal ofProductivity Analysis, 23, 383-409.

Auriol, E. and Laffont, J. J. (1992) 'Regulation by duopoly', Journal of Economics and
Management Strategy, 1, 507-533.

Bergemann, D. and Morris, S. (2005) 'Robust mechanism design', Econometrica, 73(6), 1771-
1813.

Bergemann, D. and Valimaki (2006) 'Information in mechanism design', Cowles Foundation
Discussion Paper No. 1532R.

Bhattacharyya, A., Harris, T., Narayanan, R. and Raffiee, K. (1995) 'Specification and
estimation of the effect of ownership on the economic efficiency of the water utilities',
Regional Science and Urban Economics, 25(6), 759-784.

Bhattacharyya, A., Parker, E. and Raffiee, K (1994) 'An examination of the effect of ownership
on the relative efficiency of public and private water utilities', Land Economics, 70(2), 197-
209.

Bottaso, A. and Conti, M. (2003) 'Cost inefficiency in the English and Welsh water industry: An
heteroskedastic stochastic cost frontier approach', mimeo, DIEM Universita di Genova.

Byrnes, P., Grosskopf, S. and Hayes, K. (1986) 'Efficiency and ownership: further evidence',
Review of Economics and Statistics, 68(2), 337-341.

Cameron, A. and Trivedi, P. (2005) Microeconometrics, Methods and Applications. Cambridge
University Press.

Clarke, G., Kosec, K. and Wallsten, S. (2004) 'Has private participation in water and sewerage
improved coverage? Empirical evidence from Latin America', World Bank Policy Research
Working Paper No. 3445.

Corton, M. (2003) 'Benchmarking in the Latin American water sector: the case of Peru', Utilities
Policy, 11, 133-142.









Crain, W. and Zardkoohi, A. (1978) 'A test of the property-rights theory of the firm: water
utilities in the United States', Journal ofLaw and Economics, 21(2), 395-408.

Cremer, J. and McLean, R. (1985) 'Optimal selling strategies under uncertainty for a
discriminating monopolist when demands are interdependent', Econometrica, 53(2), 345-362.

Cremer, J. and McLean, R. (1988) 'Full extraction of the surplus in bayesian and dominant
strategy auctions', Econometrica, 56(6), 1247-1257.

Cubbin, J. and 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.

Dasgupta, P. and Maskin, E. (2000) 'Efficient auctions', Quarterly Journal of Economics, 115,
341-388.

Demski, J. and Sappington, D. (1984) 'Optimal incentive contracts with multiple agents',
Journal of Economic Theory, 33, 152-171.

Demski, J., Sappington, D. and Spiller, P. (1988) 'Incentive schemes with multiple agents and
bankruptcy constraints', Journal of Economic Theory, 44, 156-167.

Eso, P. (2005) 'An optimal auction with correlated values and risk aversion', Journal of
Economic Theory, 125, 78-89.

Estache, A. and Kouassi, E. (2002) 'Sector organization, governance, and the inefficiency of
African water utilities', World Bank Policy Research Working Paper No. 2890.

Estache, A. and Rossi, M. (1999) 'Comparing the performance of public and private water
companies in Asia and Pacific region: what a stochastic costs frontier shows', World Bank
Policy Research Working Paper No. 2152.

Estache, A. and Rossi, M. (2002) 'How different is the efficiency of public and private water
companies in Asia?', The World Bank Economic Review, 16(1), 139-148.

Estache, A. and Trujillo, L. (2003) 'Efficiency effects of privatization in Argentina's water and
sanitation services', Water Policy, 5(4), 369-380.

Fabbri, P. and Fraquelli, G. (2000) 'Costs and structure of technology in the Italian water
industry', Empirica, 27, 65-82.

Fang, H. and Morris, S. (2006) 'Multidimensional private value auctions', Journal of Economic
Theory, 126, 1-30.

Faria, R. (2005) 'Public versus private water utilities: empirical evidence for Brazilian
companies', Economics Bulletin, 8(2), 1-7.









Feigenbaum, S. and Teeples, R. (1983) 'Public vs. private water delivery: a hedonic cost
approach', The Review of Economics and Statistics, 65(4), 672-678.

Fox, W. and Hofler, R. (1986) 'Using homothetic composed error frontiers to measure water
utility efficiency', S.il/el ii Economic Journal, 53(2), 461-477.

Fudenberg, D. and Tirole, J (1991) Game Theory. The MIT Press.

Galiani, S. Gertler, P. and Schargrodsky, E. (2005) 'Water for life: the impact of privatization of
water services on child mortality', Journal of Political Economy, 113(1), 83-120.

Greene, W. (2003) Econometric Analysis, 5th Edition. Prentice Hall.

Greene, W. (2005) 'Fixed and random effects in stochastic frontier models', Journal of
Productivity Analysis, 23, 7-32.

Heifetz, A. and Neeman, Z. (2006) 'On the generic impossibility of full surplus extraction in
mechanism design', Econometrica, 74(1), 213-233.

Jehiel, P. and Moldovanu, B (2001) 'Efficient design with interdependent valuations',
Econometrica, 69(5), 1237-1259.

Kim, H. Y. (1987) 'Economies of scale in multiproduct firms: an empirical analysis', Economica
(new series), 54(214), 185-206.

Kumbhakar, S. and Lovell, K. (2000) Stochastic Frontier Analysis. Cambridge University Press.

Laffont, J. J. and Tirole, J. (1986) 'Using cost observation to regulate firms', Journal of Political
Economy, 94, 614-641.

McAfee, R. and Reny, P. (1992) 'Correlated information and mechanism design', Econometrica,
60(2), 395-421.

Miller, N., Pratt, J., Zeckhauser, R. and Johnson, S. (2007) 'Mechanism design with
multidimensional, continuous types and interdependent valuations', Journal of Economic
Theory, forthcoming.

Murillo-Zamorano, L. (2004) 'Economic efficiency and frontier techniques', Journal of
Economic Surveys, 18(1), 33-77.

Neeman, Z. (2004) 'The relevance of private information on mechanism design', Journal of
Economic Theory, 117, 55-77.

Oliveira, G. and Fujiwara, T. (2005) 'Brazil's regulatory framework: predictability or
uncertainty?', Texto para Discussdo No. 147, Escola de Economia de Sao Paulo.

Parlatore, A. (1999) 'Privatization of the water utility sector in Brazil', Essay No. 8 in
'Privatization in Brazil: the case ofpublic utilities', BNDES.









Pinheiro, A. (2003) 'Regulatory reform in Brazilian infrastructure: where do we stand?', Texto
para Discussdo No. 964, IPEA.

Robert, J. (1991) 'Continuity in auction design', Journal of Economic Theory, 55, 169-179.

Saal, D. and Parker, D. (2000) 'The impact of privatization and regulation on the water and
sewerage industry in England and Wales: a translog cost function model', Managerial and
Decision Economics, 21, 253-268.

Sappington, D. (1983) 'Limited liability contracts between principal and agent', Journal of
Economic Theory, 29, 1-21.

Schmidt, P. and Sickles, R. (1984) 'Production frontiers and panel data', Journal of Business and
Economic Statistics, 2(4), 367-374.

Seroa da Motta, R. and Moreira, A. (2006) 'Efficiency and regulation in the sanitation sector in
Brazil', Utilities Policy, 14, 185-195.

Shleifer, A. (1985) 'A theory of yardstick competition', Rand Journal ofEconomics, 16(3), 314-
327.

Soares, D. (2001) 'Privatization of sanitation and water distribution in Brazil: a general overview
of the current market and outlook for private investors', Journal of Project Finance, 7(1),
35-51.

Spady, R. and Friedlaender, A. (1978) 'Hedonic cost functions for the regulated trucking
industry', Bell Journal, 9, 159-179.

Stewart, M. (1993) 'Modeling water cost 1992-93', OFWAT Research Paper No. 4.

Tangeras, T. (2002) 'Collusion-proof yardstick competition', Journal of Public Economics, 83,
231-254.

Teeples, R. and Glyer, D. (1987) 'Cost of water delivery systems: specification and ownership
effects', The Review of Economics and Statistics, 69(3), 399-408.

Tupper, H. and Resende, M. (2004) 'Efficiency and regulatory issues in the Brazilian water and
sewage sector: an empirical study', Utilities Policy, 12, 29-40.

Vargas, M. and De Lima, R. (2004) 'Concessoes privadas de saneamento no Brasil: bom negocio
para quem?, Ambiente & Sociedade, VII(2).

Wilson, R. (1987) 'Game-Theoretic Approaches to Trading Processes', in T. Bewley (ed.)
Advances in Economic Theory: Fifth World Congress, Ch. 2, Cambridge University Press,
33-77.









Zellner, A. (1962) 'An efficient method of estimating seemingly unrelated regression equations
and tests for aggregation bias', Journal of the American Statistical Association, 57, 348-
368.









BIOGRAPHICAL SKETCH

Guillermo Sebastian Sabbioni Perez was born in 1975 in Argentina. In March 2000, he

graduated with honors from the Universidad Cat6lica Argentina with a Licenciatura en

Economia (Bachelor of Arts in economics). After graduation, he worked for one year as an

Analyst in the Corporate Finance Department of the largest media group in Argentina. He then

worked for two and a half years as a Category Manager in the Commercial Department of a

supermarket chain store, also in Argentina. While getting experience in the private sector, he also

engaged in part-time teaching at the Universidad Cat6lica Argentina.

He was awarded a Fulbright Scholarship at the end of 2002; this allowed him to return to

school to pursue graduate education, this time in the United States. He started a doctorate in

economics at the University of Florida in the fall of 2003. While pursuing his doctorate, he

worked part-time as a Research Assistant for PURC, Public Utility Research Center. His work at

PURC provided him with the opportunity to present his research at international meetings.

He was the instructor of undergraduate Game Theory in the summer semester of 2006. He

also presented one of his research papers at the Second Summer School on "Economic Analysis

of Heterogeneity in Social Organizations" in CORE, Louvain-la-Neuve, Belgium, in June 2006.

As recognition of his work, he was awarded the Madelyn M. Lockhart International Travel

Award and the Walter Lanzillotti Research Grant by the Department of Economics, both in 2006.

He graduated in August 2007, his dissertation titled Theoretical and Empirical Analyses of

Incentives and Public Ownership.





PAGE 1

1 THEORETICAL AND EM PIRICAL ANALYSES OF INCENTIVES AND PUBLIC OWNERSHIP By GUILLERMO SEBASTIAN SABBIONI PEREZ 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 Guillermo Sebastian Sabbioni Perez

PAGE 3

3 To my wife Carolina, to my daughter Francisca, to my parents Jorge and Nelyta and to Inda, Maru and Vicky

PAGE 4

4 ACKNOWLEDGMENTS I thank David Sappington for his invaluable he lp and advice. I also thank Sanford Berg, Steven Slutsky, Jon Hamilton and Chunrong Ai for their comments.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 LIST OF FIGURES................................................................................................................ .........8 ABSTRACT....................................................................................................................... ..............9 CHAPTER 1 MULTI-AGENT CONTRACTS WITH UNKNOWN COST CORRELATION..................11 Introduction................................................................................................................... ..........11 The Model...................................................................................................................... .........14 Uninformed Principal: Limits on Communication.................................................................21 If Both Agents Observe the Correlation..........................................................................22 If Only One Agent Observes the Correlation..................................................................32 Uninformed Principal: No Limits on Communication...........................................................33 If Both Agents Observe the Correlation..........................................................................34 If Only One Agent Observes the Correlation..................................................................34 Limits on Communication and Exclusion..............................................................................40 Conclusions.................................................................................................................... .........43 2 EFFICIENCY IN THE BRAZI LIAN SANITATION SECTOR...........................................53 Introduction................................................................................................................... ..........53 Overview of Brazils Wate r and Sewerage Industry..............................................................56 Methodology.................................................................................................................... .......59 Results........................................................................................................................ .............64 Sensitivity Checks............................................................................................................. .....66 Conclusions.................................................................................................................... .........67 APPENDIX DERIVATION OF THE SOLU TIONS TO CHAPTER 1................................79 Limits on Communication......................................................................................................79 When Correlations are Relatively Similar.......................................................................79 When Correlations are Relatively Different....................................................................81 No Limits on Communication................................................................................................83 When Correlations are Relatively Similar.......................................................................83 When Correlations are Relatively Different....................................................................87 Limits on Communication and Exclusion..............................................................................90 LIST OF REFERENCES............................................................................................................. ..92

PAGE 6

6 BIOGRAPHICAL SKETCH.........................................................................................................97

PAGE 7

7 LIST OF TABLES Table page 2-1 Average statistics by operator-type for 2004.....................................................................70 2-2 Summary statistics for first-stage regressions....................................................................71 2-3 First-stage LSDV regression results..................................................................................72 2-4 Second-stage regression results.........................................................................................73 2-5 Ranking of firm-specifi c costs across firm-types..............................................................74 2-6 First-stage LSDV regression re sults using a balanced panel.............................................75 2-7 Second-stage regression resu lts using a balanced panel....................................................76 2-8 First-stage LSDV regression re sults excluding the Regional type....................................77 2-9 Second-stage regression result s excluding the Regional type...........................................78

PAGE 8

8 LIST OF FIGURES Figure page 1-1 Timing at [P-NOC], when both agents obs erve but do not report the correlation.............46 1-2 A solution to [P-CM] that can also be a solution to [P-NOC]...........................................47 1-3 Welfare under the two alterna tive mechanisms at [P-NOC].............................................48 1-4 Timing at [P-CO], when only agent A observes and reports the correlation.....................49 1-5 Welfare at [P-EXCL] be low welfare at [P-NOC]..............................................................50 1-6 Welfare at [P-EXCL] sometimes larger than welfare at [P-NOC]....................................51 1-7 Welfare at [P-EXCL] almost always larger than welfare at [P-NOC]...............................52

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 THEORETICAL AND EM PIRICAL ANALYSES OF INCENTIVES AND PUBLIC OWNERSHIP By Guillermo Sebastian Sabbioni Perez August 2007 Chair: David Sappington Major: Economics This dissertation includes both theoretical and empirical res earch in economic efficiency. The first chapter theoretically evaluates how to design procurement models when two agents may have better information than the principa l about their producti on costs and about the similarity (i.e., correlation) of those costs between agents. Th e results indicate that if the uncertainty about the correlation is not severe, th e principal should design a contract as if she knew that correlation is the lowest possible. The drawback with su ch mechanism is that the highcost agent earns rent if correlation is high. In cont rast, if the uncertainty about the correlation is more severe, the principal should no t tailor the solution to any of the potential correlations. This contract reduces the rent of the high-cost agen t if correlation is high by giving rent to the lowcost agent if correlation is low. Although the we lfare level may be different, the characterization of the results holds when both agents observe th e correlation and do not re port it, as well as in the case where only one agent observes and reports th e correlation to the pr incipal. Finally, when the probability of the low-correlation setting is very small, the principal may find optimal to exclude the high-cost agent from the contract if correlation is low, allowing her to achieve the socially optimal situatio n if correlation is high.

PAGE 10

10 The second chapter empirically ev aluates the relative efficiency of different type of water and sewerage operators in Brazil. The analysis consists of a cost comp arison between public and private operators, and between state-level and muni cipal-level operators. In a first stage, a cost function is estimated utilizing a fixed-effects pa nel data model. In a second-stage, the firmspecific costs from the first stage are explained by means of firm-type indicator variables. The results illustrate that water and sewerage provi sion in Brazil is characterized by substantial economies of scale, indicating that state-level provision is economically more efficient. The results also show that there is not an economi cally significant difference between the costs of private and public firms.

PAGE 11

11 CHAPTER 1 MULTI-AGENT CONTRACTS WITH UNKNOWN COST CORRELATION Introduction This chapter analyzes a multi-agent principal-agent model where the agents are privately informed not only about their produc tion costs but also about the ex tent to which those costs are correlated. The analysis is performed with a nd without limits on communication, and considers the possibility that either one or both agents ma y be perfectly informed about the correlation. The results show that full rent extraction is impossi ble if the informed agents cannot communicate the realization of the correlation to the principal. A dditionally, the results also illustrate that even with communication on the correlatio n the principal may not be able to fully extract rent from all agents if only one of them is in formed about the exact correlation. The most important finding of the analysis under limited communication is that the highcost agent earns positive rent if correlation is high. When limits on communication on the correlation prevail, the principal cannot tailor the payments to the exact correlation coefficient. As a result, the extreme lottery of payments for the high-cost agent should ensure that nonnegative expected rent is obtai ned under both potential correla tions. Extreme payments for the high-cost agent are necessary to prevent the low-cost agent from exaggerating cost. Since expected rent is calculated using two different sets of conditional probabilities (one for each correlation), and since the high payment is obtained with hi gher probability (and the low payment with lower probability) under the hi gh-correlation environment, the principal can eliminate rents for the high-cost agent only if correlation is low. Similarly, the most important finding of the analysis wit hout limits on communication is that if only one of the agents is perfectly inform ed about the extent of co st correlation, again the high-cost agent enjoys positive rent when correla tion is high. Although the rent for the high-cost

PAGE 12

12 agent if correlation is high parall els the results with blocked communication, the source of this rent is slightly different. Instead of being explained by the inabil ity of the principal to tailor payments to each correlation, when communication on the correlation is available the mechanism designer needs to prevent the high-co st agent from untruthfully underreporting the correlation. If correlation is high, the high-cost agent would claim that correlation is low because he would then be assigned a high payment more frequently (and a low payment less frequently) than what he should if the principal designe d two independent mechanisms, one for each correlation report. When limits on communication prevail, the rent for the high-cost agent if correlation is high is explained by his required pa rticipation if correlation is low combined with the fact that his rent is always higher under the high-correl ation scenario than under the low-correlation setting. A natural question is th en whether the principal should always require non-negative rent for the high-cost agent if correlation is low. Numeri cal examples illustrate that if the probability of the low correlation scenario is low enough, it c ould be optimal to excl ude the high-cost agent from the contract if correlation is low. This work is related to two streams of resear ch in the mechanism design literature. First, since the agent(s) have private information on more than one dime nsion, the results contribute to the literature on multidimensiona l mechanism design. Dasgupta and Maskin (2000) show that if bidders types are multidimensional and independ ently distributed there may be no efficient auction (i.e., one that assigns the object to the bidder that values it the most). Jehiel and Moldovanu (2001) illustrate that efficient mechanism design ma y be impossible in a social choice setting, where each agent can receive multidimensional signals from all other agents and where the signals are independently drawn for each agent. The difference between the former

PAGE 13

13 two analyses and the present study is that here the two elements of the two-dimensional type are not independently distributed. Trying to rank the revenue between a First Price and a Second Price auction, Fang and Morris (2006) consider an auction where the bidders, besides knowing their own type, also receive a signal about the other bidders t ypes. In Fang and Morris model there is a separation between each players valuation and his signal about his opponents valuation, since both realizations are drawn from two independen t probability distributions. In their model, the beliefs of the first bidder a bout the second bidder depend on the type of the second bidder rather than on the type of the fi rst bidder. In this st udy, on the other hand, the agents privately observed cost at the same time provides information about the other agents cost. Miller et al. (2007) cons ider a problem where agents types are multidimensional and continuous, but they relax Jehiel and Mol dovanus assumption that the agents private information is independently distributed. Mille r et al. show that incentive-compatible implementation of any decision rule is possible provided that the agents types satisfy one of their correlation conditions. The present work is di fferent in that the principal is not required to utilize budget-balanced transfer payments and in the discrete support of the agents types. Second, when addressing the case with limited communication on the correlation, the present chapter also makes a contribution to the literature on robust mechanism design. This stream of research attempts to relax the a ssumption of too much common knowledge at the time of contract design. Too much common knowledge refers to perf ect information about all the relevant parameters of the problem at hand, as explained by Bergemann and Morris (2005), Bergemann and Vlimki (2006) and Heifetz and Neeman (2006). The influential statement by Wilson (1987) has been the starting point of this emerging literature: Game Theory is deficient to the extent it assumes features to be comm on knowledge, such as one players probability

PAGE 14

14 assessment about anothers preferences or inform ation. One way of relaxing the assumption of too much common knowledge is by focusing on cont racts that are not fine-tuned to the exact environment faced (e.g., to the exact co rrelation between the agents costs). The origins of both streams of research can be traced back to the seminal work by Demski and Sappington (1984) and Cremer and McLean (1985, 1988), who showed that if agents are sufficiently risk-neutral and if they face no limited liability c onstraints, mechanisms can be designed to ensure costless truthf ul revelation of private inform ation provided the agents types are correlated. McAfee and Reny ( 1992) extended the analysis to the case where agents may have infinitely many possible types. This chapter starts with a description of th e model employed. First, the case with limited communication on the correlation is analyze d. Second, the situation with no limits on communication is presented. A following section ev aluates the possibility of excluding some agents from the game if their participation is too costly in terms of welfare. Conclusions are presented in the final section. The Model A principal contracts with two agents, A and B, each producing one unit of output. For each agent, the cost of producing one unit of output is given by e C The parameter } {H L is the agents initial cost, with 0 L H Meanwhile, e denotes the costreducing effort the agent exerts. Effort e generates disutility 0 ) ( e for the agent, where ) ( e is increasing and convex. The final cost C is perfectly observable by the principal, but she cannot distinguish between the cost parameter and the effort level e. Each agent knows his own initial cost, but not the cost of his counterpart.1 The principal reimburses each agent for his 1 The principal is subsequently referred as s he, while the agents are referred as he.

PAGE 15

15 observed final cost C and may provide additional payment ( t ). The utility of each (risk-neutral) agent is given by the difference between the transfer payment t and the disutility ) ( e of the cost-reducing effort exerted: ) ( e t U The agents initial costs are assumed to be positively and imperfectly correlated. The prevailing probability structure consists of HH LLp p = 2 / and HL LHp p = 2 / ) 1 ( where ijp denotes the joint probability that agent A (B) has cost i and agent B (A) has cost j with } { ,H L j i This formulation presumes that the probabi lity that either agen t faces an identical counterpart is invariant to the ag ents cost realization. For exampl e, the probability that agent A (B) has low cost is and the probability that he has high cost is 1 when agent B (A) has low cost.2 The positive and imperfect correlation between the agents costs is captured by ) 1 (2 1.3, 4 This ensures that the probability that an agent has low cost is higher when the other agent also has low cost than when the other agent has high cost: 1 if ) 1 (2 1 The parameter takes on the value 1 with probability 1 and the value 0 with probability 0 where 0 1 and 10 1 Thus, when alpha is 1 the agents costs are highly correlated. This setting is referred to as the high-correlation scenario. When alpha is 0 the agents costs are less correla ted (although still positively). This se tting is referred to as the low-correlation scenario. This information is common knowledge at th e start of the game.5 2 From Bayes rule. 3 The correlation coefficient is given by 2 1 > 0 See Es (2005). 4 The values = 0.5 and = 1 are excluded from the analysis. These values would constitute no correlation and perfect (positive) correlation, respectively. 5 The prior presented satisfies the Beliefs-Determine-Prefe rences (BDP) property from Neeman (2004). In short, a prior has the BDP property if each ag ent has different beliefs about the ot her agents types depending on his own type. Neeman shows that full rent extraction results hinge on the BDP assumption.

PAGE 16

16 The principal never observes the exact realization of the correl ation. In contrast, at least one of the agents acquires precise information a bout the correlation befo re contracting. This means that besides her informati on disadvantage about th e agents initial cost s, the principal is also less informed than (at least one of) the agen ts about the exact correlation. In addition, two alternatives are considered rega rding limits on communication. In th e first alternative setting, the agent(s) who are informed about the correlation are unable to communicate its realization. The second case consists of a framework with no lim its on communication, in which the informed agent(s) report the actual correlation observed. The principal designs a mechanism where si multaneously and indepe ndently both agents report (some or all of) th eir private information.6 Each agent always reports his privately observed cost as it is standard in the literature. Additionally, the pr incipal can also request the more informed agent(s) to repor t the actual correlation. The d ecision of whether to ask for a correlation report depends on the particular circum stances faced by the principal. As explained later in more detail, contract s that do not depend on the exact correlation between the agents types constitute one important example of th e issues addressed by the emerging literature on robust mechanism design. Agent A submits the report A r and agent B submits the report B r Based on the joint report ) (B Ar r r, the principal assigns a pair )] ( ), ( [ r C r t to each agent. A pair entails a transfer ) ( r t from the principal to the agent and a final cost ) ( r C that the agent must meet.7 Both the transfer (t) and the final cost (C) depend on all information (r) reported to the principal. Therefore, each agent knows that the ] [ C t pair received can vary with the report of his 6 Collusion possibilities are not considered here. 7 From the cost function C = e a final-cost target is equivalent to an effort target.

PAGE 17

17 counterpart. In the cases with asymmetric ag ents (i.e., when only one agent observes the correlation), the analysis assumes that the prin cipal offers the same menu of options to both agents. The reason behind this assumption is that the agents may refuse to accept different menus of contracts, on the grounds that the principal would be treating them unfairly. The agents will only contract with the princi pal if they anticipate non-negative expected rent from doing so.8 However, the computation of the e xpected rent depends on the agents knowledge about the correlation. Fo r example, if a low-cost agent observes a low correlation, he knows that the other agent has low cost with probability 0 and high cost with probability 01 On the other hand, if an agent is uninfo rmed about the correlation, he only knows the expected probability of facing each type of counterpart.9 An agent who is uninformed about the correlation only knows that the other agen t has similar cost with probability and different cost with probability 1, where 1 1 0 0 The principal maximizes the sum of the value derived from both units produced, plus the agents profits, minus the cost of social funds.10 The principal values each unit produced at S, where S is assumed constant and suff iciently large. The social f unds are the transfers and the reimbursed costs received by the agents, whic h the principal delivers at a cost of 1+ per unit.11 If the principal observed the agents costs:12 } { } {] [ ) 1 ( )] ( [ 2B A i i i B A i i i iC t C t S W 8 Opportunity rents are normalized at zero and no ex-post limited liability constraints are imposed. 9 Every time an agent is described as informed or uni nformed, it is meant from the point of view of whether the agent has perfect information on th e correlation. As a remainder, all ag ents are always perfectly informed about their own production costs. 10 See Laffont and Tirole (1986). 11 The parameter > 0 measures the distortion introduced by the excise-tax system utilized to raise social funds. 12 ( -C) is used when specific reference to the cost parameter needs to be made. Otherwise, (e) constitutes more compact notation.

PAGE 18

18 Standard manipulation yields:13 } { } {)] ( [ )] ( [ ) 1 (B A i i i B A i i ie t e e W (1-1) Equation 1-1 shows that if the principal observed the agents costs, the solution would be characterized by ) ( e t and *e e for both agents, where 1 *) ( e .14 The socially efficient level of effort *e would be delivered without any re nt being afforded to any agent. Demski and Sappington (1984) and Cremer and McLean (1985) (CM) illustrated how this result can be replicated if th e principal does not observe the agents costs but she knows exactly how these costs are correlated.15 Let ) (jk i jkC t denote the ex-post rent for an agent (A or B) that reports cost j when he has cost i and when the other agent reports cost k for k j i } { H L This problem [P-CM] constitutes the benchmark situation of this chapter: Maximize } { } { ,)] ( [ 2 )] ( [ 2 ) 1 (H L j i ij ij ij H L j i ij ij ije t p e e p W (1-2) subject to: 0 )] ( )[ 1 ( )] ( [ ij i ij ii i ii iC t C t U j i H L j i }, { (1-3)-(1-4) )] ( )[ 1 ( )] ( [jj i jj ji i ji iC t C t U j i H L j i }, { (1-5)-(1-6) 13 The constants 2S and the s are dropped for the sake of exposition. 14 From C = e an extra unit of effort reduces the fi nal cost also by one unit. Therefore, (e) = 1 denotes the equality of the marginal cost and the margin al benefit of exerting cost-reducing effort. 15 It is well documented that risk aversion and/or limited liability prevent the principal from achieving the firstbest. See Sappington (1983) for the single-agent case. See also Cremer and McLean (1988), Demski, Sappington and Spiller (1988) and Robert (1991)

PAGE 19

19 Equation 1-2 incorporates the four possible co st combinations that can arise and their respective probabilities. Equations 1-3 and 1-4 en sure non-negative expected rent for the agents. Equations 1-5 and 1-6 ensure that the agents report their costs trut hfully in equilibrium.16 The solution to [P-CM] permits payments that fully extract (expected) rent while ensuring *e eij for j i,} {H L. To accomplish this, the principal sets relatively high payments for cost reports that match and relatively low paymen ts when cost reports do not match. These two degrees of freedom allow for an infinite num ber of solutions to [P-CM]. After selecting min HH HHt t and min LL LLt t the principal finds HLt and LHt such that no rent is obtained in expectation by either agent, and so 0 L HU U. In particular, consider the bound min HH HHt t where:17 ) (min HH HHe t ] ) 1 ( [ 1 2 1HH HL ) (HHe (1-7) with ) (Hi Hie 0 ) ( Hie for } {H L i .18 The lower bound minHH HHt t prevents the low-cost agent from exaggerating his cost while ensuring that a high-cost agent ea rns non-negative rent in expecta tion. This is achieved with a high payment ) (HH HHe t if both agents report hi gh cost and a low payment ) (HL HLe t if a high-cost report is not matched with the same repor t from the other agent. If the low-cost agent lies, he is relatively likely to receive the low HLt payment and relatively unlikely to receive the 16 This is known as Bayes-Nash implementation. Stronger solutions would be obtained if Ex-Post or Dominant Strategy implementation was required. Ex-Post implementation consists on the requirement that the incentive constraints are satisfied even afte r the other agents make their (truthful) report. Dominant Strategy implementation consists on the requirement that an agent is always better off by reporting the truth, even when the other agents may lie. 17 The lower bound tLL min guarantees that a high-cost agent does not claim low cost, which is usually not constraining for the principal. 18 H is the profit earned by a low-cost agent in the single-agent case. This profit is explained by the savings in disutility of effort that the low-cost agent would enjoy if he exaggerates his cost.

PAGE 20

20 high HHt payment because, due to the positive corr elation among costs, the other agent is relatively likely to report low cost. It is important to notice that minHHt is increasing in Hi which means that minHHt is increasing in effort Hie, for } { H L i .19 The effort exerted by the high co st agent has a positive impact on the reward (i.e., payment above di sutility of effort) that an agen t receives if a high-cost report is matched by his counterpart. Similarly, the e ffort exerted by the high cost agent positively affects (the absolute value of) the penalty (i.e., payment below disutility of effort) that an agent receives if a high-cost report is not matched by the othe r agent. In other words, the larger is Hie, the higher HHt has to be (and the lower HLt has to be) to become unattractive for a low-cost agent. This feature of the payments has important implications for the solution to the principals problem when she is imperfectly informed about the correlation. The subsequent analysis focuse s on settings where the principal, unlike the agent(s), does not know exactly the extent of th e correlation between the agents co sts. In such settings, one or both agents have better inform ation than the principal about both their costs and the exact correlation. As it will be seen, the principal ca nnot achieve the first-best outcome in those settings. The ensuing analysis is performed with a nd without limits on communication about the correlation. The situation with full communication of all private information follows the standard optimal contracting approach. The case with bl ocked communication on the correlation deserves additional explanation. It is important to understand why the principa l may not utilize all available communication channels to gather as much information as possible. 19 It also means that tHH min is increasing in the cost advantage of the low-cost agent,

PAGE 21

21 Uninformed Principal: Limits on Communication Consider, first, the setting where the informed agent(s) are unable to report the correlation to the uninformed principal. In this context, it is impossible for the pr incipal to fine-tune the contract to the exact correl ation observed by the agent(s). As explained below, specific circumstances prevent the principal from making a contract conditional on the actual correlation. Consequently, the principal designs a mechanis m where both agents report only their cost: i Ar and j Br, with } { H L j i For example, this could be the situation f aced by a regulator when publicly-owned firms are privatized quickly.20 In the pre-privatization stage, the regulator may not know the identity of the operators that will take contro l of the soon-to-be privatized firm s. Alternatively, the regulator may not know exactly which production technologies the new firms will utilize. As a result of this uncertainty, the regu lator may ignore the exact correlation between the firms costs when she designs the regulatory contract.21 The inability of the agent(s) to subsequently re port the realized correlation can be a stylized means for capturi ng prohibitive costs of acquiring more precise information about the correlation at a later stage. For example, it may be too costly or it may take a long time for an expert auditor to provi de accurate information about the correlation.22 Limited communication on the correlation is a key element of the new literature on robust mechanism design. This strand of research attemp ts to relax the implicit assumption of too much common knowledge at the time of contract design.23 As such, this literature focuses on contracts 20 When a government makes a tender for the construction of highways may serve as another example. 21 In the auction setting presented by Arya et al. (2005), for example, the authors suggest that the contract designed can be utilized with different pairs of bidders (i.e., with pairs of bidders that may have different correlations among their valuations for the object auctioned). 22 In the context of a fast privatization process, political pressure sometimes requires that a regulatory framework be set up early enough. 23 See Bergemann and Morris (2005) and Heifetz and Neeman (2006).

PAGE 22

22 that are not fine-tuned to the exact environment faced (e.g., on contracts that do not rely on the exact correlation between the agents types). Acco rding to researchers, the main motivation is that real-world mechanisms seem to be simpler than what it w ould be required by a theoretically optimal contracting approach. For example, Arya et al. (2005, p. 15) sugg est that the robustness problem might help us better understand observed institutions. Additionally, Bergemann and Vlimki (2006, p. 3) say that practitioners have often been led to argue in favor of using simpler but apparently sup-optimal mechanisms Also, Bergemann and Morris (2005, p. 1) emphasize that the optimal mechanisms solvi ng the well-defined planners problem seem unreasonably complicated. Two cases are examined below: one in wh ich both agents are informed about the correlation and one in which only a single agen t is so informed. In the context of the privatization process introduced before, the situ ation with only one agen t informed about the correlation could arise when only one of the fi rms has been operating in the industry for a sufficiently long time (perhaps in other geographi cal areas). Therefore, only one of the firms would be endowed with sufficient expertise to permit an accurate assessment of the extent of cost correlation in the environment. This expe rience would allow the firm to be well informed about the correlation between its own cost and th e cost of the (less e xperienced) counterpart.24 On the other hand, if both firms have been operati ng in the industry for a similar length of time, they could both know exactly the extent to which their costs are correlated. If Both Agents Observe the Correlation The principal computes expected welfare: 24 Although plausible, this chapter does not consider the possibility that a more experienced agent could presumably be more likely to have low cost.

PAGE 23

23 } { } { ,)] ( [ 2 )] ( [ 2 ) 1 (H L j i ij ij ij H L j i ij ij ije t p e e p W (1-8) with 0 0 1 1ij ij ijp p p and where s ijp denotes the probability of joint costs i and j when s for } { ,H L j i and } 1 0 { s. The difference between Eq. 1-8 and Eq. 1-2 is explained by the fact that ijp is perfectly known by the principa l at [P-CM], whereas it can only be estimated (ijp ) when correlation is uncertain an d its communication is blocked. To ensure that both agents participate and re port their costs truthfully for all cost and correlation realizations, the next eigh t constraints have to be satisfied:25 0 )] ( )[ 1 ( )] ( [ ij i ij s ii i ii s s iC t C t U j i H L j i }, { } 1 0 { s (1-9)-(1-12) )] ( )[ 1 ( )] ( [jj i jj s ji i ji s s iC t C t U j i H L j i }, { } 1 0 { s (1-13)-(1-16) With no communication the principal solves problem [P-NOC]: Maximize Eq. 1-8 subject to Eq. 1-9 to Eq. 1-16. The only difference between the constraints at [P-NOC] and the constraints at [P-CM] is that the constraints at [P-NOC] c onsider the possibility of two pot ential correlations (alpha could be either 1 or 0 ). The timing is depicted in Figure 1-1. At the solution to [P-NOC], the two degrees of freedom that the principal had available at [P-CM] are utilized to satisfy the four additiona l constraints. There are two alternative ways in which the principal can perform this task. Which of the two mechanisms is optimal depends on the specific values of the parameters, as is later illustrated in detail.26 25 Before concluding, the last section discusses whether it is always optimal to require participation and truthful revelation by all agents in all scenarios. 26 See the Appendix for a formal derivation of all the results.

PAGE 24

24 In the first potential solution to [P-NOC], one degree of freedom is utilized to set ) (LL LLe t while the other degree of freedom is used to set HHt equal to the lower bound that [P-CM] would require if correlation is known to be low (i.e., to set HHt equal to minHHt from Eq. 17 if 0 ). This payment structure is subsequently referred to as the CM-Low mechanism, because it is identical to one of the solutions to [P-CM] if the pr incipal knows that correlation is low. The fact that ) (LL LLe t implies that ) (LH LHe t regardless of which correlation occurs. Since ) (LL LLe t the principal can do no better than to set ) (LH LHe t Setting ) (LH LHe t would imply a violation of the participati on constraints of the low-cost agent, while setting ) (LH LHe t would imply unnecessary positive rent for him. This is no different than what the principal can accomplish at [P-CM], when fully extracti ng expected (and ex-post) rent from the low-cost agent by giving him non-stochast ic transfers. These pa yments automatically guarantee that the participation constraints fo r the low-cost agent are satisfied under both potential correlations because they are also satisfied ex -post (i.e., for any cost report of the other agent). At the same time, ) (0 minHH HHt t ensures that the incentive compatibility constraints for the low-cost agent are satisfied for both potentia l correlation coefficients. This happens because the lottery of payments for the high-cost agent (HHt, HLt) tailored to the low-correlation setting is sufficiently extreme if correlation is high.27 In other words, if the payments for the high-cost agent are not attractive for the lowcost agent if correlation is low, they are even less attractive under the high-correlation setting. The reason for this result is simple. If th e low-cost agent lies 27 A lottery of payments for the high-cost agent is subsequently called more extreme when the difference between tHH and tHL gets larger.

PAGE 25

25 under the high-correlation setting, he is extremely likely to receive the low HLt payment (i.e., even more likely than in the lo w-correlation scenario). This happens because the high-correlation framework imposes on the low-cost agent a very high chance of facing a similar agent, from 0 1 Figure 1-2 depicts this solu tion, showing the CM well-known result of increasingly extreme payments for the high-cost agent when correlation becomes arbitrarily small.28 The problem with the CM-Low mech anism comes from the fact that minHHt ) (HHe as shown by Eq. 1-7. This is a key feature of the so lution to [P-CM]: in order to prevent untruthful reporting from a low-cost agent, a high-cost agen t obtains positive (ex-post) rent if he faces an identical agent. This rent is not a problem at [P-CM], because the pr incipal offsets the high payment ) (HH HHe t with a low payment ) (HL HLe t such that the high-cost agent earns no expected rent. However, the positive and negative (ex-post) rents for the high-cost agent come at a cost at [P-NOC]: they prevent the principal from fully extracting rent from the high-cost agent under both correlations, as is now illustrated. At [P-NOC], the principal does not know which correlation is realized. Therefore, she has two sets of potential weights to use when calculati ng the expected rent of the high-cost agent. As she does at [P-CM], the principal would like to average positive and ne gative (ex-post) rents such that (ex-ante) rent HU is reduced to zero. However, achie ving this goal under two sets of conditional probabilities is impossibl e. This is apparent after a r eexamination of the participation constraints for the high-cost agent: 0 ) 1 ( HL s HH s s HU U U } 1 0 { s with 0 ) ( HH HH HHe t U and 0 ) ( HL HL HLe t U from Eq. 1-7. 28 This result can also be obtained analytically by taki ng the derivative of Eq. 1-7 with respect to alpha.

PAGE 26

26 If the principal extracts all rent for the highcost agent at the high-correlation scenario, she would not provide him with at least zero rent in the low-correlation setting. That is, 01HU is impossible without violating 00HU This result is explained by the necessary positive and negative ex-post rents for the high-cost agent (0 HHU and 0 HLU), combined with 0 1 Therefore, the expected rent for the high-cost agent is always larger if correlation is high: HL HH HL HHU U U U ) 1 ( ) 1 (0 0 1 1 As a result, the principal can only extract re nt from the high-cost agent under the lowcorrelation setting, leaving him with positive rent if correlation is high:29 1 HU HH HL ) 1 ( 1 20 0 0 0 1 >0 Recalling that ) (Hi Hie 0 ) ( Hie it can be seen that rent 1 HU increases with effort level Hie, for } { H L i As it happens at [P-CM], the pa yments for the high-cost agent include rewards and penalties wh ich discourage untruthful repor ting by the low-cost agent. The problem is that these rewards and penalties increase (in absolute value) with effort Hie. That is, the higher the effort required from a high-cost ag ent, the more extreme his lottery of payments needs to be (i.e., higher HHt and lower HLt) to become unattractive for a low-cost agent. That is not a problem at [P-CM], where no rent is provided to either agen t. At [P-NOC], however, there is rent for the high-cost agent if correlation is high. This rent depends on the effort he delivers, because his effort has a positive impact on the re wards and penalties included in his payments. 29 The principal provides rent to the high-cost agent if correlation is high because she requires his participation if correlation is low. This assumption is relaxed in the last section of this chapter, wh ere the principal no longer requires that the high-cost agent earns no n-negative rent if correlation is low.

PAGE 27

27 Consequently, the principal asks for suboptimal effort levels *e eHL and *e eHH in order to reduce the rent 1 HU of the high-cost agent if correlation is high:30 1 ) 1 ( ) 1 2 ( ) ( ) 1 ( 1 ) ( '0 1 0 0 1 HH HHe 1 ) 1 ( ) 1 2 ( ) ( ) 1 ( 1 ) ( '0 1 0 0 1 HL HLe The CM-Low mechanism works we ll if correlation is low, where no rent is commanded by either agent. The problem arises if correlation is high, in which case rent is obtained by the highcost agent. The welfare loss generated by rent 1 HU is severe if the potenti al correlations are very different. The reason is that unde r the CM-Low mechanism, the pr incipal tailors one of the CM solutions to the low-correlation setting (recall ) (0 minHH HHt t ). If this scenario does not occur, however, the welfare loss depends on how wrong the principal is with respect to the exact degree of correlation (i.e., how diffe rent the actual correla tion is high with respect to what the principal thought low). Depending on this difference, therefore, the welfare loss under the CMLow mechanism calls for an alternative solution to [P-NOC]. Intuitively, this second contract should work better than the CM-L ow contract when th e potential correlations are substantially distinct with respect to the correla tion at the low-correlation setting. In the second potential solution to [P-NOC], the two degrees of freedom available at [PCM] are utilized in a different way than unde r the CM-Low mechanis m. This alternative mechanism does not consist of a CM solution ta ilored to any of the potential correlation coefficients. In this alternative contract, the two degrees of freedom available at [P-CM] are employed to guarantee ex-post truthful reporting by the low-cost agent. That is, one degree of 30 The efficient level of effort is delivered after a low cost report: eLL = eLH = e*

PAGE 28

28 freedom is used to set ) (LL LLe t = ) ( HL HLe t while the other degree of freedom is used to set ) (LH LHe t = ) ( HH HHe t As a result, both incentive compatibility constraints for the low-cost agent are satisfied because they are sa tisfied ex-post (i.e., for any cost report of the other agent): 0 )] ( [ )] ( [ ) 1 ( )] ( [ )] ( [ HH HH LH LH s HL HL LL LL se t e t e t e t } 1 0 { s This mechanism is subsequently referred to as the Ex-Post mechanism. The disadvantage of the Ex-Post contract is that ) (LL LLe t which means that the prin cipal cannot fully extract rent from the low-cost agent under both correlat ions. The reason for this result is, again, the impossibility of setting equal to zero the average of positive and negative (ex-post) rents under two possible sets of weights. At most, expected rent can be zero under only one of the potential correlations. As a consequence, under the Ex-Pos t mechanism the low-cost agent obtains rent when correlation is low: 0 LU HL HH ) 1 ( 10 0 0 1 0 1 >0 If correlation is high, however, the rent for the high-cost agent is lower than under the CMLow mechanism:31 1 HU HH HL ) 1 ( 11 1 0 1 0 1 >0 The principal cannot comp letely eliminate rent 1 HU because positive and negative (expost) rents for the high-cost agent are still required. Recall that ) (HH HHe t and ) (HL HLe t 31 The fact that rent UH 1 is lower under the Ex-Post mechanism than under the CM-Low mechanism is easily verified from the effect of 1 > 0 in the denominator of both expressions: 1 + 0 1 > 2 0 1 The term in brackets [ ] has only a seco nd-order effect on rent.

PAGE 29

29 prevent untruthful reporting from the advantaged low-cost agent: he w ould probably receive a low payment if he claims high cost. It can be seen that rents under the Ex-Pos t mechanism depend on the difference between the potential correlations (0 1 ) measured with respect to the middle correlation coefficient (10 1 ). Recalling that 1 2 denotes the correlation coefficient between costs, the middle correlation coefficient 10 1 is defined as the correlation at the middle point of the distance between 1 and 0 For example, if 1 =0.8 and 0 =0.6, the middle correlation would be the correlation coefficient if alpha was 0.7. It is important to notice that the middle correlation coefficient is not the expected correlation 1 2 which would incorporate the probabilities of each scenario in the calculation. This confirms that the Ex-Post mechanism does not consist of a CM mechanism tailored to any of the potential correlations. Like under the CM-Low mechanism, rents unde r the Ex-Post mechan ism also increase with the effort Hie delivered by the high-cost agent. The explanation resides again on the positive impact of his effort on the rewards and pe nalties that his payments include to discourage untruthful reporting by the low-cost agent. Theref ore, the principal reduc es rents under the ExPost mechanism through the suboptimal effort levels *e eHL and *e eHH:32 1 ) 1 ( ) 1 ( ) ( ) 1 ( 1 ) ( '0 0 1 1 1 0 0 1 HH HHe 1 1 ) 1 ( ) 1 ( ) ( ) 1 ( 1 ) ( '0 0 1 1 1 0 0 1 HL HLe 32 Again, the efficient level of effort is delivered after a low cost report: eLL = eLH = e*

PAGE 30

30 As it was mentioned before, which of the two alternative mechanisms yields a higher welfare level depends on the parameters of th e problem. The CM-Low m echanism is optimal for [P-NOC] if: 1 0 0 0 11 2 (1-17) Equation 1-17 can be evaluated in terms of how different the tw o potential correlations are, measured with respect to the correl ation at the low-correlation setting.33 For similar correlations (small ) 1 2 /( ) (0 0 1 ), the CM-Low mechanism yields a higher welfare level. This happens because the welfare loss that the CM-Low mech anism entails is direct ly proportional to how different the potential correlations are. As explained above, the rent that the CM-Low mechanism provides to the high-cost agent under the high-correlation scenario is t oo large if the potential correlations are very di stinct. Remember that this rent if correlation is high depends on how wrong the principal is with respect to the exact degree of correlation (i.e., how different the actual correlation is high with respect to what the pr incipal thought low). Then, the principal should utilize the CM-Low mechanism only if the potentia l correlation coefficients are relatively similar. In contrast, if the potential correlations become too distinct, the principal should employ the Ex-Post mechanism. This contra ct does not consist of a CM contract tailored to any of the potential correlation coefficients. Figure 1-3 plots the be havior of welfare under both alternative contracts as a function of how different the potential correlations are.34 Equivalently, Eq. 1-17 can also be analyzed in terms of the re lative likelihood of each correlation coefficient. When 0 is so large that Eq. 1-17 holds, the principal ensures that no rent is afforded under the (relatively likely) low-co rrelation scenario. She does so by implementing 33 Remember that 2 1 denotes the correl ation coefficient. 34 The graph assumes a constant value of 2 while 1 is shown in the horizontal axis.

PAGE 31

31 the CM-Low mechanism, which provides no rent to either agent if correlation is low. Although this payment structure provides rent 1 HU to the high-cost agent if correlation is high, the principal still applies the CM-Low mechanism if the high-correlation setting is relatively unlikely (i.e., if 1 is so low that Eq. 1-17 holds). On the other hand, if 1 is so large that Eq. 1-17 does not hold, the principal reduces the (relatively likely) rent 1 HU as much as possible by means of the Ex-Post mechanism. She cannot completely eliminate this rent because ) (HH HHe t and ) (HL HLe t still preven t untruthful reporting from the low-cost agent. The drawback to the Ex-Post mechanism is that the low-cost agent obtains rent 0 LU if correlation is low. Yet, the principal is willing to accept rent 0 LU if the low-correlation scenario is relatively unlikely (i.e., if 0 is so low that Eq. 1-17 does not hold). The results obtained are similar to those of Arya et al. (2005). In an auction environment, they also find a condition that determines whic h of two mechanisms the principal should apply. Furthermore, the proposed mechanisms and th e circumstances under which they should be employed are similar to the ones presented in this study. If the uncertain ty about the correlation is not severe, the principal should ut ilize a CM-Low type of mechanism.35 On the other hand, the principal should employ an Ex-Post type of mechan ism if the uncertainty about the correlation is more severe.36 Additionally, after considering the parameter values assumed by Arya et al., the condition that determines optimality in th eir paper is the same as Eq. 117 in this study. Instead of a discrete support for the parame ter alpha, they assume a uniform distribution between a lower bound 5 0 and 1. Given such a probability structur e, they find that the principal should 35 Bayesian-Nash is the name given by Arya et al. to such a contract. 36 Dominant-Strategy is the name given by Arya et al. to such a contract.

PAGE 32

32 employ the CM-Low type of mechanism if 3 / 2 Now notice the result of substituting 5 01 0 and 11 into Eq. 1-17: 5 0 5 0 1 2 10 0 Manipulation yields 3 / 20 which means that their condition can be considered a special case of Eq. 1-17 in this chapter.37 If Only One Agent Observes the Correlation Consider now a situation where communication of the correlation is still blocked, but where only agent A observes the exact correlation be fore contracting. This is a framework where only agent A is endowed with a strong knowledge and expertise about the industry. Meanwhile, agent B and the principal remain ignorant abou t the correlation. They never acquire such privileged information, knowing only the probabilities of facing either 1 or 0 Returning to the previous privatization example, this asymmetr y of agents could arise when firm A has been operating in the industry for a long time (perhaps in other market s), while firm B is relatively new in the industry. In such a cont ext, it is plausible th at not all the agents are perfectly informed about the correlation between their costs. Since it is assumed that the principal does not design a different menu of options for each agent even when she distinguishes the experience d agent A from the inexperienced agent B, the solution to this problem is identical to the solutio n to [P-NOC]. The explanation for this result is simple. To ensure As participation, the principa l must guarantee that participation and incentive 37 There is only one difference between their results and th e ones in the present study. Unlike here, the Ex-Post type of mechanism from Arya et al. does not provide any rent to the low-valuation bidder (which is the equivalent of the high-cost agent). (See their Corollary 2, on page 11). The linearity of the agents utility in their model is the conjectured explanation for such a difference (i.e. there is no convexity like the one coming from the disutility of effort in the present model). A formal proof is beyond the scope of this study.

PAGE 33

33 compatibility constraints are satis fied under both potential correlations. Theref ore, the constraints are automatically satisfied for th e inexperienced agent B, who co mputes expected rent not only across both potential types of c ounterpart, but also across both pot ential correlations. Formally, if the constraints from [P-NOC] are satisfied, then the equivalent c onstraints for an agent that only observes are also satisfied: 0 )] ( )[ 1 ( )] ( [ ij i ij ii i ii iC t C t U j i H L j i }, { )] ( )[ 1 ( )] ( [ jj i jj ji i ji iC t C t U j i H L j i }, { As a consequence, the solution to [P-NOC] also applies if only one of the agents observes the correlation but the principal offers th e same menu of options to both of them. Uninformed Principal: No Limits on Communication In the second case of interest, nothing preven ts the principal from asking for a correlation report from the informed agent(s). Unlike in the case with limits on communication, the Revelation Principle applies in this context. Ther efore, the principal can restrict attention to truthful and direct mechanisms where the agen ts report all their priv ate information. This situation is interesting because an agent who is informed a bout the correlation becomes a twodimensional type of agent. The (simple) situa tion where both agents observe the correlation is presented first. Next, the (more interesting) setting with only one ag ent informed about the correlation is investigated.

PAGE 34

34 If Both Agents Observe the Correlation It is well known that if some element of private information is common knowledge for both agents, the principal can elic it that information at no cost.38 She can do so by designing a mechanism where both agents report their indi vidual cost and the realized correlation: A r ) (s i and B r ) (h j with } { ,H L j i and } 0 1 { h s. Since it is impossible to have different (and truthful) correlation reports, the principal can design a f orcing contract that threatens the agents to very large negativ e rents if the correlation reports differ.39 Therefore, even when the principal has imperfect information about the correlation, the fi rst-best situation can still be replicated if both ag ents can communicate the actual correlation observed. Hence, the solution to this problem is iden tical to the solution to [P-CM].40 If Only One Agent Observes the Correlation If the correlation is known only by agent A, the principal designs a mechanism where the inexperienced agent B only reports his cost, ) (i Br while the experienced agent A reports his cost and the realized correlation, ) (s j Ar with } { ,H L j i and } 1 0 { s. The timing of the game is depicted in Figure 1-4. As explained above, both agents are offered the same menu of contracts even when the principal knows about their asymme try in terms of their information on the exact correlation. Let ) (s jk i s jkC t denote the ex-post rent for an ag ent (A or B) that reports cost j when he has 38 See Fudenberg and Tirole (1991). 39 In the work of Tangeras (2002), for example, the el ement of private information that is common knowledge for both agents is the industry portion of their costs, which has to be added to each agents idiosyncratic cost. Their cost structure follows the model proposed by Auriol and Laffont (1992). 40 There also exists a Nash equilibrium in which both agen ts lie about the correlation. As pointed out by Fudenberg and Tirole, the possibility of multiple equilibria in shoot-them-all mechanisms gave rise to a large literature on unique Nash implementation (1991, p. 293).

PAGE 35

35 cost i when the other agent reports cost k and when the experienced agent A reports that alpha is s for } { H L k j i and } 1 0 { s The principal computes expected welfare: } 1 0 { } { } { ,)] ( [ 2 )] ( [ 2 ) 1 (s H L j i s ij s ij s ij H L j i s ij s ij s ij se t p e e p W (1-18) The difference between Eq. 1-18 and Eq. 1-8 is that the principal now assigns different contracts for different correlations. To ensure that the experienced agent participates and reports his cost truthfully for all cost and correlation realizations, conditional on a truthful correlation report, the next constraint s have to be satisfied: 0 )] ( )[ 1 ( )] ( [ s ij i s ij s s ii i s ii s s iC t C t U j i H L j i }, { } 1 0 { s (1-19)-(1-22) )] ( )[ 1 ( )] ( [s jj i s jj s s ji i s ji s s iC t C t U j i H L j i }, { } 1 0 { s (1-23)-(1-26) Additionally, conditional on a trut hful cost report, the experi enced agent A should not lie about the correlation: )] ( )[ 1 ( )] ( [h ij i h ij s h ii i h ii s s iC t C t U j i H L j i }, { } 1 0 { h s (1-27)-(1-30) Finally, agent A should not simultaneously mi sreport his cost a nd the correlation: )] ( )[ 1 ( )] ( [h jj i h jj s h ji i h ji s s iC t C t U j i H L j i }, { } 1 0 { h s (1-31)-(1-34)

PAGE 36

36 With only one agent informed and with commun ication of the correl ation, the principal solves problem [P-CO]:41 Maximize Eq. 1-18 subject to Eq. 1-19 to Eq. 1-34. Intuition could suggest that once a correlati on report becomes available, the principal should again be able to achieve the first-best by means of two different CM mechanisms, one for each correlation. That is not th e case, however. The problem is th at at [P-CO], the agent who is informed about the correlation may not have the in centive of truthfully reporting its realization. In particular, a high-cost agent would gain by clai ming that correlation is low when it is actually high. An intuitive explanation follows. Consider a payment structure that consists of two independent CM mechanisms (i.e., one for each correlation). Based on the correlation repo rt, the principal could offer non-stochastic payments to the low-cost agent and an extreme enough lottery (i.e., low s HLt and high s HHt) to the high-cost agent, for } 1 0 { s. Under each correlation, the high-co st agents lottery would extract all his rent and it would also pr event a low-cost agent from exa ggerating his cost. The problem of these two independent CM mechanisms is that if the experienced agent has high cost, he would enjoy rent under the high-correlation setting if he claims that correlation is low. The explanation behind this incentive follows. After the high-cost agent unde rreports the correlation, he would be assigned the high payment 0 HHt with probability 1 (instead of with probability 0 ) and he would be assigned the low payment 0 HLt with probability 11 (instead of with probability 01 ). Since 1 >0 this average of positive and negative ex-post rents yields rent for the high-cost agent if correlation is 41 As in the case with no communication and only one agent informed, it is not necessary to impose participation and incentive compatibility constraints for the inexperi enced agent B. Those constraints are automatically satisfied when the principal offers the same menu of contracts to both agents.

PAGE 37

37 high. Therefore, a high-cost agent would want to report that correlati on is low when the highcorrelation environment arises. Although they prev ent untruthful cost repo rting by the low-cost agent, extreme payments to the high-cost agent generate a new incentive if there is asymmetric information about the correlation. A high-cost agent would earn pos itive rent if correlation is high by reporting that correlation is low. As a result of this incen tive, full rent extraction is not possible at [P-CO]. The two-dimens ional type of the informed ag ent does not allow the principal to consider the problem as if they were two independent CM problems. The rent for the high-c ost agent if correlati on is high is unavoidable.42 This result is similar to what happens at [P-NOC], although the source of the rent at [P-CO] is different. At [P-NOC], the rent for the high-cost agent at the high-correl ation setting occurs beca use the principal cannot tailor his payments to the actual correlation. This comes as a result of designing only one set of payments that has to work well regardless of th e correlation. At [P-CO] however, the principal can design different payments for different correla tions. Yet, the rent for the high-cost agent if correlation is high still prevails. At [P-CO], this rent exists to discourag e him from claiming that correlation is low when it is actually high.43 The solution to [P-CO] is similar to the solution to [P-NOC] in that Eq. 1-17 again determines the optimal mechanism. If the low-co rrelation scenario is rela tively likely (i.e., if 0 is so large that 1-17 holds), the principal only affo rds rent to the high-cost agent in the (relatively unlikely) event that correlation is high: 42 The incentive of the high-cost agent to claim that correlation is low when it is actually high is always binding at [P-CO]. 43 If [P-CO] could be considered as two independent CM problems, there would be four degrees of freedom (recall that a standard CM problem has two degrees of freedom). Since one degree of freedom is utilized to prevent untruthful correlation reporting from the high-cost agent if correlation is high, problem [P-CO] has three degrees of freedom instead of four.

PAGE 38

38 1HU 0 0 0 0 0 0 1) 1 ( 1 2HH HL >0 The principal provides rent to the high-cost agent if correlation is high by increasing the payment 1 HHt (without reducing 1 HLt accordingly) above the 1 HHt that an independent CM mechanism would require (i.e., one designed specifically for the high-correlation setting).44 As a result, the lottery (0HHt, 0HLt) under the low-correlation scenario is relatively less attractive than the lottery (1HHt, 1HLt) obtained if truthfully revealing that correlation is high. Consequently, the rent provided by the lottery (1HHt, 1HLt) prevents the high-cost agent from underreporting the correlation.45 To reduce the rent of the high-cost agent if co rrelation is high, the pr incipal also requests less than the efficient effort from the high-cost ag ent if correlation is low. This suboptimal effort 0 Hie<* e reduces how extreme his payments need to be if correlation is low, because it reduces the rewards and penalties includ ed in those payments as a means to avoid untruthful cost reporting by the low-cost agent:46 1 ) 1 ( ) 1 2 ( ) ( ) 1 ( 1 ) ( '0 0 0 0 1 0 0 1 0 HH HHe ; and 1 ) 1 ( ) 1 2 ( ) ( ) 1 ( 1 ) ( '0 0 0 0 1 0 0 1 0 HL HLe 44 Equation 1-7 would determine this value, for = 1. 45 Since the lottery for the high-cost agent if correlation is high is even more extreme than what an independent CM mechanism would require, a low-cost agent does not have the incentive of claiming high-cost under the highcorrelation environment. He also does not have the incentive of claiming high cost and low correlation. The reason is as follows. If the high-cost/low-correlation lotte ry is unattractive for a low-cost agent if correlation is low, it is even more unattractive for a low-cost agent if correlation is high (the low payment tHL 0 would arise even more frequently, from 1 > 0). 46 Since the effort exerted by the high-co st agent in the high-correlation scenario does not have any impact on rents, its socially efficient level is required: eHH 1 = eHL 1 = e* Also, eLL 1 = eLH 1 = eLL 0 = eLH 0 = e*

PAGE 39

39 As a consequence of this effort distortion, the less extreme payments if correlation is low become less attractive for a high-cost agent w ho would untruthfully unde rreport the correlation. Making the lottery (0 HHt, 0 HLt) under the low-correlation scenario less extrem e contributes to the rent reduction for the high cost agent if correla tion is high, because the required increase in 1 HHt is smaller than what an independent CM mechanism would dictate. The contract just described is simila r to the CM-Low mechanism at [P-NOC].47 As such, welfare decreases substantially if the principal applies this mechanism when the correlation coefficients are very different (with respect to the correlation at the low-correlation setting).48 The reason comes again from the rent for the high-cost agent, which increases with the difference between both potential co rrelations. Therefore, if Eq. 1-17 does not hold, the high-cost agent commands smaller rent in the (relativel y likely) event that correlation is high: 1 HU 0 1 0 1 0 1 0 1) 1 ( 1HL HH >0 Like at [P-NOC], the principal reduces rent 1 HU at the cost of affording rent to the lowcost agent in the (relatively unlikel y) event that correlation is low: 0 LU ) 1 ( 10 0 0 0 0 1 0 1HL HH >0 The suboptimal effort levels in this case are as follows 1 )] 1 ( [ ) 1 ( ) ( ) 1 ( 1 ) ( '0 0 0 1 1 0 0 1 0 0 1 0 HH HHe ; and 1 ) 1 ( ] ) 1 ( [ ) 1 ( ) ( ) 1 ( 1 ) ( '0 0 0 1 1 0 0 1 0 0 1 0 HL HLe 47 Although it is not required due to the degrees of freedom available, the principal could offer non-stochastic payments to the low-cost agent under both correlations: tLL 1 = tLH 1 = tLL 0 = tLH 0 = (e*) 48 Again, recall that 2 -1 denotes the correlation coefficient.

PAGE 40

40 This contract resembles the Ex-Post mechan ism at [P-NOC]. Rent s again depend on the difference between the potential correlations (0 1 ) with respect to the middle correlation coefficient (10 1 ), which refers to the correlation at the middle point of the distance between 1 and 0 Another similarity with the Ex-Post mechanism from [P-NOC] resides on the fact that the incentive constraints that prev ent the low-cost agent from exaggerating cost are binding under both correlations instead of only at the low-correlation setting.49 Limits on Communication and Exclusion The analysis so far has assumed that the prin cipal never excludes any agent from the game. That is, participation and truthf ul revelation of private informa tion has been required from all agents under both potential correl ations. The results from the optimization problems performed have illustrated the welfare loss imposed by th is assumption. In part icular, problem [P-NOC] showed that rent for the high-cost agent is alwa ys higher under the high-co rrelation scenario than under the low-correlation setting (0 1H HU U ). Recall that the expl anation resides on the implications of the link between 0 HHU and 0 HLU (which prevent unt ruthful reporting by the low-cost agent) and 0 1 As a consequence, the best the principal could do is reduce rent 0HU to zero and allow positive rent 1HU A natural question then arises: should the principal always require non-negative rent 00HU for the high-cost agent if correlation is low? The answer depends on the welfare achieved with and without the participation of the high-cost agent if correlation is low. Intuitively, if the probability of the low-correlation scenario is low enough, it may be optimal to exclude the hi gh-cost agent from the contract. Then, consider 49 Recall that this is the reason why this mechanism was given the Ex-Post name.

PAGE 41

41 the following problem, where the principal desi gns a mechanism that would provide negative rent for the high-cost agent if correlation is low. The principal computes expected welfare: } { 1 } { } { 1 } {)] ( [ 2 )] ( [ 2 )] ( [ 2 )] ( [ 2 ) 1 (H L j Hj Hj Hj H L i Li Li Li H L j Hj Hj Hj H L i Li Li Lie t p e t p e e p e e p W (1-35) Equation 1-35 arbitrarily exclude s the high-cost agent from th e game if correlation is low (i.e., if alpha is 0 ). To ensure that the high-cost agent doe s not participate (whe ther truthfully or lying) if correlation is low, the next two constraints have to be satisfied: 0 )] ( )[ 1 ( )] ( [0 0 0 HL H HL HH H HH HC t C t U (1-36) 0 )] ( )[ 1 ( )] ( [0 0 LL H LL LH H LHC t C t (1-37) Like at [P-NOC], to ensure that the low-co st agent participates and reports his cost truthfully for both correlations, the next four constraints have to be satisfied: 0 )] ( )[ 1 ( )] ( [ LH L LH s LL L LL s s LC t C t U } 1 0 { s (1-38)-(1-39) )] ( )[ 1 ( )] ( [HH L HH s HL L HL s s LC t C t U } 1 0 { s (1-40)-(1-41) Like at [P-NOC], to ensure that the high-co st agent participates and reports his cost truthfully if correlation is high, the ne xt two constraints should be verified: 0 )] ( )[ 1 ( )] ( [1 1 1 HL H HL HH H HH HC t C t U (1-42) )] ( )[ 1 ( )] ( [1 1 1LL H LL LH H LH HC t C t U (1-43)

PAGE 42

42 With no communication on the correlation and excluding the high-cost agent from the game if correlation is low, the principal solves problem [P-EXCL]: Maximize Eq. 1-35 subject to Eq. 1-36 to Eq. 1-43. At the solution to [P-EXCL], the socially efficient effort level is restored: e eij for j i ,} { H L No rent is provided to the lowcost agent under any correlation (00 1 L LU U ) and no rent is earned by the high-cost agent if correlation is high (01HU ). As expected, the highcost agent would obtain negative rent if correlation is low (00HU ), so he does not participate in that scenario. The principal need only ensure that HLt is low enough: 1 *) (1 0 1 e tHL Like in the previous problems, by setting HLt small enough the principal guarantees that a low-cost agent does not exagge rate his cost. Additionally, HLt small enough ensures that the high-cost agent earns negative rent if correlation is low (00HU ) because all rent is extracted if correlation is high (01HU ). Recall from problem [P-NOC] that 0 1H HU U implies that 00HU if 01HU The welfare loss at the solution to [P-EXCL] is attributed to the zero effort delivered by the high-cost agent if correlation is low. Theref ore, if the probability of the low-correlation setting is sufficiently low, the solution to [P -EXCL] may be an alternative to the Ex-Post mechanism obtained at [P-NOC]. Recall that the Ex-Post contract is optimal at [P-NOC] when the low-correlation setting is relatively unlikely (i.e., when 0 is so low that Eq. 1-17 does not hold). Some conclusions can be drawn from numerical examples.

PAGE 43

43 Figure 1-5 shows that welfare at the solution to [P-EXCL] may be still below the welfare achieved under the Ex-Post mechanism (which c onstitutes the solution to [P-NOC]), even for a relatively small probability of the low-correlation setting (5% in the example). Figure 1-6 shows that a slight decrease of that probability (to 3%) can raise welf are at the solution to [P-EXCL] above the welfare achieved at the solution to [P -NOC]. Finally, Figure 1-7 illustrates that when the probability of the low-correlation scenario dr ops substantially (to 1% ), the welfare under the Ex-Post mechanism is almost always below the welfare attained at th e solution to [P-EXCL]. Therefore, depending on the parameters of the problem, the figures show that it may not be optimal to require participation and truthful revelation of privat e information from all agents under all scenarios.50 Conclusions This chapter contributes to the recent and growing literature on both multidimensional and robust mechanism design. The results from [P-NOC ] show that full rent extraction is impossible if two informed agents cannot communicate the re alization of the correla tion to the principal. This finding coincides with the results obtai ned by other researchers on robust mechanism design, like Arya et al. (2005) Bergemann and Morris (2005), Heifetz and Neeman (2006) and Neeman (2004). Additionally, the solution to pr oblem [P-CO] illustrates that even with communication on the correlation the principal may not be able to fully extract rents from the more informed agents. These results coincide with the findings by authors investigating multidimensional mechanism design, like Miller et al. (2007). When limits on communication on the correlatio n prevail, the principal cannot tailor the payments to the exact correlation coefficient. As a result, the high-cost agent earns positive rent 50 Although the analysis was not performed, it is conjectured that the Ex-Post type of mechanism that constitutes a solution to [P-CO] if Eq. 1-7 does not hold could also be replaced by a contract that excludes the high-cost agent from the game if correlation is low.

PAGE 44

44 if correlation is high. This re sult is the most impo rtant finding of the analysis under limited communication. The reason behind the rent for the high-cost agent under the high-correlation setting is that his extreme lotter y of payments should ensure that non-negative exp ected rent is obtained under both potential correlations. Since e xpected rent is calculat ed using two different sets of weights (one for each correlation), and since the high payment is obtained with higher probability (and the low payment with lowe r probability) under the high-correlation environment, the principal can re duce rents for the high-cost agent to zero only if correlation is low. As a remainder, extreme payments for the high-cost agent prevent the low-cost agent from exaggerating cost. When full communication is restored, and if onl y one agent is perfectly informed about the extent of cost correlation, the principal is still unable to fully extract rent from both agents under both correlations. Even when the mechanism desi gner can tailor payments to each correlation coefficient, the high-cost agent still obtains positiv e rent if correlation is high. This is the most important result of the analysis without limits on communication. Although the rent for the highcost agent if correlation is high parallels the re sults with blocked comm unication, the source of this rent is slightly different. Instead of being explained by the in ability of the principal to tailor payments to each correlation, when communicati on on the correlation is available the principal has to worry about a high-cost agent untruthfull y underreporting the correlation. He would do so because he would obtain a high payment more frequently (and a low payment less frequently) than what he should if the principal designe d two independent CM mechanisms, one for each correlation report. The rent for the high-cost agent if correlation is high is driv en by the fact that non-negative rent is required for him under th e low-correlation setting. If the principal could ex clude the high-

PAGE 45

45 cost agent from the game if correlation is low, the socially efficient effort level can be restored without affording any rents to the (participating) agents. The cost of such a contract is that no effort is delivered by the high-cost agent when co rrelation is low, because he does not participate due to the negative rent he would obtain. If th e probability of the low-correlation scenario is small enough, numerical examples show that the principal can achieve higher welfare by designing a contract that excludes the high-cost agent from the game when correlation is low. Finally, the usual criticis m of the risk neutrality assumpti on also applies to this study. The solutions obtained make use of that assumption, since agents here voluntarily participate in the mechanism proposed by the principa l even when they could potent ially obtain large negative expost rent. Therefore, the impossibi lity of the full-rent-extraction result would probably be more pronounced if risk aversion or limited liability constraints were imposed.

PAGE 46

46 All parties learn 1, 0and 1, 0Each agent learns his own cost parameter The principal offers two contracts Each agent reports his cost A specific (t,C) pair is assigned to each agent depending on both reports The agents produce, transfers are made and costs are reimbursed Both agents observe the correlation but do not report it Figure 1-1. Timing at [P-NOC], when both agents observe but do not report the correlation.

PAGE 47

47 Parameters: 0.55 0.01 0.99 0.60 0.20 (e) 1e44 Payments under the CM-Low contract-0.80 -0.60 -0.40 -0.20 0.00 0.20 0.40 0.60 0.80 1.00 0.560.600.630.670.710.740.780.820.850.890.930.961.0 0 tLL(CM-Low) tHH(CM-Low) tHL(CM-Low) tLH(CM-Low) Figure 1-2. A solution to [P-CM] that can also be a solution to [P-NOC].

PAGE 48

48 Parameters: 0.55 0.70 0.30 0.60 0.20 (e) 1e44 Welfare at P-NOC2.24 2.26 2.28 2.30 2.32 2.34 2.36 2.38 2.40 2.42 0.560.600.630.670.710.740. 780.820.850.890.930.961.0 0 CM-Low P-NOC Ex-Post P-NOC Figure 1-3. Welfare under the two al ternative mechanisms at [P-NOC].

PAGE 49

49 All parties learn 1, 0and 1, 0Each agent learns his own cost parameter The principal offers four contracts Each agent reports his cost and agent A reports the correlation A specific (t,C) pair is assigned to each agent depending on both reports The agents produce, transfers are made and costs are reimbursed Agent A observes the correlation Figure 1-4. Timing at [P-CO] when only agent A observes a nd reports the correlation.

PAGE 50

50 Parameters: 0.55 0.05 0.95 0.60 0.20 (e) 1e44 Welfare at P-NOC vs P-EXCL2.31 2.32 2.33 2.34 2.35 2.36 2.37 2.38 2.39 2.40 0.560.600.630.670.710.740.780.820.850.890.930.961.0 0 Ex-Post P-NOC Altern P-EXCL Figure 1-5. Welfare at [P-EXC L] below welfare at [P-NOC]

PAGE 51

51 Parameters: 0.55 0.03 0.97 0.60 0.20 (e) 1e44 Welfare at P-NOC vs P-EXCL2.32 2.33 2.34 2.35 2.36 2.37 2.38 2.39 2.40 0.560.600.630.670.710.740. 780.820.850.890.930.961.0 0 Ex-Post P-NOC Altern P-EXCL Figure 1-6. Welfare at [P-EXCL] some times larger than welfare at [P-NOC]

PAGE 52

52 Parameters: 0.55 0.01 0.99 0.60 0.20 (e) 1e44 Welfare at P-NOC vs P-EXCL2.32 2.33 2.34 2.35 2.36 2.37 2.38 2.39 2.40 0.560.600.630.670.710.740. 780.820.850.890.930.961.0 0 Ex-Post P-NOC Altern P-EXCL Figure 1-7. Welfare at [P-EXCL] almost always larger than welfare at [P-NOC]

PAGE 53

53 CHAPTER 2 EFFICIENCY IN THE BRAZI LIAN SANITATION SECTOR Introduction Early work on the relative performance of water and sewerage (WS) utilities by Crain and Zardkoohi (1978) tried to determine whether private U.S. water utilities attained a more efficient level of operation than public ones. Since then, a number of pa pers have been published on the efficiency of WS utilities. Some authors have al so focused on the private vs. public issue, while others have tested other hypot heses, like the existence of economies of scale, economies of scope, or the possible homogeneity or homot heticity in the production technology. Data availability limited the types of studies: early pa pers focused mostly on utilities in the U.K. and U.S. because these countries pioneered the co llection and publication of data on WS firms. Until a decade ago, little rese arch was conducted into the effi ciency of WS utilities in developing countries. Then, studies began to address the performance of water systems using quantitative techniques. A numb er of papers focused on Asia n and African water utilities, most of them supported by the World Bank.1 These studies provided insights for countries implementing World Bank policies to increase cove rage and quality of WS services in their regions. Politically potent and economi cally important, WS utilities generate public concern over efficiency. Using data from Br azil, this study extends the sta ndard approach by comparing not only the performance of public and private WS firm s, but also the performance of different types of public WS operators. There are four different types of WS providers in Brazil. The first type consists of regional public operators, which provide serv ices at the state level. The ot her three types of WS operators 1 For example, Estache and Ro ssi (1999), Estache and Rossi (2002), and Estache and Kouassi (2002).

PAGE 54

54 provide services at the lo cal (municipal) level. The first local type consists of private firms that have signed concession contracts wi th the municipalities where they operate. The other two local types both consist of pub licly-owned operators, but they differ in their legal status. One type consists of local public providers that are organized similarly to a corporate business. They are called public-corporative operators throughout this chapter. The ot her type consists of local public providers that are run like not-for-pro fit organizations. They are called public-noncorporative providers throughout this study. Besides the public vs. private discussion, c ontroversy exists in Brazil about whether municipalities or states should be responsible for WS provisi on. The Association of State-Owned Sanitation Firms (AESBE), for example, argues that WS services should be provided at the state level for two reasons: one, the larger scale of operation permits scale economies; two, there is a possibility of cross-subsidization be tween poorer and rich er municipalities.2 In contrast, the National Association of Municipal Sanitation Services (ASSEMAE) favors municipal provision on the grounds that WS services are an essential necessity for the population.3 ASSEMAE does not explain in detail why the essential nature of WS services calls for municipal and not state provision, but standard arguments are based on local control, responsivene ss to citizen concerns and awareness of local conditions. This study finds that WS provision in Braz il is characterized by economies of scale. Therefore, since an increase in output genera tes a less-than-proportional increase in (operating) costs, WS provision at the state level should be preferred. The pot ential efficiency gains are not trivial when one recalls that Braz il has a population of 187-million people. 2 Associao das Empresas de Sa neamento Bsico Estaduais. See http://www.aesbe.org.br (last visit: March 26, 2007). 3 Associao Nacional dos Servios Municipais de Saneamento. See http://www.assemae.org.br (last visit: March 26, 2007).

PAGE 55

55 In a first stage, a fixed-effects panel data model with data for 2000-2004 is employed. A cost function is proposed, identifying firm-s pecific (operating) costs which account for inefficiency and other unobserved heterogeneity. In a second stage, those firm-specific costs are explained by means of firm-type and other time-i nvariant indicator variables. The results show that regional operators have lower firm-speci fic costs than local-private and local-public providers. This finding indicates th at the efficiency gains of the state providers from their scale of operation are augmented by their lower firm-specific costs. This work also shows that the local-public vs. local-private comparison depends on how the local public provider is organized. Private and public-corporative providers have lower firmspecific costs than public-non-corporative providers. This finding indicates that the WS operators organized as not-for-profit organizations have the highest firm-specific costs in Brazil. In spite of the firm-specific cost differen ces found, it is worth mentioning that these differences represent a small portion of operating cost s. The first-stage regressions illustrate that the output produced, input prices and other technol ogical factors explain most of the variation of operating cost, regardless of the firm-type. As a result, the firm-specifi c cost differences found are not substantial from an economic point of view. Quantifying the relative efficiency of the Br azilian regional operators, Tupper and Resende (2004) use Data Envelopment An alysis (DEA) with data for 1996-2000. The efficiency scores obtained are considered in the construction of a proposed linear reimbursement rule that constitutes a yardstick mechanism. However, the authors acknowledge that its implementation is constrained by the weak current regulatory fr amework. Utilizing DEA with data for 1998-2002, Seroa da Motta and Moreira (2006 ) argue that the government leve l at which conceding authority resides is not a crucial barrier to the Brazilian san itation sectors development when looking at

PAGE 56

56 the operators performance. Unlike this study, they find that ownership does not matter for productivity gains for municipal services. Like th is chapter, they find that regional operators benefit from larger scale economies. Evidence on the beneficial effects of privat e sector participati on (PSP) in the Latin American sanitation sector is not conclusive. In Argentina, Bolivia and Brazil, for example, Clarke et al. (2004) fi nd that even when connection rates to piped water improved following the introduction of PSP, connection rates similarly improved in the control regions that never privatized. In contrast, Galiani et al. (2005) find that child morta lity in Argentina fell 8% in the areas that privatized their water services. They conclude that while privatization is associated with reductions in deaths from infectious dise ases, it is uncorrelated with deaths from causes unrelated to water conditions. The ambiguity on the beneficial effects of PSP in the Latin American sanitation sector coincides with the results from other regions.4 This chapter first presents an overview of Brazils water and sewerage industry. The study then illustrates the two-stage methodology uti lized. A following section present the results obtained. After performing some sensitivity checks, conclusions are provided in a final section. Overview of Brazils Water and Sewerage Industry In 1971, Brazil created a national plan for WS provision (PLANASA).5 This plan delegated authority for the provi sion of WS services to twenty seven newly born state-owned companies. According to PLANASA, these public companies were the only sanitation entities authorized to obtain financing from the Nationa l Housing Bank (Banco N acional de Habitaao 4 For the US, Bhattacharyya, Parker and Raffiee (1994) found evidence of greater efficiency in public utilities, Crain and Zardkoohi (1978) found evidence in favor of private operators, and Byrnes et al. (1986), Feigenbaum and Teeples (1984) and Fox and Hofler (1986) found no difference between public and private operators. For Asia, Estache and Rossi (1999) found evidence in favor of private operators while in a later study Estache and Rossi (2002) found no difference. For Africa, Estache and Kouassi (2002) found evidence in favor of private firms. 5 Plano Nacional de Saneamento. See So ares (2001) for a detailed description.

PAGE 57

57 BNH). This feature made PLANASA attractive for the municipalities that were interested in expanding their sanitation systems.6 About 3,200 municipalities jo ined the new plan, awarding concessions to the state-owned companies for 20 to 30 years.7 On the other hand, about 1,800 municipalities never adhered to PLANASA, providi ng WS services on their own ever since. The three types of local operators (private, public-corporative and public-non-corporative) provide WS services within the municipalities that never joined the system. The PLANASA model started to decline in th e 80s. After 1986, the BNH was unable to finance the required expansion of the WS sector due to a weak fiscal situation of the federal government.8 Anti-inflationary policies may have also pl ayed a role, since the government at that time pressed for low water tariffs to keep in flation under control. The 80s decade was also characterized by an emphasis on decentralization, best illustrated in the constitutional reform of 1988. The centralizing concept of PLANASA, on the other hand, wa s more in accordance to the military regimes of earlier decades. There have recently been some attempts to de fine a new framework for the WS sector. Bill 4147/2001, for example, intended to allow for more private participation. This bill defined the states to be the conceding authority in metropol itan areas. However, the constitutional reform of 1988 granted to municipalities the right to make concessions for public services of local 6 See Faria (2005). 7 In many cases, however, there was never any formal contract between the municipality and the state-owned companies. 8 PLANASA formally extinguished in 1992. See Parlatore (1999).

PAGE 58

58 interest.9 Due to that controversy about the interpretation of the Constitution, Bill 4147 never became law.10, 11 A new Bill 5296/2005 also attempted to redefine the rules for the WS sector. A Parliamentary Commission approved it on July 2 006 after many modifications and it recently became Law 11445 in January 2007. This new bill specifies that municipalities have the conceding authority over services of local interest. Nevertheless, the implications of the new Law 11445 are yet to be seen. Due to the heated debate about where the conceding authority re sides, only some municipalities that never adhere d to PLANASA have made con cessions to private operators.12 These private companies provide WS serv ices to less than 4% of the population.13 The Brazilian Association of Private Water and Sewage Operators (ABCON) suggests that only through a more active private participation will the WS s ector meet the high inve stment levels required.14 According to 2004 data, approximately 112,000 people are directly employed in the WS sector, almost 90% by the state-owned companies. The national coverage for water services is roughly 85%, although the sewage coverage is below 60%. On average, almost 30% of the treated water produced is unaccounted for, due to leaking through broken pipes and illegal 9 Water distribution and sewerage collection are defined as services of local interest. On the other hand, water catchment and water and sewerage treatment are defined to be of local interest only in case of exclusive use by the municipality. See Ministrio das Cidades, http://www.cidades.gov.br (last visit: March 26, 2007). 10 There was also a strong opposition from the public and representative institutions, like the Brazilian Association of Sanitary and Environmental Engi neering (Associao Brasileira de Engenharia Sanitria e Ambiental ABES). See http://www.abes-dn.org.br (last visit: March 26, 2007). 11 A Nacional Water Agency (Agencia Nacional de Agua s ANA) was also created in 2000. However, the main function of ANA is to monitor the utilization of water resources. Its role as a regulator is yet to be defined. 12 See Vargas and De Lima (2004). 13 This figure contrasts with the situation in other infrastructure sectors like tel ecommunications, railroads and electricity, where private participati on is much more active. See Oliveira and Fujiwara (2005) and Pinheiro (2003). 14 Associao Brasileira de Concessionrias de Servios Pblicos de gua e Esgoto. See http://www.abcon.com.br (last visit: March 26, 2007).

PAGE 59

59 connections. Furthermore, only 50% of the sewa ge volume collected receives some type of treatment. Table 2-1 shows these statistics by operator-type. Methodology Duality theory implies that the production tec hnology of a firm can be modeled with a cost function, where the firms cost depends on its ou tput level and the prices of the inputs employed in production. Other factors may also affect th e firms technology and hence the firms costs. Specifically: ) ( z w q c c (2-1) where c denotes cost, q denotes output level, w denotes input prices and z includes other control variables.15 To empirically estimate Eq. 2-1, a panel data framework is adopted: it i it itu X Y ', (2-2) where itY denotes the dependent variable for individual i at time t itX denotes the vector of explanatory variables, iu accounts for time-invariant heteroge neity at the individual level and it denotes random statistical noise. Heterogene ity is the denomination of the observed and unobserved unique individua l characteristics. Fixed or random effects models can be adopt ed for panel data. The fixed effects model allows unobserved heterogeneity to be correlated with the explanat ory variables. In contrast, the random effects model assumes that any unobserved he terogeneity is distribut ed independently of the covariates. In the context of this study, correlation betwee n unobserved heterogeneity and the explanatory variables is hard to rule out. Such a correla tion would exist, for example, if the firm 15 Control variables have been sometimes denominated hedonic measures, referring to th e approach introduced by Spady and Friedlaender (1978) for the trucking industry. They emphasize that the service dimensions provided by the firm should enter the cost function as explanatory variables. Feigenbaum and Teeples (1983) first introduced the hedonic approach to the water sector.

PAGE 60

60 can modify its output level based on private in formation about its unobserved inefficiency. Therefore, a fixed effects model for panel data is employed. The fixed effects formulation allows unobserved inefficiency to be capture d by the firm-specific coefficients.16 In a panel data framework, the cost function from Eq. 2-1 takes the following form: i it it z it w it q itu z w q c 0, (2-3) where the s are parameters to be estimated and iu denotes cost inefficiency and any additional unobserved heterogeneit y. It is assumed that the it are i.i.d and uncorrelated with the covariates. In contrast, the iu are allowed to be potentially correlated with the explanatory variables.17 The least squares dummy vari ables (LSDV) estimator is utilized, including also a year-specific effect:18 it t it z it w it q i itz w q c (2-4) where the firm-specific intercepts i iu 0 account for inefficiency and any other unobserved heterogeneity. Utilizing this formulati on, Schmidt and Sickles (1984) proposed the measure ) min(* i i i to construct a ranking of relativ e inefficiency. Their approach permits the computation of individual inefficiency te rms relative to the most efficient firm in the sample.19 That calculation might be appropriate when one is concerned about efficiency at the individual-firm level. For example, a regulator co uld set the tariff of firm A based partly on its 16 According to Greene (2005), assuming that inefficiency is time-invariant is not a problem in short panels. This is especially true in the water industry, which is characterized by low technological change. 17 It is unnecessary to make any distribu tional assumption on the inefficiency term i. If one is willing to make distributional assumptions on the i, Maximum Likelihood would theoretically allow for more efficient estimates than both fixed and random effects models. Neverthele ss, Kumbhakar and Lovell (2000) and Murillo-Zamorano (2004) mention several papers that after performing em pirical comparisons of the three approaches generate similar efficiency rankings, especially at the top and bottom of the distribution. 18 The LSDV estimator is equivalent to the within-groups estimator. 19 Ashton (2000) constructs an efficiency rank ing of British water firms utilizing that measure.

PAGE 61

61 relative efficiency with respect to firm B, as suggested by Shle ifer (1985). However, the focus of this study is not yardstick comparison betw een individual operator s but rather between different types of firms. The goal of this chapter is to identify whether regional, local-public or local-private firms are relatively more efficien t in providing WS services in Brazil. Thus, an alternative analysis is pursued. After estimating Eq. 2-4, the predicted firm-specific costs i are computed. A high i indicates an inherently high cost for firm i even when controlling for output, input prices and other exogenous variables. Once the predicted fi rm-specific costs are obtained, an additional regression is performed. In this second-stage regression, the depe ndent variable is the predicted firm-specific cost (per unit of output), while fi rm-type indicators and regional dummies are the explanatory variables:20 i i i i iq Region Type (2-5) The vectors and contain the coefficients for each of the Type and Region indicator variables. Meanwhile, iq represents the average output of firm i for the period unde r analysis. If the coefficients for the firm-type dummy variables are sta tistically significant in Eq. 2-5, there will be evidence of relatively distinct firm-specific costs between the different operator-types. Following the extant literature, Operating Cost is utilized to re present the dependent variable c on Eq. 2-4.21 Wage is employed to represent input prices w since they account for 20 Although output is present in the first stage regression the firm-specific costs are going to be correlated with output by the nature of the fixed-effects model. That is the explanation behind the utilization of firm-specific costs per unit of output. 21 Although it is also conceivable to use total cost as the dependent variable, that would require data on the price of capital, since depreciation charges constitute a large share of total costs. Since reliable data on the price of capital are unavailable, this study focuses only on operating cost, which excludes depreciation.

PAGE 62

62 more than 40% of the operating cost. 22 Wages were calculated as the ratio of total labor expenses divided by the number of employees as it is standard in the literature.23 Although the volume of water produ ced seems like the most appealing output variable, the number of connections has also been widely used by researchers.24 Thus, both Volume and Connections are employed as two alternative measures of output q As control variables, this study includes Network Length ,25 the Percentage of Urban Population ,26 a Metering Index,27 a Fluorination Index28 and a Sewerage Dummy that equals 1 if the firm also provides sewerage collection (not all operato rs provide both services).29 22 Data on other input prices are limited. Some data on energy consumption suggests that energy is the second most important input, representing around 20% of operating cost. 23 More detailed data on input prices would theoretically allow for more efficient estimates utilizing the Seemingly Unrelated Regressions (SUR) model proposed by Zellner (1962). This model cons ists of a multivariate regression system. Besides the cost function, the input-demand share-equations are utilized, enhancing the efficiency of the estimation because the same coefficients participate not only in the cost function but also in the input-demand share-equations. 24 The number of connections is employed by Ashton (2 000), Estache and Rossi (1999) Estache and Rossi (2002) and Teeples and Glyer (1987). The volume of water produced is utilized by Antonioli and Filippini (2001), Aubert and Reynaud (2005), Bhattacharyya, Harris et al (1995), Bhattacharyya, Parker and Raffiee (1994), Bottaso and Conti (2003), Corton (2003), Crain and Zardkoohi (1978), Cubbin and Tzanidakis (1998), Estache and Rossi (1999), Estache and Rossi (2002), Fabbri an d Fraquelli (2000), Fox and Hofler (1986), Kim (1987), Stewart (1993) and Teeples and Glyer (1987). The number of customers is used by Antonioli and Filippini (2001), Aubert and Reynaud (2005), Fabbri and Fraquelli (2000) and Saal and Parker (2000). 25 The length of pipes is utilized by Antonioli and Filippini (2001), Bottaso and Conti (2003), Corton (2003), Cubbin and Tzanidakis (1998), Fox and Hofler (1986), Kim (1987) and Stewart (1993). 26 A proxy of density is used by Bottaso and Conti ( 2003), Fabbri and Fraquelli (2000) (ratio between population served and the length of pipelines) and Teeples and Glyer (1987) (connections per mile of line). The percentage of non-domestic consumers is employed by Bottaso and Conti (2003), Estache and Rossi (1999), Estache and Rossi (2002), Fox and Hofler (1986), Kim (1987) and Stewart (1993). 27 The percentage of metered connections is used by Cubbin and Tzanidakis (1998), Estache and Rossi (2002), Feigenbaum and Teeples (1983) and Teeples and Glyer (1987). 28 A proxy for quality is used by Antonioli and Filippini (2001) (dummy indicating if water has to be chemically treated before distribution), Estache and Rossi (1999) (continuity), Estache and Rossi (2002) (continuity), Feigenbaum, and Teeples (1983) (water treatment index), Fox and Hofler (1986) (tests of water quality and tests of organic contamination), Saal and Parker (2000) (percentage of water that is compliant with key parameters relative to the compliance percentage for England and Wa les) and Teeples and Glyer (1987) (water treatment index). 29 Other control variables have been also employed in previous literature. The percentage of water losses is used by Antonioli and Filippini (2001) and Bhattacharyya, Harris et al. (1995). The storage capacity is used by Feigenbaum and Teeples (1983), Fox and Hofler (1986) and Teeples and Glyer (1987). A dummy indicating if the utility has to purchase water from other utility is employed by Aubert and Reynaud (2005), Feigenbaum and

PAGE 63

63 Earlier literature suggests that a longer network should be as sociated with higher costs due to its maintenance (fixing leaks, for example). Previous research also suggests that a higher metering index should be associated with higher costs due to the reading and maintenance of the meters. The fluorination index shoul d also be associated with higher costs due to a more intense chemical treatment of water before delivery. Th e sewerage dummy is also expected to show a positive sign, capturing the higher operating cost of providing both water and sewerage services. Finally, the effect of a higher proportion of urban population in the area served is difficult to predict. On the one hand, many researchers argue that having customers densely located in a small area reduces costs. On the other hand, Fe igenbaum and Teeples (1 983) argue that we should expect that it is more costly to supply more densely developed service areas, which requires more hydrants, higher wate r pressure and greater peak cap abilities for fire protection.30 The main source of data is the National System of Sanitation Information (SNIS) of Brazil.31 Operators voluntarily join the SNIS, which started collecting data in 1995. The number of firms providing data has increas ed each year ever since. This study utilizes an unbalanced panel for 2000-2004. There are approximately 180 observations for 2000 and 340 observations for 2004, with almost 1200 observations in total.32 The SNIS is part of the Modernization Teeples (1983), Fox and Hofler (1986) and Teeples and Glyer (1987) (they use the proportion of water that is purchased). The number of districts served is used by Corton (2003). A dummy indicating if the utility obtains water from surface sources is used by Aubert and Reynau d (2005), Bhattacharyya, Harr is et al. (1995), Bottaso and Conti (2003) (they use river sources), Estache and Rossi (1999) and Fox and Hofler (1986). I ignore the reasons that support the inclusion of a proxy for capital stock, as done by Antonioli and Filippini (2001) (number of water wells), Aubert and Reynaud (2005) (average net base rate divided by the estimated price of capital), Bhattacharyya, Harris et al. (1995) (r esidual of the revenue less variable costs) and Bottaso and Conti (2003) (replacement costs of net tangible assets). Such a variab le would be an appropriat e covariate in a production function, but not in a cost function 30 Feigenbaum and Teeples (1983, p.674). They confirm that result in their paper. 31 Sistema Nacional de Informaoes sobre Saneamento. See www.snis.gov.br (last visit: May 7, 2007). 32 Missing data on some variables explain the variation in the number of observations across the two models presented in the next section. The online data are split into several files, each containing only a certain group of variables (financial, descriptive, op erational, etc) and a certain type of firm (regional, local, etc). These spreadsheets were pooled together for this work.

PAGE 64

64 Program of the Sanitation Sector (PMSS), whic h Brazil started in 1992 with financial support from the World Bank.33 To adjust monetary figures for infl ation, data on the Brazilian Consumer Price Index (IPCA) are utilized.34 Results Summary statistics of the variables used in the first-stage LSDV regressions are presented in Table 2-2, discriminated by operator type. Th e size difference between the regional and local operators is evident when observing the average output, network length a nd operating cost of each type of firm. For example, from the point of view of the number of connections, the average regional operator is 14 time s bigger than the average public-corporative provider. Table 2-3 presents the results from both LSDV regressions, according to Eq. 2-4. Alternatively, both Connections and Volume are positive and statistically significant. The hypotheses that their coefficients are equal to 1 is rejected which provides evidence of increasing returns to scale regardless of the out put variable chosen. For example, a 10% increase in the volume of water produced generates only a 0.98% increase in operating cost.35 This confirms the argument made by AESBE, which favor s state-level provision due to economies of scale. As expected, Wage also has a positive and statisti cally significant effect on cost. From the set of control variables, the Metering Index shows the expected positive sign and statistical significance. This means that when the fraction of metered connections is higher, 33 Programa de Modernizaao do Setor de Saneamento. The PMSS resides on the sphere of the National Secretariat of Environmental Sanitation (Secretaria Nacional de Saneamento Ambiental), which depends on the Ministry of Cities (Ministrio das Cidades). See www.cidades.gov.br (last visit: May 7, 2007). 34 The ndice Nacional de Preos ao Consumidor Amplo (IPCA) is constructed by the Brazilian Institute of Geography and Statistics (Instituto Brasileir o de Geografia e Estatistica IBGE). See www.ibge.gov.br (last visit: May 7, 2007). 35 To test for economies of scale that may vary with output, specifications including the square of the output variable were run. The results were not satisfactory. The Volume2 variable was not statistically significant, while Volume remained statistically significant with little change in the value of its coefficient. Meanwhile, both Connections and Connections2 became statistically insignificant.

PAGE 65

65 operating cost increases. This is in line with earlier research suggesting that reading and maintaining the meters has a positive impact on cost. The Sewerage Dummy coefficient is positive and statistically significant. This means th at collecting and treating sewage increases the operating cost of a water provider. Finally, Network Length the Percentage of Urban Population and the Fluorination Index are statistically insignificant. The first-stage LSDV regressions in Table 2-3 also illustrate th at output, input prices and other technological factor s explain most of the variation of operating cost.36 As a result, the firmspecific costs provided by the first-stage regressions represent a small portion of the operating cost (less than 1%). After running the first-stage LSDV regres sions, the predicted firm-specific costs i are obtained. Following Eq. 2-5, a second-stage re gression is then performed, where the firmspecific cost (per unit of output) plays the role of the dependent variable. Firm-type and five region-indicators are the time-invariant explanat ory variables in the second-stage regression. Dummies for the five different regions in which Brazil is divided are included because geographic heterogeneity may differently a ffect the cost of access to raw water.37 Even when the specific effect of regional heteroge neity is not the focus of this st udy, it is important to control for that time-invariant characteristic. If the firm -type dummy variables are statistically significant in the second-stage regression, there will be evidence of relatively distinct firm-specific costs between the different types of operators. Table 2-4 presents the results of the second-st age regression. The nega tive and statistically significant coefficient of the Regional dummy indicates that regiona l providers have lower firm36 The high explanatory power of both models remains even if the firm fixed-effects (not reported) are excluded. In that case, the R2 is still above 0.95. 37 These regional dummies could partially capture different energy prices as well.

PAGE 66

66 specific costs than all other operators. At the local level, meanwhile, the firm-specific cost comparison between private and public operato rs depends on how the public provider is organized. The positive and statisti cally significant coefficient of the Public-Non-Corp dummy variable indicates that the loca l-public-non-corporative operators ha ve higher firm-specific costs than both the local-private and the local-public-c orporative providers. Tabl e 2-4 also shows that there is not a statistically signi ficant difference between the firm-s pecific cost of local-private and local-public-cor porative operators. Table 2-5 illustrates the value of the firm-speci fic costs, obtained from the results of the second-stage regression. The table shows that the region al public operators are the lowest-cost WS providers, while the local-public-non-corp orative operators are the highest-cost WS providers in Brazil. Although the differences are substantial and statisti cally significant, it is worth recalling that the firm-specific costs are not a significant portion of the operating costs. Sensitivity Checks As a first sensitivity check, a balanced panel was used. Since the sample size increases over the years, it is important to check that inco rporating new firms does not affect the analysis. The results are presented in Table 2-6. The positive and significant effect of both output variables ( Connections and Volume ) remains. The same is true for the Wage variable across both specifications. Table 2-7 contains the results of the second stage re gressions. It can be verified that the lower firm-specific cost for the Regional type is confirmed when using a balanced panel, regardless of the output variable chosen. Howeve r, the higher firm-specific cost for the publicnon-corporative type is not obtai ned when a balanced panel is employed. The reason behind this statistically imprecise result could be the loss of observations when utilizing a balanced panel. That is, most of the observations dropped when using a balanced panel correspond to local firms (which include the public-non-corporativ e type) rather than Regional ones.

PAGE 67

67 As a second sensitivity check, and utilizing agai n the larger unbalanced panel, the Regional type was excluded from the sample. The reason fo r this sensitivity check is that the size differential between the Regional and Local operators could affect the conclusions. Table 2-8 shows the results of the first stage LSDV regre ssion and Table 2-9 contains the results of the second stage regression. The firs t stage regressions show that the positive and statistically significant effect of output, wage and the meteri ng variable are again verified, along with the sewerage dummy. Furthermore, the coefficients obtained are similar to those in Table 2-3 and Table 2-4. For example, the measure of economies of scale indicates that a 10% increase in the volume of water produced generates a 0.94% in crease in operating cost. The second stage regressions show that the higher firm-specific cost for the public -non-corporative type is again confirmed, providing confidence about the co nclusions drawn earlier. The statistically insignificant difference between the private and p ublic-corporative operators is also verified. Conclusions Brazil is a country that lacks perfect access to WS services. Efficiency improvements could free up funds for network expansion, which would c onstitute a step toward s a desired full service situation. Therefore, greater attention for cost-containment is needed, regardless of the jurisdictional and owne rship/organizational st atus of the WS operators. Improving our understanding of relative performa nce can help policy makers focus on the sources of differential cost patterns. The results of this study suggest that, at least for Brazil, evidence of economies of scale is enough to claim that WS provision at the state level is more effi cient than WS provision at the municipal level. Economies of s cale generate substantial cost savings, which far outweigh any potential differential in firm-specific costs. As such, the argument ma de by AESBE seems more compelling than the argument made by ASSEMAE.

PAGE 68

68 In addition, this chapter finds evidence of inherently lower fi rm-specific costs (per unit of output) for regional WS firms than for all other types of WS operators in Brazil. These lower costs reinforce the efficiency gains the regional firms achieve through actual economies of scale. Finally, this study shows evidence of higher firm -specific costs (per un it of output) for localpublic-non-corporative providers than for local-p rivate and local-public -corporative providers. Future research could examine what features generate the in trinsic cost differences among operator types. In particular, the higher firm-s pecific cost for the public-non-corporative type deserves further attention. It may be important to check whether the not-for-profit motive of those organizations actually drives their higher firm-specific costs. Even when cost differences between the differe nt types of WS operators were found, it is worth noting that these differences represent a small portion of operating costs. The first-stage regressions illustrated th at output, input prices and other technological f actors explain most of the variation of operating cost, regardless of the firm-type. As a result, the firm-specific cost differences presented are significant from a statistical point of vi ew, but less significant from an economic perspective. Clearly, much work remains. For the purpose of rewarding good performance and penalizing weak performance, scholars and practit ioners need to develop efficiency-measuring procedures that can pass legal ch allenges. The process must con tinue to build on the pioneering research of those whose work is cited in the references. In part icular, the publication of league tables is one way to put pressure on the weakest performing WS utilities. Similarly, the managers of WS utilities in the top 20 percent might be awar ded some share of the cost savings that can be attributed to their efforts. T hose promoting improvements in WS sector performance can take steps to reduce production costs and free up cash flows for netw ork rehabilitation and expansion.

PAGE 69

69 Identifying, implementing, and evaluating good incentive systems represent a challenge for regulators. A final issue that also deserves future research follows. The analysis in this chapter only considered relative measures of efficiency. The goal was to identify sources of cost differences between the different types of operators. However, cost savings for the en tire industry could also be estimated utilizing the results obtained. For example, reducing water losses would also free up funds for network expansion. Table 2-1 shows th at water losses in 2004 stood at almost 30% on average, while they were 49% for the state ope rators. Thus, a reduction of water losses by 10% should not be hard to achieve. Yet, it could represent an almo st 1% lower operating cost. Some researchers suggest that the onl y explanation behind the lack of incentives for reducing water losses is that it may be cheaper to produce more water instead. Although geography might validate this statement in some cases, the issue deserves further exploration.

PAGE 70

70 Table 2-1. Average statistics by operator-type for 200438 Type of firm # Connect. Empl. Water cover. Sewer. cover. Water losses Sewer. Treatm. Private 31 30470 88.9 80.1% 50.4% 29.1% 54.4% Public Non-Corp 296 18851 99.2 86.3% 63.0% 26.6% 46.5% Public Corp 11 75180 400.3 98.8% 68.5% 41.3% 36.6% Regional 25 1104748 2978.6 71.3% 33.8% 48.7% 74.5% Total 369 91660 300.7 85.1% 58.5% 28.9% 51.3% 38 The sample also includes six Microregional operators, wh ich are not the focus of the analysis. These are public operators that are neither regional nor local, since th ey provide services to just a few municipalities.

PAGE 71

71 Table 2-2. Summary statistic s for first-stage regressions39 Variable Private N=62 Public Non Corp. N=913 Public Corp. N=42 Regional N=121 Operating Cost 9,102,144 5,088,097 25,661,470 239,346,200 (12,500,360) (13,124,270) (34,681,050) (338,851,600) Connections 50,539 24,251 74,573 1,021,909 (67,902) (39,282) (67,960) (1,234,135) Volume 22,239 8,552 34,899 396,122 (41,240) (17,315) (33,649) (583,563) Wage 15,959 11,512 18,082 31,147 (6,953) (6,256) (9,349) (10,613) Network Length 642 294 1,005 11,398 (840) (457) (1,044) (12,930) Dummy sewerage 0.82 0.56 0.98 0.94 (0.39) (0.50) (0.15) (0.23) Urban % 0.88 0.77 0.95 0.77 (0.09) (0.21) (0.05) (0.12) Metering % 0.88 0.77 0.86 0.74 (0.19) (0.32) (0.21) (0.28) Fluorination % 0.33 0.30 0.35 0.20 (0.46) (0.44) (0.47) (0.36) N=1163 (Standard deviations in parenthesis) 39 Volume is in 1000m3/year and Network length is in Km. Operating cost and Wage are in Reais/year deflated using 2000 as the base year. For the Volume variable, summary statistics are for 1172 observations. The 1163 observations include 25 observations for the Microregional category, which are not reported.

PAGE 72

72 Table 2-3. First-stag e LSDV regression results40 Dependent Variable: Operating Cost Connections Volume Connections 0.427 (0.112)*** Volume 0.098 (0.040)** Wage 0.150 0.157 (0.043)*** (0.040)*** Network Length 0.024 0.089 (0.073) (0.088) Dummy sewerage 0.122 0.138 (0.058)** (0.067)** Urban % 0.070 0.082 (0.246) (0.258) Metering % 0.382 0.409 (0.177)** (0.165)** Fluorination % 0.001 0.005 (0.035) (0.035) Constant 8.496 11.274 (1.133)*** (0.713)*** Observations 1163 1172 R-squared 0.99730 0.99729 Year and firm fixedeffect not reported Standard errors clustered at the state-level significant at 10% ** significan t at 5% *** significant at 1% 40 Operating Cost, Volume, Connections, Wage and Network length are in ln form. The statistical significance of all coefficients is very similar when the standa rd errors are clustered at the Region level.

PAGE 73

73 Table 2-4. Second-st age regression results Dependent Variable: (ln) Firm-Specific Cost per Unit of Output (from LSDV regressions) Connections Volume Public Non Corp 0.331 0.294 (0.127)*** (0.124)** Public Corp -0.009 -0.073 (0.217) (0.176) Regional -1.154 -0.270 (0.190)*** (0.158)* Constant -0.960 2.640 (0.124)*** (0.116)*** Observations 380 380 R-squared 0.31726 0.04428 Omitted type: Private Region fixed-eff ect not reported Robust standard errors in parenthesis significant at 10% ** significan t at 5% *** significant at 1%

PAGE 74

74 Table 2-5. Ranking of firm-sp ecific costs across firm-types. Firm type Firm-Specific cost per unit of output Index (Regional=100) ($/Connection) ($/1000m3) Regional 0.12 10.70 100 100 Public Corp 0.38 13.03 314 122 Private 0.38 14.01 317 131 Public Non Corp 0.53 18.80 441 176

PAGE 75

75 Table 2-6. First-stage LSDV regres sion results using a balanced panel Dependent Variable: Operating Cost Connections Volume Connections 0.631 (0.152)*** Volume 0.127 (0.043)*** Wage 0.185 0.186 (0.022)*** (0.026)*** Network Length 0.014 0.082 (0.074) (0.104) Dummy sewerage 0.034 0.046 (0.048) (0.045) Urban % -0.158 -0.150 (0.135) (0.139) Metering % -0.002 -0.009 (0.203) (0.188) Fluorination % 0.023 0.023 (0.028) (0.030) Constant 7.894 13.287 (1.155)*** (1.053)*** Observations 758 766 R-squared 0.99713 0.99711 Year and firm fixedeffect not reported Standard errors clustered at the state-level significant at 10% ** significan t at 5% *** significant at 1%

PAGE 76

76 Table 2-7. Second-stage regressi on results using a balanced panel Dependent Variable: (ln) Firm-Specific Cost per Unit of Output (from LSDV regressions) Connections Volume Public Non Corp 0.176 0.265 (0.219) (0.179) Public Corp -0.005 0.056 (0.301) (0.166) Regional -1.966 -0.339 (0.259)*** (0.182)* Constant -2.702 2.795 (0.328)*** (0.249)*** Observations 170 170 R-squared 0.54194 0.11137 Omitted type: Private Region fixed-eff ect not reported Robust standard errors in parenthesis significant at 10% ** significan t at 5% *** significant at 1%

PAGE 77

77 Table 2-8. First-stage LSDV regression results ex cluding the Regional type Dependent Variable: Operating Cost Connections Volume Connections 0.439 (0.130)*** Volume 0.094 (0.040)** Wage 0.145 0.152 (0.045)*** (0.041)*** Network Length 0.022 0.083 (0.072) (0.087) Dummy sewerage 0.119 0.135 (0.058)* (0.066)* Urban % 0.062 0.077 (0.252) (0.264) Metering % 0.424 0.460 (0.194)** (0.179)** Fluorination % 0.004 0.005 (0.039) (0.039) Constant 8.433 11.37 (1.326)*** (0.721)*** Observations 1042 1047 R-squared 0.99579 0.99575 Year and firm fixedeffect not reported Standard errors clustered at the state-level significant at 10% ** significan t at 5% *** significant at 1%

PAGE 78

78 Table 2-9. Second-stage regression results excluding the Regional type Dependent Variable: (ln) Firm-Specific Cost per Unit of Output (from LSDV regressions) Connections Volume Public Non Corp 0.331 0.286 (0.127)*** (0.120)** Public Corp -0.037 -0.077 (0.220) (0.170) Constant -1.005 2.711 (0.124)*** (0.113)*** Observations 354 354 R-squared 0.1322 0.01567 Omitted type: Private Region fixed-eff ect not reported Robust standard errors in parenthesis significant at 10% ** significan t at 5% *** significant at 1%

PAGE 79

79 APPENDIX DERIVATION OF THE SOLUTIONS TO CHAPTER 1 Limits on Communication When Correlations are Relatively Similar If Eq. 1-17 holds, solve [P-NOC] imposing only Eq. 1-9 (multiplier denoted with 1LP), Eq. 1-10 (multiplier denoted with 0LP ), Eq. 1-12 (multiplier denoted with 0HP ) and Eq. 1-14 (multiplier denoted with 0LI ). Solving from the first order conditions with respect to payments yields: 01 1 LP, 0 1 20 0 1 1 0 0 LP, 0 1 20 0 1 1 0 LI and 0 1 2 10 0 1 1 0 0 HP. Setting equal to zero the first order conditi ons with respect to the effort vector: 0 ) ( ) ( ) ( ) ( ] [ )] ( 1 ][ )[ 1 (0 0 0 0 1 1 0 0 1 1 0 0 1 1 LL L LL L LL L LL LL LLe I e P e P e e e L Substituting the value of the multipliers and simplifying: 1 ) ( LLe 0 ) ( ) 1 ( ) ( ) 1 ( ) ( ) 1 ( ) ( )] 1 ( ) 1 ( [ )] ( 1 )][ 1 ( ) 1 ( )[ 1 (0 0 0 0 1 1 0 0 1 1 0 0 1 1 LH L LH L LH L LH LH LHe I e P e P e e e L Substituting the value of the multipliers and simplifying: 1 ) ( LHe 0 ) ( ) 1 ( ) ( ) ( ] [ )] ( 1 ][ )[ 1 (0 0 0 0 0 0 1 1 0 0 1 1 HH L HH H HH HH HHe I e P e e e L Substituting the value of the multipliers and simplifying:

PAGE 80

80 1 ) ( ) 1 ( ) 1 2 ( ) 1 ( 1 ) ( '' 0 0 1 1 0 1 0 0 1 HH HHe 0 ) ( ) ( ) 1 ( ) ( )] 1 ( ) 1 ( [ )] ( 1 )][ 1 ( ) 1 ( )[ 1 (0 0 0 0 0 0 1 1 0 0 1 1 HL L HL H HL HL HLe I e P e e e L Substituting the value of the multipliers and simplifying: 1 ) 1 ( ) 1 2 ( ) 1 ( 1 ) ( '' 0 0 1 1 0 1 0 0 1 HL HLe Solve from the binding constrai nts to obtain the payments. From Eq. 1-9 and Eq. 1-10: ) (*e tLL and ) (*e tLH Substituting in Eq. 1-14 and simplifying: HL HH HL HHt e e t0 0 0 01 ) ( ) ( 1 Rearranging terms in Eq. 1-12: HL HL HH HHt e e t0 0 0 01 ) ( 1 ) ( Equating the right-hand-sides of the last two equations: HL HH HL HLe t 0 0 0 0) 1 ( 1 2 ) ( Substituting back in Eq. 1-12: HL HH HH HHe t 0 0 0 0) 1 ( 1 2 1 ) ( The excluded constraints (Eq. 1-11, Eq. 1-13, Eq. 1-15 and Eq. 1-16) are satisfied (not binding) by the payments above. In particular, su bstituting the payments in Eq. 1-11 yields the rent for the high-cost agen t if correlation is high:

PAGE 81

81 0 ) 1 ( 1 20 0 0 0 1 1 HL HH HU When Correlations are Relatively Different If Eq. 1-17 does not hold, solve [P-NOC] imposing only Eq. 1-9 (multiplier denoted with 1LP), Eq. 1-12 (multiplier denoted with 0HP ), Eq. 1-13 (multiplier denoted with 1LI) and Eq. 114 (multiplier denoted with 0LI ). Solving from the first order conditions with respect to payments: 0 1 ) 1 2 ( ) (0 1 0 0 0 1 1 1 LI, 00 0 LI 0 1 1 20 1 0 1 LP and 0 1 1 20 1 1 0 HP Setting equal to zero the first order conditi ons with respect to the effort vector: 0 ) ( ) ( ) ( ) ( ] [ )] ( 1 ][ )[ 1 (0 0 1 1 1 1 0 0 1 1 0 0 1 1 LL L LL L LL L LL LL LLe I e I e P e e e L Substituting the value of the multipliers and simplifying: 1 ) ( LLe 0 ) ( ) 1 ( ) ( ) 1 ( ) ( ) 1 ( ) ( )] 1 ( ) 1 ( [ )] ( 1 )][ 1 ( ) 1 ( )[ 1 (0 0 1 1 1 1 0 0 1 1 0 0 1 1 LH L LH L LH L LH LH LHe I e I e P e e e L Substituting the value of the multipliers and simplifying: 1 ) ( LHe 0 ) ( ) 1 ( ) ( ) 1 ( ) ( ) ( ] [ )] ( 1 ][ )[ 1 (0 0 1 1 0 0 0 0 1 1 0 0 1 1 HH L HH L HH H HH HH HHe I e I e P e e e L Substituting the value of the multipliers and simplifying:

PAGE 82

82 1 ) ( ) 1 ( ) 1 ( ) 1 ( 1 ) ( '' 0 0 1 1 0 0 1 1 0 1 0 1 HH HHe 0 ) ( ) ( ) ( ) 1 ( ) ( )] 1 ( ) 1 ( [ )] ( 1 )][ 1 ( ) 1 ( )[ 1 (0 0 1 1 0 0 0 0 1 1 0 0 1 1 HL L HL L HL H HL HL HLe I e I e P e e e L Substituting the value of the multipliers and simplifying: 1 ) 1 ( ) 1 ( ) 1 ( ) 1 ( 1 ) ( '' 0 0 1 1 0 0 1 1 0 1 0 1 HL HLe Solve from the binding constraints to obtain the payments. From Eq. 1-9: )] ( [ 1 ) (1 1 LH LH LL LLe t e t From Eq. 1-12: )] ( [ 1 ) (0 0 HL HL HH HHe t e t Substituting the last two expressions into Eq. 1-13 and Eq. 1-14 yields: HL HH HL HLe t 1 1 0 1 0) 1 ( 1 ) ( HL HH LH LHe t ) 1 ( 1 ) (0 0 0 1 1 Substituting LHt in Eq. 1-9 yields: HL HH LL LLe t ) 1 ( 1 1 ) (0 0 0 1 1 Substituting HLt in Eq. 1-12 yields: HL HH HH HHe t 1 1 0 1 0) 1 ( 1 1 ) (

PAGE 83

83 The excluded constraints (Eq. 1-10, Eq. 1-11, Eq. 1-15 and Eq. 1-16) are satisfied (not binding) by the payments above. In particular, substituting the payments in Eq. 1-10 and Eq. 111 yields the rent for the high-cost agent if correlation is high and the rent for the low-cost agent if correlation is low: 0 ) 1 ( 11 1 0 1 0 1 1 HL HH HU 0 ) 1 ( 10 0 0 1 0 1 0 HL HH LU No Limits on Communication When Correlations are Relatively Similar If Eq. 1-17 holds, solve [P-CO] imposing only Eq. 1-19 (multiplier denoted with 19 ), Eq. 1-20 (multiplier denoted with 20 ), Eq. 1-21 (multiplier denoted with 21 ), Eq. 1-23 (multiplier denoted with 23 ) and Eq. 1-30 (multiplier denoted with 30 ). Solving from the first order conditions with respect to payments yields: 0 1 20 0 1 1 23 0 1 2 ) 1 ( ) 1 2 (0 0 1 1 0 0 20 0 1 20 0 1 1 0 19 01 21 and 01 30 Setting equal to zero the first order conditi ons with respect to the effort vector: 0 0 0 0 0 0) 1 ( ) ( LL LLe e L) ( '0 0 19 LLe ) ( '0 0 23 LLe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '0LLe 0 0 0 0 0 0) 1 ( ) ( HH HHe e L ) ( '0 0 20 HHe

PAGE 84

84 ) ( ) 1 (0 0 23 HHe ) ( '0 1 30 HHe =0 Substituting the value of the multipliers and simplifying: 1 ) 1 2 ( ) ( ) 1 ( ) 1 ( 1 ) ( '0 0 0 1 0 0 0 1 0 HH HHe ) 1 ( ) 1 ( ) ( ) 1 (0 0 0 0 0 0 LH LHe e L ) ( ) 1 (0 0 19 LHe ) ( ) 1 (0 0 23 LHe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '0LHe ) 1 ( ) 1 ( ) ( ) 1 (0 0 0 0 0 0 HL HLe e L ) ( ) 1 (0 0 20 HLe ) ( '0 0 23 HLe ) ( ) 1 (0 1 30 HLe =0 Substituting the value of the multipliers and simplifying: 1 ) 1 2 ( ) ( ) 1 ( ) 1 ( 1 ) ( '0 0 0 1 0 0 0 1 0 HL HLe 1 1 1 1 1 1) 1 ( ) ( LL LLe e L ) ( '1 1 21 LLe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1LLe 1 1 1 1 1 1) 1 ( ) ( HH HHe e L ) ( '1 1 30 HHe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1HHe ) 1 ( ) 1 ( ) ( ) 1 (1 1 1 1 1 1 LH LHe e L) ( ) 1 (1 1 21 LHe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1LHe

PAGE 85

85 ) 1 ( ) 1 ( ) ( ) 1 (1 1 1 1 1 1 HL HLe e L ) ( ) 1 (1 1 30 HLe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1HLe Solve from the binding constraints to obtain the payments. The five equations are as follows: 0 )] ( )[ 1 ( )] ( [0 0 0 0 0 0 LH LH LL LLe t e t (1-19) 0 )] ( )[ 1 ( )] ( [0 0 0 0 0 0 HL HL HH HHe t e t (1-20) 0 )] ( )[ 1 ( )] ( [1 1 1 1 1 1 LH LH LL LLe t e t (1-21) 0 )] ( ) ( )[ 1 ( )] ( ) ( [0 0 0 0 0 0 0 0 0 0 HH HH LH LH HL HL LL LLe t e t e t e t (1-23) 0 )] ( ) ( )[ 1 ( )] ( ) ( [0 0 1 1 1 0 0 1 1 1 HL HL HL HL HH HH HH HHe t e t e t e t (1-30) The system has an infinite number of so lutions. Selecting arbitrary values for 1LHt 1HLt and 0LHt a solution can be characterized as follows: )] ( [ ) 1 ( ) (0 0 0 0 0 0LH LH LL LLe t e t 0 0 0 0 0 0 0 0) 1 ( 1 2 ) (HH HL HL HLe t 0 0 0 0 0 0 0 0) 1 ( 1 2 1 ) (HH HL HH HHe t )] ( [ ) 1 ( ) (1 1 1 1 1 1LH LH LL LLe t e t 0 0 0 0 0 1 0 1 1 1 1 1 1 1) 1 ( ) 1 2 ( )] ( [ ) 1 ( ) (HH HL HL HL HH HHe t e t

PAGE 86

86 Some of the excluded constraints are automati cally satisfied by the payments above. For the rest of the excluded constr aints to be satisfied, the following bounds on the arbitrary payments should hold: 1 0 1 0 1 0 1 1 1 1 1 1 0 0 0 0 0 0 1 1 1 1 1 1) 1 ( 1 ) ( ; ) 1 ( ) 1 ( 1 2 1 2 ) ( min ) (LL LH LH LL LH HH HL LH LH LHe e t e where 0 ) ( ) ( s Li s Li s Lie e for } { H L i and } 0 1 { s 0 0 0 0 0 1 0 0 1 0 1 0 1 0 1 0 1 1 0 0 0 0 0 1 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 1) 1 ( ) 1 2 )( 1 ( ) )( 1 ( ) 1 ( 1 ) ( ; ) 1 ( ) 1 2 )( 1 2 ( ) )( 1 ( ) 1 ( 1 2 ) ( min ) 1 ( 1 2 ) (HH HL HH HL HL HH HL HH HL HL HL HH HL HLe e t e 0 1 0 1 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0) 1 ( ) 1 ( 1 2 1 ) ( ; ) 1 ( 1 2 ) ( min ) (LL LH HH HL LH LL LH LH LH LHe e t e In particular, substituting the payments into Eq. 1-22 yields the rent for the high-cost agent if correlation is high: 0 ) 1 ( ) 1 2 (0 0 0 0 0 0 1 1 HH HL HU

PAGE 87

87 When Correlations are Relatively Different If Eq. 1-17 does not hold, solve [P-CO] impos ing only Eq. 1-20 (multiplier denoted with 20 ), Eq. 1-21 (multiplier denoted with 21 ), Eq. 1-23 (multiplier denoted with 23 ), Eq. 1-30 (multiplier denoted with 30 ) and Eq. 1-33 (multiplier denoted with 33 ). Solving from the first order conditions with respect to payments yields: 1 1 21 0 1 20 >0, 1 1 21 0 0 21 >0, 0 23 >0, 1 30 >0 and 1 ) 1 2 ( ) (1 0 0 0 0 1 1 33 >0 Setting equal to zero the first order conditions with respect to the effort vector: 0 0 0 0 0 0) 1 ( ) ( LL LLe e L ) ( '0 0 23 LLe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '0LLe 0 0 0 0 0 0) 1 ( ) ( HH HHe e L ) ( '0 0 20 HHe ) ( ) 1 (0 0 23 HHe ) ( '0 1 30 HHe ) ( ) 1 (0 1 33 HHe =0 Substituting the value of the multipliers and simplifying: 1 ) 1 ( ) ( )] 1 ( [ ) 1 ( 1 ) ( '0 1 0 0 1 0 0 1 1 0 0 0 HH HHe ) 1 ( ) 1 ( ) ( ) 1 (0 0 0 0 0 0 LH LHe e L ) ( ) 1 (0 0 23 LHe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '0LHe ) 1 ( ) 1 ( ) ( ) 1 (0 0 0 0 0 0 HL HLe e L ) ( ) 1 (0 0 20 HLe

PAGE 88

88 ) ( '0 0 23 HLe ) ( ) 1 (0 1 30 HLe ) ( '0 1 33 HLe =0 Substituting the value of the multipliers and simplifying: 1 ) 1 ( ) ( ) 1 ( ] ) 1 ( [ ) 1 ( 1 ) ( '0 1 0 0 1 0 0 1 1 0 0 0 HL HLe 1 1 1 1 1 1) 1 ( ) ( LL LLe e L ) ( '1 1 21 LLe ) ( '1 1 33 LLe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1LLe 1 1 1 1 1 1) 1 ( ) ( HH HHe e L ) ( '1 1 30 HHe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1HHe ) 1 ( ) 1 ( ) ( ) 1 (1 1 1 1 1 1 LH LHe e L) ( ) 1 (1 1 21 LHe ) ( ) 1 (1 1 33 LHe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1LHe ) 1 ( ) 1 ( ) ( ) 1 (1 1 1 1 1 1 HL HLe e L ) ( ) 1 (1 1 30 HLe =0 Substituting the value of the multipliers and simplifying: 1 ) ( '1HLe Solve from the binding constraints to obtain the payments. The five equations are as follows: 0 )] ( )[ 1 ( )] ( [0 0 0 0 0 0 HL HL HH HHe t e t (1-20) 0 )] ( )[ 1 ( )] ( [1 1 1 1 1 1 LH LH LL LLe t e t (1-21) 0 )] ( ) ( )[ 1 ( )] ( ) ( [0 0 0 0 0 0 0 0 0 0 HH HH LH LH HL HL LL LLe t e t e t e t (1-23)

PAGE 89

89 0 )] ( ) ( )[ 1 ( )] ( ) ( [0 0 1 1 1 0 0 1 1 1 HL HL HL HL HH HH HH HHe t e t e t e t (1-30) 0 )] ( ) ( )[ 1 ( )] ( ) ( [0 0 1 1 1 0 0 1 1 1 HH HH LH LH HL HL LL LLe t e t e t e t (1-33) The system has an infinite number of so lutions. Selecting arbitrary values for 1 LHt 1 HLt and 0 LHt a solution can be characterized as follows: )] ( [ ) 1 ( ) (1 1 1 1 1 1 LH LH LL LLe t e t 0 1 0 1 0 1 0 0 0) 1 ( 1 ) (HL HH HL HLe t 0 1 0 1 0 1 0 0 0) 1 ( 1 1 ) (HL HH HH HHe t ) ( 1 ) 1 ( ) 1 ( ) ( ) (0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 LH LH HL HH LL LLe t e t ) ( ) 1 ( ) 1 ( ) 1 ( ) (1 1 1 1 0 1 0 1 0 1 1 0 1 1 1 HL HL HL HH HH HHe t e t Some of the excluded constraints are automati cally satisfied by the payments above. For the rest of the excluded constr aints to be satisfied, the following bounds on the arbitrary payments should hold: 0 0 0 0 0 1 1 0 0 1 0 1 0 1 0 0 0 1 0 1 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0) 1 ( ) 1 ( ) ( ; ) 1 ( 1 1 ) ( ) ( ; ) 1 ( ) 1 2 )( 1 ( ) )( 1 ( ) 1 ( 1 2 ) ( minHL HH LH LH LL HL LH HL HH LL LH LH LHe e e t

PAGE 90

90 0 1 0 1 0 1 0 1 1 0 1 0 1 0 1 0 1 0 0 1 1 0 1 0 1 1 0 1 0 1 1 1 1 1 1 1 1 1 1) 1 ( ) 1 ( ) ( ; ) 1 ( 1 1 ) ( ; ) 1 ( ) 1 2 )( 1 ( ) )( 1 ( ) 1 ( 1 2 ) ( minHL HH HL HL HH HH HL HL HH HL HH HL HLe e e t ) 1 ( ) 1 ( ) ( ; ) 1 ( 1 ) ( ; ) 1 ( ) 1 2 )( 1 ( ) ( ) 1 ( 1 2 ) ( min0 0 0 0 0 1 1 1 1 0 1 0 1 0 1 1 0 1 0 1 1 0 1 0 1 1 1 1 1 1 1 1 1 1 HL HH LH LL LH LH HL HH LH LL LH LHe e e t In particular, substituting the payments into Eq. 1-19 and Eq. 1-22 yields the rent for the low-cost agent if correlation is low and the rent for the high-cost agent if correlation is high: 0 ) 1 ( ) 1 ( ) (0 0 0 0 0 1 0 1 0 HL HH LU 0 ) 1 ( ) 1 (0 1 0 1 0 1 0 1 1 HL HH HU Limits on Communication and Exclusion Solve [P-EXCL] imposing only Eq. 1-38 (multiplier denoted with 38 ), Eq. 1-39 (multiplier denoted with 39 ) and Eq. 1-42 (multiplier denoted with 42 ). Solving from the first order conditions with respect to payments yields: 0 38 >0, 1 39 >0 and 1 42 >0. Setting equal to zero the firs t order conditions with respec t to the effort vector and substituting the value of the multipliers yields 1 ) ( ije for }. { H L j i Solving from the binding constraints yields the payments. The solution is given by:

PAGE 91

91 ) (LH LHe t ) (LL LLe t and )] ( [ ) 1 ( ) (1 1HL HL HH HHe t e t with HLt arbitrary. Some of the excluded constraints are automati cally satisfied by the payments above. For the rest of the excluded c onstraints to be satisfied, the following upper bound on HLt should hold: )] ( *) ( [ 1 *) ( )]; ( *) ( [ 1 2 *) ( *); ( min1 0 1 1 1e e e e e e e tHL

PAGE 92

92 LIST OF REFERENCES Antonioli, B. and Filipini, M. (2001) The use of a variable cost function in the regulation of the Italian water industry, Utilities Policy 10 181-187. Arya, A., Demski, J., Glover, J. and Liang, P. (2005) Quasi-robust multiagent contracts, Carnegie Mellon Working Paper. Ashton, J. (2000) Cost efficiency in the UKs water and sewerage industry, Applied Economics Letters 7 455-458. Aubert, C. and Reynaud, A. (2005) The impact of regulation on cost e fficiency: An empirical analysis of Wisconsin water utilities, Journal of Productivity Analysis 23 383-409. Auriol, E. and Laffont, J. J. (1992) Regul ation by duopoly, Journal of Economics and Management Strategy 1 507-533. Bergemann, D. and Morris, S. (2005) Robust mechanism design, Econometrica 73 (6), 17711813. Bergemann, D. and Vlimki (2006) Informa tion in mechanism design, Cowles Foundation Discussion Paper No. 1532R. Bhattacharyya, A., Harris, T., Narayanan, R. and Raffiee, K. ( 1995) Specification and estimation of the effect of ownership on the economic efficiency of the water utilities, Regional Science and Urban Economics 25 (6), 759-784. Bhattacharyya, A., Parker, E. and Raffiee, K (1994) An examination of the effect of ownership on the relative efficiency of public and private water utilities, Land Economics 70 (2), 197209. Bottaso, A. and Conti, M. (2003) Cost inefficien cy in the English and Welsh water industry: An heteroskedastic stochastic cost frontier appr oach, mimeo, DIEM Universit di Genova. Byrnes, P., Grosskopf, S. and Hayes, K. (1986) Efficiency and ownership: further evidence, Review of Economics and Statistics 68 (2), 337-341. Cameron, A. and Trivedi, P. (2005) Microeconometrics, Me thods and Applications. Cambridge University Press. Clarke, G., Kosec, K. and Wallsten, S. (2004) H as private participation in water and sewerage improved coverage? Empirical evidence from Latin America, World Bank Policy Research Working Paper No. 3445. Corton, M. (2003) Benchmarking in the Latin American water sector: the case of Peru, Utilities Policy 11 133-142.

PAGE 93

93 Crain, W. and Zardkoohi, A. (1978) A test of the property-rights theory of the firm: water utilities in the United States, Journal of Law and Economics 21 (2), 395-408. Cremer, J. and McLean, R. (1985) Optimal selling strategies under uncertainty for a discriminating monopolist when demands are interdependent, Econometrica, 53 (2), 345-362. Cremer, J. and McLean, R. (1988) Full extracti on of the surplus in bayesian and dominant strategy auctions, Econometrica, 56 (6), 1247-1257. Cubbin, J. and Tzanidakis, G. (1998) Regress ion versus data envelopment analysis for efficiency measurement: an application to the England and Wales regulated water industry, Utilities Policy 7 75-85. Dasgupta, P. and Maskin, E. (2000) Efficient auctions, Quarterly Journal of Economics 115 341-388. Demski, J. and Sappington, D. (1984) Optimal incentive contracts with multiple agents, Journal of Economic Theory 33 152-171. Demski, J., Sappington, D. and Sp iller, P. (1988) Incentive schemes with multiple agents and bankruptcy constraints, Journal of Economic Theory, 44 156-167. Es, P. (2005) An optimal auction with correlated values and risk aversion, Journal of Economic Theory, 125 78-89. Estache, A. and Kouassi, E. ( 2002) Sector organization, governa nce, and the inefficiency of African water utilities, World Bank Polic y Research Working Paper No. 2890. Estache, A. and Rossi, M. (1999) Comparing the performance of public and private water companies in Asia and Pacific region: what a stochastic costs frontier shows, World Bank Policy Research Working Paper No. 2152. Estache, A. and Rossi, M. (2002) How different is the efficiency of public and private water companies in Asia?, The World Bank Economic Review 16 (1), 139-148. Estache, A. and Trujillo, L. ( 2003) Efficiency effects of priva tization in Argentinas water and sanitation services, Water Policy 5 (4), 369-380. Fabbri, P. and Fraquelli, G. (2000) Costs a nd structure of technology in the Italian water industry, Empirica 27 65-82. Fang, H. and Morris, S. (2006) Multid imensional private value auctions, Journal of Economic Theory 126 1-30. Faria, R. (2005) Public versus private wa ter utilities: empirical evidence for Brazilian companies, Economics Bulletin 8 (2), 1-7.

PAGE 94

94 Feigenbaum, S. and Teeples, R. (1983) Publ ic vs. private water delivery: a hedonic cost approach, The Review of Economics and Statistics 65 (4), 672-678. Fox, W. and Hofler, R. (1986) Using homothet ic composed error frontiers to measure water utility efficiency, Southern Economic Journal 53 (2), 461-477. Fudenberg, D. and Tirole, J (1991) Game Theory The MIT Press. Galiani, S. Gertler, P. and Schargrodsky, E. (2005) Water for life: the imp act of privatization of water services on child mortality, Journal of Political Economy 113 (1), 83-120. Greene, W. (2003) Econometric Analysis 5th Edition. Prentice Hall. Greene, W. (2005) Fixed and random eff ects in stochastic frontier models, Journal of Productivity Analysis 23 7-32. Heifetz, A. and Neeman, Z. (2006) On the gene ric impossibility of fu ll surplus extraction in mechanism design, Econometrica 74 (1), 213-233. Jehiel, P. and Moldovanu, B (2001) Efficient design with interdependent valuations, Econometrica 69 (5), 1237-1259. Kim, H. Y. (1987) Economies of scale in multiproduct firms: an empirical analysis, Economica (new series) 54 (214), 185-206. Kumbhakar, S. and Lovell, K. (2000) Stochastic Frontier Analysis Cambridge University Press. Laffont, J. J. and Tirole, J. (1986) Usi ng cost observation to regulate firms, Journal of Political Economy 94 614-641. McAfee, R. and Reny, P. (1992) Correla ted information and mechanism design, Econometrica 60 (2), 395-421. Miller, N., Pratt, J., Zeckhauser, R. a nd Johnson, S. (2007) Mechanism design with multidimensional, continuous types a nd interdependent valuations, Journal of Economic Theory forthcoming. Murillo-Zamorano, L. (2004) Economic efficiency and frontier techniques, Journal of Economic Surveys 18 (1), 33-77. Neeman, Z. (2004) The relevance of pr ivate information on mechanism design, Journal of Economic Theory, 117 55-77. Oliveira, G. and Fujiwara, T. (2005) Brazils regulatory fr amework: predictability or uncertainty?, Texto para Discusso No. 147, Escola de Economia de Sao Paulo. Parlatore, A. (1999) Privatization of the wate r utility sector in Brazil, Essay No. 8 in Privatization in Brazil: the case of public utilities, BNDES.

PAGE 95

95 Pinheiro, A. (2003) Regulatory reform in Brazilian infrastructu re: where do we stand?, Texto para Discusso No. 964, IPEA. Robert, J. (1991) Continu ity in auction design, Journal of Economic Theory 55 169-179. Saal, D. and Parker, D. (2000) The impact of privatization and re gulation on the water and sewerage industry in England and Wale s: a translog cost function model, Managerial and Decision Economics 21 253-268. Sappington, D. (1983) Limited liability co ntracts between prin cipal and agent, Journal of Economic Theory, 29 1-21. Schmidt, P. and Sickles, R. (1984) Production frontiers and panel data, Journal of Business and Economic Statistics 2 (4), 367-374. Seroa da Motta, R. and Moreira, A. (2006) Effici ency and regulation in th e sanitation sector in Brazil, Utilities Policy 14 185-195. Shleifer, A. (1985) A theory of yardstick competition, Rand Journal of Economics 16 (3), 314327. Soares, D. (2001) Privatization of sanitation and water distributi on in Brazil: a general overview of the current market and ou tlook for private investors, Journal of Project Finance 7 (1), 35-51. Spady, R. and Friedlaender, A. (1978) Hedoni c cost functions for the regulated trucking industry, Bell Journal 9 159-179. Stewart, M. (1993) Modeling water cost 1992-93, OFWAT Research Paper No. 4. Tangeras, T. (2002) Collusion-proof yardstick competition, Journal of Public Economics 83 231-254. Teeples, R. and Glyer, D. (1987) Cost of wate r delivery systems: specification and ownership effects, The Review of Economics and Statistics 69 (3), 399-408. Tupper, H. and Resende, M. (2004) Efficiency a nd regulatory issues in the Brazilian water and sewage sector: an empirical study, Utilities Policy 12 29-40. Vargas, M. and De Lima, R. ( 2004) Concessoes privadas de sa neamento no Brasil: bom negocio para quem?, Ambiente & Sociedade VII (2). Wilson, R. (1987) Game-Theoretic Approaches to Trading Processes, in T. Bewley (ed.) Advances in Economic Theory: Fifth World Congress Ch. 2, Cambridge University Press, 33.

PAGE 96

96 Zellner, A. (1962) An efficient method of estim ating seemingly unrelated regression equations and tests for aggregation bias, Journal of the American Statistical Association 57 348 368.

PAGE 97

97 BIOGRAPHICAL SKETCH Guillermo Sebastian Sabbioni Perez was born in 1975 in Argentina. In March 2000, he graduated with honors from the Univer sidad Catlica Argentina with a Licenciatura en Economa (Bachelor of Arts in economics). After gr aduation, he worked for one year as an Analyst in the Corporate Finance Department of the largest media group in Argentina. He then worked for two and a half years as a Category Manager in the Commerci al Department of a supermarket chain store, also in Argentina. While getting experience in the private sector, he also engaged in part-time teaching at the Universidad Catlica Argentina. He was awarded a Fulbright Scholarship at the end of 2002; this allowed him to return to school to pursue graduate education, this time in the United States. He started a doctorate in economics at the University of Florida in the fall of 2003. While pursuing his doctorate, he worked part-time as a Research Assistant for PURC, Public Utility Research Center. His work at PURC provided him with the opportunity to presen t his research at in ternational meetings. He was the instructor of undergraduate Game Th eory in the summer semester of 2006. He also presented one of his research papers at the Second Summer School on "Economic Analysis of Heterogeneity in Social Or ganizations" in CORE, Louvain-la-Neuve, Belgium, in June 2006. As recognition of his work, he was awarded th e Madelyn M. Lockhart International Travel Award and the Walter Lanzillotti Research Gran t by the Department of Economics, both in 2006. He graduated in August 2007, his dissertation titled Theoretical and Empirical Analyses of Incentives and Public Ownership