<%BANNER%>

The Impact of Historical and Regional Networks on Trade Volumes in the Western Hemisphere

Permanent Link: http://ufdc.ufl.edu/UFE0042322/00001

Material Information

Title: The Impact of Historical and Regional Networks on Trade Volumes in the Western Hemisphere A Gravity Model Analysis
Physical Description: 1 online resource (207 p.)
Language: english
Creator: Sandberg, Harry
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: colonialism, ftaa, gravity, neocolonialism, networks, regionalism, rtas, trade
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The purpose of this study is to analyze the effects of historical and regional networks on trade volumes in the Western Hemisphere by using the gravity model of international trade. These network effects are attributed to former colonial relationships and to the enactment of regional trading agreements. After reviewing and outlining the evolution of the gravity model, two empirical specifications of the model are fitted to three data-sets covering the bilateral trade transactions of the countries in the Americas. One data-set encompasses aggregate bilateral trade volumes, and the remaining two data-sets disaggregate bilateral trade volumes on the agricultural level and the manufactured goods level, respectively. The evidence suggests that the gravity model should be estimated by sector using disaggregated trade data. In particular, the determinants of agricultural trade volumes are found to be different from the determinants of trade in other product categories. The results indicate that historical and regional networks have significantly shaped the trade behavior of the countries in the Americas by influencing trade volumes. Imperial-based trade relationships are found between former colonies and their former metropolitan rulers. Such distortions are especially prevalent between the U.K. and her former colonies and tend to be stronger for trade in agricultural products. When it comes to regionalism, the evidence suggests an inverse relationship between economic size and regional dependency. Smaller economies, especially those located in the Caribbean basin, tend to cooperate more extensively than larger, more self-sufficient ones. The Caribbean Community and Common Market (CARICOM) and the Central American Common Market (CACM) have had significant effects on the trade behavior of their members. These effects are more prevalent for agricultural goods trade than for trade in manufactured products. Conversely, after controlling for economics, geography, and history, the postulated effects of the Mercado Comun del Sur (MERCOSUR) and the North American Free Trade Agreement (NAFTA) diminish empirically. In order to proceed with a potential Free Trade Area of the Americas (FTAA), the Western Hemisphere needs to consider both its recent geo-political history of enacting regional trade agreements and its former imperial history.
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 Harry Sandberg.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Seale, James L.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-06-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042322:00001

Permanent Link: http://ufdc.ufl.edu/UFE0042322/00001

Material Information

Title: The Impact of Historical and Regional Networks on Trade Volumes in the Western Hemisphere A Gravity Model Analysis
Physical Description: 1 online resource (207 p.)
Language: english
Creator: Sandberg, Harry
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: colonialism, ftaa, gravity, neocolonialism, networks, regionalism, rtas, trade
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The purpose of this study is to analyze the effects of historical and regional networks on trade volumes in the Western Hemisphere by using the gravity model of international trade. These network effects are attributed to former colonial relationships and to the enactment of regional trading agreements. After reviewing and outlining the evolution of the gravity model, two empirical specifications of the model are fitted to three data-sets covering the bilateral trade transactions of the countries in the Americas. One data-set encompasses aggregate bilateral trade volumes, and the remaining two data-sets disaggregate bilateral trade volumes on the agricultural level and the manufactured goods level, respectively. The evidence suggests that the gravity model should be estimated by sector using disaggregated trade data. In particular, the determinants of agricultural trade volumes are found to be different from the determinants of trade in other product categories. The results indicate that historical and regional networks have significantly shaped the trade behavior of the countries in the Americas by influencing trade volumes. Imperial-based trade relationships are found between former colonies and their former metropolitan rulers. Such distortions are especially prevalent between the U.K. and her former colonies and tend to be stronger for trade in agricultural products. When it comes to regionalism, the evidence suggests an inverse relationship between economic size and regional dependency. Smaller economies, especially those located in the Caribbean basin, tend to cooperate more extensively than larger, more self-sufficient ones. The Caribbean Community and Common Market (CARICOM) and the Central American Common Market (CACM) have had significant effects on the trade behavior of their members. These effects are more prevalent for agricultural goods trade than for trade in manufactured products. Conversely, after controlling for economics, geography, and history, the postulated effects of the Mercado Comun del Sur (MERCOSUR) and the North American Free Trade Agreement (NAFTA) diminish empirically. In order to proceed with a potential Free Trade Area of the Americas (FTAA), the Western Hemisphere needs to consider both its recent geo-political history of enacting regional trade agreements and its former imperial history.
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 Harry Sandberg.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Seale, James L.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-06-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042322:00001


This item has the following downloads:


Full Text

PAGE 1

1 THE IMPACT OF HISTORICAL AND REGIONAL NETWORKS ON TRADE VOLUMES IN THE WESTERN HEMISPHERE: A GRAVITY MODEL ANALYSIS By HARRY MIKAEL SANDBERG 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 2010

PAGE 2

2 2010 Harry Mikael Sandberg

PAGE 3

3 To my parents

PAGE 4

4 ACKNOWLEDGMENTS I would like to extend my deepest gratitude and respect to my mentor and advisor, Dr. James L. Seale, Jr of the Food and Resource Economics department at the University of Florida. His supervision and encouragement has been instrumental in the completion of this project. I would like to thank the remaining members of my doctoral committee, namely Dr. Timothy Taylor, Dr. Jeffrey Burkhardt, and Dr. Elias Dinopoulos all members of the faculty at the University of Florida, for th eir guidance and support I feel fortunate to have collaborated with such a talented and accomplished group of scholars. I would also like to acknowl edge Dr. Thomas Spreen and Ms. Jessica Herman for their encouragement over the years. Over the past decade, I have been both a graduate student and a member of the faculty at the University of Florida. During my years as a graduate student I encountered an eclectic group of fellow students who have become li felong friends and colleagues. In particular, I would like to mention Dr. Marisa Zansler, Dr. Grigorios Livanis, Dr. Lurleen Walters, Dr. Jione Jung Mr. Garfield Love, and Mr. Karl Cerullo. I would also like to thank my colleagues in the Food and Res ource Economics d epartment and my students over the years for making my time at the University of Florida filled with re warding and memorable experiences I am indebted to the University of Florida for financial support during my graduate studies via a fellowship. Without this gener ous award, my endeavors at the doctoral level would not have been possible.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ...............................................................................................................4 LIST OF TABLES ...........................................................................................................................7 LIST OF FIGURES .........................................................................................................................9 ABSTRACT ...................................................................................................................................10 CHAPTER 1 INTRODUCTION AND PRELIMINARIES .........................................................................12 Introduction .............................................................................................................................12 Networks and Trade Volumes ................................................................................................13 Context of Study .....................................................................................................................18 Aim and Scope ........................................................................................................................20 2 THE GRAVITY MODEL: LITERATURE AND METHODOLOGY ..................................27 Introduction to the Model .......................................................................................................27 The Contemporary Gravity Model of International Trade .....................................................28 Early Gravity Analysis ...........................................................................................................33 Modern Gravity Models .........................................................................................................38 A Gravity Model: The Dutch School ..............................................................................38 A Gravity Model: The Finnish School ............................................................................41 A Gravi ty Model: The Probabilistic Approach ...............................................................42 In Search of a Theory .............................................................................................................44 Pre 1979 ..........................................................................................................................44 Anderson (1979) ..............................................................................................................46 Bergstrand (1985) ............................................................................................................52 Bergstrand (1989) ............................................................................................................58 Further Forays into Economic Theory ............................................................................63 Transaction Costs and Remoteness .................................................................................68 Helpman (1987) ......................................................................................................................74 Remarks ..................................................................................................................................79 3 CONCEPTUAL FRAMEWORK AND DATA .....................................................................83 Conceptual Framework ...........................................................................................................83 Data .........................................................................................................................................89 Data Sources and Sample ................................................................................................89 Problem of Zero Observations ......................................................................................92

PAGE 6

6 4 EMPIRICAL ESTIMATION .................................................................................................97 Introductory Remarks .............................................................................................................97 Aggregation Issues ..................................................................................................................98 Testing the Similarity of Variances .................................................................................99 Testing the Similarity of Estimated Parameters ............................................................103 Analysis of Results: Preliminary Considerations .................................................................106 Performance of the Gravity Variables ..................................................................................107 Gravity Variab les in the MH Specification ...................................................................107 Gravity Variables in the MTPLAB Specification .........................................................111 Network Effects ....................................................................................................................116 Interpretation of Binary Variables .................................................................................116 Analysis of Results: Historical Networks ......................................................................117 Analys is of Results: Regional Networks .......................................................................120 Empirical Remarks ...............................................................................................................123 5 CONCLUDING REMARKS ................................................................................................140 APPENDIX A SUPPLEMENTAL TABLES ...............................................................................................146 B GRAPHICAL PRESENTATION OF HISTORICAL EFFECTS ........................................153 C GRAPHICAL PRESENTATI ON OF REGIONAL EFFECTS ...........................................157 D MEXICO CENTERED ESTIMATIONS .............................................................................161 E SPECIFICATION TESTING ...............................................................................................180 Preliminaries .........................................................................................................................180 The J Test .............................................................................................................................182 The Cox Test .........................................................................................................................184 The Likelihood Dominance Criterion ...................................................................................188 LIST OF REFERENCES .............................................................................................................195 BIOGRAPHICAL SKETCH .......................................................................................................207

PAGE 7

7 LIST OF TABLES Table page 11 Current regional trading agreements in the Western Hemisphere .....................................22 12 Intra regional export shares (as % of total bloc exports) ...................................................23 13 Colonial history of the Americas .......................................................................................24 14 Percentage share of total exports dest ined to the former metropolitan ruler in Western Europe ..................................................................................................................25 31 Dummy variables in vector w ............................................................................................95 32 Countries included in the stud y ..........................................................................................96 41 Regression index ..............................................................................................................125 42 MH specification: aggregate trade (UNCTAD total) .......................................................126 43 MH specification: agricultural trade (UNCTAD categories 1 & 2) ................................128 44 MH specification: manufactured goods trade (UNCTAD category 5) ............................130 45 MTPLAB specification: aggregate trade (UNCTAD total) .............................................132 46 MTPLAB specification: agricultural trade (UNCTAD categories 1 & 2) .......................134 47 MTPLAB specification: manufactured goods trade (UNCTAD category 5) ..................136 48 Effects of binary variables, historical networks ...............................................................138 49 Effects of binary variables, regional networks ................................................................139 A 1 F test for similarity of variances, MH specification ........................................................146 A 2 F test for similarity of variances, MTPLAB specification ..............................................147 A 3 F test for similarity of parameters, MH specification ......................................................148 A 4 F test for similarity of parameters, MTPLAB specification ............................................149 A 5 Estimated standard errors of parameters for the population variables, aggregate trade ..150 A 6 Estimated standard errors of parameters for the population variables, agricultural trade ..................................................................................................................................151

PAGE 8

8 A 7 Estimated standard erro rs of parameters for the population variables, manufactured goods trade .......................................................................................................................152 D 1 Regression index for Mexicocentered estimations .........................................................167 D 2 MH specification: aggregate trade (UNCTAD total), Mexico centered data ..................168 D 3 MH specification: agricultural trade (UNCTAD categories 1 & 2), Mexico centered data ...................................................................................................................................170 D 4 MH specification: manufactured goods trade (UNCTAD category 5), Mexico centered data ....................................................................................................................172 D 5 MTPLAB specification: aggregate trade (UNCTAD total), Mexico centered data ........174 D 6 MTPLAB specification: agricultural trade (UNCTAD categories 1 & 2), Mexico centered data ....................................................................................................................176 D 7 MTPLAB specification: manufactured goods trade (UNCTAD category 5), Mexico centered data ....................................................................................................................178 E 1 Results from the J test ......................................................................................................192 E 2 Results from the Cox test .................................................................................................193 E 3 Results from the Likelihood Dominance Criterion ..........................................................194

PAGE 9

9 LIST OF FIGURES Figure page 11 A regional and historical network view of the Americas ..................................................26 21 A trade matrix ....................................................................................................................82 B 1 Es timated historical effects: MH specification, aggregate trade .....................................154 B 2 Estimated historical effects: MH specification, agricultural trade ...................................154 B 3 Estimated historical effects: MH specification, manufactured goods trade ....................155 B 4 Estimated historical effects: MTPLAB specification, aggregate trade ............................155 B 5 Estimated historical effects: MTPLAB specification, agricultural trade .........................156 B 6 Estimated historical effects: MTPLAB spec ification, manufactured go ods trade ...........156 C 1 Estimated regional effects: MH specification, aggregate trade .......................................158 C 2 Estimated regional effects: MH specification, agricultural trade ....................................158 C 3 Estimated regional effects: MH specification, manufactured goods trade ......................159 C 4 Estimated region al effects: MTPLAB specification, aggregate trade .............................159 C 5: Estimated reigonal effects: MTPLAB specification, agricultural trade ...........................160 C 6 Estimated regional effects: MTPLAB specification, manufactured goods trade ............160

PAGE 10

10 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 THE IMPACT OF HISTORICAL AND REGIONAL NETWORKS ON TRADE VOLUMES IN THE WESTERN HEMISPHERE: A GRAVITY MODEL ANALYSIS By Harry Mikael Sandberg December 2010 Chair: James L. Seale, Jr. Major : Food a nd Resource Economics The purpose of this study is to analyze the effects of historical and regional networks on trade volumes in the Western Hemisphere by using the gravity model of international trade. The se network effects are attribu ted to former co lonial relationships and to the enactment of regional trading agreements. After reviewing and outlining the evolution of the gravity model, two empirical specifications of the model are fitted to three data set s covering the bilateral trade transactions o f the countries in the Americas O ne dataset encompasses aggregate bilateral trade volumes and the remaining two dataset s disaggregate bilateral trade volumes on the agricultural level and the manufactured goods level respectively. The evidence sugge sts that the gravity model should be estimated by sector using disaggregated trade data. In particular, the determinants of agr icultural trade volumes are found to be different from the determinants of trade in other product categories The results indi cate that historical and regional networks have significantly shaped the trade behavior of the countries in the Amer icas by influencing trade volumes. I mperial based trade relationships are found between former colonies and their former metropolitan ruler s Such

PAGE 11

11 distortions are especially prevalent between the U.K. and her former colonies and tend to be stronger for trade in agricultural products When it comes to regional ism t he evidence suggests an inverse relationship between economic size and regi onal dependency. S maller economies, especially those located in the Caribbean basin, tend to cooperate more extensively than larger more self sufficient ones. The Caribbean Community and Common Market ( CARICOM ) and the Central American Common Market ( CA CM ) have had significant effects on the trade behavior of their member s These effects are more pr evalent for agricultural goods trade than for trade in manufactur ed products Conversely, after controlling for economics, geography, and history, the postulated effects of the Mercado Comun del Sur ( MERCOSUR ) and the North American Free Trade Agreement ( NAFTA) diminish empirically. I n order to proceed with a potential Free Trade A rea of the Americas (FTAA), the Western Hemisphere needs to consider both its recent geo political history of enacting regional trade agreements and its former imperial history.

PAGE 12

12 CHAPTER 1 INTRODUCTION AND PRE LIMINARIES Introduction Ever since the success of economic integration in Western Europe, other parts of the world have l ooked at economic integration and free trade as ve hicles for economic prosperity, both rhetorically and in practice. While some of these regional trade agreements have materialized into toothless paper tigers failing to increase intraregional commerce an d moving their member countries toward a more liberal ized trade environment (e.g ., the ones found on the Africa n continent ), others have indeed been successful (e.g ., the ones in Western Europe North America, and South Ea st Asia). Despite their inherent second best nature, regional trading agreements can offer a path in the direction of free trade by using the integration process as a stepping stone towards a more pareto optimal environment.1 The Free Trade Area of the Americas (FTAA) is currently bei ng negotiated among 34 sovereign countries in North America, Latin America, and the Caribbean, with the ambition of a free trade area spanning the entire Western Hemisphere. These FTAA economies include Antigua and Barbuda, Argentina, the Bahamas, Barbados, Belize, Bolivia, Brazil, Canada, Chile, Columbia, Costa Rica, Dominica, Dominican Republic, Ecuador, El Salvador, Grenada, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, St Kitts and Nevis, St Lucia, St Vincent a nd the Grenadines, Suriname, Trinidad and Tobago, United States, Uruguay, and Venezuela (Free Trade Area of the Americas Administrative Secretariat 2002). There are, however, a couple of notable exclusions such as Cuba the current dependencies of the U nited Kingdom (U.K .) France, and the Netherlands, and current U.S. 1 It should be n oted, however, that regional trading agreements are sometimes pursued not only to liberalize trade, but also to enhance regional cooperation and to promote peace and stability. The latter was certainly part of the motivation behind the European Union (Mol le 2006).

PAGE 13

13 territories. The potential FTAA is rather unique due to its pancontinental nature and would represent the largest integrated geographic area in the world. Furthermore, the Western Hemis phere is comprised of a rather heterogeneous group of countries, both in terms of economics and sheer physical size. The FTAA proposal was put forth by then United States Vice President Albert Gore during a speech in Mexico City in December 1993 and it w as subsequently announced during the Summit of the Americas in Miami in December 1994 that official negotiations were to commence (Wrobel 1998). Negotiations were scheduled to be complete by the end of 2005. Needless to say, an FTAA is yet to emerge. Th e currently stalled negations and the FTAAs eventual implementation continue to be sensitive issues, both within the United States and throughout the region. Given its contentious nature, it is imperative that the discussion is based on facts and not pro paganda. A good starting point for analyzing the FTAA is to develop an understanding of the trade behavior of the countries in the Western Hemisphere. Similar sentiments were, at least partially, echoed by Zahniser et al. (2002). Networks and Trade Volum es It can be postulated that the trade behavior in the Western Hemisphe re is influenced by relationships resulting from different types of linkages based on either regionalism or histor y Eichengreen and Irwin (1998) suggested that current global trade pa tterns are highly influenced by hysteresis, where historical events (and relationships) permanently bias or distort trade patterns. Such historical events include past colonial legacies, trade negations, and the implementation of trade agreements. T he re are currently numerous regional trading agreements in the Western Hemisphere. S uch agreements clearly establish linkages and would in all likelihood inf luence the trade behavior of member states The major regional trade agreements are presented with their

PAGE 14

14 respective members and years of enactment in Table 1 1. Almost every economy in the hemisphere is participating in some form of a n integrated area. The oldest regional trading agreement in the Western Hemisphere is the Central American Common Mark et (CACM), enacted in 1959. Other early agreements include the Andean Pact sometimes referred to as the Andean Community or the Andean Group, and the Caribbean Community and Common Mar ket (CARICOM). The Andean Pact was formally enacted in 1969 and CARIC OM was established in 1973. More recent integration effort s include Mercado Comun del Sur (MERCOSUR), formed in 1991 among the countries in the Southern cone of Latin America, the North American Free Trade Agreement (NAFTA), enacted in 1994 among Canada, the United States, and Mexico, and the Group of Three, enacted in 1995 among Mexico, Colombia, and Venezuela. As these agreements g enerally follow geographic boundaries they divide the hemisphere into different subregions. It should be noted that the se agreements are often over lapping. In particular, t he Latin American Free Trade Agreement (LAFTA), enacted in 1960, which in 1980 became the Latin American Integration Association (LAIA) totally encompasses the Andean Pact, MERCOSUR and the Group of Three. There is also over lap between NAFTA and the Group of Three with Mexico being a member of both agreements. The remaining two members of the Group of Three, Colombia and Venezuela, are also participating in the Andean Pact. Frankel, Stein, and Wei (1997) pointed out that LAIA has been rather ineffective in increasing intra regional trade and cooperation beyond what has been accomplished via MERCOSUR and the Andean Pact, even though they were formally enacted later. The only two agreements without overlapping memberships are CACM and CARICOM These two agreements consist of

PAGE 15

15 the smaller economies in the hemisphere, all located in the central region and in the Caribbean Basin. While economic integration in Western Europe has followed a structured, formal process with clearly stated policy goals throughout its evolution, agreements in the Western Hemispher e ha ve developed via a less stylized often ad hoc manner. T he official names of these agreements may be misleading, as they are not necessarily indicative of the depth or the nature of current integration. For instance, CACM and MERCOSUR are not common markets in the formal sense, as their names imply, with free movement of goods, services, labor, capital, a nd other factors of production in the E uropean tradition. R ather they are preferential trading agreements with liberalization pursued in selected industries or s ectors. T he Group of Three, while not successful in increasing intraregional trade between members in a meaningful way, has pursue d harmonization of institutions, intellectual propert y laws, and trade in services. The broader pragmatic nature of this agreement has caused other nations in the region, such as Ecuador and Chile, to express an interest in the grouping, at least rhetori cally (Frankel, Stein, and Wei 1997). Recent data reveal preliminary s upport for the existence of regional trade relati onships. Table 1 2 presents intra regional trade shares for some of the regi onal groupings for the past decades The intra regional tr ade shares in Table 1 2 are defined as the percentage of total exports of a particular grouping tha t is destined for other member states. On the one hand, CARICOM MERCOSUR and NAFTA have all experienced increases in intraregional trade over the past de cade. However, intraregional trade within the Andean Pact increased initially during the nineties, but seem s to have reached its pe ak in 1998 with a sharp decline in subsequent years. The members of CACM exhibit a similar trend, but the peak was reached earlier Intra regional trade among the Group of Three has remained low both before and after

PAGE 16

16 the formal enactment in 1995. As a comparison, intra regional trade in the European Union (EU) has remained high over the past two decade s Intra regional co mmerce within LAIA is relatively stable, but its explicit influence may be difficult to discern due to the overlapping memberships with the Andean Pact, Group of Three, and MERCOSU R Other relationships are the consequences of historical legacies. Almos t all FTAA economies are former European colonies. As a lingering effect, there are often neo colonial trade ties present with large volumes of trade between the former dependency and the former metropolitan ruler in Western Europe. Rauch (1999) postulat ed that colonial ties often result in lasting trade relationships and Grier (1999) pointed out that the establishment of such trade linkages was one of the major motivations behind imperialism. Table 1 3 presents the colonial history of the economies in the Western Hemisphere. A useful way of approaching neoimperial trade is to observe the percentage of a former dependencys total exports that is s hipped to the former ruler. These export shares are presented in Table 1 4. The corresponding trade share for the Caribbean economies, most of which are former British dependencies, range between 8 percent to upwards of 70 percent (!) of total exports over the past decades. Interestingly, no analogous relationship exists for most former Spanish and Portuguese colonies as imperial export shares linger in the lower single digits Grier (1999) suggested that the U.K. has maintained a closer relationship with its former colonies after their independence vis vis France, Spai n, or Portugal As Brysk, Parsons, a nd Sandholtz (2002) pointed out, Great Britains post colonial relationships have focused on trade linkages and offering preferential market access to products produced in her former dependencies These policy distortions, particularly prevalent for agric ultural commodities, would naturally bias trade volumes; e.g., they would result in higher trade levels between the

PAGE 17

17 U.K. and her former colonies post independence These higher trade volumes are not necessarily in response to comparative advantages but ar e rather the outcome s of deliberately distortive policy. As a contrast, Spains relationships with its former dependencies emphasize the provision of foreign aid and Frances post colonial influence has mostly been in the military arena (Brysk, Parsons, a nd Sandholtz 2002) The impact of historical ties was also discussed b y Anderson and Norheim (1993), Eichengreen and Irwin (1998), Frankel, Stein, and Wei (1997), Hamilton and Winters (1992), Kleiman (1976, 1977), Linnemann (1966), Livingstone (1976) San dberg and Martin (2001), and Sandberg, Seale, and Taylor (2006) Lasting economic relationships can be characterized as networks. Podolny and Page (1998) defined a network as any collection of actors that pursue repeated, enduring exchange relations w ith one another (p. 59). One can henceforth view a group of countries engaging in large volumes of trade on a regular basis as being representative of a network. Such networks may have institutional arrangements, familiarity with customs and commercial procedures, and regulatory environments as to facilitate trade between participants. Rauch (1999, 2001) and Combes, Lafourcade, and Mayer (2005) discussed in great detail th e importance of networks for trade. Eichergreen and Irwin (1998) and DeGroot et al. (2004) postulated similar frameworks without explicitly using the term networks For instance, DeGroot et al. (2004) referred to the idea of networ ks as institutions Figure 1 1 illustrates the idea that hemispheric trade is being influenced by existing networks. The outlined ovals in Figure 11 are not drawn to scale and their sizes are not intended to infer anything about the size of the agreements. Rather, the shapes are used for illustrative purposes to give the reader an idea of the overla pping nature of current trade agreements. Frankel, Stein, and Wei (1997) provided a similar framework concerning

PAGE 18

18 regionalism in the Western Hemisphere, which they referred to as a web of agreements (p. 7). It should be noted, however, that these networ ks may in fact facilitate the integration process rather than hindering it. As suggested by Smith (1999), such relationships might provide the structure ne cessary to accomplish wider integration. While intuitive arguments can made in support of the pres ence of networks in t he Americas one has to be cautious of venturing down the path of attributing existing trade behavior to history or regionalism. Observed trade relationships may simply be part of the natural trade behavior that would prevail wit hou t aforementioned linkages and may not represent any true distortions. Rather they may naturally occur due to the economic and geographic cir cumstances of trading partners; countries within a particular region may exhibit higher trade levels internally vi s vis external nations due to cultural similarities or geographic proximity. Similarly, the presumed neo colonial trade distortion might just be in response to the relative market size of the Western Europe an countries Context of Study It becomes in teresting to empirically investigate the extent to which these networks influence trade behavior. However, g iven the timely nature of the topic and the controversy surrounding the FTAA, empirical assessments of the trade patterns in the Americas are remar kably scant. On the other hand, t here is a plethora of qualitative discussions regarding the practical implications of the FTAA and the political realities surrounding its implementation. These appraisals range from cautiously optimistic (or at the very least neutral ) to adversely pessimistic: see, for example, Berna l (1994), Bouzas (2005), Eichengreen (2004), Ker r (2002), Lee (1995), Nicholls et al. (2000), Read (2004), Rivas Campo, Antonio, and Juk (2003), von Roozendall (2006), Ruiz (2007), Salazar Xir inachs and Tavares de Araujo (1999), Schott and Hufbauer (1999), Smith (1999), and Wrobel (1998). What emerges from this qualitative

PAGE 19

19 literature is the idea that the potential success of the FTAA, both in terms of economics and its political survival, rests with the agreements ability to generate and promote intra regional trade and cooperation. If this cannot be convincingly argued, the FTAA runs the risk of losing speed mid stream and may suffer the fate of becoming yet another ineffective paper agreement in the economic integration arena. Some empirical studies regarding the FTAA have been conducted via the utilization of general equilibr ium models with mixed to ambiguous results regardin g the welfare implications of the proposed agreement : see, for i nstance Brown, Kiyota, and Stern (2005), Das and Andriamananjara (2006), Diao and Somwaru (2000, 2001), Herter, Hummels, Ivanic, and Roman (2007), Kehoe and Kehoe (1994), Kouparitsas (1997), and Rutherford and Martinez (2000). These types of models are u seful for assessing the overall welfare effects of trade. However, the determinants of trade volumes are not explicitly analyzed. Moreover, given the diverse nature of the economies in the Americas, both in terms of sheer economic size and the level of development, the findings of general equilibrium models might be difficult to generalize to such groupi ng s (Sandberg, Seale, and Taylor 2006). T he gravity model of international trade is a more appropriate methodology since it allows for a targeted analys is of the determination of trade behavior T here are indeed some gravity based investigations using data for the Western Hemisp here in the literature The evidence provided via the gravity models estimated by Cheng and Tsai (2008), Frankel, Stein, and We i (1995), Frankel, Stein, and Wei (1997), Frankel, Stein, and Wei (1998), Garman, Peterson, and Gilliard (1998), Grant and Lambert (2008), Gould (1998), Sandberg Taylor, and Seale (2006), Soloaga and Winters (2001) Thoumi (1989a, 1989b) Vollrath, Hallah an, and Gehl har (2006), and Zahniser et al. (2002) suggest ed that regional trade agreements and (or) historical linkages

PAGE 20

20 in the Americas have had significant effects on trade volumes. While these studies certainly provide justification for the questions posed, their data were not inclusive of the entire hemisphere. Rather, these studies were regionally focused or included only the larger (or smaller) economies. Hence, there is a clear gap in the empirical literature which the current study is intended t o help fill. Aim and Scope The aim of this study is to empirically investigate the determinants of trade volumes in the Western Hemisphere by using the gravity model of international trade. Particularly, the focus is on the extent to which economic integration and colonial history influence trade behavior. The gravity model is applied to three data set s encompassing the ten year period 19922001. To capture any sectoral differences, the estimation is implemented using both aggregate trade data (total bilateral trade) and disaggregated trade data ( agricultural bilateral trade versus manufactured goods bilateral trade) Differences in outcomes can be used to assess implications across sectors. Does it matter whether one is utilizing aggregate trade data or disaggregated trade data? S hould the gravity model be estimated economy wide using aggregate data or should it be estimated by sector? These are issues that will be addressed. This analysis proceed s as follows. The next chapter presents an in de pth discussion of the gravity model Rigorous attention will be given its historical development and evolution. Se veral theoretical foundations and derivations of the model are reviewed as to provide the proper context within the field of international e conomics. Chapter 3 presents the conceptual framework with an outline of the model spec ifications to be estimated. This chapter also discusses the data and related empirical issues. Chapter 4 reports the empirical results in detail and discusses repercu ssions. Chapter 5 concludes with an epilogue and remarks A couple of appendices contain supplemental tables (Appendix A) and graphical presentations (Appendix B

PAGE 21

21 and Appendix C). Appendix D provides Mexicocentered estimations of the two models, as rati onalized in Chapter 4. Appendix E provides empirical specification testing of the model s estimated in Chapter 4 Where appropriate, this appended material is referenced to in the text.

PAGE 22

22 Table 1 1. Current regional trading agreements in the Western Hemisphere Source s : Frankel, Stein, and Wei (1997), Caribbean Community Secretariat (2001) a) The name Latin American Free Trade Association (LAFTA) applies before 1980; the name Latin American Integration Association (LAIA) applies after 1980. Name of agreement Year enacted Current members Andean Community (or Andean Pact) 1969 Bolivia, Colombia, Ecuador, Peru, and Venezuela Caribbean Community and Common Market (CARICOM) 1973 Antigua and Barbuda, the Bahamas, Barbados, Belize, Dominica, Grenada, Guyana, Haiti, Jamaica, Montserrat, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Suriname, and Trinidad and Tobago Central American Common Market (CACM) 1959 Costa Rica, El Salvador, Guatemala, Honduras, and Nicaragua Group of Three 1995 Colombia, Mexico, and Venezuela Latin American Free Trade Association (LAFTA) /Latin American Integration Association (LAIA)a 1960 Argentina Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Paraguay, Peru, Uruguay, and Venezuel a Mercado Comun del Sur (MERCOSUR) 1991 Argentina, Brazil, Paraguay, and Uruguay North American Free Trade Agreement (NAFTA ) 1994 Canada, Mexico, the United States

PAGE 23

23 Table 1 2. Intra regional export shares (as % of total bloc exports) 1980 1990 1993 1994 1995 199 6 1997 1998 1999 2000 2001 2002 2003 2004 Average a Andean Pact 3.8 3.8 9.8 10.5 11.8 10.4 10.1 11.9 9.3 8.5 10.9 9.5 7.8 8.7 9.1 CACM 24.4 15.3 16.9 16.7 14.1 15.7 13 14.5 11.6 12.4 15 11.1 13 20 15.3 CARICOM 4.2 7.8 8.8 3.7 3.8 12.9 13.8 17.1 15.3 15 1 3.3 12.5 12.4 12.5 10.9 Group of 3 1.8 2.0 NA NA 3.2 NA 2.8 2.7 1.8 1.7 2.1 1.8 1.6 2.3 2.2 LAFTA/ LAIA 13.7 10.6 16.3 16.4 16.6 16.5 17.2 16.7 13 12.9 13 11.1 11.4 12.6 14.1 MERCOSUR 11.6 8.9 18.5 19.2 20.2 22.8 25.4 25.1 20.5 20.8 17.3 11.6 11.9 12.6 1 7.6 NAFTA 33.6 41.4 45.8 47.9 46.2 47.5 49.1 51.7 54.6 55.7 55.5 56.7 56.1 55.9 49.9 For comparison purposes : EU 61 66 61.7 62.1 62.4 61.5 55.4 55.2 62.6 62.1 61.3 60.6 61.1 60.7 61 Source: World Bank (1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006). Data refers to merchandise trade only, as reported in World Development Indictors 1998 through 2006. Please refer to Table 1 1 for member ship information. a) Defined as the arrhythmic average of the colonial trade shares for years in which data were available.

PAGE 24

24 Table 1 3. Colonial history of the Americas Country Colonial ruler Year of independence Antigua and Barbuda U.K. 1981 Argentina Spain 1816 The Bahamas U.K. 1973 Barbados U.K. 1966 Belize U.K. 1981 Bolivi a Spain 1825 Brazil Portugal 1822 Chile Spain 1810 Colombia Spain 1810 Costa Rica Spain 1821 Dominica U.K. 1978 Dominican Republic Spain 1865 Ecuador Spain 1822 El Salvador Spain 1821 Grenada U.K. 1974 Guatemala Spain 1821 Guyana U.K. 1966 Hait i France 1804 Honduras Spain 1821 India U.K. 1947 Jamaica U.K. 1962 Mexico Spain 1810 Nicaragua Spain 1821 Panama Spain 1821 Paraguay Spain 1811 Peru Spain 1821 St Kitts and Nevis U.K. 1983 St Lucia U.K. 1979 Singapore U.K. 1965 St Vincent and The Grenadines U.K. 1979 Suriname The Netherlands 1975 Trinidad and Tobago U.K. 1962 Uruguay Spain 1825 Venezuela Spain 1811 Source: Central Intelligence Agency (2007) https://www.cia.gov/library/publications/the world factbook/index. Note: the Uni ted States and Canada are not treated as former colonies.

PAGE 25

25 Table 1 4. Percentage share of total exports destined to the former metropolitan ruler in Western Europe Country 1980 1985 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Average a Argentina 2 .34 2.54 2.74 4.02 4.09 3.85 3.74 3.41 3.09 2.36 3.19 4.12 3.29 Barbados 8.99 8.87 18.34 10.49 25.99 20.48 25.53 21.82 21.35 22.30 20.21 16.78 18.43 Belize NA NA NA NA 28.58 0.00 36.19 46.10 45.65 33.19 34.45 27.77 35.99 Bolivia 0.05 0.38 0.34 0.18 0.31 0.82 0.13 0.21 0.26 0.73 0.52 0.18 0.34 Brazil NA 0.53 0.69 0.71 0.72 0.70 0.69 0.89 0.68 0.77 0.86 0.69 0.72 Chile NA 2.00 3.20 3.88 3.64 2.61 1.94 1.91 1.78 2.01 1.89 2.00 2.44 Colombia 3.82 2.21 1.68 2.09 2.11 1.69 1.95 1.83 1.48 1.29 1.39 1.29 1.90 Costa Rica NA NA 0.49 0.67 0.79 0.59 0.96 0.96 1.00 1.70 2.10 1.07 1.03 Dominica NA 52.02 51.55 56.11 NA 51.74 48.60 39.47 36.33 32.84 NA 28.94 44.18 Dominican Republic NA NA NA NA 0.70 0.71 0.47 0.53 0.38 0.26 NA NA 0.51 Ecuador 0.59 0.14 2.49 3.97 3 .72 2.72 2.98 3.44 2.72 2.45 3.33 2.75 2.61 El Salvador NA NA 0.43 0.45 0.28 0.23 0.68 0.91 0.60 0.64 0.14 0.25 0.46 Grenada NA 26.49 23.95 22.73 NA 14.53 14.90 11.54 8.74 5.72 2.06 4.03 13.47 Guatemala NA NA 0.21 0.18 0.50 0.37 0.43 0.62 0.37 0.49 0.44 0.39 0.40 Haiti NA NA 2.71 2.05 6.36 1.98 3.06 4.61 5.08 4.66 NA NA 3.81 Honduras NA NA 1.75 1.80 1.47 2.06 3.60 3.86 2.95 2.39 1.62 1.83 2.33 India 6.09 4.81 6.55 6.36 6.53 6.20 6.42 6.31 6.12 6.02 5.57 NA 6.09 Jamaica 19.75 17.79 15.25 17.58 17.08 1 3.83 13.46 13.57 12.95 13.76 NA NA 15.50 Mexico NA NA 5.47 4.21 2.67 1.69 1.43 0.99 1.00 0.86 0.61 0.60 1.95 Nicaragua NA NA 3.60 0.97 2.07 0.91 3.87 7.76 11.10 9.01 4.24 2.61 3.85 Paraguay NA 3.17 1.51 1.79 1.14 0.72 0.68 1.03 0.69 0.43 0.48 0.64 1.12 Peru 0.72 0.38 1.22 1.09 1.70 1.72 1.74 2.24 2.40 2.33 2.66 2.99 4.24 Saint Kitts and Nevis NA NA NA NA NA 50.29 57.18 45.14 42.33 37.26 NA NA 46.44 Saint Lucia NA 64.10 55.78 53.97 57.14 55.17 58.83 61.10 69.70 63.11 63.53 63.45 60.53 Saint Vincent an d the Grenadines NA NA NA NA NA 42.49 33.69 42.43 41.99 33.62 44.05 44.17 40.35 Suriname NA NA 28.86 25.94 28.86 NA 27.72 27.80 18.07 20.32 21.27 NA 24.86 Trinidad and Tobago 1.56 3.84 2.83 2.23 1.71 2.02 1.87 2.50 1.89 2.37 1.98 2.32 2.26 Uruguay NA 0. 82 0.81 1.48 2.02 1.18 1.74 2.53 1.58 2.15 2.01 3.11 1.77 Venezuela NA 1.55 0.80 0.89 0.59 0.38 0.62 0.41 0.86 0.82 1.42 0.97 0.85 Source: United Nations Common Database, United Nations (2004) The trade shares are defined as percentage share of total exports heading to the former metropolitan ruler in Western Europe. NA: s ome countries are missing due to poor data availability. a) Defined as the arrhythmic average of the colonial trade shares for years in which data were available.

PAGE 26

26 Figure 1 1. A regional and historical network view of the Americas Neo imperial networks European Union FTAA NAFTA CARICOM CACM Group Of Three LAFTA / LAIA MERCOSUR Andean Pact FTAA Neo imperial networks European Union

PAGE 27

27 CHAPTER 2 THE GRAVITY MODEL: LITERATURE AND METHODOLOGY Introduction to the Model Even though the gravity model is commonly associated with international trade analysis, similar models have been used extensively by other social scientists in a variety of disciplines Over the past century the model has been proven useful in the investigations of social interactions between regions or countries, such as traffic flows, travel, tourism, telephone calls, the movement of people, and, of course, trade. The models origin can be traced to the use of so cial physics during the late nineteenth century and to the early writings on the economics of location and transport ation costs. Social physics was an intellectual movement where social scientists used analogs to the laws of natural science to ex plain social phenomena (Stewart 1950). A common theme for all gravity models is a framework similar to Newtonian gravity with two bodies of different mass at a specified distance being subject to a force acting upon them. This force, F was originally formulated by Sir Isaac Newton (1686) and can be expressed as 2 ijGMM F d (2 1) where Mi is the mass of body i and Mj is the mass of body j d is the str aight line distance between the two bodies and G is a gravitational constant (Stewart 1950). For the social scientist, geographical units (such as cities, regions, or countries) are the analogs of the physical bodies in Equation 21. The mass, or size, of these bodies, Mi and Mj respectively, are commonly measured by their incomes or populations. The force in Equation 21, F is usually a flow of a certain nature from one geographic unit to another, and d is simply the

PAGE 28

28 geographic straight line distance between them exerting a resistance to the force or the social interaction in question. In the realm of international economics, the bilateral trade of goods from one economic unit to another is representat ive of the gravitational force. The size of that flow would be determined by the economic mass, or trading capacity, of the two countries and the distance between them would act as a resistance to that flow. Todays cont emporary gravity model is, at least to a certain extent, based on economic theory and encompasses many factors. The Contemporary Gravity Model of International Trade The basic idea behind the gravity model of international trade is that bilateral trade volumes from one country to another can be explained by : a) factors that capture the potential of a countr y to export goods and services; b) factors that capture the propensity of a country to import goods and services; and by c) other forces that either attract or inhibit bilateral trade (Pyhnen 1963a, 1963b; Pulliainen 1963). To illustrate the models basic structure, it is helpful to consider a trade matrix Figure 2 1, as adopted from Savage and Deutsch (1960), Pyhnen (1963a), and Taplin (1967). Assume that the world consists of n different economic units (i.e., countries). The trade matrix subsequently illustrates world trade in a given time period. Each Xij entry represents the value of exports from country i to country j or equivalently the value of country j s imports from country i Thus, each flow of exports is also a flow of imports. The diagonal terms, Xii, represent domestic trade within a given country i and would not be treated as international trade. As such, one has (n2 n) exp ort (or import) flows per time period. If each row is summed over all columns, one obtains the value of country i s total exports to the rest of the world, Ei, 1 n i ij jEX (2 2)

PAGE 29

29 and likewise, if each column is summed over all rows, the value of country j s imports from the rest of the world, Mj, is obtained, 1 n j ij iMX (2 3) Tw is the total value of world trade. The gravity model explains the size of the individual Xij entries by treating them as endo genous. E ach non diagonal entry in the trade matrix would represent one observation in a gravity model sample for a particular time period. Thus, there will be a total of n(n 1) unique country pairs. There is no particular reason why all the economic units in the world should be included in the analysis. One can limit the matrix to a p articular subset of countries in the world economy. T he interpretation of the individual Xij, however, does not change. It should be noted that some researchers define (Xij + Xji ) as one observation, and not as two separate ones. Such a definition implies that the model estimates the determinants of the total amount of trade between two countries regardless of direction. In the case where (Xij + Xji ) is utilized rather than Xij, there will be n(n 1)/2 unique trading country pairs. Although there were earlier attempts by economists to utilize a similar co ncept (e.g ., Reilly 1929; Zipf 1946), the introduction of the contemporary econometric gravity model is usuall y attributed to Tinbergen (1962) and Pyhnen (1963a, 1963b) who independently and concurrently explored similar models. The model was refined by the subsequent research of Pulliainen (1963), Linnemann (1966), and Aitken (1973). It should be noted, however, that none of these studies actually referred to the model as a gravity model even though Pulliainen (1963), Pyhnen (1963a, 1963b) and Linnemann (1966) recognized the appeal to social physics. While Hewett (1976), Barker (1977), and Sattinger (1978 ) were the first to explicitly

PAGE 30

30 use the term gravity model to describ e these econometric models surprisingly few studies cite the latter three empirical papers. While there are some variations on the theme, a typical gravity model of international trade takes the following form : 035 124 ijtu ijtititjtjtijXeYYNNDee w (2 4) where Xijt is the bilateral exports from country i to country j in period t Yi, Yj are measures of the economic sizes of country i and country j in time period t respectively, most commonly measured by their GDPs, Ni, Nj are measures of the physical sizes of country i and country j in time period t most often measured by their respective populations, Dij is the physical, or geographic, distance between the two countries, and w is vector o f variables capturing any resistance factors or facilitative factors to trade. For simplicity, it is assumed that these variables enter the model exponentially. Following convention, u is a normally distributed error component c apturing any random influe nces. and are parameters. Economic size, or GDP, of an exporting country measures its productive capacity. As such, it is also an indicator of the range of product varieties available for export since richer economies tend to have a more sophisticated productive base. An importers GDP serves as an indicator of the absorptive capacity of imported goods. Naturally, both countries incomes would have positive effect s on bilateral trade levels (Hewett 1976; Linnemann 1966; Pyhnen 1963a, 1963b; Pulliainen 1963; Tinbergen 1962). It is postulated that countries with large r populations tend to have diversified economies and tend be more self sufficient, and would thus trade less (Brada and Mendez 1983; Hewett 1976; Linnemann 1966). However, countries with large populations also tend to have a larger industrial base and are able to capture more economies of scale in production than would smaller

PAGE 31

31 economies. Hence, the sign of the impact of physical size, or population, on bilateral trade levels should be amb iguous (Brada and Mendez 1983; Linnemann 1966) In a broader sense, as crude as the generalization might be, one can view the exporters income and population as indicators of supply factors, or potential export supply, and the importers income and popul ation as indicators of demand factors or potential import demand (Aitken 1973; Linnemann 1966). Hewett (197 6) referred to this notion as . the potential size of the foreign sector in each country . (p. 2). Distance is included as a proxy for t ransaction costs and would as such have a negative impact on the volume of trade (Linnemann 1966; Pyhnen 1963a, 1963b; Tinbergen 1962). It is further argued by Linnemann (1966) that the geographic distance is also an indicator of Beckermans (1956) asse rtion of the psychic distance between countries. It is common to augment the gravity model with variables that either increase or reduce trade (Aitken 1973; Hewett 1976; Linnemann 1966; Pyhnen 1963a, 1963b; Pulliainen 1963; Tinbergen 1962). Such vari ables are included in vector w from Equation 24. In previous research, such variables have for instance, included different measures of the price level in each country, the exchange rate, trade policy proxies, binary variables for assessing the impact o f participation in regional trading agreements1, of sharing a common commercial language2, of sharing a common border3, and of historical colonial ties4. Naturally, i t can be argued that participation in economic integration agreements alters trade patte rns. Furthermore, sharing a boarder or a common language would reduce transaction 1 A binary variable is introduced where a value of 1 is assigned if both trading partners of a particular country pair participate in a regional trading agreement, 0 otherwise. 2 For this binary variable, a value of 1 is assigned if both trading partners of a particular country pair share a common language, 0 otherwise. 3 This binary variable undertakes a value of 1 if both trading partners of a particular country pair are adjacent, 0 otherwise. 4 Here a value of 1 is assigned if there is a colon ial history between the two trading partners, 0 otherwise.

PAGE 32

32 costs and countries sharing a common language o ften have cultural similarities. This translate s into stronger trade ties since countries tend to interact more intensively with countries with whom they have similarities ceteris paribus Conversely, c ountries with historical colonial ties usually have strong trade relationships established between the former colony and the former metropolitan ruler and would thus trade more than they would with other countries. By explicitly controlling for these qualitative factors, the direct impact on trade patterns can be assessed. An appealing aspect of using binary variables to capture the effects of regional trading agreements, lin guistic ties, adjacency, and colonial history on bilateral trade is that one is able to assess how trade flows under the presence of such influence s diffe r from presumably normal or b aseline, trade patterns ( Hewett 1976; Linnemann 1966; Tinbergen 1962). In this context, the normal (or baseline) bilateral trade flows would be the ones where the above mentioned factors do not exert an influence. Thus, the benchmark trade behavior is established as the case where the binary variables are jointly equal t o zero. Consequently, the se presumably normal trade flows are dependent on the continuous variables only. This would follow since the binary variables in this case enter the model as a factor of unity (the exponential of zero is equal to one). For examp le, when two trading partners both participate in a regional trading agreement, the binary variable would take on a value of unity and the parameter of that particular binary variable would thus enable an analysis of by what factor s trade flows between cou ntries within a particular integ rated area differ from normal ones not affected by the agreement. One is henceforth able to isolate the effects of qualitative facto rs in a straightforward manner. This procedure is explained in further detail in Chapter 4.

PAGE 33

33 Taking the natural logs of both sides of Equation 24 while assuming that m variables are present in vector w yields 012345 1ln lnlnlnlnlnM ijt it jt it jt ijmmijijt mXYYNNDWu (2 5) Since the model is linearized in logarithms, statistical estimation is relatively straightforward an d the estimated coefficients of the continuous variables (i.e., the s) can be interpreted as elasticities. In addition to the ple thora of empirical analyses of trade flows in a variety of context s (i.e., assessing the impact of regional trading agreemen ts, colonial ties, cultural similarities, foreign direct investment, intellectual property rights, network effects, border effects, transportation costs, and different exchange rate regimes ) much work has been devoted to providing a theoretical justification for the model. Most noticeably, Anderson (1979), Anderson and van Wincoop (2003), Bergstrand (1985, 1989), Deardorff (1998), Evenett and Keller (2002), Helpman (1987), Helpman and Krugman (1985), and, if one applies a more liberal standard, Linnemann (1966) have demonstrated the models basis in economic theory. Today it is commonly accepted that the gravity model can be derived from a partial general equilibrium framework or an international expenditure system and that the model has a basis in both traditional Heckscher Ohlin (H O) trade models as well as in models of increasing returns. Early Gravity Analysis According to the followers of social physics social occurrences are analogs of the phenomena in the physical sciences and can thus be explained by the techniques and methods of natural science. According to Stewart (1950), the term social physics was introduced in 1836 by Adolphe Quetelet ( who incidentally was the first to realize that the normal statistical distribution had wide applica tions in the social sciences ) It should be noted that two centuries

PAGE 34

34 earlier however, John Graunt s Natural and Political Observations Made u pon the Bills of Mortality (1665) and Sir William Pettys Two Essays in Political Arithmetick (1687) dealt with the empirical measurement of economic and social phenomena (Spiegel 1995; Stewart 1950). One of first proponents of the direct social application of the law of gravitation was H.C. Carey in Principles of Social Science (1858). He considered humans the m olecules of society and since molecules and matter are subject to physical laws, these physical principles were applicable to social science as well. He noted that humans are subject to the laws of gravitation, with people having the tendency to g ravitate towards larger cities since they offer better economic opportunities He noted that bigger cities exerted a greater attraction on the population than smaller ones. Similar conclusions were reached by Ravenstein (1885). A simple mathematical formulation of this principle can be found in Young (1924), were he suggested that the movement of farm workers from one community to another was a direct function of the attraction of the destination community and an inverse function of the distance between that com munity and the current location. Reilly (1929) extended the discussion to retail flows where under normal conditions two cities draw retail trade from a smaller intermediate city or town in direct proportion to some power of the population of these tw o larger cities and in an inverse proportion to some power of the distance of each of the cities from the smaller intermediate city (1929, p. 16). Reilly (1929) proposed a deterministic equation that explained the law of retail gravitation which somew hat resembled a gravity relation stating that Nn aaa bbbBPD BPD (2 6) where Ba is the amount of business that city A attracts from the intermediate city, Bb is the amount of business that city B attracts from the intermediate city, Pa a nd Pb are the populations

PAGE 35

35 of cities A and B respectively, Da is the distance from city A to the intermediate city, and Db the distance from city B to the intermediate city. He assumed N to be unity. After obtaining values for Ba, Bb, Pa, Pb, Da, and Db using retail data from the state of Texas into Equation 26, he solved for n and found that in about 35% of the cases, the estimated value of n fell in between 1.5 and 2.5 (Reilly 1929). Interestingly, these estimates are similar to the distance elasticites found in modern gravity models. Sociologists capitaliz ed on the usefulness of gravity style modeling in studying the movement and interaction of people. It was among these social scientists that the concept was to be explored on a grand scale. A detai led survey is provided by Carrothers (1956) Stewart (1948) introduced the demographic gravitational force between regions i and j Fij, as defined as 2 ij ij ijPP F D (2 7) w here, similarly to Equation 2 1, Pi, Pj are their respectiv e populations and Dij represents the straightline distance between them. Stewart (1941, 1942, 1947, 1948, 1950) made the observation that the number of undergraduate students from a certain region at a given university was proportional to the total population of that area and inversely proportional to the distance from the college (1941, p.89). Over the next two decades, social scientists applied the gravity concept to a wide array of interactions and flows between regions, such as intercity telephon e calls (Hammer and Ikle 1957), tourism (Al caly 1967; Glejser and Dramais 1969), traffic, commuting and automobile travel (Alcaly 1967; Glejser and Dramais 1969; Hammer and Ikl 1957; Ikl 1954; Zipf 1946), migration (Beals, Levy, and M oses 1967; Glejser and Dramais 1969), other types of travel, such as railway or airplane (Alcaly 1967; Zipf 1946), and the transportation of goods (Glejser and Dramais 1969; Zipf 1946).

PAGE 36

36 Zipf (1946) empirical study is noteworthy as he applied the methodology to the movement of people between regions via highway traffic, railroad travel, and airway travel, and to domestic trade within the U.S. This might be the first explicit application of the gravity concept to trade flows, if one for the moment overlooks the more novel at tempt by Reilly (1929). Zipf (1946) used a potential energy relation to explain the flow of goods betwee n U.S. cities as m easured by railroad shipments, formulated as ij ij ijPP X D (2 8) where Xij is the volume of railroad shipm ents from city i to city j (i.e., trade), and the other variab les follow convention with indices i and j referring to cities. After taking logs of Equation 28 and a dding a constant term, Zipf (1946) regressed loglogij ij ijPP X D (2 9) In teresting, h is results indicated that the coefficient of the bracketed term, was not significantly different from unity.5 Cavanaugh (1950), who recognized the clear ties these model s had with the writings of Carey (1858) and Ravenstein (1885), found si milar results. Economists have long been intrigued by the impact of distance on trade flows. The notion that the geographic distance between two economic units has a negative impact on the volume of trade is a stylized fact. Consider two economic regi ons, A and B located with a given distance between them. Since transportation across large distances is costly, the transportation cost of exports from A to B would be proportional to t he distance between them (Ohlin 1933). 5 A striking feature of Zipf (1946) was that the model allowed for the existence of trade imbalances between regions by suggesting that such imbalances would be accompanied by capital flows as to balance the actual amount of money being transferred. The allowance for trade imbalances would not be revisited until Tinbergen (1962) and Anderson (1979).

PAGE 37

37 That is, two economic regi ons located further away have higher cost of transportation between them, and would consequently trade less than would two regions within a shorter distance, ceteris paribus Studies by Beckerman (1956), Isard and Peck (1954), and Isard (1954, 1956) among others, documented a strong negative impact of distance on the volume of both domestic trade and international trade. Beckerman (1956) and Isard and Peck (1954) recognized that distance alone does not explain the volume of trade, nor does it fully account for transaction costs, and needs to be considered in a broader context with the consideration of such factors as culture and language barriers, which naturally also influence transaction costs. Beckerman (1956) introduced the term psychic distance to i llustrate this point, an issue later revisited by Linnemann (1966). Garnaut (1994) referred to this idea as the subjective resistance to trade. A n intuitive extension of gravity models to the field of international trade was about to emerge. This renaissance, fueled by plethoric applications and theoretical explorations, would last well into the next millennium. Isard (1956) noted that gravity models provide an appealing framework under which the distance factor can be incorporated in the analysis of international trade, although he did not provide an empirical application of his own. Beckerman (1956) also provided a clear set up for the use of gravity models, although he did not operationalize the concept explicitly. Over the next two decades, there were three main approaches taken in the formal introduction of gravity models t o international economics; the Dutch School via Tinbergen (1962) and Linnem ann (1966); the Finnish School via Pyhnen (1963a, 1963b) and Pulliainen (1963); and the Probabilist ic Approach by Savage and Deutsch (1960), Goodman (1963), and Leamer and Stern (1970). These three frameworks are reviewed below.

PAGE 38

38 Modern Gravity Models A Gravity Model: The Dutch School Many economists (perhaps inaccurately) attribute the formal intro duction of the gravity model in international trade to Tinbergen (1962) and Pyhnen (1963a, 1963b). Jan Tinbergen, a Dutch economist, initiated the concept in Econometrics (1951). Based on a Walrasian market framework, he defined the concept of a turnover equation (1951, p. 2627). In a given market there is both a supply function and a demand function. The idea was that the interaction of these market forces determines the equilibrium quantity sold in that particular market, which is what the turnover equation illustrates (Tinbergen 1951; Linnemann 1966). In an appendix to Shaping the World Economy (1962), Tinbergen present ed an empirical study entitled An Analysis of Trade Flows In his own words: The purpose of the present analysis is to deter mine the normal or standard volume of international trade that would prevail in the absence of discriminating trade impediments (1962, p. 262). He suggested a turnover equation to explain the bilateral trade flow from country i to country j as dependent on factors specific to these countries in the following deterministic formulation, again altered slightly for notational consistency: 3 12ijijijXYYD (2 10) where Xij is the value of bilateral exports from country i to country j in a g iven time period, Yi and Yj a re their respective incomes, Dij is the geographical distance between the two countries is a constant, and the s are parameters of the model (i.e., elasticities). National income, or GDP, he maintained, measured an economys ability supply exports or demand imports. The actual model estimated again in deterministic form, was 012345log logloglogloglogij i j ij ij ijXYYDAP ,6 (2 11) 6 Here 0 = log ( ).

PAGE 39

39 where Aij is a binary variable to capture the effect of adjacency, and Pij is a binary variable indicating whether or not there is preferential agreement between the two countries. Note that Tinbergen (1962) defined the binary variables in logarithms.7 Tinbergen estimated the model using bilateral export data between 42 countries. As expected, both preferential treatment and adjacency had positive effects on trade flows. Crucial to Tinbergens analysis was the assumption that any deviations from normal trading patterns, that is by the model predicted trade patterns, were due to stochastic impediments or stochastic preferences. He calculated the deviation between actual and fitted trade for each country with respect to the res t of the world as an aggregate unit. He found that certain countries displayed quite a deviation from the normal trade behavior Towards the end of his investigation, however, he admitted that some of these deviations might have been due to international capital flows. Thus, t he trade balance merely adjusts the current account (Tinbergen 1962). Linnemann (1966) ( incidentally Jan Tinbergens graduate student) cited the works of Isard (1956) and Beckerman (1956), but he explicitly rejected the relevance of social physics as a motivation for the model. He maintained that bilateral trade volumes were determined by the potential supply and the potential demand of trading partners via a quasi Walrasian general equilibrium framework. As pointed out by Li nnemann (1966), the system had to be considered quasi Walrasian since in a true Walrasian framework bilateral transactions are not explicitly addressed. Assum ing a situation with n countries and that each country only exports one good, there are n traded goods in the world economy Assume further that the trade potential for a given country, Wi, can be expressed as 7 Thus, Aij is equal to 1 in logarithms (or 10 in numerical form) if country i and j share a common border, 0 in logarithms (or 1 in numerical form) otherwise, and Pij is equal to 1 in logarithms (or 10 in numerical form) if there was a preferential agreement between country i and j 0 in logarithms (or 1 in numerical form) otherwise.

PAGE 40

40 iv iiiWYN (2 12) where represent the domestic market creating effect of population and is a scale factor Observe that a countrys trade potential is determined by its income ( Y ) and population ( N ). L et this function be the same for all countries. T he demand for count ry i s products by country j s consumers can be expressed as D ijjiijXWpt (2 13) where tij is a trade resistance measure (or transaction costs), pi is the supply price in the exporting country, is a constant, is the trade pote ntial coefficient of demand is the price coefficient of demand, and is the trade resistance coefficient of demand. Let a countrys export supply be defined as the difference between domestic production and domestic consumption of the own good. Thus county i s export supply can be expressed as S iiiXWp (2 14) where is a constant, represents the trade potential coefficient of supply and is the price coefficient of supply Linnemann (1966) a ssume d further tha t trade was balanced. W ith respect to each country, there are ( n + 1 ) equations in the system. There are ( n 1 ) import demand equations for country i s good, one export supply equation, and a trade balance equation, n SD i ij jiXX (2 15) Together, this system determines ( n 1) Xij D, Xi S, and pi. Using the balanced trade condition, one can solve for the price level as follows 1 n i i jij jipWWt (2 16)

PAGE 41

41 Then substituting Equation 216 into Equation 213, Xij D, yiel ds 1 n ij jijjij jiX WWtWt (2 17) Using the expression s for Wi and Wj, Equation 212, in Equation 217 and assuming that the trade resistance factor can be defined as ijijijtDP (2 18) where Pij is a preferential treatme nt binary variable, Dij is the bilateral distance, and and are parameters, one obtains 1j iij iii jjjijijX YNYNDP jn jjjij jiYNt (2 19) After simplifying the parameters, a standard gravity model results 356 124ijiijjijijXYNYNDP (2 20) A Gravity Model: The Finnish School Independently and contemporaneously to Tinbergen (1962) Pyhnen (1963a 1963b), a Finnish economist, also introduced a model of bilateral trade flows. As pointed out by Pyhnen (1963a, 1963b), his research was not intended to c ompete with Tinbergen (1962) Pyhnens trade flow equation was based on an input output framework and takes a slightly different form: 12 301ij ijij ijYY X D (2 21) w here 0 is a constant i and j are country specific effects 1 and 2 are income parameters, t he parameter represents the transportation cost unit of distance and the 3 parameter an isolation factor (Pyhnen 1963a). While not explicitly referring to his equation as a gravity

PAGE 42

42 model Pyhnen recognized that his model had its roots in the gravity formula from physical science (1963a, p. 99). This awareness became furt her evident in Pyhnen (1963b). Pulliainen (1963) continued the emphasis on the analog with physical gravity. He suggested that the volume of international trade bilaterally was determined by forces of push factors and pull factors of the respective countries. The trading capacity of each country was said to be indicative of th ese forces. Pulliainen (1963) added a factor measuring the regional difference in temperatures as weighted by regional income to control for resource differences and he treated the distance variable conventionally. While the models proposed by Pyhnen ( 1963a, 1963b) and Pulliainen (1966) performed well empirically, neither one of them devoted much attention to the theoretical foundations of the models. Rather, the model was treated as a purely empirical exercise. One can thus far talk about two diffe rent approaches: the Dutch approach of Tinbergen (1962) and Linnemann (1966), and the Finnish approach of Pyhnen (1963a, 1963b) and Pulliainen (1963). However, around the same time, there was a third approach to gravity models; an approach that used a probabilistic model to explain trade flows. A Gravity Model: The Probabilistic Approach A more obscure, rarely cited justification for the use of gravity style models was based on a probabilistic model of international trade as described in Savage and Deut sch (1960), Goodman (1963), and Leamer and Stern (1970) .8 The idea was that exporters and importers are assigned to each other in the world economy by a random process. In short, it was assumed that world trade consists of a large number of independent t ransactions called consignments (Savage and Deutch 1960; Goodman 1963). Let Pi be the probability associated with the event 8 It should be noted that this probability approach was casually introduced by Dodd (1950).

PAGE 43

43 that country i is the exporter, and Qj the probability that country j is the importer. Let S be a correction factor to exclude domestic trade as a possibility, such that 11kk kSPQ (2 22) where k is the number of countries involved in world trade. Then the probability that there is a movement of goods from country i to country j Pij, is ijijPSPQ (2 23) It was suggested that if T was total world trade, equivalently the sum of all consignments, and Xi the sum of all consignments originating in country i and Mj the sum of all consignments destined to country j then Pi and Qj cou ld be approximated by i iX P T (2 24) and j jM Q T (2 25) respectively. As indicated by Leamer and Stern (1970), Pi and Qj are also measures of country i s and country j s respective shares of world trade. Leamer and Stern (1970) used this concept to develop a gravity model. Keeping with the previous notation and ignoring domestic trade, they defined the probability that a trade flow travels from country i to country j as ijijpPQ (2 26) i.e., the joint probability of each country participating in world trade. Assume that total world trade is made up of N consignments and that each bilateral trade transaction is of size Then total world trade can be expressed as

PAGE 44

44 TN (2 27) The expected volume of exports from country i to country j Xij, can consequently be expressed by taking into account both Equation 226 and Equation 227 such that 2 ijij ij ijXMXM XNpT TT (2 28) which represent s the product of country i s export sector and country j s desire to import in pro portion to world trade. I f Yi and Yj were substituted for total exports and total imports and a trade resistance term is amended multiplicatively, a gravity type mo del would result. Sattinger (1978) employed a similar justification for the use of the gravity model, a terminology he explicitly used. In Search of a Theory Pre 1979 While these three frameworks, the Dutch School, the Finnish School, and the P robabili stic A pproach, certainly provide rationales for the use of gravity models, a sound theoretical anchoring was lacking. Geraci and Prewo (1977) provided an algebraic manipulation of an errors in variati ons system that resulted in a gravity model. They assumed that bilateral trade volumes are determined by a system of export supply functions and import demand functions similar to Linnemann (1966). In general form, bilateral trade flows from i to j were described as ,,ijijijXfYYR (2 29) where Yi and Yj represent income s and Rij represents any resistance to trade. Rij is not directly observable and as such has to be proxied. To operationalize this factor, it had to be decomposed. The following variables were assumed to influence the aforementioned resistance: the average tariff rate of the importing country, Zj, a binary variable indicating whether both trading partners

PAGE 45

45 are members of a preferential trading area, Gij, a binary variable indicating whether the trading partners share a co mmon language, Lij, a binary variable indicating whether the countries share a common border, Bij, and finally, a transport cost factor, Tij *, which was a function of the geographic distance between the trading partners, Dij, and the average unit value of exports from i Vi. The inclusion of Vi stemmed from the assumption that handling fees in conjunction with shipping, i.e., port char ges, terminal fees, and so forth, are fixed costs. It should be noted, however, that the Tij function cannot be observed directly. Tij, the ratio of the observed landed value of exports, as measured in the importing country, to the observed value at the port of export was used as a proxy for Tij *. Geraci and Prewo (1978) specified a system of three functional relationshi ps *,,,,,,ijijjijijijijXfYYZGLBT (2 30) ijijTgT (2 31) and *(,)ijijiThDV (2 32) They assumed multiplicative natures of Xij, Tij, and Tij *, but allowed Xij to be exponential in the binary variables and the tariff rate. Taking logs of Equations 230, 231, and 232 results in 01234567ln lnln lnij i jijijijij ijijXYYZGLBTu (2 33) *lnlnijijijTT (2 34) and 012ln lnlnij ij iijTDV (2 35) where u, and are random disturbance terms. Substituting Tij (Equation 235) into Xij (Equation 233) and Tij (Equation 234) yields a two equation errors in variables system

PAGE 46

46 0123456ln lnlnij i jijijijijXYYZGLB 7012lnlnij iijijDVu (2 36) and 012ln lnlnij ij iijijTDV (2 37) By some algebraic manipulation the system becomes 0123456ln lnlnij i jijijijijXYYZGLB 12lnlnij iijDV (2 38) and 012lnlnlnij ij iijTDVw (2 39) where 0070171272 7, and ijijij ijijijvuw A gravity like structur e i s henceforth obtained. Anderson (1979) Anderson (1979) is often credited with providing the first rigorous theoretical derivation of the gravity model. His derivation was based on an international expenditure system. Anderson (1979) assumed that each of two countries, i and j produced two goods: a traded good and a nontraded good. He also initially postulated that no transaction costs were present Preferences were assumed to be homothetic and identical across countries. The traded goods share o f each countrys national product was defined as a function of each countrys income ( Y ) and population ( N ) such that ,iiiFYN (2 40) and ,jjjGYN (2 41)

PAGE 47

47 i represents the expenditure share on country i s traded good by country j as a percentage of country j s income. Anderson (1979) suggested that the demand for country i s exported good by country j Mij, would equal ijijjMY (2 42) where Mij is the value of j s imports from i Allowing for trade imbalances by mi, a current account correction factor, a trade balance relation can be obtained iii jji jmYY (2 43) This relation suggests that the value of i s aggregate imports and expenditures on its own nontraded goods has to be equal to country i s aggregate exports to the rest to the world plus the expenditures on its own nontraded goods, allowing for temporary current account imbalances by the correction factor mi. Solving out for the expenditure share from the trade balance constraint yields iii i jj jmY Y (2 44) Substituting Equation 244 into the demand relation, Equation 242, results in a simple gravity model iiijj ij jj jmYY M Y (2 45) It should be noted that Anderson (1979) did not append the current account factor, mi, until this final stage. Nothing is lost, however, by incorporating this factor in the initial definition of the trade balance constr aint.

PAGE 48

48 Anderson (1979) went on to show that the standard gravity equation will result by allowing for transportation costs. After relaxing the assumption that each country produces only two goods he introduced transaction costs by factor ijk, encompassing both transportation costs and trade barriers, for shipments of good k from country i to country j Red efining ik as the share of country i s expenditure on country j s tradable good k as a function of the transaction cost factor specific to country j j, results in ikikj (2 46) Assuming that the transaction cost factor is of the iceberg type, see Samuelson (1954, 1983), the landed value of any shipment of good k from country i to country j Mijk, can be expressed as 1ijk ikjj ijkMY (2 47) where ijk is the specific transaction cost factor for shipment of good k from i to j Consequently as transaction costs are in proportion to the quantity shipped, some of the shipment melts away duri ng transi t from country i to country j hence the term iceberg cost (Samuelson 1954, 1983). A ggregate imports of country j from country i can be found by summing over industries k such that 1ij ijkjj ik kk ijkMMY (2 48) The trade balance constraint, Equation 243, subsequently becomes 1iii ij jj ik jjk ijkmYMY (2 49)

PAGE 49

49 Assuming that the transaction cost factor ijk, is constant across all commodities for any pair of trading partners i.e., country i and country j it can be expressed as a function of the bilateral distance between the two countries, dij, such that ijkijijfd (2 50) Usin g Equation 250, country j s demand for country i s traded goods Equation 248, can be rewritten as 1ijjj ik k ijMY fd (2 51) and the trade balance condition Equation 249, similarly becomes 1iii jj ik jk ijmYY fd (2 52) Ande rson (1979) divided through each term of the above trade balance equation by the sum jj jY By using the resulting expression, Anderson (1979) subsequently isolated and solved for the expenditure share, 1 1iii ik k jj j jj j jj ij jmY Y Y Y fd (2 53) After s ubstituting Equation 253 into the import demand equation, Equation 251, the following model obtains

PAGE 50

50 11 iiijj ij jjij j jj j jjij jmYY M Yfd Y Yfd (2 54) Assuming that the number of countries is large and that the average distance between them does not vary in any significant manner, one can ignore the last bracketed term since this term would be the same across countries (And erson 1979). Furthermore, the denominator in the first can be treated as a constant as well as this term is fairly congruent ac ross countries. It was assumed that the current account factor and the traded g oods shares were functions of a countrys income and its population. A log linear functional form was imposed Following Anderson (1979), the current account factor, mi, can be explicitly expressed as 12imiimkYN (2 55) and the traded goods shares as 12iiikYN (2 56) and 12jjjkYN (2 57) S ubstituting Equations 255, 256, and 257 into the import demand equation, Equation 254, results in 11221211111ijm ijMkkkYNYN fd (2 58) If one simplifies the parameters ( and ) and imposing a log linear functional form for the transaction cost term, an expression resembling the standard gravity model is obtained. W ithout altering the right hand side of the equation, one can interchange between imports and exports on the left hand side, thus Mij can be interchanged for Xij to obtain the conventional notation. The

PAGE 51

51 count ry index is not affected as count ry i is still the origin country and country j still the destination country. Expanding Andersons framework : Following Andersons (1979) framework of a linear expenditure system, Oguledo and MacPhee (1994) derived the gravity model along similar lines. While they explicitly featured prices as exogenous variables, their assumptions were the same as in Anderson (1979) T he notation and methodology below are altered slightly for consistency with Anderson (1979) Oguledo and MacPhee (1994) maintained that the demand for imports from country i by country j could be expressed as ()ijijiijjjMTTY (2 59) where Mij represents the landed value of imports from country i in country j Tij represents transaction cost s ( encompassing both transpor t costs and trade barriers and is assumed t o be of the iceberg type ) i is country j s expenditure share on country i s traded good expressed as a percentage of country j s income which is functionally dependent upon Tij, j repr esents country j s traded goods share expressed as a percentage of country j s income, and Yj represents the income of country j Defining the transport factor as ijijjTDt (2 60) where tj represents country j s ad valorem tar iff imposed on imported gods and Dij is the bilateral distance between country i and country j the trade balance equation can be expressed as ii ijjjiij jYTYT .9 (2 61) Solving out for the expenditure share i(Tij) from Equation 261 while using Equation 260 yields 9 Note that unlike Anderson (1979), Oguledo and MacPhee (1994) did not allow for the current account factor m

PAGE 52

52 ii iij ijjjj jY T DtY (2 62) Substituting Equation 262 into the import demand function, Equation 259, yields iijjijj ij jjijj jYYDt M YDt (2 63) Under the assumption that the traded goods shares depends not only on income ( Y ) and population ( N ) but also on domestic prices ( P ) Oguledo and MacPhee (1994) imposed explicit functions for 3 12iiiiYNP (2 64) and 3 12jjjjYNP (2 65) where and a re parameters. After substituting these expressions into the import demand function, Equation 263, simplifying the constant term and the parameters, and interchanging import data for export data, they obtained and equation similar to the standar d gravity model 35678 1240 ijiiijjjijjXYNPYNPDt (2 66) where the s are parameters. Bergstrand (1985) Bergstrand (1985, 1989) provided two frequently cited derivations.10 Common to both approaches is the assumption that the underlying demand conditions can be modeled by constant elasticity of substitution (CES) utility functions and that the supply conditions can be 10 An almost identical derivation was provided in Gould (1994), which was described as a modification of Bergstrand (1985).

PAGE 53

53 modeled by constant elasticity of transformation (CET) technologies. He assumed that the world economy consists of N countries and that all consumers share identical utility functions of CES nature where the choice is between two types of goods: imported goods and domestically produced goods. The utility function for consumers in country j would take the following form 1 1 1j j j jN j kj jj k kjUXX (2 67) where Xkj is the amount of goods produced in country k demanded by country j s consumers. Subsequently, Xjj is the amount of domestically produced goods demanded in j Parameter is defined as 1j j j (2 68) where is the CES between domestically produced goods and imported goods. Similarly parameter is defined as 1j j j (2 69) where is the CES among imported goods. This specification al lows consumers to first choose whether to consume domestically produced goods or imported goods and then, for the imported goods consumers will choose among the potential suppliers on the world market (Bergstrand 1985) Consumer choices are constrained by the following budget constraint based on the expenditure approach,

PAGE 54

54 1 N jkjkj kYPX (2 70) given that kjkjkj kj kjPTC P E (2 71) w here Yj represents country j s income, Pkj is the currency price of country k s go od in country j Tkj is a tariff factor imposed by country j on country k s products, Ckj is a transport ation cost factor, and Ekj is the exchange rate as measured as the number of country k s currency units per one unit of country j s currency. Clearly, for the special case of domestically produced goods, Tjj, Cjj, and Ejj are all unity. Maximizing the utility function Equation 267, subject to the expenditure constraint Equation 270, yields N(N 1) import demand functions, 1 1 1 1 1 1 1 1 1 1jj j j j j j j jN kjk kj D ij ijj N kj jj k kjP XYP PP ,1,..., () ijNij (2 72) O ne can ignore the N demand functions for the domestically produced goods intended for home consumption, XD jj, since these goods would not enter into the gravity relation. On the supply side of the economy firm s in each country i are assumed to be profit maximizers and do so in accordance with 1 N iikikii kPXWR i = 1, N (2 73)

PAGE 55

55 where i denotes profit of firms in country i Ri is the amount used of a single internationally immobile productive resource, and Wi is the money wage rate (or rental rate) for that factor in country i s currency. The endowment of R available in each country i s utiliz ed in production according to a CET technology as follows 1 1 1i i i iiN i ik ii k kiRXX i = 1 N (2 74) Parameter is expressed as 1i i i 0i, (2 75) where i represents the CET between p roduction for home consumption or foreign consumption. Similarly, p arameter is expressed as 1i i i 0i, (2 76) where i represents the CET among international markets. Substit uting the availability of the productive resource, Ri from Equation 274, into the profit function, Equation 2.73, and optimizing will result in N2 first order conditions which can be solved for N(N 1) bilateral export supply relations 1 1 1 1 1 1 1 1 1 1ii i j i i i i iN ik k ki S ijiij N ik ii k kiP XYP PP ,1,..., and () ijNij (2 77)

PAGE 56

56 Analogously, one can ignore the supply functions for the domestically produced goods intended for home consumption, XS ii, since these goods would not enter into the gravity relation. Similarly to Linneman n (1966), Bergstrand (1985) utilized N2 equilibrium conditions such that DS ijijijXXX (2 78) Equation 278 simply constrains trade to be balanced on the bilateral level. Applying the derived import demand functions and export supply functions to the above constraint results in N2 partial subsystems of a general equilib rium setup, each one consists of four equations with four unknown variables ( Xij, Xij D, Xij S, Pij). Since the gravity model explains the value of the exports from one country to another, one needs solutions for Xij and Pij. By using the N2 equilibrium conditions Bergstrand (1985) solved out for Xij and Pij resulting in the following 1 1 1 1 1 11ijj j j jjij jii i ijij iijN kj ijij k kj ij N ijij ik k kiYYEP X CTP 1 11 11 1 1 11 111 ij ji ii ii j j iiNN kj ik ii jj kk ki kjPPPP (2 79) and

PAGE 57

57 1 11 1 1 1 11 111 1 1 1 1 1jj ii i ii j j j i ii j j j jj jNNN kj jij ik ik ii kkk ki kj ki ij N kj iijij jj k kjYEP P PP P YCTPP 1ij (2 80) Assuming that utility functions and technologies are congruent across countries, i.e., the CET parameters in Xij and Pij are the same for all country pairs, in combination with a small country assumption, thereby exogenizing for eign pri ces on the world market multiplying Xij and Pij yields the value of bilateral exports from country i to country j such that ijijijPXPX (2 81) Equation 281 i mplies the following general ized gravity model 1 1 1 11 1 1 1 1 11 1 1 N kj ijij k kj ij N ij ik k kjYYEP PX CTP 11 11 11 1 11 1 11 1 NN kj ik ii jj kk ki kjPPPP (2 82)

PAGE 58

58 (Bergstrand 1985). After simplifying expo nents and parameters, albeit in quite the restrictive manner, Equation 282 simplifies to a gravity like structure with prices and the exchange rate included as exogenous. The rather complex price terms, i.e., the denominator of the gravity model, were proxied by the use of the GDP deflator; however, the empirical performance of these price approximat ions was ambiguous. Illustrative of said ambiguity, Bergstrand (1985) obta ined positive coefficients for the exporters income and negative coefficients of the exporter s GDP deflator thus supporting his conclusion that the elasticity of substitution among imported goods was greater than unity. He also found that for an import ing country, the CES among imported goods exceeded the CES among imports and domestically produced goods, which consequently was concluded to be less than unity. Bergstrand (1989) Bergstrand (1989) was still puzzled as how to theoretically justify the inclusion of population as an exogenous variable and was bothered by the rather restrictive assumption of a single productive factor. Given a monopolistically competitive framework he derived a generalized gravity model unique to each industry. Bergstrand (1989) operated under a multi country, multi industry (two sectors ), two factors framework. It was assumed that each consumer l in country j maximizes the following nested CobbDouglas CES Stone Geary utility function subject to a budget constraint 1 11 11 11AB An Bn ABHH NN jl Ahnjl Bhnjl B nh nhUXXX (2 83) where 1, and 01.AB A, B, and are parameters, and BX represents the minimum consumption requirement of the good produced in sector B required by any consumer (Bergstrand 1989). Regarding the indices in Equation 283, A refers to the manufacturing s ector,

PAGE 59

59 B to the nonmanufacturing sector, h identifies the firm, n the country of production, j the country of consumption, and l the individual worker or consumer. As such, XAhnjl would represent the quantity consumed by consumer l of the good produced i n the manufacturing sector, A from firm h in country n Similarly, XB hnjl would represent the quantity consumed by consumer l of the good produced in the manufacturing sector, B from firm h in country n. Consequently, each consumers utility depends on all goods consumed, both manufactured and nonmanufactured, produced domestically or imported from abroad. Consumers maximize their utility subject to their income. Bergstrand (1989) assumed that all consumers are identical and therefore aggregated across consumers to obtain country j s inverse demand curve for the output of a particular firm h, in a particular sector A or B in country i Accord ingly, the inverse demand function by country j s consumers for the output of firm g in the manufacturing sector, A produced in country i can be expressed by 11 1 1 1 1 1 111AA A A A An Aj j ij Aij H N AhjnAnj AgijAij nh njYyE P PT XT E (2 84) where yj is the per capita income of country j in terms of the minimum consumption of the nonmanufactured good, and E P and T have similar interpretations as in Bergstrand (1985). It should be noted, however, that industry identifier subscripts on P and T and firm identifier subscripts on P were added. On the firm level, each firm h in each industry produces a uniquely differentiated product under incre asing returns that is sold in monopolistic ally competitive markets (Bergstrand 1989). An appealing feature of this approach is that products are differentiated on the firm level and not merely by country of origin. This was also pointed out by Deardorff (1998). Production is

PAGE 60

60 undertaken using two factors, labor ( L ) and capital ( K ) Technology is assumed to be linear according to agiLaLaagiLX (2 85) and agiKaKaagiKX (2 86) 1,...,; ,; and 1,...,aigHaABiN where, the s represent fixed set up requirements of each factor, the s are the unit input requirements for the output in industry a (i.e., A or B ) by firm g in country i (Bergstrand 1989). The output is divided among domestic consumption and exports to foreign mar kets according the CET surface 1 1a aN agi ainagin nXCX (2 87) 1; 1,...,; ,; and 1,...,a aigHaABiN where C is a transport ation cost factor. Given the labor wage rate, w and the rental rate of capital, r each firm then determines their export supply to N foreign markets by maximizing the following profit function 1 N agiainaginiLaiKaiLaagiiKaagi nPXwrwXrX 1 11a aNN ainaginiLaiKaiLa ainagin nnPXwrwCX 1 1a aN iKa ainagin nrCX (2 88) 1; 1,...,; ,; and 1,...,a aigHaABiN

PAGE 61

61 Firms supply exports on the world market according to their ma rginal cost functions derived from the profit maximization process given the demand functions derived from utility maximization. Solving for the reduced forms and summing across all firms in a particular industry, Bergstrand (1989) arrived at an industry level generalized gravity model explaining the volume of exports from i to j of industry A s good, such that 1 1 11* *1A AAA AA A A AA AAK ij Aij LBKB KALBKBLA i iYY PX K L 1 1 11 11 1A AA AA AA AAA AAAAij j Aij AijE y CT 1 1 1 1 1 1 111AA A AA A AA A ANN Ain AinAin nn Ain njP PT CE (2 89) 1; 1,...,; ,; and 1,...,a aigHaABiN or in simpler functional relation notation 11,,,,,,,,,''NN AnjAnj K i Ain Aijij jAijAijAij nn i Ain njPT KP PXfYYyCTE parameters L CE (2 90) where Yi k is country i s income expressed in units of capital, Ki */Li is the capital labor ratio of the exporting country net of resources required for set up costs, and yj is the per capita income of country j in terms of the minimum consumption of the nonmanufactured good.

PAGE 62

62 Empirically, Bergstrand (1989) estimated the gravity model at the single digit Standard Industrial Trade Classifications (SITC) level, thus resulting in nine different industry level estimations. The exporters income in terms of capital units was proxied by the exporters GDP, and its capital labor ratio was proxied by its GDP per capita. The importers income per capita in terms of the required amounts of nonmanufactured goods was approximated by the importers GDP per capita. Remaining variables were proxied similarly to Bergstrand (1985). Bergstrand (1989) appealed to both the HO theorem and the Rybczynski theorem. The HO theorem states that a country has a comparative advantage in the export of a good that uses its relatively abundant resource relatively intensively in production, while the Rybczynski theorem states that an increase of a particular resource will lead to an increase in the output of the goods that uses t hat resource relatively intensively in production. Thus, if a country saw its GDP per capita increase (which proxies the exporters capital labor ratio) the gravity model should predict, if consistent with the Rybczynski theorem, that production and expor ts of capital intensive goods would increase. Indeed, Bergstrand (1989) observed positive coefficients for the exporters GDP per capita for the presumably capital intensive industries. This em pirical result validated the use of GDP per capita as a prox y for the capital labor ratio in the ex porting country. Conversely, Bergstrand (1989) obtained a negative coefficient for the exporters per capita income in the presumably labor intensive industries, thus further lending support to the results. From Ber gstrands findings one can then infer that per capita income would be a better indicator of supply capabilities than just the absolute level of national income. One could also make a similar argument on the demand side; per capita income would give a bett er indication of absorptive capacity of imported goods (Bergstrand 1989) T here are also econometric motivations for using per capita income rather than income in absolute terms

PAGE 63

63 when estimating gravity models These considerations will be discussed in Chapter 3 and Chapter 4. Further Forays into Economic Theory Additional derivations were provided by Deardorff (1998) and Evenett and Keller (2002). Evenett and Keller (2002) assumed that two goods are being produced, Z1 and Z2, with n varieties of each. Suppose that all production takes place under increasing returns to scale (IRS) and that complete specialization will prevail in each variety. Let output price be the same for all varieties of a good. Consequently, t he output of any given country j can be expressed by 11 2 212 jj jzzjzj zYpnzpnz (2 91) where 1zp represents the output price of good 1 regardless of variety 2zp represents the output price of good 2 regardless of variety 1j Zn and 2j Zn represent the number of varieties good 1 and good 2 respectively in country j 1 jz and 2 jz are the outputs of each variety of good 1 and good 2 respectively in country j Assume further that preferences are identical and homothetic across countries that free trade prevails, that all transaction costs are zero, and that trade takes place in final goods only (Evenett and Keller 2002) In this model, countries will import goods in proportion to their national incomes. Define si as the share of country i s national income in world income. It is then postulated that country i s consumers demand a fraction, si, of the world economy output. As such, (1 si) of the output pr oduced in i is exported. Similarly, one can think of sj as the share of any countrys domestic output exported to country j The bilateral imports of country j from country i or exports from i to j Xij, can henceforth be expressed as 11 2 212 ijji ij zzjzjj zXspnzpnzsY (2 92)

PAGE 64

64 If trade is assumed to be balanced ij ijjjiiXsYXsY (2 93) and using the definition of the income shares, bilateral exports can consequently be expressed as ij ij wYY X Y (2 94) where Yw is world income. Hence, a simple version of the gravity model results. This model allows for intra industry trade in that countries are trading in different varieties of the same goods. Evenett and Keller (2002) referred to this gravity model as a pure I RS model. Along similar lines, but chronologically earlier, Helpman and Krugman (1985) and Helpman (1987) pointed out, as a consequence of algebra, that the total volume of trade between any given country pair (the sum of the exports from i to j and the exports from j to i ) can be expressed as i j ij ij xjzj xiziijjijiVspXpZspXpZXXsYsY (2 95) Evenett and Keller (2002) suggested that the gravity model also could be derived from three other scenarios. They maintained that an identical gravity model could be derived from a multicone HO model with perfect specialization. They applied the assumption of homothetic and identical preferences. Assume a two country two factors, two goods world. Assume further two homogenous goods, Z1 and Z2, are produced under constant returns to scale (CRS). Let country i be relatively capital abundant and let good Z1 be relatively capital intensive in production. Consequently, let country j and good Z2 be labor abundant and labor intensive in production, respectively. The HO theore m predicts that country i will specialize in the production of good Z1, and country j will specialize in the production of Z2. If perfect (i.e., complete) specialization prevails, world production of each good, Z1w and Z2w respectively, can be expressed as the

PAGE 65

65 domestic output of the good in the country that experiences a comparative advantage in its production, such that 11 iwZZ (2 96) and 22 jwZZ (2 97) where Z1i is the output of good 1 in countr y i which equals world production of good 1, and Z2j is the output of good 2 in country j which equals world production of good 2. Using Equation 296 and Equation 297, national income (i.e., GDP) of each country can be defined as 11 ziipZY (2 98) and 22 zjjpZY respectively, where 1Zp and2Zprepresent output prices of good 1 and good 2. If trade is assumed to be balanced and no transportation cost or trade barr iers are present, bilateral trade flows can be expressed as 2121 iijj ijzjjziijiXspZsYspZsYX (2 99) where, si is the share of country i s national income in world income and sj is the share of country j s national income in world income, a simple gravity model is obtained ij ij wYY X Y (2 100) (Deardorff 1998; Evenett and Keller 2002). Equation 2100, the multi cone HO model, is identical to Equation 294, the pure IRS model. Deardorff (1998) referred to Equation 2100 as the simp le frictionless gravity equation, since there are no transaction costs present and the geographical distance would as such not enter as a factor.

PAGE 66

66 A similar gravity model can be obtained from an IRS/ uni cone HO model with incomplete specialization, but a s will be shown, the volume of bilateral trade will be lower than under perfect product spe cialization (Evenett and Keller 2002). Maintain that good Z2 is labor intensive in production and produced under CRS and assume the same factor abundancies as under the multicone HO model, but let good Z1 be capital intensive and produced under IRS. Consequently, country i produces both goods while country j produces only Z2. Thus only the capital intensive good is specialized in production. Defining i as the share of the labor intensive good in country i s GDP, 22 1222 12 zizi i i zizipZpZ Y pZpZ (2 102) implies that (1 i) represents the share of the capital intensive good in country i s GDP. Evenett and Keller (2002) assumed trade to be balanc ed and consequently defined the bilateral exports from i to j as 11 11 j jj ijjizi ii ii w Y XXspZsYY Y (2 103) Thus, the gravity model from the IRS/ uni cone HO framework becomes 1ij ij i wYY X Y (2 104) where bilateral trade now also depends on the relative size of the capital intensive sector in country i (1 i ) (Evenett and Keller 2002). By algebraic observation, it is obvious that the volume of trade suggested by Equation 2104 is less than that suggested by Equation 294 or Equation 2100. Finally, Evenett and Keller (2002) demonstrated that the gravity model could be derived from a uni cone HO model with incomplete specialization in both goods. They assumed that

PAGE 67

67 both goods were homogenous and produced under CRS with Z1 being capital intensive and Z2 labor intensive. Country i is capital abundant and would thus export the capital intensive good and subsequently country j is labor abundant and would export the labor intensive good. However, both countries maintain production of the other good, thus resulting in incomplete specialization. The bilateral exports from country i to country j Xij, Evenett and Keller (2002) argued, can be expressed as 11ij ij i j wYY X Y (2 105) which would simplify to ij ijji wYY X Y (2 106) and a gravity model results. By inspection, whe n comparing the four above derived gravity models 1ij ij ij iji wwwYYYYYY YYY (2 107) the volume of trade suggested by Equation 2106 is less than by Equation 2104, which was previously stated to produce less trade than Equation 294 and Equation 2100. This results shows, Evenett and Keller (2002) maintained, the higher the degree of product specialization, the larger the volume of bilateral trade They found that industry level data from 58 countries supported that the gravity model stems from t he IRS/u ni cone HO model (Equation 2104) and the uni cone HO model (Equation 2106) while only marginal empirical support was found for the pure IRS model (Equation 294) The multi cone HO model (Equation 2100) was rejected as a theory explaining the gravity model.

PAGE 68

68 Trans action Costs and Remoteness Deardorff (1998) derived a gravity model by assuming that trans action costs are of the iceberg type and that preferences are identical across countries based on a Cobb Douglas utility function. Following similar basic principles as Evenett and Keller (2002), the value of exports from country i to country j with transaction costs factor tij was expressed as ij ij ijwYY X tY (2 108) Clearly, the value of bilateral exports varies inversely with the size of tij. However, Deardorff (1998) was not satisfied with this formulation due to its restrictive nature and instead introduced a derivation similar in scope to the expenditure system approach of Anderson (1979) and included a relative di stance measure as an explanatory variable. It was assumed that each country produces one good and that goods are differentiated by country of origin. The consumers in the importing country, i.e., country j are assumed to be maximizing the following CES utility function given products from N potential trading partners, 1 1 1 N j iij iUx (2 109) where is a distribution parameter and is the CES. This utility function is seemingly more restrictive than that of Bergstrand (1985, 1989) in that Deardorff (1998) assumed that consumers are indifferent between imported and domestically produced goods.11 Given that pi is the money price of goods produced in country i tijpi is the price of imported goods from country i in country j inclusive of any transaction cost s 11 It should be recalled that Bergstrand (1985, 1989) proposed a twostep structure. Fir st consumers chose between imports and domestically produced goods and then among imported goods consumers chose among potential suppliers.

PAGE 69

69 Taking into account a linear income constraint, the amount of goods produced in country i demanded by utility maximizing consumers in country j can be expressed as 11iji ij ji I iji jtp xY tpp (2 110) pj I is a price ind ex for imported goods from country i in country j such that 1 1 11 1 N I j iiji iptp (2 111) Consequently the landed value of imports from country i in country j can be written as, 1 iji ijji I jtp XY p (2 112) To obtain an e xpression for the distribution parameter i, Deardorff (1998) used the definition of country i s share of world income si, and Equation 2112 such that 11 111NN iji iji ii i ijj ij II jj ww j jtp tppx s px s YYp p (2 113) i can explicitly be solved for from Equation 2113 as 1 11i i N w ijj j I j jY Y tp s p (2 114) Substi tuting Equation 2114 into Equation 2112, Xij, yields 1 1 1 ij I j ij ij N w ih h I h ht p YY X Y t s p (2 115)

PAGE 70

70 By normalizing money prices, pi and pj, to equal unity, the price index term, pI j, Equation 2111, simplifies to 1 1 1 1 N s j iij it (2 116) where s j can be interpreted as a measure of country j s average distance, or relative distance, from its trading partners. Deardorff (1998) referred to ij ij s jt (2 117) as the relative distance bet ween country i and country j Thus, using the concept of relative distance, Deardorffs (1998) gravity model of the landed value of exports from country i in country j becomes 1 1 1 ijij ij N w hih hYY X Y s (2 118) The relative distance of a country, sometimes called the remoteness of a c ountry (Frankel, Stein, and Wei 1997), was further explored theoretically by Anderson and van Wincoop (2003). Relative distance, or as Anderson and van Wincoop (2003) called it, multilateral resistance, was the centerpiece of their model. In par ticular, they showed that theory predicts that trade resistance between two countries does not only consist of the bilateral trade barriers and resistance between country i and country j but also incorporates country i s resistance vis vis the rest of t he world, and country j s resistance vis vis the rest of the world (Anderson and van Wincoop 2003). The latter two factors are the so called multilateral resistance.

PAGE 71

71 Anderson and van Wincoop (2003) assumed that each country specializes in t he producti on of one good only, which was noticeably more restrictive than both Anderson (1979) and Bergstrand (1985, 1989) and that preferences are identical and homothetic across countries and follow a CES utility relation. Uj, the utility derived of consumers in country j can be expressed as 11 1 1 N j iij iUc (2 119) where cij is the consumption of goods originating in country i and, again, is a distribution parameter. Let the price of country i s goods to country j s consumers be represented by ijiijppt (2 120) where pi is the supply price in country i and tij is the transaction cost for trade between i and j Thus, the value of bilateral exports can be expressed as ijijijXpc (2 121) Assuming that consumers maximize their utility subject to a linear income constraint, t he following demand function results 1 iiij ij j jpt XY P (2 122) w here, similarly to Deardorff (1998), the Pj factor is a price index of country j such that 1 1 1 1 N j iiij iPpt (2 123) Imposing marketclearing behavior in combination with the derived demand for foreign goods, the follo wing was obtained

PAGE 72

72 1 11 NN iijj i ij j jj jtp YXY P 1 1 1 N ij ii j j jt pY P (2 124) Similar to Deardorf (1998), Anderson and van Wincoop (2003) solved out for the distribution parameter While Deardorf (1998) assumed all prices to equal unit y for simpl icity, Anderson and van Wincoop (2003) were less restrictive and allowed prices to vary. However, to incorporate varying prices, ipi, i.e., scaled prices in country i was solved for rather than by itself ( Anderson and van Wincoop 2003). Substituting the solution into the bilateral export demand function, Equation 2122, yields 1 ijij ij wijYYt X YP (2 125) given that 1 1 1 1 N ij ij j jt s P (2 126) and 1 1 1 1 N ij ji i it Ps (2 127) Imposing symmetric trade barr iers on the bilateral level, tij = tji, the following holds iiP (2 128) with the subsequent implication 1 11 1 N j iiij iPPst (2 129) (Anderson and van Wincoop 2003). Substituting Pi for i in the bilateral export demand function yields a gravity like structure

PAGE 73

73 1 ijij ij wijYYt X YPP (2 130) which depends not only on bilateral trade resistance factors, but also on the trade resistance with respect to all trading partners thro ugh the Pi and Pj terms. Thus, multilateral resistance should be taken into account when estimating gravity models. Upon the empirical suggestions by Polak (1996) Deardorff (1998), and Anderson and van Wincoop (2003) many recent studies utilize relati ve distance, or remoteness, measures While there are variations on the theme, remoteness is commonly measured by the average geographic distance of a country to all its trading partners weighted by their respective GDPs or GDP shares of world GDP. This remoteness of country i vis vis its trading partners can be defined as itY Yjt Remotenessd ij ji wt (2 131) where Yw is aggregate world GDP in time period t (Frankel, Stein, and Wei 1997) The remoteness of country j can be defined analogously Due to the ad hoc nature of the concept, there are some variations e mpirically The remoteness framework have been used in some permutation by among others, Anderson and van Wincoop (2003), DiMauro (2000), Frankel, Stein, and Wei (1997), Frankel and W ei (1998), Helliwell (1997), Polak (1996), Sandberg, Seale, and Taylor (2006), a nd Soloaga and Winters (2001). Why would the economic remoteness of a country matter? A country that is located r elatively remote with respect to its trading partners will tra de more extensively with a country located within a particular distance than would an economy that is less remote in the relative sense. Frankel, Stein, and Wei (1997) provided an intuitive explanation: The distance between

PAGE 74

74 Australia and New Zealand is t he same as the distance between Spain and Poland. Spain and Poland have lots of other natural trading partners close at hand, but Australia and New Zealand do not. One might thus expect the antipodean pair, who have less in the way of alternatives, to tr ade more with each other, holding other influences constant, than the European pair (p. 143). As per their framework, remoteness should have a positive impact on bilateral exports after controlling for bilateral distances and other factors Polak (1996) provided a similar argument. Helpman (1987) A n alternative specification of the gravity model and quite possibly a theoretically more correct formulation, was provided by Helpman (1987) based on the extensive work of Helpman and K rugman (1985). The ex position below is altered slightly for notational consistency and is highly abbreviated. In its simplest form, assume a two country (country i and country j ), two goods ( Z1 and Z2), and two inputs ( L and K ) world, where L denotes labor and K denotes capit al. Assume further that good 1 ( Z1) is capital intensive in production and that good 2 ( Z2) is labor intensive in production and that country i is relatively capital abundant and that country j is relatively labor abundant. As per conventional HO theory country i will export (and country j will import) Z1 and country j will export (and country i will import) Z2. Assuming that the two goods are homogenous and that tastes homothetic and constant across countries, the value of bilateral exports from count ry i to country j can be expressed as 1_ 11 i ijzXpZsZ (2 132) where Z1 is the domestic output of good 1 in country i si is the share of country i s spending (or country i s GDP) of world spending (or world GDP), and 1Z is the world output level of good Z1. Similarly, the value of bilateral exports from country j to country i can be expressed as

PAGE 75

75 2_ 22 j jizXpZsZ (2 133) where Z2 is the domestic output of good 2 in country j sj is the share of country j s spending (or country j s GDP) of world spending (or world GDP), and jZ is the world output level of good Z2. In a two country world, world GDP is simply the bilateral sum of the two countries respective GDPs, such that i i ijYs YY (2 134) and j j ijY s YY (2 135) If trade is restricted to be balanced, the bilateral trade can be expressed as 12__ 1122 ij ijijjiz zVXXpZsZpZsZ (2 136) It was geometrically shown, using factor price equalization, that the larger the bilateral differences in resource endowments, the larger is the volume of trade, Vij, between the two countries (Helpman and Krugman 1985; Helpman 1987). This is consistent with HOtheory ; countries wit h different resource endowments would engage in trade more extensively. Helpman and Krugman (1985) also analyzed the case where the two goods are differentiated and produced under increasing returns. Under product differentiation, it is assumed tha t each firm, n, is producing a unique variety of the differentiated good. The output per variety is the world output level of a particular good divided by the number of varieties. Under this scenario, there is two way trade in both products, i.e., both countr ies export and impor t both goods The value of bilateral trade between country i to country j i.e., the sum of

PAGE 76

76 bilateral exports from country i to country j and bilateral exports from country j to country i can be expressed as 11122 122 ijjjii ijVspZpZspZpZ (2 137) where 1 jZ is the output of good 1 in country j 2 jZ is the output of good 2 in country j 1 iZ is the output of good 1 in country i 2 iZ is the output of good 2 in country i and p1 and p2 are output prices (Helpman and Krugman 1985; Helpman 1987). This model was referred to as the pure IRS model previously in this chapter Considering that the GDP of country i can be expressed as 1122 ii iYpZpZ (2 138) and that the GDP of country j can be expressed as 1122 jj jYpZpZ (2 139) Equation 2137 simplifies to ij ijjiVsYsY (2 140) Algebraically, Equation 2140 is equivalent to 2ij ij ijVssYY (2 141) It is evident from Equation 2141 that the volume of trade in differentiated products is dependent on combined economic size of trading partners ( Yi + Yj) and the relative economic sizes of the two economies, si and sj (Helpman and Krugman 1985). Clearly, combined economic size, or the bilateral sum of GDPs, would have a positive impact on trade volumes. The impact relative economic size, or economic dispersion, varies with relative size of the two countries GDPs. Numerically, the si and sj are both bound by 0 and 1. Bilateral trade between country i and country j will be the greatest in volume when both the economies are of equal relative size, or

PAGE 77

77 when both si and sj are equal to 0.5. Helpman and Krugman (1985) also showed the meridian scenario of one good being homogenous and one good being differentiated and produced under increasing returns to scale. In his empirical study, Helpman (1987) emphasized intra industry trade e.g., the two way trad e of differentiated products sold in monopolistically competitive markets. Given the framework of Helpman and Krugman (1985), the goal in Helpman (1987) was to see how resource endowments, combined economic size, and economic similarity affected the volum e of bilateral intra industry trade. The model was explicitly rationalized to be applicable to the study of the trade behavior of subsets of countries. Let Sij k k, with k being the commodity indicator represent an index of the volume of bilateral intra industry trade as a percentage of total bilateral trade 2min,ijji kk ij k kk ijji kk kXX S XX (2 142) where Xij k denotes bilateral exports of commodity k from country i to country j in a given time period. Helpman (1987) suggested, while not proving its functional form per se, a gravity type relation for explaining the magnitude of this intra industry trade index, such that 22 01 2 3log log log1 jj ij ii kk ij ij ijij YY YY S YY NN YYYY (2 143) The first bracketed term is the absolute value of the differences in GDP per capita. Helpman (19 87) argued that this variable was a proxy for the similarities in factor endowments and countries with similar resource endowments would, as the HOtheory suggests, trade less. However, as was pointed out The share of intra industry trade in bilateral tr ade flows should be larger for countries with similar incomes per capita ( Helpman 1987; p. 74). The second term

PAGE 78

78 simply captures the effect of combined economic size. Clearly, larger economies would trade more than smaller ones. The last bracketed term was crucial in his analysis. It is a dispersion index measuring the relative economic size of the two countries ( Helpman and Krugman 1985; Helpman 1987) This index varies from 0 to 0.5, with the limiting case of 0 indicating complete dissimilarity in in come and 0.5 suggesting identical economic sizes. This dispersion index should have a positive effect on bilateral trade volumes as greater varieties of differentiated goods would be available (Breuss and Egger 1999). While Helpman (1987) did not pursue this, the model can be amended, as appropriate, with additional variables capturing any friction or enhancing influences on bilateral trad e such as the bilateral distance and binary variables to control for qualitative factors By using the value of bil ateral exports, Xij, as the dependent variable, rather than the volume of intra industry trade, empirical congruency vis vis previous gravity models is achieved. As such the theoretical consistency of the Helpman model is merged with the empirical eleg ance of the more commonly used specification in the tradition of Tinbergen (1962) Pyhnen (1963a, 1963b), and Linnemann (1966). Over the past decade, Breuss and Egger (1999), Egger (2000, 2002) and Antonucci and Manzocchi (2006) estimated augmented ve rsions of the Helpman (1987) model using the value of bilateral exports as the endogenous variable. These authors also provided additional rational e for the variables in Helpman (1987) model, or at least a de facto justification for their inclu sion. DiMa uro (2000) fitted a similar specification with both bilateral foreign direct investment ( FDI ) flows and exports as endogenous variables with satisfactory results. Such modified versions of Helpman s (1987) model provide interesting options for assessing t rade behavior Even though only a s mall minority of studies employs such model s they perform well empirica lly. One

PAGE 79

79 obvious appeal is the intuitive interpretation of the parameters for the variables in Equation 2143, i.e., 1, 2, and 3, and the clear theoretical foundations of the model. Remarks T he literature identifies two families of gravity model specifications; i.e., a traditional specification of the model in the tradition of Tinbergen (1962) Pyhnen (1963a, 1963b), Linnemann (1966) and othe r researchers, and a more recently derived specification by Helpman (1987) emphasizing intra industry traded goods produced under increasing returns to scale based on the work by Helpman and Krugman (1985) While models a la Tinbergen (1962) are more comm only used, the Helpman (1987) model may have a more straightforward basis in economic theory, particularly when it comes to justifying its variables. However, little research has been devoted to determining which one of the two specifications should be ut ilized. There is no formal assessment in the literature regarding which specification of the model provides a better fit to the data or which model more accurately assesses trade patterns. Nevertheless, Breuss and Egger (1999) suggested that the Helpman (1987) model is closer to the HO theory of international trade vis vis the more commonly applied formulation of the model, at least conceptually. It also seems that the Helpman (1987) model is more appropriate when studying groups of countries or when t he sample only contains a portion of the global economy. Over the past century, t he most common use of the gravity model is to assess the impact of regional trading agreements on bilateral trade flows, or how trade flows within the integrated area differ from presumably normal trade flows not affected by said regional agreement; see, for example, Breuss and Egger (1997, 1999), Coe and Hoffmeister (1998), Egger (2002), Frankel, Sten and Wei (1995, 1997, 1998), Garman, Peterson, and Gilliard (1998), Grant and Lambert (2008), Hassan (2001), Nilsson (2000), Sandberg, Seale, and Taylor (2006), Soloanga and Winters (2001), Vollrath, Hallahan, and Gehlhar (2006), Wall (2003), Wei and Frankel

PAGE 80

80 (1997), and Zahniser et al. (2002) as well as classical papers by Abrams (1980), Aitken (1973), Aitken and Obutelewicz (1976), Barker (1977), Brada and Medez (1983), Hewett (1976), Prewo (1978), and Sapir (1981). This list, of course, is by no means complete. Other uses of the gravity model include the impact of immigration on bilateral trade flows. The rationale is that immigrants will likely have strong ties with their former home countries thus fostering trade relationships between the current and former country of residence (Gould 1994; Head and Ries 1998). The impac t of colonial legacies on bilateral trade patterns, with neocolonial trade ties evident between the former colonies and the former metropolitan ruler, was investigated by Eichengreen and Irwin (1998), Rauch (1999), Sandberg and Martin (2001), Sandberg, Se ale, and Taylor (2006), among others. While most studies include a proxy for sharing cultural and linguistic ties, the issue was investigated in depth by Boisso and Ferrantino (1997). That adjacency would increase bilateral trade levels is pretty self ex planatory even though the explicit empirical effects still present some puzzles (Anderson and van Wincoop 2003; McCallum 1995). The extensive derivations reviewed in this chapter emphasize an important point: there are theoretical foundations for the gravity model. If one manipulates any of a collection of theoretical frameworks long enough, one obtains something resembling a gravity equation. It is important to keep in mind that economists were searching for a theoretical basis for an already operationalized concept. Consequently, any theoretical derivations are therefore strictly de facto in nature, in that a theory was searched for to fit the model. While it is interesting to take note of the models theoretical foundations, the focus however ought to be placed onto its empirical elegance and illuminating results. As summarized by Taplin (1967): Its value lies in its ability

PAGE 81

81 to identify extreme cases of artificial barriers to trade, the role of distance, and the effects of membership in various cus toms union and trade pref erence groups (p. 442). Indee d.

PAGE 82

82 0 X 12 X 1j X 1n E 1 X 21 0 X 2j X 2n E 2 X i1 X i2 0 X in E i X n1 X n2 ... X nj 0 E n M 1 M 2 M j M n T w Figure 2 1. A trade matrix

PAGE 83

83 CHAPTER 3 CONCEPTUAL FRAMEWORK AND DATA Conceptual Framework Chapter 2 outlined the evolution of the gravity model of international trade and presented various theoretical foundations and derivations. While it is a fair claim that the model has a basis in a variety of theor etical frameworks quite elaborate manipulations and restrictions are necessary in order for a gravity equation to result. The model has been used with great empirical success over the past century in the analysis of trade behavior in a wide array of cont exts This study will utilize the gravity model to establish the presumably normal or baseline, trade behavior of the countries in the Western Hemisphere. By employing qualitative binary variables, any distortions or biases resulting from establishe d trade networks due to history and regionalism will be identified and analyzed. As outlined in the previous chapter, e mpirical specifications of the gravity model can be argued to fall into either one of two categories. The most commonly employed spec ification follows the tradition of Tinbergen (1962) Pyhnen (1963a, 1963b) and Linnemann (1966), with theoretical foundations provided by Anderson (1979) and Bergstrand (1985, 1989) among others While there are numerous theoretical derivations and ju stifications for the gravity model other than Anderson (1979) and Bergstrand (1985, 1989), as the previous chapter suggests, Anderson (1979) and Bergstrand (1985, 1989) provided the first thorough and most commonly accepted derivations of the model. Most studies explicitly cite these works to justif y the models foundation in economic theory. Therefore, this specification of the model will be referred to as the Modified Tinbergen PyhnenLinnemann AndersonBer gstrand (MTPLAB) specification for the purpos es of this study The term modified is added as the model is augmented by

PAGE 84

84 binary variables to enable the current analysis of interest and the remoteness variables as to comply with current empirical practice. The second category of empirical gravity m odels is the formulation based on the theoretical works of Helpman and Kru gman (1985) and Helpman (1987) This family of models will be referred to as the Modified Helpman (MH) specif ication The same caveat applies regarding the use of the term modifie d. The reader should note, however, that there is no formal designation of the two specifications in the literature. Rather the use of the terminology MT P LAB specification and MH specification is the authors own nomenclature. The MT P LAB specification in general, states that 012345ln lnlnlnlnijt it jt it jtijXYYNND 67 1ln lnM it jtmmijijt mRemotenessRemotenessWu (3 1) where Xijt represents the bilateral exports from country i to country j in time period t Yi t is nominal GDP for country i in time period t Yj t is nominal GDP for country j in time period t Ni t is country i s population in time period t Nj t is country j s population in time period t and Dij is the bilateral distance in kilometers between the capital of country i and the capital of country j Remotenessit represents the remoteness of the exporter in time period t which in this study is defined as itY D Yjt Remoteness ij ji wt (3 2) where Ywt is aggregate world GDP in time period t while Remotenessjt represents the remoteness of the i mporter in time period t, analogously defined in this study as

PAGE 85

85 jtY D Yit Remoteness ij ij wt (3 3) The reader should note that these remoteness measures incorporate the trading partners relative shares of world GDP Ywt, e.g., the ( Yjt/ Ywt)s for exporting country i and the ( Yi t/ Ywt)s for importing country j Since the trading partners relative shares of global GDP vary from year to year, the calculated remoteness varies from year to year as well. Thus, the calculated remoteness of an exporter (or an importer) is therefore applicable in the time period t for which it is computed.1 Wmijs are m binary variables present in vector w corresponding to any qualitative relationships between country i and country j and ijtu is a no r mally distributed error term. One can consequently define the parameters from the MTPLAB specification in vector form as where 071'(,...,,,...,)m Per capita GDPs ( Yi t/Ni t, Yj t/Nj t) may more accurately capture the trading capacity of the two countries as compared to absolute GDPs ( Yi t, Yj t). As B ergstrand (1985, 1989) suggested, GDP per capita of the exporting country can proxy its capital labor ratio and the GDP per capita of the importing country indicates its ability to absorb imports. Inclusion of per capita incomes becomes particularly appealing when considering that trading partners populations already cap ture economies of scale and sheer physical size. Using per capita GDPs rather than absolute GDPs in the estimation also has econometric rationale. As pointed out by Breuss and Egger (1997), aggregate income and population are likely highly correlated in that countries with large populations tend to have larger GDPs, ceteris 1 By the construction of these remoteness measures, relatively large count ries, e.g., the United States and Brazi l would appear relatively less remote v is vis their trading partners as these larger economies would comprise a larger relative portion of the world economy. Conversely, relatively sm all countries, e.g., Barbados and Guatemala, would appear relatively more remote vis vis their trading partners as these smaller economies would comprise a smaller relative portion of the world economy.

PAGE 86

86 paribus This would inherently result in multi collinearity between income and population when estimating the gravity model. Multicollinearity leads to higher standard errors and t herefore renders statistically insignificant parameter estimates. Sandberg, Seale, and Taylor (2006) shared this concern. It should be noted, however, that neither the overall F statistic nor the r square are affected by correlated regressors as the sums of squared errors are impervious; only the statistical significance of the estimated parameters is impacted. Per capita GDP is less likely to be highly correlated with population. Some countries with large populations have relative low per capita incom es, while other countries with relatively small populations conversely have high per capita incomes. By employing per capita GDPs instead of absolute GDPs, the aforementioned multi collinearity problem should be averted (Breuss and Egger 1997). This issue is formally addressed in the following chapter when the empirical results are discussed. Following Breuss and Egger (1997), one can easily incorporate per capita GDPs with some simple algebraic operations performed on Equation 31 resulting in ** 01 2 345ln lnlnlnlnjt it ijt it jtij it jtY Y X NND NN 67 1ln lnM it jtmmijijt mRemotenessRemotenessWu (3 4) where 3 3 1 and 4 4 2. Alternatively, Equation 34 can be expressed as 01231 425ln lnln ln lnijt it jt it jtijXYYNND 67 1ln lnM it jtmmijijt mRemotenessRemotenessWu (3 5) A similar manipulation of population parameters was used by Garman Petersen, and Gilliard (1999), Sandberg, Seale, and Taylor (2006), and S mith (1999). Equation 34 serves as the MTPLAB specification

PAGE 87

87 The second gravity model formulation of interest is the MH specification, which states that 22 01 2 3ln ln ln ln1jt jt it it ijt itjt itjt itjt itjt YY YY X YY NN YYYY 45 6 1lnln lnM ij it jtmmijijt mDRemotenessRemotenessW (3 6) where Xijt represents the bilateral exports from country i to country j in time period t Yi t is nominal GDP for country i in time period t Yj t is nominal GDP for country j in time period t Ni t is country i s population in time period t Nj t is country j s population in time period t and Dij is the bilateral distance in kilometers between the capital of country i and the capital of country j Remotenessit represents the remoteness of the exporter in time period t defined as itY D Yjt Remoteness ij ji wt (3 2) where Ywt is aggregate world GDP in time period t, while Remotenessjt represents the remoteness of the importer in time period t analogously defined as jtY D Yit Remoteness ij ij wt (3 3) Wmijs are m binary variables in vector w corresponding to any qualitative relationships between country i and country j and ijt is a normally distributed disturbance. One can express the parameters from the MH specification Equation 3 6, in vector form as where 061'(,...,,,...,)m As summarized in Chapter 2, t he first bracketed term in Equation 36 is the absolute value of the differences in GDP per capita, a proxy for the similarities in factor

PAGE 88

88 endowments and countries with similar resource endowments would, as HO theory su ggests, trade less The second term represents combined economic size or combined market size effect The third bracketed term is a dispersion index measuring the relative economic size of the two countries (Helpman 1987) For the purposes of this st udy, a total of 11 binary variables are included in vector w (thus m = 11) in Equation 34 and Equation 36. Table 3 1 presents and defines these variables. These binary variables are intended to capture any network effects due to regional and historical linkages after the continuous variables in both models have established the presumably normal trade behavior that would prevail in the absence of these effects Variable W1 addresses adjacency and its influence, variables W2 through W5 control for hist orical linkages and cultural similarities due to colonial heritage and sharing a common commercial language, variables W6 through W10 measures the impact of regional linkages resulting from participation in regional trade agreements (i.e., participation in the Andean Pact, CACM, CARICOM MERCOSUR and NAFTA respectively). Note that no binary variables are introduced to control for participation in the Group of Three or LAIA. It would be empirically impossible to include dummy variables for these two agreements due to the perfect overlap between MERCOSUR and LAIA (all MERCOSUR members are also participating in LAIA) and the Group of Three economies are also participating in NAFTA and the A ndean Pact (Mexico is a member of NAFTA and LAIA; Colombia and Venezuela are members of the Andean Pact and LAIA). Explicitly including binary variables for the Group of Three and LAIA would subsequently result in perfect collinearity and empirical estimation would not be feasible. A final binary variable, W11, controls f or the importing country being a member of the European Union. This variable is intended to capture any preferential treatment (e.g., policy distortions) extended by the European

PAGE 89

89 Union to goods imported from the economies in the Western Hemi sphere It is imperative to con trol for such market access, as the colonial variables otherwise will likely pick up that distortive effect and thereby bias the results The continuous variables in both models are referred to as the gravity variables and the binary variables capturing historical and regional linkages in w are referred to as network effects T he gravity variables establish a baseline or presumably normal trade behavior prevailing without any interfering network effects. These qualitative effec ts would subsequently capture an y deviations from the baseline trade behavior cau sed by history or regionalism. Data Data Sources and Sample There is no ex ante reason why historical and regional networks will exert the same effects on the agricultural se ctor vis vis the manufacturing sector of the economies in the Americas. Trade in agricultural commodities may be more or less susceptible to neo colonial trade biases or regional preferences relative to trade in manufactured goods and vice versa. It is an appealing inquiry to appraise whether such differences are present. To capture any sectoral differences in these effects, the gravity model is estimated using three different dataset s. The first dataset encompasses aggregate trade data; the second includes trade in agricultural products and primary commodities only ; and the third dataset represents trade in manufactured goods. Hence, each of the two model specifications is estimated three times for a given year, once for each of the three data se t s Due t o the structure of the available data, there are a total of ten annual cross sections, covering the years 1992 through 2001. As a result, there are sixty regressions to be estimated (ten annual cross sections, three data set s, two model specific ations).

PAGE 90

90 An inherent characteristic of the data, covering 1992 through 2001, which can be argued to be both a weakness and a strength of the current analysis, is that any changes in trade behavior post the 9/11 terrorist attacks are not addressed. For the years immediately following 2001 trade volumes may be significantly distorted as trade patterns in all likelihood were altered while coming to terms with new world conditions. These issues are consequently not investigated in this study. Thus, the tr ade behavior captured is not influenced by the post 9/11 environment. There are a total of 63 co untries (or economic units) included in the sample They are presented in Table 3 2. As evident of the sample, t he dataset consists of the FTAA economies plus the majority of the economies of the Organization for Economic Cooperation and Development (OECD)2, China, India, Israel and Singapore, South Africa, and others. Note that Cuba, the hemispheres current Dutch dependencies ( such as Aruba St Martin, and the Netherland s Antilles ) and t he current French dependencies ( such as French Guyana, Guadeloupe Martinique, and St Barthelmey ) are omitted from the sample, as are Puerto Rico and both the British and U.S. Virgin Islands. Since the emphasis is on the trade patterns of the economies in the Western Hemisphere, a particular characteristic of the dataset is its unbalanced nature in favor of the thirty four FTAA countries. This implies that the bilateral exports between FTAA members are included, as are t he FTAA countries bilateral exports to and from nonFTAA economic units. However, the bilateral exports between non FTAA economies are excluded. For example, consider the four countries Australia, Barbados, Canada, and South Africa, where Barbados and C anada are both part of the FTAA group. Bilateral exports between Barbados and Canada are included as are Barbados and Canadas bilateral export flows to and from Australia and South Africa. 2 Norway (!) is missing from the sample due to lack of reported data.

PAGE 91

91 However, purely external bilateral exports, as those between Au stralia and South Africa, are not included. This unique feature enables any noise from the global trade environment unrelated to the Western Hemisphere to be removed. C onsidering that 40% of all world commerce takes place within the European continent (Feenstra and Taylor 2008) it is important that influences of intra European trade patterns are eliminated from this study as to enable a more pure analysis of trade in the Americas. Bilateral export data for the dependent variable are obtained from the I nter American Development Banks DATA Intal CD ROM Version 4.0 (2003). The data are disaggregated based on the United Nations Conference on Trade and Development (UNCTAD) nomenclature. The six UNCTAD product categories are: Category 1: All Food Products; Category 2: Agricultural Raw Materials; Category 3: Minerals and Metals; Category 4: Fuels; Category 5: Manufactured Goods; and Category 6: Other Products. Bilateral exports of UNCTAD Category 1 (Food products) and Category 2 (Agricultural raw materials) are combined into a data set representing agricultural trade. Export data from Category 5 (Manufactured goods) are used for trade in manufactures. The dataset for aggregate trade consists of the sum of all six nomenclatures as presented on the CD ROM. Consequently, the aggregate trade data include more than simply the arithmetic sum of agricultural trade and trade in manufactured goods; it also includes the remaining three nomenclatures, i.e., minerals and metals, fuels, and other products. Compiling the data sets required for any gravity based study can be a cumbersome process. As the number of possible c ountry pairs easily grows large, the data sets are usually comprehensive. The data were retrieved from the CD ROM using the Data Intal software. Upon extracting the data, they were exported into Microsoft Excel where the data had to be

PAGE 92

92 manipulated extensively. Since multiple time periods are considered, one possibly way of arranging the data is by exporting country, importing country, and by yea r. Thus, the dataset has two country identifier columns, one for the exporter and one for the importer. For each possible country pair, the data is sorted by year and a third column indicates the temporal aspect. Each one of the explanatory variables i s subsequently assigned a unique column, corresponding to the respective exporter, importer and the year in question. A tenacious process of matching the trade data for given country pairs and the corresponding explanatory data ensued. Due to data avai lability, the number of unique bilateral trade observations (or unique country pairs) in a given year varies from 1554 to 2657. The uneven availability of bilateral trade data from year to year is a common problem plaguing gravity studies incorporating de veloping economies as trade data are simply not always readily collected and available. The number of observations is remarkably lower for 1992 and 1993, as compared to the rest of the sample period due to fewer countries being represented. It should be noted that the data for 1992 and 1993 are missing much of intra CARICOM trade. GDP and population data were obtained from the United Nations Common Database. The distance data and information about colonial ties, language, and adjacency were obtained f rom Mayer and Zignago (2006). Information regarding participation in regional trading agreements was obtained from Frankel, Stein, and Wei (1997). Problem of Zero Observations Before analyzing the results, it is important to note that the sample conta ins a noninsignificant number of zero observations i.e., in some instances the reported bi lateral exports are indeed zero. Some country pairings simply do not trade in certain products and the observed bilateral export s would in such a case equal zero More specifically, for the dataset encompassing agricultural trade (UNCTAD categories 1 and 2), 1.1% of the available bilateral

PAGE 93

93 observations are zero (or 198 instances out of the 17,872 data points available), for the data set encompassing manufactured goods trade (UNCTAD category 5), the corresponding percentage is 0.86% (or 170 instances out of the 19,554 available data points), and for the aggregate trade dataset 8% of the data points are zero (or 1801 observations out of the 22,426 available data points ). These reported zero observations pose an empirical problem, as the dependent variable is the natural logarithm of bilateral exports O ne cannot take the natural logarithm of zero. Frankel, Stein, and Wei (1997), Bikker (1987) and Brada and Me ndez (1985) suggested to simply drop these observations and estimate the model with the remaining data. However, this procedure would result in biased parameter estimates as t here may indeed be legitimate reasons as to why certain bi lateral trade flows ar e zero. Nevertheless, Frankel, Stein, and Wei (1997) claimed that the results were not sensitive to the omission of these observations. Following Boisso and Ferrantino (1997), Eichengreen and Irwin (1998), Head and Ries (1998), Sandberg and Martin (2001) and Sandberg, Seale, and Taylor (2006), a value of one is added to the dependent variable before the natural logarithm is taken. Thus, the dependent variable becomes Xijt = Xijt +1. In the cases where there is no observed bilateral trade, i.e., where Xijt = 0, Xijt = 1, and lnXijt = ln(1) = 0. Subsequently, in the cases where Xijt > 0 the dependent variable becomes ln Xijt = ln( Xijt +1), which should not be significantly different from ln( Xijt). The dependent variable in the two proposed empirical specifications, Equation 34 and Equation 36, ln Xijt, is replaced with ln Xijt resulting in the following two models to be estimated: ** 01 2 345ln lnlnlnlnjt it ijt it jtij it jtY Y X NNDNN 67 1ln lnM it jtmmijijt mRemotenessRemotenessWu (3 7)

PAGE 94

94 which, represents the MTPLAB specification and 22 01 2 3ln ln ln ln1jt jt it it ijt itjt itjt itjt itjt YY YY X YY NN YYYY 45 6 1lnln lnM ij it jtmmijijt mDRemotenessRemotenessW (3 8) which, analogously, represents the MH specification. The next chapter discusses, in detail, the results from the empirical estimations. It also addresses the rest rictions imposed i n Equation 37 (and Equation 34) via the inclusion of per capita GDP instead of absolute GDP as well as the data aggrega tion issues encountered in estimating the models While the MTPLAB model is the conventionally most commonly employed specification, the MH model has defi nite appeal for this study. Given the inquires posed and the nature of the data used, the MH model seems to be a theoretically more consistent model as it was derived explicitly for the study of groups of countries. The data sets are FTAA centered, as ex plained above, and thusly do not represent a global sample. Consequently, in the following empirical analysis, the MH specification is treated as the first model, or the primary model, and the MTPLAB specification is treated as the second model, or the secondary model. The MTPLAB is still utilized as it is more commonly used empirically and it also serves as an appealing avenue of assessing the robustness of the binary network effects from the MH model

PAGE 95

95 Table 3 1. Dummy variables in vector w Variabl e Variable name Variable definition Contingency/Adjacency: W 1 Common border Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. Culture and historical linkages: W 2 Common language Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. W 3 Colonial linkage: U.K. Equal to 1 if one of the trading partners is the U.K. and the other a former British colony in the Western Hemisphere, 0 otherwise. (Note that neither the United States nor Canada are considered former colonies in this context). W 4 Colonial linkage: Spain Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere, 0 otherwise. W 5 Colonial linkage: Portugal Equal to 1 if one of the trading partners is Portugal and the other a former Portuguese colony in the Western Hemisphere, 0 otherwise. Regional trade agreements: W 6 NAFTA Equal to 1 if both trading partners are members of NAFTA, 0 otherwise. W 7 CARI COM Equal to 1 if both trading partners are members of CARICOM, 0 otherwise. W 8 MERCOSUR Equal to 1 if both trading partners are members of MERCOSUR, 0 otherwise. W 9 Andean Pact Equal to 1 if both trading partners are members of the Andean Pact, 0 other wise. W 10 CACM Equal to 1 if both trading partners are members of CACM, 0 otherwise. Effects of the European Union being the importer: W 11 EU Importer Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 96

96 Table 3 2. Countries included in the study Country Antigua and Barbuda a Iceland Argentina a India Australia Ireland Austria Israel The Bahamas a Italy Barbados a Jamaica a Belgium Luxembourg Japan Belize a Mexico a Bolivia a The Netherlands Brazil a New Zeeland C anada a Nicaragua a Chile a Panama a China Paraguay a Colombia a Peru a Costa Rica a Poland Czech Republic Portugal Denmark Singapore Dominica a Slovak Republic Dominican Republic a South Africa Ecuador a South Korea El Salvador a St Kitts and Nevis a Finlan d St Lucia a France St Vincent and the Grenadines a Germany Sweden Greece Switzerland Grenada a Trinidad and Tobago a Guatemala a Turkey Guyana a United Kingdom Haiti Uruguay a Honduras a United States a Hong Kong Venezuela a Hungary a ) Part of the FTAA gro up.

PAGE 97

97 CHAPTER 4 EMPIRICAL ESTIMATION Introductory Remarks All estimation s for this study are performed using GAUSS econometric software (Aptech Systerms Inc. 2001). As discussed in Chapter 3, two econometric models are considered The MH model serves as the primary model and is used to establish the baseline trade behavior. The MTPLAB model provides a robustness check of the estimated binary effects, as these variables are the same across the two specifications. The primary model, the MH specificatio n, is defined as Equation 38, or 22 01 2 3ln ln ln ln1jt jt it it ijt itjt itjt itjt itjt YY YY X YY NN YYYY 45 6 1lnln lnM ij it jtmmijijt mDRemotenessRemotenessW (4 1) where Xijt represents the bilateral exports from country i to country j in time period t plus 1, Yi t is nominal GDP for country i in time period t Yj t is n ominal GDP for country j in time period t Ni t is country i s population in time period t Nj t is country j s population in time period t and Dij is the bilateral distance in kilometers between the capitals of country i and country j Remotenessit and Re motenessjt are defined as itY D Yjt Remoteness ij ji wt (4 2) and jtY D Yit Remoteness ij ij wt (4 3)

PAGE 98

98 respectively. Wmijs are m binary variables in vector w corresponding to any qualitative relationships between country i and country j (as defin ed in Table 3 1), and ijtv is a normally distributed error term. Naturally, Equations 42 and 43 are the direct equivalents of Equ ations 32 and 33. As explained previously, the first bracketed term in Equation 41 is the absolute value of the differences in GDP per capita, the second term represents combined economic size, and the third bracketed term is a dispersion index measuring the relative economic size of the two countries. The secondary model, or the M TPLAB specification is defined as Equation 37, or ** 01 2 345ln lnlnlnlnjt it ijt it jtij it jtY Y X NNDNN 67 1ln ln ,M it jtmmijijt mRemotenessRemotenessWu (4 4) where Xijt *, Yi t, Yj t, Ni t, Nj t, Dij, Remotenessit, Remotenessjt, and Wmij are defined as above, and ijtu is a normally distributed disturbance. Aggregation Issues The first issue to be considered is whether the gravity model should be estimated using aggregate trade data or by using disaggregated trade data by sector (i.e., the agricultural sector versus the manufactur ing sector). In the co ntext of this study, this issue is addressed by formally testing whether the agricultural trade data set and manufactured goods trade dataset can be aggregated or pooled. Two simple F tests can be implemented to test the empirical appropriateness of ag gregating trade data when estimating gravity models. The first test follows a conceptually similar F test applied by Seale (1985 1990), who estimated production functions with two different dataset s. In the context of this study, this first test compar es the estimated variances of

PAGE 99

99 the residuals from the gravity model obtained by using agricultural trade data to the estimated variances of the residuals from the gravity model obtained by using manufactured goods trade data. If the variances are found to be statistically the same, aggregation of the data may be appropriate. If the variances are found to be statistically different, the gravity model should be estimated sector by sector with disaggregated trade data. Following Seale (1985, 1990), a second F test addresses whether the estimated parameters are the same across the dataset s. Again, in the context of this study, if the estimated parameters are statistically different, the dataset s should not be aggregated, which would lend further support to estimating the model sector by sector. Testing the Similarity of V ariances Let 2 A represent the variance of the residuals from any specification of the model obtained by using agricultural trade data and let 2 M represent the variance of the residuals from the corresponding specification of the model obtained by using manufactured g oods trade data. As suggested by Seale (1985 1990), an F test is appropriate for testing whether the two variances are statistically the same. This becomes evident when considering the structures of 2 A and 2 M. Let eA equal the vector of estimated residuals associated with any specification of the gravity model obtained by us ing agricultural trade data. L et eM equal the vector of estimated residuals associated with the corresponding specification obtained by using manufactured goods trade data. The sum of squared errors for each estimation is found by AAee and 'MMee respectively. Assuming that eA and eM are normally distributed, AAee and 'MMee would both follow the c hi square dist ribution as the product of two normally distributed variables is chi squa re distributed. The calculated test statistic considers a ratio of two chisquare distributed

PAGE 100

100 variables and given the fact that th e ratio of two chi square distributed variables follows the F distribution, an F test is the appropriate test (Wackerly, Mendenhall, and Scheaffer 2002) The sample variance for the residuals obtain ed from the model estimated using the agricultural trade data set 2AS can be calculated as 2' AA A AAS nk ee (4 5) where nA represents the number of observations in the agricultural trade dataset and kA represents the number of independent regressors in the model. 2AS follows t he chi square distribution. Similarly, the sample variance for the residuals obtained from using the manufactured goods trade dataset 2MS can be calculated as 2' MM M MMS nk ee (4 6) where nM represents the number of observations in the manufactures trade dataset and km represents the number of independent regressors in the model. Consequently 2MS also follows the chi square distribution. Generalizing from above notation, let 2 1 S equal the sample variance for the residuals obtained from one dataset and let 2 2 S equal the sample variance for the residuals obtained from the other data set Assume that 2 1 S is larger th an 2 2 S The F statistic can then be constructed as follows 122 1 2 2 dfdfS F S (4 7) where df1 equals the degrees of freedom for the numerator, n1 k1, and df2 equals the degrees of freedom for the denominator, n2 k2, where again n refers to the number of observations and k

PAGE 101

101 is the number of regressors This particular F ratio tests whether 2 1 and 2 2 are statistically the same. More formally, the following null hypothesis, 22 012: H is tested against the alternative hypothesis, 22 12:aH Specifically, one tests whether 2 A and 2 M are equal, or 22 0:AMH against 22:aAMH Table A 1 in Appendix A presents the sums of squared errors and the calculated F statistic for the MH specification of the gravity model using agricultural trade data and manufactured goods trade data, respectively. In the table one can also find the number of observations, nA and nM, and the number of regressors, k It should be noted that kA = kM. For the MH specification AAee exceeds 'MMee during the years 1992 through 1998. Thus, 2AS is treated as 2 1 S and 2MS is treated as 2 2 S for 1992 through 1998. However, for the remaining three years, 1999 through 2001, 'MMee exceeds AAee and consequently 2MS is treated as 2 1 S and 2AS is treated as 2 2 S in 1991 through 2001. The interchangeable nature of 2 1 S and 2 2 S depending on their magnitudes was outlined by Wackerly, Mendenhall, and Scheaffer (2002). Thus, for 1992 through 1998, the F statistic is calculated as 1,' 'AAMMAA AA nknk MM MMnk F nk ee ee (4 8)

PAGE 102

102 and for 1999 through 2001, the F statistic is calculated as ,' 'MMAAMM MM nknk AA AAnk F nk ee ee (4 9) Table A 2 in Appendix A presents the corresponding values associated with the MTPLAB specification. The estimates convey that AAee exceeds 'MMee for all years for the MTPLAB specification so 2AS is treated as 2 1 S and 2MS is treated as 2 2 S The F ratio is therefore calculated as 1,' 'AAMMAA AA nknk MM MMnk F nk ee ee (4 8) for all years. For both 2AS and 2MS the degrees of freedom are quite large, ranging from 1217 to 2290. In terms of obtaining critical F values, it is appr opriate to consider infinity as the degrees of freedom of both the numerator and the denominator and consequently the critical F value is equal to 1. As the results from the F test convey, one can reject the null hypothesis that 2 A = 2 M at the one percen t alpha level for both the MH specification and the MTPLAB specification T he variances obtained from using the agricultural trade data set are statistically different from the variances obtained by using the manufactured goods trade data set This concl usion holds for all years. Likewise, for the MTPLAB specification one can reject the null hypothesis that 2 A = 2 M at the one percent alpha level for all years. Both the MH specification and the MTPLAB specification of the gravity model should be estimated with disaggregated trade data,

PAGE 103

103 e.g ., agricultural products versus manufactured goods. Before discussin g repercuss ions, however, it is useful to consider a second F test. Testing the Similarity of E stima ted P arameters A related F test is to examine whether the estimated parameters are the same across sectors. Seale (1985 1990) suggested and performed a s imilar test. From an econometric standpoint, this second F test assumes that the previously conducted F test is unable to reject the null hypothesis that the estimated variances are statistically the same across the agricultural trade dataset and the man ufacture d goods trade data set. However, as indicated above, the previous F test rejects this null hypothesis While the null hypothesis is rejected, it is not rejected by much as the calculated F ratios are relat ively close to unity, the critical F valu e for the test. T his would, at least heuristically legitimize proceeding with the second F test. A stacked data set is created by simply stacking the agricultural trade data and corresponding regressors with the manufactured goods trade data and its corresponding regressors. This newly created stacked data set consequently has roughly twice the number of observations as the respective disaggregated data set s. The term roughly is used since the data set s are unbalan ced ( f or a given year the numbe r of observations for bilateral agricultural exports is not necessarily the same as bilateral exports of manufactured goods ) Auxiliary regressions of the gravity model using the stacked data set are estimated These estimations represent the restri cted model as the parameters are constrained to be equal across the two sectors The estimations from the disaggregated dataset s are jointly considered the unrestricted model as the parameters are not restricted to be the same across sectors. By obta ining the sums of squared errors from the restricted and unrestricted models, an F test can be applied to determine if the restrictions hold.

PAGE 104

104 The estimated residuals from the restricted stacked model can be defined in vector form as Re obtained from the auxiliary regressions discussed above. T he sum of squared errors is calculated as RRee Conversely, the estimated residuals from the unrestricted model(s) can be defined in vector form as UR e ,where URAM eee (4 10) Subsequently, the sum of squared errors for this unrestricted model is found by ''' URURAAMM eeeeee. (4 11) The F ratio is constructed as follows ,2'' 2AMRRURUR rnnk URUR AMr F nnk eeee ee (4 12) where r is the number of restrictions and k is th e number of regressors in the restricted model. In this case, r = k since the restricted model constrains all parameters to be equal across product groups. The degrees of freedom for the numera tor can thusly be expressed as either r or k whereas the degrees of freedom for the denominator equal ( nA + nM 2 k ). The F distribution applies since the calculated statistic is the ratio of two chisquare distributed variables. This F ratio tests t he validity of the null hypothesis that all estimated parameters from the gravity model are statistically the same for trade in agricultural products as for trade in manufactured goods against the alternative hypothesis that they are statistically differen t. Restated formally, one tests whether 0: H Estimated parameters are the same across sectors against :aH Not 0H

PAGE 105

105 Since one can define the parameters from the MH specification in vect or form as where 06111'(,...,,,...,) and the parameters from the MTPLAB specification in vector form as where 07111'(,...,,,...,) the hypotheses can be re written as 0:AMH :aAMH for the MH specification, and 0: AM H :aAMH for the MTPLAB model Table A 3 in Appendix A presents the calculated F ratios for the MH specification and Table A 4 presents the corresponding F ratios for the M TPLAB specific ation. The restricted and unrestricted sums of squared errors are presented as are nA, nM, and k The evidence suggests that the parameters are statistically different across product groups at the one percent alpha level as the critical F value is equal to 2.04 in 1992 and 1993 and equal to 1.88 in 1994 through 2001. This conclusion holds for all years for both specification of the gravity model. One can consequently reject the null hypotheses that the estimated parameters from the gravity model are the same for agricultural trade as for manufactured goods trade. The results from this second F test are interesting considering the concern outlined above where the previous F test produced F ratios relatively close to the critical F values. While the similarity of variances across datasets can be rejected with a relatively small margin one can more confidently conclude that the estimated parameters are different across the two datasets. This should lay to rest any concerns regarding the first F test. Thus, i t seems that the two dataset s should not be aggregated when performing gravity estimations at least not for this

PAGE 106

106 sample These two F tests lend support to the idea of estimating the gravity model separately for each sector. Nevertheless one should be cautious of simply discarding the parameter estimates obtained using the aggregate trade dataset First, the gravity literature overwhelming utilizes aggregate trade data. Second, the aggregate dataset also contains the remaining three UNCTAD nomenclatures in addition to agricultural products (Categories 1 and 2 respectively) and manufactured goods (Category 5), namely Category 3: Minerals and Fuels; Category 4: Fuels; and Category 6: Other products. The aggregate dataset would obviously capture t he trade patterns encompassed via these omitted categories. Such relationships would be unaccounted for if the aggregate dataset is abandoned. Henceforth, rather than disposing of the aggregate trade estimates, these F tests provide encouraging support for utilizing disaggregated trade data and aggregate trade data in estimating the gravity model. Analysis of Results: Preliminary Considerations Due to the results from the above conducted F tests, estimations for all three data set s are presented. The outputs from the si xty regressions are offered in T ables 4 2 through 47. Table 41 indexes these regressions for easier reference. Since the dataset encompasses countries of vastly different sizes and subsequent ly great variations in bilateral trade volumes, heteroskedasticity is present. To correct for the occurrence of heteroskedasticity, Whites robust standard errors are estimated. These are obtained using 11 2 1 '''n iii iSE exx (4 13) where, in generic form, represents parameters, X represents a general set of regressors and ei represents the estimated residual for observation i ( Mittelhammer, Judge, and Miller 2000). By obtaining these robust standard errors, the heteroskedasticity problem is avoided as th ey

PAGE 107

107 sufficiently correct for the weight of each error term. The estimated Whites consistent standard errors calculated via Equation 413 are presented along with the parameter estimates in T ables 4 2 through 47. Before addressing the empirical results from the two models, it is useful to cons ider the organization of the following discussion. The econometric performance of the gravity variables in the two models is presented first. Since the MH model, Equation 41, is the primary model, its empirical results are discussed first. After an empirical discussio n about the gravity variables in the M H specification attention is tu rned to the gravity variables in the MTPLAB specification Equation 44, which represents the secondary model Following the di scussion of the gravity variables, the focus shifts to the estimated network effects, or the impact of historical and regional linkages, across both models via the performance of the binary variables in w Given that both models encompass identical w vect ors, the results are presented jointly in a compare and contrast fashion with the empirical performance of the MTPLAB specification serving as a robustness check. Performance of the Gravity Variables Gravity Variables in the MH Specification The evidence suggests that economic gravity does a good job at accounting for hemispheric trade patterns by using the MH model. The model fits the data well with high overall explanatory power (satisfactory r squares and significant F ratios), see Tables 4 2 through 44. The null hypotheses that all estimated parameters are jointly equal to zero are with confidence rejected beyond the one percent significance level using the F test throughout the sample. This conclusion holds for all three data sets. The F test is the appropriate test since the linear restrictions are tested as a ratio of two chi square distributed sums of squared errors. Of course, one has to assume normally distributed error terms.

PAGE 108

108 Judging from the calculated r squar ed values the explanatory variables in the MH model account for 51 to 71 percent of the variation in bilateral trade volumes of the FTAA economies. These values are marginally higher for manufactured goods trade. The manufactured goods dataset also produces higher F ratios, indic ating that the MH model has higher explanatory power when applied to trade in differentiated manufactured goods vis vis the application to agricultural trade data. These results should come as a surprise, as Helpman (1987) initially derived the model f or predicting intra industry trade. Turning to the performance of the individual variables, t he estimated parameters for the absolute differences in GDP per capita, or the difference in relative factor endowments, are remarkably consistent for both aggre gate trade (ranging from 0.588 to 0.831 in magnitudes) and agricultural trade (corresponding range of parameters being 0.234 to 0.527). Similar parameter estimates were found by Egger (2000, 2002) using aggregate trade data. Helpman (1987) obtained negat ive parameter estimates for industry level data; however, he used an intra industry trade index as a dependent variable instead of bilateral exports. The effect of bilateral resource endowment differences diminishes, in both significance and magnitude, for manufactured goods, with parameter estimates of roughly half the magnitude or less. The evidence suggests that agricultural production in the Americas, or at least the agricultural commodities intended for the world market, is more dependent on comparati ve resource endowments, whereas the production of industrial output is less so, relatively speaking. The combined economic mass of the trading partners is captured by the bilateral sums of their respective GDPs. Positive statistically significant param eter estimates suggest, as in previous studies by Breuss and Egger (1999), Egger (2000, 2002), and Helpman (1987), that larger combined economic size results in more trade, ceteris paribus Nevertheless, the

PAGE 109

109 estimated coefficients are somewhat larger in m agnitude as compared to previous investigations, particularly compared to Helpman (1987). The effect of combined economic mass is larger for manufactured goods (2.156 to 2.359) as compared to agricultural commodities (1.812 to 2.142). Trade in manufactures is therefore more dependent on the combined market size, as richer economies tend to buy more manufactured goods which consequently tend to be produced in capital abundant countries. Consequently, large volumes of manufactured goods trade require high income levels of both the exporter and the importer. Although somewhat puzzling, the magnitudes grow marginally larger for aggregate trade (2.794 to 3.011) vis vis the disaggregated trade data. The evidence suggest that the UNCTAD product categories not included in the agricultural or manufactured goods dataset s are more sensitive to combined economic market size. Most noticeably, these products would include minerals, metals, and fuels; trade in these goods is dominated by large economies. The est imated parameters found are of larger magnitude than previous studies. It is worth mentioning, however, that previous MH model investigations are focusing on OECD countries or former Soviet bloc countries in Eastern Europe. It is plausible that intra Eur opean trade flows are vastly different from intra American and cross Atlantic ones. Given that this study excludes pure intra European transactions, these effects would not be captured and different parameter estimates would result. The econometric perfo rmance of the dispersion index suggests a strong positive effect between relative economic sizes and trade volumes. Intriguingly, there are no discernable disparity between agricultural commodities and manufactures in this regard. Nevertheless, a somewha t larger effect is observed for manufactured goods, with estimated parameters ranging from 1.14 to 1.355, whereas the corresponding range for agricultural goods is 0.991 to 1.228.

PAGE 110

110 These minor variations aside, the results do suggest that similar economies have stronger trade relationships, ceteris paribus If one generalizes, the results convey that capital abundant countries trade more extensively with other capital abundant countries and labor abundant countries trade more with other similar countries. As such, much of intra American trade is intra industry based. For aggregate trade, the distance elasticity is relatively close to negative 2 for most years, thus lending support to the social physics rationale for economic gravity, recall Equation 21. However, the effect of distance diminishes when isolating manufactured goods trade and weakens further for agricultural trade. Hence, agricultural trade patterns seem less sensitive to distance and thereby transportation (and transaction) costs. The esti mated parameters for the remoteness variables are statistically vague for aggregate trade. However, for agricultural commodities a statistically significant relationship exists between the relative distance of a country to its trading partners and trade f lows. The exporters relative remoteness positively influences trade volumes (agricultural trade and manufacturing) while the importers has an adverse eff ect when it comes to agricultural trade and to a lesser extent manufactured goods trade. It seems that remote economies are forced to develop highly competitive industries to overcome any obstacles resulting from higher transportation and transaction costs in order to compete globally. This competitive shrewdness, relative to similar less remote econo mies, results in more goods being exported in spite of the natural spatial disadvantage. Conversely, a remote importing country would be forced to be relatively more self sufficient and would, ceteris paribus import lower quantities. Particularly, this relationship tends to be observed when it

PAGE 111

111 comes to agricultural products, which tend to be perishable and bulky and thus expensive to transport. The evidence suggests that adjacency loses its effect after controlling for other variables; the null hypothes is that the estimated parameter is statistically zero cannot be rejected for the vast majority of years. Sharing a common language, or the influence of cultural similarities, has a statistically positive influence on trade volumes. Interestingly, the estimated parameters for the language variable are larger for aggregate trade than for agricultural or manufactured goods trade. Given that the aggregate trade data include more goods than simply agricultural and manufactures, it seems that linguistic common alities exert a stronger impact on trade in minerals, metals, an d fuels than they do for agricultural goods or manufactured products. Before considering the binary network effects in w attention is turned to the econometric performance of the MTPLAB mo del. As mentioned earlier, the w vector is perfectly overlapping between the two models. Gravity V ariables in the MTPLAB S pecification Absolute GDP versus per capita GDP in the MTPLAB s pecification : Before discussing the econometric performance of individual variables, attention is give n to the parameter restriction necessary to include per capita GDP in the MTPLAB specification Specifically, should the MTPLAB gravity mode l be estimated using absolute GDP or should per capita GDP be utilized? The inco rporation of GDP per capita requires the following restriction to be placed on the estimated population elasticities 3 3 1 and 4 4 2, see Equations 34 and 35. The MTPLAB model is estimated twice; once with the restriction in pl ace (i.e., by using per capita GDPs for incomes) and once without the restriction present (i.e., by using absolute GDPs for inc omes). Since highly correlated regressors cause larger standard errors, one can make a determination regarding the appropriateness of the restriction by simply comparing the standard

PAGE 112

112 errors for the estimated coefficients of the population variables from b oth estimations. It should be noted that the standard errors of all other estimated parameters are unaffected by this operation. Tables A 5 through A 7 in Appendix A present the standard errors of the parameters for the natural log of the exporters population and for the natural log of the im porters population obtained when using GDP per capita for income and when using absolute GDP for income. T he standard errors obtained by using the aggregate trade data set are presented in Table A 5. T ables A 6 and A 7 present the corresponding standard errors obtained by using agricultural trade data and manuf actured goods trade data respectively The calculated standard errors obtained from using GDP per capita are also reported in Tables 4 2, 43, and 44. S ince the standard errors for all other est imated parameters are unchanged, they are not presented in Tables A 5 through A 7 to save space. The evidence suggests that the standard errors for the estimated parameters for the exporters population and the i mporters population obtained when using absolute GDP exceed those obtained when using per capita GDP. This conclusion holds uniformly for all three dataset s. One can therefore argue that some degree of multicollinearity is present between absolute GDP and population. Henceforth, utilizing per capita GDP would result in higher statistical significance of the population variables. Similar conclusions were made by Breuss and Egger (1997) and Sandberg, Seale, and Taylor (2006). G iven the relatively high r square values and overall statistically signif icant parameters from Tables 4 5, 46, and 47, evident of lower collinearity, per capita GDP is the preferred variable for income. Performance of gravity variables in the MTPLAB s pecification : The MTPLAB model fits the data well, with high over all explanatory power. In addition to producing significant F -

PAGE 113

113 ratios, the explanatory variables in the MTPLAB model accounts for 53 to 73 percent of the variation in bilateral trade volumes of the FTAA economies. The evidence suggests that the fit to manufactured goods trade data is superior to the fit to agricultural trade data. As c onve yed in Tables 4 5 through 47, the gravity variables in the MTPLAB specification (GDPs per capita, populations, and distance) perform in compliance with a priori expectations. Beyond the one percent significance level, per capita incomes and populations exert statistically significant positive effects on international trade flows and the geographic distance a statistically signi ficant negative effect. The estimated elasticities with respect to per capita income of the exporter range from 1.846 to 2.241 for aggregate trade, from 1.01 to 1.236 for agricultural trade, and from 1.60 to 1.824 for trade in manufactured goods. While t here is no clear over all temporal trend in these variations, the larger values are observed for the early years. As stated in the previous chapter the sample of countries is different for 1992 and 1993, most noticeably missing much of the intra CARICOM trade due to data availability. The estimated elasticities with respect to per capita income of the importer range from 1.051 to 1.309 for aggregate trade, from 0.965 to 1.098 for agricultural trade, and from 0.86 to 0.938 for trade in manufactured goods. The estimated parameters for the distance variable suggest a corresponding relationship vis vis the MH specification The distance elastiticies are estimated to be around negative 2 for aggregate trade and they are subsequently diminishing for manufa ctured goods and yet again even more so for agricultural commodities. Even though Polak (1996), among others, argued for their inclusion, the remoteness (or relatively distance) variables perform ambiguously throughout the sample. Unlike when using the M H model, t here is n o statistical evidence that remoteness affects bilateral trade flows after controlling for other factors. This is in contrast with Frankel,

PAGE 114

114 Stein, and Wei (1997) who found statistically significant results using a variation of the MTPLA B specification The results convey that t rade in agricultural products is more sensitive to the per capita income of the importer (absorptive capacity) than to that of the exporter (productive capacity). This would indicate that agricultural trade is r elatively demand driven with economic conditions of the importer exerting greater influence vis vis the economic conditions of the exporter. Much of the exports from the regions developing economies i.e., the Caribbean and Latin America, consist of agricultural products granted favorable treatment by the United States, Canada, and the European Union, via, for example, the Gene ral System of Preferences (GSP) Their exports ( primarily agricultural in nature) enter the North American and European markets with relatively low, if any, import duties. Consequently, the importers income should e xert a greater influence on agricultural trade, as the United States, Canada, and Western Europe have relatively high income levels relative to the countries in La tin America and the Caribbean. Furthermore, considering the diminished impact of distance on agricultural trade, as these often bulky products are presumably shipped large distances to reach the developed markets, the absorptive capacity of the importer o utweighs the resista nce provided by geographic location. Manufactured goods trade in the Americas conversely, is more sensitive to per capita income of the exporter with statistically higher elasticities observed throughout the sample. These results conf orm to HOtheory as manufactured goods tend to be more capital intensive and capital abundant countries tend to export capital intensive goods.1 I t logically follows that industrial output has higher distance elasticity, as there are fewer preferential access dist ortions observed. Manufactured products are therefore more subjected to the gravitational resistance of 1 Recall that Bergstrand (1985, 1989) suggested that the per capita income of the exporter serves as a proxy for its capital labor ratio.

PAGE 115

115 geography and traditional economic factors, which would be the case for, say, industrial exports from Chile or Brazil shipped to the Unit ed St ates or Europe These consignment s are not granted the same GSP treatment as primary exports originating in the Caribbean basin. Statistically significant throughout the sample, the estimated parameters for the population variables, all positive, are gre ater in magnitude for the exporter than for the importer This holds for both aggregate trade and manufactured goods trade. The evidence suggests that trade in manufactured goods is more sensitive to the economies of scale of the exporter. For agricultu ral trade, however, the estimated population parameters are not statistically different across exporters and importers; in fact the estimated parameters fall within the same confidence intervals. Perhaps relative scale economies are less common, or certai nly less influential, in the agricultural sector vis vis manufacturing. Considering the dominance of the agricultural sector in the smaller Caribbean economies, the results are not all that surprising An islandeconomy, or microstate, has a natural bo undary of how much economies of scale can be realized given its limited physical size and smaller population. On the other hand, large economies like the United States and Brazil, with plentiful land, significant capital stocks, and sizeable populations, have virtually limitless potentials for economies of scale, albeit constrained by available technology. The evidence suggests that the effects of adjacency and cultural similarities (as proxied by sharing a common language) are similar to those found usi ng the MH model. Particularly adjacency loses its effect after controlling for other variables. The null hypothesis that the estimated parameter is statistically zero cannot be rejected for the vast majority of years. This conclusion holds for all thre e dataset s. Given that significant portion of the hemispheric trade flows are transported via sea (island nations, cross oceanic trade, etc), adjacency should lose some of its proposed influence. Sharing a common language, or the influence of cultural

PAGE 116

116 s imilarities, has a statistically positive effect on trade volumes. The effects are robust across product groups. Network Effects Interpretation of Binary Variables The discussion will now shift to the empirical performance of the binary variables in ve ctor w for both specifications of the model. Since the w vectors are congruent across the two models, the results from the MTPLAB specification serve as a check for robustness. The binary effects of history and regionalism measure how any trade volume de viates from a presumably normal or ba seline, trade behavior; they measure the magnitu de of any distortions or biases. The baseline trade behavior is where all eleven binary variables in vector w are jointly equal to 0. In th is study the benchmark behavior is consequently that between two countries in the sample that do not share a common border, do not share a common language, with no imperial history, are not members of the same trade agreement ( NAFTA CARICOM MERCOSUR Andean Pact, or CACM ), and t he importing country is not a member of the European Union. Thus, an example of the baseline trade behavior would, for instance, be the commercial interaction between Canada and Argentina or that between Brazil and Australia. Reassuringly, the estimated parameters from Tables 4 2 through 47 suggest that the effects of historical and regional linkages are robust across both model specifications. For interpretive purposes, it useful to consider the average parameter estimate for any particular variable. The average parameter estimates as presented in Tables 4 8 and 49, are the arithmetic mean of the statistically non zero estimates. These average parameter values for the MH specification are calculated as follows: m n m n (4 13)

PAGE 117

117 where m indicates a given binary variable and n is the number of years for which the parameter estimate is statistically different from zero. The marginal effect of these binary variables is subsequently equal to the baseline trade volume multiplied times a factor of me for the MH specification For the benchmark trade behavior trading partners do not have any of thes e qualitative factors in w in common such that Wm = 0 for all m This effect would enter the M H g ravity model as (0) 0 1 m ee which indicates that the baseline volume should be multiplied by a factor of 1, or the bilateral trade volume simply represents the baseline volume itself. For the analysis pertaining to the MTPLAB specification the is interchanged for in the above notation. An alternative way of analyzing the estimated historical and regional effects is to consider the estimated parameters for the variables in w graphically over time. Graphs in Appendix B and Appendix C are providing this representation. A similar presentation was utilized by Sandberg, Seale, and Taylor (2006). Analysis of Results: Historical Networks The average parameter values and marginal effects of the historical binaries are presented in Table 4 8 and the graphical representation can be found in Figures B 1 through B 6 in Appendix B. Both formulations of the model suggest that former colonial linkages have a clear influence on hemispheric trade patterns. The effects of neo colonial trade distortions are robust across the two models. T he evidence conveyed by both specifications is similar albeit the estimates from the MH specification are marginally larger. Differences are rather observed across sectors There is strong evidence of neo colonial trade distortions between the former British dependencies in the Western Hemisphere and the former metropolitan ruler, with estimated

PAGE 118

118 parameters found to be positive and statistically different from zero. Former British colonies trade with the U.K by quite the large magnitude (up to a staggering 40 times the baseline volume !). It should nevertheless be noted that most of the estimated marginal effects are in the range of 3.2 to 18.7 t imes the baseline volume The magnitude s, however, are significantly lower t han was found by Sandberg, Seale, a nd Taylor (2006), who fitted a variation of the MTPLAB model to a Caribbean centered dataset a group of countrie s that all have neo colonial trade ties to the U.K Even if the baseline trade behavior is somewhat obtuse, it does not diminish the clear neo colonial trade distortion exerted by the U.K. The favorable treatment extended by the U.K. to its former dependencies, relative to other nonEuropean trading partners, seems to greatly impact trade volumes, especially w hen it comes to agricultural products. This effect, however, does diminish for manufactured goods as compared to agricultural commodities. One can certainly make the argument that Commonwealth preferential treatment is counter productive in that it reinf orces the dependence upon agricultural exports. Most of the countries in the Caribbean basin ( mostly former British colonies) are reliant upon primary exports and these preferences quite possibly increase the opportunity cost of structural transformation and industrialization of their economies, particularly when considering the stronger effect observed for agriculture. The estimates suggest that former Spanish colonies (i.e., most of Central and South America) trade between 1.7 to 2.9 times as much with Spain relative to the baseline trade volume Spains neocolonial influence seems to be marginally stronger for the MH specification, with the notable exception being agricultural goods for which the MTPLAB formulation results in a larger corollary. For the former Portuguese colony in the sample (i.e.,

PAGE 119

119 Brazil) there is no neo colonial distortion present, except for agricultural commodities. On the contrary, for aggregate trade, Portugal is less engaged with its former colony vis vis the baseline trade volume (statistically significant negative parameters for both specifications) It should be noted that the Portuguese network effect for the MH specification using manufactured goods coincides with the baseline reference, since the estimated parameters are not statistically different from zero. A possible explanation for these difference s and the particularly strong British linkage could be that Englands former colonies gained independence rela tively recently (see Table 1 3). M any of the Caribbean Is lands gained independence as recent ly as the mid part of the past century, as compared to the former Iberian depend encies who were granted autonomy centuries ago. Addi tionally, British trade policy is more specifically targeted toward eng aging its former dependencies. As a result, stronger neo co l onial trade linkages would be observed for former and current members of the Commonwealth. This evidence supports the postulations provided by Brysk, Parsons, and Sandholz (2002) and Grier (1999) that the U.K. h as maintained, and indeed encouraged, stronger post colonial trade relationships with its former colonies relative to Spain, Portugal, and France. Across the two models, the historical effects seem to be marginally stronger for the MH specification. It s eems that a portion of these neocolonial influences is already captured by the gravity variables of the MTPLAB specification F or instance, per capita income might measure some of this effect, especially when it comes to the importing party; t he former i mperial rulers in Europe enjoy higher per capita incomes than their former dependencies. Consequently, one of t he appeal s of the MH specification is that the gravity variables do not interfere with the binary

PAGE 120

120 effects and the MH model may provide a mor e ac curate, or pure assessment of these hist orical networks, particularly as one attempts to isolate and identify their effects. Analysis of Results: Regional Networks Turning the attention to the effects of regionalism, Table 4 9 presents the calculated av erage parameter values and the marginal effects. The graphical presentation of the parameters is found in Figures C 1 through C 6 in Appendix C. When considering the empirical performance of the regional variables across the two models, the results are again, surp risingly robust. H owever, one can discern from the evidence that the trend is reversed as compared to the historical effects with the MTPLAB model suggesting marginal ly larger regional effects than the MH model For agricultural trade and man ufactured goods trade, the NAFTA binary performs poorly. Only for aggregate trade does the parameter become significantly different from zero with a negative average estimated parameter Startlingly, it seems that the gravity variables (i.e., economic ci rcumstances and geographic proximity) already capture the presumed effects of NAFTA in that the U.S.CanadaMexico tradenexus appears strong enough given natural conditions that any statistical effect from the NAFTA agreement disappears (or even becomes negative!). For agricultural trade and manufactured goods trade, the NAFTA members behave in accordance to the baseline volume Given the disparity observed regarding the aggregate data set i t is plausible that the NAFTA countries even over trade when it comes to minerals, metals, and fuels product categories also included in the aggregate trade dataset Indeed, Canada is the United States biggest supplier of fuels (Inter Ame rican Development Bank 2003). A striking phenomenon is that smaller ec onomies tend to stick together, with stronger effects observed for CARICOM CACM and the Andean Pact than for NAFTA or MERCOSUR Furthermore, the evidence suggests stronger regional tendencies in the Central

PAGE 121

121 region of the hemisphere. CARICOM CACM and the Andean Pact have greater influence for agricultural commodities than for manufactured goods and aggregate trade, thus supporting the idea that many of these economies are highly specialized in, and thereby dependent upon, agricultural exports. By far, CARICOM seems to exert the greatest influence on the trade behavior of its members. Intra CARICOM trade volumes are significantly biased on a magnitude of 16.64 to 72.3 times that of the models presumed normal patterns. The argument can certainly be mad e that small island economies are more trade dependent vis vis larger more self sufficient economies. It seems that geography, economic s, resources, politics, and policy cause these countries to exchange larger volumes. Statistically significant positive parameters were also found by among others, Sandberg, Seale, and Taylor (2006) and Thoumi (1989a, 1989b). MERCOSUR exerts a statistically significant effect on agricultural trade for both specifications, indicating a modest relative distortion of trade patterns. This effect is also present for manufactured goods using the MH model For the remaining three instances the effect diminishes to zero indicating that intra MERCOSUR trade conforms to the baseline trade volume The effect of the An dean Pact, consisting of medium sized economies, is somewhere in between the ambiguous effects of NAFTA and MERCOSUR and the stronger effects of CARICOM and CACM thus lending further support to the inverse relationship between the economic size of members and regional dependency. The MTPLAB specification provides marginally larger effects for the regional linkages whereas the MH specification suggests marginally lower effects. It seems that the MH formulation already accounts for some of the regional effect s, whereas the MTPLAB specification s gravity variables do not. Given MH models more formal foundation in intra -

PAGE 122

122 industry trade, this supports the idea that the goods traded regionally within the hemisphere would be intra industry traded goods. Consequen tly, regional effects would appear smaller if intra industry trade is already accounted for via the MH specification. This should be contrasted with the finding that the MTPLAB specification already accounts for some of the historical linkages by controll ing for the absorptive capacity of the importer and the capital labor ratio s (e.g., per capita incomes) of the trading partners Inter regional trade would subsequently be inter industry trade. An argument can be made that the goods exported to the U.K. by its former colonies in the Caribbean basin are products not produced within the European Union. These trade flows, as a result, would be inter industry trade. Using a variation of the MTPLAB gravity model and disaggregated trade data Vollrath, Halla han, and Gehlhar (2006) found that NAFTA has not exerted a statistically significant effect on bilateral trade in agricultural commodities; however, they did find a positive statistically significant relationship for processed food products. Furthermore, Vollrath, Hallahan, and Gehlhar (2006 ) found that MERCOSUR has exerted a positive influence on trade patterns over the past decade with a marginally larger effect observed for processed food products vis vis agricultural commodities. It should be noted that the magnitudes of the estimated parameters for agricultural commodities found in this study are similar to those of Vollrath, Hallahan, and Gehlhar (2006). Statistically ambiguous parameters for NAFTA MERCOSUR and the Andean Pact were obtai ned by Gr ant and Lambert (2008) for both agricultural and nonagricultural trade using the MTPLAB model Soloaga and Winters (2001) however, suggest ed positive relationships for NAFTA MERCOSUR Andean Pact, and CACM fitting the MTPLAB specification to aggregate trade data.

PAGE 123

123 Empirical Remarks The fact that the NAFTA variables perform ambiguously in this study is somewhat discerning. Anecdotal evidence provided by Table 12 along with stylized knowledge suggest that trade between the U.S. and Mexico has increased since NAFTAs formation. Most analysts would argue that agricultural trade has increased as a result of NAFTA. It is possible that the empirical effect of NAFTA gets lost in the noise contained in the data sets. By using an FTAA centered dataset, of which Mexico is only a fraction, the influence of NAFTA might be hard to disentangle. To remedy this potential problem, three Mexico centered datasets are created from the original data. By the term Mexico centered means that any transactions not invo lving Mexico ar e omitted from the data sets; all observations contain Mexico as either the exporter or the importer. As an empiri cal exercise, the models are reestimated using these newly created datasets. The results are reported in Appendix D. The evidence provided by these Mexico centered datasets is intriguing. Using the agricultural trade data, the NAFTA binary effect is positive and statistically significant for all years when re estimating the MH model ( and for all years except 1994 when estim ating the MTPLAB model) However, when re estimating the models using aggregate trade and manufactured goods trade, the results are ambiguous. This indicates that NAFTA has had a significant effect o n Mexicos bilateral trade behavior at least when it c omes to agricul ture. A more detailed discussion is provided in Appendix D. D oes the empirical specification of the gravity model matter? To a certain extent it does not: the MH specification produces robust binary network effects as the estimated eff ects obtained from the MTPLAB specification are similar to those from the MH specification. Nevertheless, there are subtle differences. Appendix E provides statistical testing to assess which econometric specification of the gravity model is appropriate given these data. However,

PAGE 124

124 as Seale (1994) suggested, when discriminating between non nested specifications of a model, the model should be chosen based on economic theory as an alternative to empirical testing when faced with nonnested specifications. When considering a group of countries (e.g., not a global sample of countries), it seems that the MH model is the more relevant specification Helpman (1987) derived the model to be applicable when considering a sub set of the world economy. Given that t he sample for this study consists of 64 counties and that the data are FTAA centered (see Chapter 3 for details), the MH specification is a more suitable model.

PAGE 125

125 Table 4 1. Regression index Table Empirical specification Data set Table 4 2 MH specifica tion Aggregate trade (UNCTAD total) Table 4 3 MH specification Agricultural trade (UNCTAD cat. 1 & 2) Table 4 4 MH specification Manufactured goods trade (UNCTAD cat. 5) Table 4 5 MTPLAB specification Aggregate trade (UNCTAD total) Table 4 6 MTPLAB spe cification Agricultural trade (UNCTAD cat. 1 & 2) Table 4 7 MTPLAB specification Manufactured goods trade (UNCTAD cat. 5)

PAGE 126

126 Table 4 2. MH specification: aggregate trade (UNCTAD total) Dependent variable: ln(exports from country i to country j in U.S. dol lars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Constant 51.007*** 41.612*** 29.648*** 32.625*** 39.603*** 39.27*** 36.027*** 43.294*** 43.158*** 49.167*** (6.814) (6.859) (5.689) (6.151) (5.930) (5.379) (4.827) (5.309) (5.032) (4.94 7) Bilateral GDP difference a 0.831*** 0.779*** 0.588*** 0.593*** 0.607*** 0.635*** 0.629*** 0.629*** 0.638*** 0.617*** (0.128) (0.126) (0.089) (0.088) (0.089) (0.081) (0.083) (0.085) (0.083) (0.085) Bilateral sum of GDPs b 2.975*** 3.011*** 2.832*** 2.8 53*** 2.877*** 2.897*** 2.845*** 2.836*** 2.794*** 2.832*** (0.085) (0.081) (0.056) (0.056) (0.060) (0.054) (0.056) (0.058) (0.056) (0.060) Bilateral similarity of GDPs c 1.772*** 1.926*** 1.704*** 1.730*** 1.713*** 1.698*** 1.743*** 1.648*** 1.685*** 1. 690*** (0.092) (0.093) (0.049) (0.051) (0.054) (0.044) (0.048) (0.051) (0.048) (0.050) Distance d 2.041*** 1.714*** 1.991*** 2.228*** 2.292*** 2.409*** 1.947*** 1.914*** 2.074*** 2.011*** (0.194) (0.201) (0.130) (0.126) (0.136) (0.118) (0.132) (0.134) (0.125) (0.137) Remoteness exporter e 0.829** 0.128 0.467* 0.217 0.735*** 0.929*** 1.257*** 1.692*** 1.876*** 2.213*** (0.331) (0.345) (0.262) (0.256) (0.274) (0.257) (0.246) (0.250) (0.242) (0.251) Remoteness importer e 0.154 0.584 1.424*** 0.682 0.440 0.609 1.594*** 1.258*** 1.145*** 0.973** (0.611) (0.597) (0.506) (0.560) (0.511) (0.475) (0.418) (0.449) (0.426) (0.400) Common border f 0.461 1.148*** 0.492 0.063 0.429 0.978** 0.045 0.005 0.212 0.113 (0.370) (0.390) (0.338) (0.290) (0.276) (0.460) (0.290) (0.269) (0.369) (0.267) Common language g 1.772*** 1.937*** 1.041*** 1.012*** 0.820*** 0.731*** 0.929*** 1.001*** 0.681*** 0.935*** (0.274) (0.301) (0.212) (0.213) (0.215) (0.197) (0.199) (0.209) (0.202) (0.205) Colonial linkage: U.K. h --4.450*** 4.111*** 4.309*** 3.920*** 2.945*** 3.048*** 3.373*** 3.493*** --(0.469) (0.478) (0.511) (0.561) (0.950) (0.658) (0.584) (0.759) Colonial linkage: Spain k 0.466 0.505 0.898*** 0.865*** 1.066*** 1.490*** 1.109*** 0.933 *** 1.239*** 1.010*** (0.345) (0.373) (0.294) (0.299) (0.308) (0.303) (0.296) (0.289) (0.268) (0.270) Colonial linkage: Portugal l 1.776*** 1.712*** 0.287 0.459 0.386 0.069 0.137 0.239 0.030 0.046 (0.809) (0.628) (0.857) (0.734) (0.500) (0.674) (0.559) (0.477) (0.427) (0.460) NAFTA m --3.375*** 2.920*** 2.463*** 2.697*** 2.855*** 2.488*** 2.522*** 2.738*** --(0.767) (0.795) (0.722) (0.689) (0.678) (0.683) (0.646) (0.673) CARICOM m --4.373*** 3.947*** 3.887*** 4.009*** 2.629 *** 3.773*** 3.543*** 2.981*** --(0.492) (0.450) (0.487) (0.411) (0.545) (0.529) (0.508) (0.595)

PAGE 127

127 Table 4 2. Continued. Dependent variable: ln(exports from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 MERCOSUR m 0.355 0.046 0.246 0.341 0.106 0.077 0.219 0.222 0.067 0.497 (0.692) (0.663) (0.594) (0.577) (0.588) (0.660) (0.530) (0.541) (0.576) (0.538) Andean Pact m 0.337 0.346 0.457 0.273 0.739** 0.847** 0.915*** 1.135*** 1.089*** 1.140*** (0.316) (0.346) (0.347) (0.394) (0.333) (0.372) (0.314) (0.319) (0.319) (0.278) CACM m 1.871*** 1.972*** 1.372*** 1.040*** 1.387*** 1.357*** 1.738*** 1.850*** 1.522*** 1.389*** (0.401) (0.427) (0.372) (0.342) (0.344) (0.359) (0.344) (0.353) (0.330) (0.325) EU importer n 0.561 0.483 2.121*** 1.573*** 1.497*** 1.983*** 2.235*** 1.927*** 1.721*** 1.814*** (0.396) (0.341) (0.306) (0.324) (0.312) (0.291) (0.266) (0.275) (0.251) (0.251) R squared 0.527 0.552 0.593 0.583 0.590 0.596 0.591 0.594 0.605 0.631 F statistic 114.728 126.953 190.615 186.034 177.049 216.323 205.442 186.652 211.690 210.935 N 1554 1560 2374 2408 2236 2657 2582 2315 2507 2243 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms *) Statistically significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level. a) lnY Y jt it NN itjt b) ln YY itjt c ) 22 ln1 Y Y jt it YYYY itjtitjt d) Bilateral distance in kilometers between the capitals of trading partners, e) Remoteness it Y Yjt d ij ji wt Binary variables: f) Equal to 1 if the two trading partners are contingent, i.e., they sha re a common border, 0 otherwise. g) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. h) Equal to 1 if one of the trading partners is the U.K. and the other a former British colony in the Western Hemisphere, 0 otherw ise. (Note that the United States and Canada are not considered former colonies). k) Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere, 0 otherwise. l) Equal to 1 if one of the trading par tners is Portugal and the other a former Portuguese colony in the Western Hemisphere, 0 otherwise. m) Integration/regional dummies: Equal to 1 if both trading partners are members a specific trading agreement, (NAFTA, CARICOM, MERCOSUR, Andean Pact, or C ACM respectively), 0 otherwise. n) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 128

128 Table 4 3. MH specification: agricultural trade (UNCTAD categories 1 & 2) Dependent variable: ln(exports of agricultural products from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Constant 27.077*** 32.445*** 25.701*** 29.718*** 31.023*** 38.017*** 31.503*** 30.208*** 28.340*** 40.295*** (4.167) (3.993) (3.591) (3.747) (3.763) ( 4.180) (3.779) (3.756) (3.168) (4.070) Bilateral GDP difference a 0.350*** 0.447*** 0.356*** 0.415*** 0.393*** 0.505*** 0.527*** 0.331*** 0.352*** 0.234*** (0.078) (0.073) (0.057) (0.057) (0.056) (0.070) (0.066) (0.061) (0.051) (0.066) Bilateral sum of GDPs b 1.812*** 1.832*** 1.847*** 1.928*** 1.939*** 2.142*** 2.013*** 1.921*** 1.813*** 2.051*** (0.055) (0.052) (0.038) (0.040) (0.043) (0.050) (0.048) (0.042) (0.037) (0.052) Bilateral similarity of GDPs c 0.991*** 1.091*** 1.061*** 1.092*** 1.124*** 1. 228*** 1.154*** 1.036*** 1.034*** 1.110*** (0.054) (0.053) (0.034) (0.033) (0.042) (0.041) (0.042) (0.034) (0.032) (0.043) Distance d 1.265*** 1.382*** 1.337*** 1.324*** 1.571*** 1.575*** 1.567*** 1.177*** 1.322*** 1.533*** (0.132) (0.131) (0 .081) (0.086) (0.098) (0.102) (0.092) (0.095) (0.080) (0.112) Remoteness exporter e 1.621*** 1.678*** 1.341*** 1.604*** 1.799*** 1.974*** 1.949*** 2.112*** 1.824*** 2.425*** (0.258) (0.256) (0.213) (0.208) (0.215) (0.237) (0.220) (0.198) (0.190) (0.237) Remoteness importer e 0.898*** 0.300 0.784*** 0.869*** 0.677** 0.695** 1.029*** 1.462*** 0.905*** 0.630** (0.319) (0.313) (0.284) (0.303) (0.284) (0.308) (0.281) (0.296) (0.235) (0.267) Common border f 0.845*** 0.659*** 0.395 0.277 0.078 0.29 9 0.055 0.338 0.322* 0.055 (0.262) (0.245) (0.211) (0.194) (0.236) (0.226) (0.219) (0.225) (0.194) (0.239) Common language g 0.319 0.171 0.605*** 0.862*** 0.462*** 0.765*** 0.614*** 0.915*** 0.666*** 0.545*** (0.229) (0.225) (0.151) (0.154) (0.172) (0 .183) (0.174) (0.156) (0.140) (0.181) Colonial linkage: U.K. h --3.480*** 3.201*** 3.459*** 3.338*** 3.017*** 2.009*** 2.452*** 2.503*** --(0.336) (0.320) (0.432) (0.424) (0.438) (0.429) (0.401) (0.612) Colonial linkage: Spain k 0.573* 0.794*** 0 .577** 0.599** 0.771*** 0.808*** 0.946*** 0.522** 0.697*** 0.949*** (0.302) (0.306) (0.255) (0.247) (0.253) (0.274) (0.287) (0.250) (0.224) (0.243) Colonial linkage: Portugal l 0.864*** 1.310*** 1.010*** 1.030*** 1.125*** 0.979*** 1.231*** 0.782* 1.266** 1.517*** (0.228) (0.280) (0.160) (0.160) (0.432) (0.356) (0.228) (0.436) (0.283) (0.231) NAFTA m --0.419 0.487 0.492 0.644 0.194 0.206 0.272 0.163 --(0.434) (0.446) (0.435) (0.492) (0.476) (0.494) (0.433) (0.510) CARICOM m --3.188*** 2 .882*** 2.225*** 3.291*** 3.227*** 3.289*** 2.850*** 3.114*** --(0.293) (0.330) (0.398) (0.431) (0.360) (0.385) (0.351) (0.377)

PAGE 129

129 Table 4 3. Continued. Dependent variable: ln(exports of agricultural prod ucts from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 MERCOSUR m 0.702* 0.676 0.954** 1.205 1.067** 0.953* 1.308*** 1.591*** 1.238*** 1.351*** (0.387) (0.440) (0.446) (0.458) (0.494) (0.504) (0.470) (0.453) (0.451) (0.475) Andean Pact m 0.250 0.624 0.619 0.904*** 0.607 0.832*** 1.297*** 1.481*** 1.192*** 1.595*** (0.481) (0.473) (0.329) (0.295) (0.528) (0.312) (0.306) (0.293) (0.257) (0.323) CACM m 2.190*** 2.131*** 1.789*** 2.021*** 1.797*** 2.130*** 2.089*** 2.626*** 2.108*** 2.395*** (0.338) (0.307) (0.271) (0.242) (0.269) (0.303) (0.268) (0.280) (0.242) (0.286) EU importer n 0.095* 0.515** 0.028 0.061 0.030 0.059 0.179 0.092 0.129 0.250 (0.233) (0.218) (0.203) (0.212) (0.202) (0.220) (0.217) (0.1 66) (0.171) (0.170) R squared 0.509 0.526 0.618 0.622 0.604 0.569 0.570 0.589 0.621 0.594 F statistic 84.313 92.183 161.788 169.041 148.223 151.996 150.745 148.227 181.990 150.497 N 1233 1263 1815 1869 1769 2091 2061 1883 2021 1867 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms. *) Statistically signif icant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level. a) lnY Y jt it NN itjt b ) ln YY itjt c ) 22 ln1 Y Y jt it YYYY itjtitjt d ) Bilateral distance in kilometers between the capitals of trading partners, e ) Remoteness it Y Yjt d ij ji wt Binary variables: f ) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. g) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. h) Equal to 1 if one of the trading partners is the U.K. and the other a f ormer British colony in the Western Hemisphere, 0 otherwise. (Note that the United States and Canada are not considered former colonies). k) Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemispher e, 0 otherwise. l) Equal to 1 if one of the trading partners is Portugal and the other a former Portuguese colony in the Western Hemisphere, 0 otherwise. m) Integration/regional dummies: Equal to 1 if both trading partners are members a specific trading agreement, (NAFTA, CARICOM, MERCOSUR, Andean Pact, or CACM respectively), 0 otherwise. n) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 130

130 Tab le 4 4. MH specification: manufactured goods trade (UNCTAD category 5) Depe ndent variable: ln(exports of manufactured goods from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Constant 28.542*** 26.304*** 18.049*** 22.337*** 25.294*** 25.767*** 20.340*** 22.216*** 20.144*** 31.198*** (3.932) (4.083) (3.580) (3.541) (3.691) (3.897) (3.689) (3.985) (3.238) (3.997) Bilateral GDP difference a 0.262*** 0.264*** 0.101* 0.094* 0.178*** 0.179*** 0.138** 0.069*** 0.0004 0.144** (0.069) (0.074) (0.054) (0.052) (0.055) (0.066) (0.06 2) (0.062) (0.053) (0.063) Bilateral sum of GDPs b 2.187*** 2.221*** 2.156*** 2.210*** 2.229*** 2.344*** 2.287*** 2.300*** 2.166*** 2.359*** (0.049) (0.047) (0.034) (0.035) (0.037) (0.043) (0.041) (0.041) (0.036) (0.046) Bilateral similarity of GDPs c 1. 164*** 1.247*** 1.170*** 1.183*** 1.185*** 1.267*** 1.269*** 1.266*** 1.140*** 1.335*** (0.047) (0.053) (0.030) (0.027) (0.032) (0.034) (0.034) (0.036) (0.029) (0.042) Distance d 1.215*** 1.293*** 1.406*** 1.454*** 1.471*** 1.563*** 1.376*** 1.24 5*** 1.250*** 1.303*** (0.119) (0.119) (0.079) (0.084) (0.082) (0.095) (0.093) (0.102) (0.076) (0.111) Remoteness exporter e 0.665*** 1.241*** 1.443*** 1.300*** 1.172*** 0.969*** 0.688*** 0.611*** 0.556*** 0.023 (0.210) (0.210) (0.171) (0.1 62) (0.167) (0.190) (0.181) (0.190) (0.206) (0.201) Remoteness importer e 0.548 0.870** 0.470 0.703** 0.836*** 0.454 0.426 0.457 0.348 0.136 (0.338) (0.361) (0.312) (0.310) (0.312) (0.333) (0.314) (0.341) (0.255) (0.315) Common border f 0.537* 0.813 *** 0.360 0.067 0.102 0.143 0.0003 0.283 0.402 0.260 (0.306) (0.283) (0.254) (0.282) (0.271) (0.365) (0.374) (0.275) (0.323) (0.273) Common language g 1.052*** 0.944*** 0.717*** 0.795*** 0.818*** 0.894*** 0.822*** 0.881*** 0.689*** 0.835*** (0.206) (0.192) (0.134) (0.126) (0.129) (0.142) (0.144) (0.147) (0.121) (0.169) Colonial linkage: U.K. h --1.496*** 1.510*** 1.190** 1.270*** 1.447*** 1.415*** 1.304** 2.587*** --(0.395) (0.335) (0.481) (0.424) (0.378) (0.411) (0.561) (0.532) Colonial l inkage: Spain k 0.168 0.320 0.697*** 0.692*** 0.600** 0.902*** 0.649*** 0.604** 0.801** 0.840*** (0.263) (0.249) (0.217) (0.240) (0.268) (0.289) (0.241) (0.243) (0.394) (0.239) Colonial linkage: Portugal l 0.013 0.237 0.488 0.243 0.168 0.577 0.558 0.551 0.825 0.595 (1.557) (1.514) (1.466) (1.321) (1.104) (1.242) (0.984) (1.057) (1.490) (0.931) NAFTA m --0.396 0.048 0.231 0.324 0.308 0.329 0.100 0.642 --(0.513) (0.636) (0.616) (0.634) (0.591) (0.591) (0.894) (0.566) CARICOM m --3.063 *** 2.601*** 2.661*** 2.667*** 2.601*** 2.676*** 2.499*** 2.705*** --(0.311) (0.347) (0.342) (0.359) (0.348) (0.362) (0.286) (0.458)

PAGE 131

131 Table 4 4. Continued. Dependent variable: ln(exports of manufactured goods from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 MERCOSUR m 0.783 0.553 0.656 0.609 0.854* 0.776 1.008** 1.007** 0.923 1.323*** (0.478) (0.448) (0.435) (0.426) (0.442) (0.503) (0.466) (0.437) (0.667) (0.437) A ndean Pact m 0.496** 0.589*** 0.503 0.350 0.847*** 0.710** 0.805*** 0.937*** 0.988** 1.113*** (0.249) (0.228) (0.334) (0.358) (0.273) (0.304) (0.295) (0.267) (0.502) (0.227) CACM m 2.571*** 2.150*** 1.968*** 2.005*** 2.153*** 2.119*** 2.259*** 2.378*** 2. 145*** 2.173*** (0.322) (0.332) (0.289) (0.281) (0.291) (0.316) (0.320) (0.323) (0.518) (0.312) EU importer n 1.425*** 1.325*** 1.565*** 1.657*** 1.347*** 2.008*** 2.171*** 2.139*** 2.095*** 2.165*** (0.218) (0.217) (0.201) (0.186) (0.181) (0 .213) (0.178) (0.191) (0.165) (0.205) R squared 0.632 0.635 0.694 0.712 0.694 0.647 0.658 0.659 0.688 0.661 F statistic 152.733 156.519 252.330 278.950 245.328 233.122 238.580 216.593 267.517 215.069 N 1350 1364 2025 2051 1963 2308 2250 2033 2204 2006 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms *) Statistically significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at th e 1 % level. a) lnY Y jt it NN itjt b) ln YY itjt c ) 22 ln1 Y Y jt it YYYY itjtitjt d ) Bilateral distance in kilometers between the capitals of trading partners, e) Remoteness it Y Yjt d ij ji wt Binary variables : f) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. g) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. h) Equal to 1 if one of the trading partners is the U.K. and the other a former British colony in the Western Hemisphere, 0 otherwise. (Note that the United States and Canada are not considered former colonies). k) Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere, 0 otherwise. l) Equal to 1 if one of the trading partners is Portugal and the other a former Portuguese colony in the Western Hemisphere, 0 otherwise. m) Integration/regional dummies: Equal to 1 if both trading partners are members a specific trading agreement, (NAFTA, CARICOM, MERCOSUR, Andean Pact, or CACM respectively), 0 otherwise. n) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 132

132 Table 45. MTPLAB specification: a ggregate trade (UNCTAD tot al) Dependent variable: ln(exports from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Constant 55.099*** 46.949*** 56.838*** 58.040*** 58.914*** 60.092*** 57.846*** 53.689*** 59.040*** 61.900*** (6 .251) (6.335) (4.487) (5.057) (5.081) (4.665) (4.359) (4.730) (4.246) (4.307) GDP per capita exporter 2.241*** 2.167*** 1.975*** 2.038*** 1.987*** 1.999*** 1.873*** 1.840*** 1.846*** 1.937*** (0.088) (0.085) (0.067) (0.066) (0.066) (0.063) (0.064) (0.06 3) (0.060) (0.065) GDP per capita importer 1.309*** 1.243*** 1.203*** 1.136*** 1.133*** 1.191*** 1.269*** 1.103*** 1.109*** 1.051*** (0.099) (0.095) (0.068) (0.074) (0.072) (0.067) (0.069) (0.068) (0.066) (0.069) Population exporter 1.619*** 1.676*** 1 .879*** 1.864*** 1.843*** 1.893*** 1.833*** 1.845*** 1.884*** 1.832*** (0.061) (0.058) (0.040) (0.041) (0.043) (0.037) (0.039) (0.043) (0.040) (0.043) Population importer 0.897*** 0.930*** 0.999*** 1.015*** 1.017*** 1.067*** 1.091*** 1.016*** 0.993*** 1 .063*** (0.072) (0.066) (0.029) (0.030) (0.032) (0.028) (0.033) (0.029) (0.029) (0.030) Distance a 1.789*** 1.485*** 2.079*** 2.306*** 2.339*** 2.483*** 2.052*** 1.966*** 2.165*** 2.136*** (0.173) (0.180) (0.109) (0.107) (0.113) (0.101) (0.11 9) (0.116) (0.107) (0.120) Remoteness exporter b 0.060 0.670 0.211 0.187 0.285 0.343 0.425 0.946** 0.433 3.412*** (0.577) (0.567) (0.413) (0.475) (0.445) (0.413) (0.375) (0.411) (0.360) (0.256) Remoteness importer b 1.753*** 1.096*** 2.066*** 2.009 *** 2.132*** 2.099*** 2.313*** 2.587*** 2.838*** 0.756** (0.329) (0.330) (0.257) (0.255) (0.257) (0.258) (0.244) (0.240) (0.237) (0.352) Common border c 0.671* 1.230*** 0.280 0.182 0.544* 1.290*** 0.444 0.282 0.607* 0.498* (0.355) (0.369) (0.3 49) (0.307) (0.294) (0.411) (0.279) (0.268) (0.346) (0.261) Common language d 1.798*** 2.018*** 0.792*** 0.738*** 0.560*** 0.629*** 0.721*** 0.843*** 0.575*** 0.719*** (0.267) (0.291) (0.182) (0.182) (0.183) (0.172) (0.182) (0.184) (0.174) (0.176) Colo nial linkage: U.K. e --3.365*** 2.990*** 3.273*** 2.854*** 1.841* 2.062** 2.240*** 2.316*** --(0.700) (0.648) (0.751) (0.690) (1.097) (0.808) (0.711) (0.880) Colonial linkage: Spain f 0.847** 1.029*** 0.809** 0.803** 1.006*** 1.190*** 1.030*** 0 .718** 0.923*** 0.839*** (0.352) (0.357) (0.317) (0.321) (0.329) (0.335) (0.304) (0.286) (0.261) (0.264) Colonial linkage: Portugal g 1.979*** 1.984*** 0.810 1.010 0.945 0.869 0.733 0.614 0.567 0.332 (0.335) (0.547) (0.590) (0.754) (0.894) (0 .752) (0.655) (0.753) (0.939) (0.699) NAFTA h --2.434*** 1.825** 1.490** 2.054** 2.159*** 2.010** 1.972** 2.357*** --(0.709) (0.819) (0.739) (0.880) (0.819) (0.825) (0.798) (0.797)

PAGE 133

133 Table 4 5 Continued Dependen t variable: ln(exports from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 CARICOM h --5.226*** 4.791*** 4.467*** 4.717*** 3.430*** 4.159*** 4.141*** 3.317*** --(0.448) (0.408) (0.453) (0.375) (0.531) (0.498) (0.473) (0.552) MERCOSUR h 0.663 0.323 0.968 1.047 0.786 0.451 0.285 0.072 0.273 0.043 (0.722) (0.687) (0.674) (0.677) (0.681) (0.724) (0.625) (0.628) (0.666) (0.609) Andean Pact h 0.630* 0.653** 0.783** 0.640* 0.943*** 1.092*** 1.285*** 1.226*** 1.339*** 1.236*** (0.318) (0.329) (0.334) (0.368) (0.323) (0.345) (0.306) (0.314) (0.311) (0.304) CACM h 2.588*** 2.745*** 2.413*** 2.091*** 2.190*** 2.263*** 2.755*** 2.364*** 2.231*** 1.927*** (0.359) (0.376) (0.337) (0.315) (0.318) (0.342) (0.322) (0.309) (0.303) (0.299) EU importer k 0.773* 0.836** 0.098 0.415 0.315 0.084 0.508* 0.344 0.109 0.259 (0.466) (0.401) (0.308) (0.327) (0.324) (0.298) (0.267) (0.281) (0.260) (0.256) R squared 0.573 0.589 0.665 0.652 0.652 0.661 0.630 0.650 0.673 0.682 F statistic 137.767 147.254 259.568 248.303 230.573 286.088 242.451 236.601 284.616 265.645 N 1554 1560 2374 2408 2236 2657 2582 2315 2507 2243 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms *) Statistically significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level. a) Bilateral distance in kilometers between the capitals of trading partners. b) Re moteness it Y Yjt d ij ji wt Binary variables: c) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. d) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. e) Equal to 1 if one of the trading partners is the U.K. and the other a former British colony in the Western Hemisphere, 0 otherwise. (Note that the United States and Canada are not considered former colonies). f) Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere, 0 otherwise. g) Equal to 1 if one of the trading partners is Portugal and the other a former Portuguese colony in the Western Hemisphere, 0 o therwise. h) Integration/regional d ummies: Equal to 1 if both trading partners are members a specific trading agreement, ( NAFTA CARICOM MERCOSUR Andean Pact, or CACM respectively), 0 otherwise. k) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 134

134 Tabl e 4 6. MTPLAB specification: a gricultural trade (UNCTAD categories 1 & 2) Dependent variable: ln(exports of agricultural products from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Constant 25.782*** 30.393 *** 34.232*** 35.091*** 34.532*** 44.085*** 35.200*** 33.148*** 33.426*** 44.625*** (3.928) (3.864) (3.486) (3.613) (3.650) (3.964) (3.661) (3.543) (3.188) (3.982) GDP per capita exporter 1.208*** 1.191*** 1.131*** 1.137*** 1.075*** 1.236*** 1.0 94*** 1.117*** 1.010*** 1.141*** (0.054) (0.055) (0.049) (0.053) (0.052) (0.062) (0.053) (0.055) (0.046) (0.061) GDP per capita importer 1.098*** 1.088*** 1.003*** 1.055*** 0.980*** 1.083*** 1.033*** 0.965*** 0.982*** 1.088*** (0.064) (0.063) (0.055) (0.052) (0.052) (0.063) (0.060) (0.052) (0.049) (0.063) Population exporter 0.724*** 0.761*** 0.956*** 0.961*** 0.982*** 1.117*** 1.033*** 1.014*** 0.945*** 1.067*** (0.037) (0.037) (0.034) (0.031) (0.036) (0.039) (0.036) (0.034) (0.029) (0.039) Popula tion importer 0.882*** 0.870*** 0.880*** 0.926*** 0.932*** 1.031*** 0.981*** 0.873*** 0.864*** 0.975*** (0.044) (0.043) (0.024) (0.026) (0.030) (0.032) (0.030) (0.025) (0.025) (0.033) Distance a 1.074*** 1.097*** 1.275*** 1.229*** 1.457*** 1.465*** 1.437*** 1.108*** 1.265*** 1.523*** (0.129) (0.129) (0.078) (0.084) (0.094) (0.101) (0.088) (0.091) (0.079) (0.113) Remoteness exporter b 1.694*** 1.626*** 1.754*** 1.838*** 1.934*** 2.242*** 2.037*** 2.362*** 2.009*** 2.668*** (0.256) (0.261) (0. 230) (0.236) (0.239) (0.255) (0.233) (0.219) (0.210) (0.259) Remoteness importer b 1.265*** 0.693** 0.468* 0.664** 0.500* 0.452 0.818*** 1.434*** 0.648*** 0.453 (0.310) (0.312) (0.278) (0.289) (0.277) (0.306) (0.283) (0.300) (0.239) (0.283) Co mmon border c 1.068*** 0.911*** 0.414* 0.308 0.050 0.256 0.007 0.352 0.291 0.021 (0.259) (0.248) (0.216) (0.199) (0.246) (0.236) (0.229) (0.224) (0.201) (0.249) Common language d 0.519*** 0.460** 0.497*** 0.721*** 0.329* 0.647*** 0.462*** 0.827*** 0.562 *** 0.438** (0.229) (0.233) (0.146) (0.152) (0.172) (0.181) (0.172) (0.152) (0.138) (0.179) Colonial linkage: U.K. e --2.981*** 2.724*** 2.921*** 2.832*** 2.594*** 1.632*** 2.064*** 2.133*** --(0.298) (0.278) (0.388) (0.377) (0.394) (0.393) (0. 370) (0.558) Colonial linkage: Spain f 0.353 0.430 0.716*** 0.781*** 0.895*** 0.916*** 1.107*** 0.588** 0.814*** 1.060*** (0.294) (0.299) (0.253) (0.249) (0.245) (0.276) (0.283) (0.246) (0.216) (0.239) Colonial linkage: Portugal g 0.719*** 1.051*** 0.935 *** 0.905*** 1.004* 0.678 1.026*** 0.767 1.268*** 1.597*** (0.252) (0.229) (0.236) (0.205) (0.541) (0.531) (0.310) (0.640) (0.397) (0.336) NAFTA h --0.476 0.684 0.770 0.402 0.058 0.265 0.321 0.193 --(0.470) (0.544) (0.474) (0.570) (0.527) (0. 540) (0.481) (0.543)

PAGE 135

135 Table 4 6 Continued Dependent variable: ln(exports of agricultural products from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 CARICOM h --3.543*** 3.178*** 2.560*** 3.732*** 3.608*** 3.402*** 3.095*** 3.221*** --(0.284) (0.319) (0.393) (0.412) (0.346) (0.377) (0.345) (0.361) MERCOSUR h 0.659* 0.733* 0.765* 1.055** 0.990** 0.856* 1.311*** 1.566*** 1.185** 1.257** (0.384) (0.441) (0.459) (0.459) (0.487) (0.518) (0.475) (0.466) (0.467) (0.492) Andean Pact h 0.423 0.821* 0.885*** 1.156*** 0.799 1.051*** 1.478*** 1.639*** 1.441*** 1.749*** (0.475) (0.464) (0.326) (0.303) (0.521) (0.320) (0.310) (0.301) (0.259) (0.335) CACM h 2.672*** 2.808*** 2.520*** 2. 756*** 2.434*** 2.902*** 2.804*** 3.070*** 2.663*** 2.770*** (0.333) (0.296) (0.258) (0.242) (0.264) (0.305) (0.264) (0.275) (0.246) (0.281) EU importer k 0.055 0.438* 0.264 0.156 0.323 0.467** 0.122 0.381** 0.327* 0.402** (0.246) (0.228) (0.205) (0.2 14) (0.202) (0.225) (0.214) (0.169) (0.174) (0.176) R squared 0.527 0.532 0.614 0.614 0.593 0.559 0.556 0.587 0.613 0.593 F statistic 90.451 94.355 158.614 163.346 141.800 145.987 142.120 147.587 176.116 149.810 N 1233 1263 1815 1869 1769 2091 2061 1883 2021 1867 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms *) Statistically significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically signif icant at the 1 % level. a) Bilateral distance in kilometers between the capitals of trading partners. b) Remoteness it Y Yjt d ij ji wt Binary variables: c) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. d) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. e) Equal to 1 if one of the trading partners is the U.K. and the other a former British colony in the Western Hemisphere, 0 otherwise. (Note that the United States and Canada are not considered former colonies). f) Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere, 0 otherwi se. g) Equal to 1 if one of the trading partners is Portuga l and the other a former Portuguese colony in the Western Hemisphere, 0 otherwise. h) Integration/regional dummies: Equal to 1 if both trading partners are members a specific trading agreement, ( NAFTA CARICOM MERCOSUR Andean Pact, or CACM respectively) 0 otherwise. k) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 136

136 Tabl e 4 7. MTPLAB specification: manufactured goods trade (UNCTAD category 5) Dependent variable: ln(exports of manufactured goods from country i to c ountry j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 Constant 37.932*** 36.450*** 35.183*** 38.082*** 38.651*** 44.027*** 39.035*** 38.983*** 36.060*** 48.715*** (3.483) (3.536) (3.032) (3.058) (3.057) (3.460) (3.142) (3.442) (2.738) (3.502) GDP per capita exporter 1.714*** 1.779*** 1.600*** 1.625*** 1.651*** 1.773*** 1.724*** 1.766*** 1.634*** 1.824*** (0.049) (0.048) (0.044) (0.042) (0.042) (0.050) (0.048) (0.050) (0.040) (0.053) GDP per capita importer 0.881*** 0 .870*** 0.867*** 0.860*** 0.849*** 0.991*** 0.938*** 0.889*** 0.889*** 0.898*** (0.054) (0.055) (0.046) (0.044) (0.045) (0.052) (0.051) (0.051) (0.046) (0.055) Population exporter 1.211*** 1.247** 1.299*** 1.299*** 1.324*** 1.461*** 1.443*** 1.450*** 1. 366*** 1.509*** (0.033) (0.033) (0.026) (0.024) (0.026) (0.033) (0.030) (0.032) (0.024) (0.034) Population importer 0.751*** 0.741*** 0.945*** 0.976*** 0.945*** 0.989*** 0.946*** 0.924*** 0.894*** 0.966*** (0.038) (0.038) (0.023) (0.023) (0.025) (0.02 6) (0.025) (0.024) (0.022) (0.030) Distance a 1.308*** 1.340*** 1.629*** 1.669*** 1.671*** 1.870*** 1.683*** 1.578*** 1.559*** 1.679*** (0.103) (0.101) (0.073) (0.076) (0.073) (0.087) (0.085) (0.088) (0.072) (0.102) Remoteness exporter b 0.482* 0.002 0.049 0.047 0.093 0.420** 0.698*** 1.013*** 0.802*** 1.658*** (0.227) (0.223) (0.192) (0.179) (0.174) (0.198) (0.186) (0.196) (0.182) (0.235) Remoteness importer b 0.679** 0.920*** 0.809*** 0.999*** 1.007*** 0.885*** 0.092 0.298 0.082 0.017 (0.296) (0.303) (0.258) (0.260) (0.261) (0.294) (0.271) (0.296) (0.226) (0.277) Common border c 0.551** 0.835*** 0.201 0.096 0.225 0.497 0.360 0.039 0.124 0.105 (0.236) (0.224) (0.200) (0.217) (0.205) (0.316) (0.316) (0.212) (0.186) (0.220) Comm on language d 1.052*** 1.080*** 0.615*** 0.726*** 0.697*** 0.800*** 0.758*** 0.824*** 0.661*** 0.721*** (0.182) (0.174) (0.123) (0.117) (0.120) (0.131) (0.130) (0.128) (0.117) (0.147) Colonial linkage: U.K. e --1.224*** 1.236*** 0.953*** 0.909** 1.048 *** 0.959** 0.993*** 1.986*** --(0.321) (0.284) (0.334) (0.365) (0.329) (0.418) (0.366) (0.656) Colonial linkage: Spain f 0.004 0.034 0.650*** 0.606** 0.561** 0.817*** 0.520*** 0.453** 0.641*** 0.712*** (0.249) (0.217) (0.180) (0.240) (0.269) (0. 296) (0.197) (0.201) (0.192) (0.202) Colonial linkage: Portugal g 0.198 0.068 0.224 0.080 0.187 0.079 0.123 0.292 0.490 0.338* (0.895) (0.770) (0.805) (0.656) (0.388) (0.390) (0.174) (0.292) (0.353) (0.185) NAFTA h --0.270 0.139 0.423 0.328 0. 190 0.318 0.189 0.727 --(0.543) (0.725) (0.711) (0.786) (0.717) (0.716) (0.664) (0.698)

PAGE 137

137 Table 4 7 Continued Dependent variable: ln(exports of manufactured goods from country i to country j in U.S. dollars) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 CARICOM h --3.482*** 2.969*** 2.924*** 3.059*** 3.033*** 2.884*** 2.704*** 2.843*** --(0.296) (0.325) (0.323) (0.337) (0.325) (0.335) (0.332) (0.432) MERCOSUR h 0.331 0.098 0.091 0.073 0.305 0.161 0.412 0.395 0.391 0.671 (0.459) (0.439) (0.400) (0.395) (0.417) (0.496) (0.444) (0.455) (0.406) (0.438) Andean Pact h 0.658*** 0.767*** 0.693** 0.483* 0.945*** 0.916*** 1.028*** 1.109*** 1.167*** 1.262*** (0.206) (0.192) (0.273) (0.292) (0.213) (0.24 1) (0.235) (0.227) (0.224) (0.178) CACM h 2.715*** 2.403*** 2.326*** 2.269*** 2.381*** 2.488*** 2.598*** 2.502*** 2.283*** 2.292*** (0.256) (0.260) (0.237) (0.222) (0.225) (0.257) (0.255) (0.247) (0.219) (0.242) EU importer k 0.234 0.196 0.508*** 0.5 44*** 0.236 0.752*** 0.883*** 1.017*** 1.057*** 0.980*** (0.220) (0.219) (0.187) (0.178) (0.185) (0.206) (0.173) (0.181) (0.161) (0.196) R squared 0.703 0.718 0.739 0.754 0.745 0.700 0.714 0.726 0.751 0.724 F statistic 210.910 229.334 316.234 345.714 316.047 296.491 310.168 296.182 366.507 289.738 N 1350 1364 2025 2051 1963 2308 2250 2033 2204 2006 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms *) Statistically significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level. a) Bilateral distance in kilometers between the capitals of trading partners. b) Remoteness it Y Yjt d ij ji wt Binary vari ables: c) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. d) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. e) Equal to 1 if one of the trading partners is the U.K. and the other a former British colony in the Western Hemisphere, 0 otherwise. (Note that the United States and Canada are not considered former colonies). f) Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony i n the Western Hemisphere, 0 otherwise. g) Equal to 1 if one of the trading partners is Portugal and the other a former Portuguese colony in the Western Hemisphere, 0 o therwise. h) Integration/regional dummies: Equal to 1 if both trading partners are members a specific trading agreement, ( NAFTA CARICOM MERCOSUR Andean Pact, or CACM respectively), 0 otherwise. k) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 138

138 Table 4 8. Effects of binary variables, h istorical n etworks Average parameter a Marginal effect b MH specification: a ggregate trade U.K. colony 3.706 40.696 Spain colony 1.076 2.934 Portugal colony 1.744 0.175 MH specification: a gricultural trade U.K. colony 2.932 18.772 Spain colony 0.724 2. 062 Portugal colony 1.111 3.039 MH specification: m anufactured goods trade U.K. colony 1.527 4.606 Spain colony 0.723 2.061 Portugal colony 0 1.000 MTPLAB specification: a ggregate trade U.K. colony 2.618 13.703 Spain colony 0.544 1.723 Po rtugal colony 1.982 0.138 MTPLAB sp ecification: a gricultural trade U.K. colony 2.485 12.003 Spain colony 0.860 2.362 Portugal colony 1.063 2.895 MTPLAB specification: m anufactured goods trade U.K. colony 1.164 3.201 Spain colony 0.620 1.8 59 Portugal colony 0.338 1.402 pertains to parameters from the M H specification of the model. a) Average Parameter is calculated as m nn where m indicates a given binary variable (e.g., colonial linkages for the U.K. Spain, and Portugal) and n is the numbe r of years for which parameter estimates statistically different from zero were found. 0 indicates that no values were statistically different from zero. b) Marginal Effects are calculated as ( me ), where m equals the Average parameter estimate for binary variable m B aseline/reference observation: w here all 11 binary variables in vector w are jointly equal to 0. For the analysis pertaining to the MTPLAB specification, the is interchanged for in the above notation.

PAGE 139

139 Table 4 9. Effects of binary variables, r egional networks Average p arameter a Marginal effect b MH specification: a ggregate trade NAFTA 2.757 0.063 CARICOM 3.643 38.197 MERCOSUR 0 1.000 Andean Pact 0.978 2.658 CACM 1.550 4.711 MH specification: a gricultural trade NAFTA 0 1.000 CARICOM 3.008 20.252 MERCOSUR 1.146 3.144 Andean Pact 1.066 2.903 CACM 2.128 8.395 MH specification: m anufactured goods trade NAFTA 0 1.000 CARICOM 2.684 14.645 MERCOSUR 1.048 2.852 Andean Pact 0.811 2.249 CACM 2.192 8.954 MTPLAB specific ation: a ggregate trade NAFTA 2.038 0.130 CARICOM 4.281 72.313 MERCOSUR 0 1.000 Andean Pact 0.983 2.672 CACM 2.357 10.556 MTPLAB specification: a gricul tural trade NAFTA 0 1.000 CARICOM 3.292 26.907 MERCOSUR 1.038 2.823 Andean Pact 1.144 3.140 CACM 2.740 15.485 MTPLAB specification: m anufactured goods trade NAFTA 0 1.000 CARICOM 2.987 19.831 MERCOSUR 0 1.000 Andean Pact 0.903 2.466 CACM 2.426 11.310 pertains to parameters from the MH specification of the model. a) Average Parameter is calculated as m nn where m indic ates a given binary variable, (e.g ., NAFTA CARICOM MERCOSUR, Andean Pact, or the CACM ) and n is the numbe r of years for which parameter estimates statistically different from zero was found. 0 indicates that no values were statistically different from zero. b) Marginal Effects are calculated as ( me ), where m equals the Average parameter estimate for binary variable m B aseline/reference observation: w here all 11 binary variables in vector w are jointly equal to 0. For the analysis pertaining to the MTPLAB specification, the is interchanged for in the above notation.

PAGE 140

140 CH APTER 5 CONCLUDING REMARKS Chap ter 1 introduces a framework in which the trade behavior of the countri es in the Western Hemisphere is influenced by networks resulting from historical legacies and regional t rading agreements The presence of these networks is postulated to influence the hemispheric wide in tegration process of the potential F ree Trade Area of the Americas (F TAA) This study employs the gravity model of international trade t o investigate the impact of history and regionalism on the trade volumes in the Western Hemisphere. Chapter 2 proceeds with a rigorous review of gravity models with substantial attention given to the ir methodological evolutio n and the different theoretical frameworks use d to justify their use As shown, gravity like structure s can be derived from numerous theoretical frameworks. I t should be noted, however, that these theoretical foundations are rest rictive and de facto in nature in that economist were searching for theories to fit a mo del already in widespread use. While there are some variations on the theme, t he literature outlines two main nonnested specifications of the model : the MTPLAB specification and the MH specification. The MTPLAB specification refers to t he traditionally used gravity model as operationalized by Tinbergen (1962) among others It is shown that this model has its roots in the use of social physics and the Newtonian notion of gravity There is a plethora of empirical applications and theor etical derivations. Three of the most cited derivations were provided by Anderson (1979) and Bergstrand (1985, 1989). The MH specification refers to a less commonly used intra industry trade model as derived by Helpman and Krugman (1985) and Helpman (1987). L ittle attention, if any, has been devoted to determining which econometric specification is preferred for empirical estimation

PAGE 141

141 Th e framework of this study as Chapter 3 discusses is to fit these two specifications of the gravity model to three da ta set s encompassing the bilateral trade volumes of the economies in the Americas The first dataset represents aggregate bilateral trade flows and the remaining two dataset s are disaggregate d on the sectoral level capturing trade in agricultural products and manufactured goods respectively. D isaggregation is made via the UNCTAD nomenclature. The MH model was derived for being applicable to trade between a subset of countries in the world ec onomy whereas the MTPLAB model seems to be more applicable for global studies of trade flows. Since the dataset s are FTAA centered, the MH specification is treated as the primary model and the MTPLAB specification serves as the secondary model, used primarily as a robustness check of the binary network effects ( since the binary variables are congruent across the two models) Both the MH specification and the M TPLAB s pecification of the gravity model perform well empirically Chapter 4 presents and discusses the empirics In general, the estimated results from both specification s confor m to a priori expectations and support the stylized facts in the literature Two F tests reveal that one can reject the hypothesis that the variances of the error terms obtained by estimating the gravity model using the agricultural trade data set and the variances of the e rror terms obtained by estimating the model using the manufactured goods trade data set are equal. Likewise, one can reject the null hypothesis that estimated parameters are the same for trade in primary produ cts as for trade in manufactures. Th ese tests indicate that the data should not be pooled and the model should be estimated sector by sector via disaggregated data. In particular, t he evidence suggests one should estimate the gravity model for agricultur al trade separately as the determinants of primary trade patterns are fundamentally different from trade in manufactured (and other ) goods.

PAGE 142

142 O ne should be cautious however, of simply ignoring estimations performed using aggregate trade data. The UNCTAD nomenclature includes product categories not incorporated into the two disaggregated data sets, such as minerals, metals, and fuels. Clearly, these commodities are important in global commerce. The influence of these product groups is therefore captured, at least partially, by the aggregate trade data. The aggregate trade data estimates are not necessarily equal to the averages of those obtained by using agricultural trade data and manufactured goods trade data. Instead, the estimated parameters obtaine d from the aggregate data are more similar to those obtained by using the manufactured goods data. I t seems that trade in minerals, metals, and fuels is determined by the same underlying factors as trade in manufactured goods. The evidence suggests that both history and regionalism have significantly shaped the trade behavior of the Western Hemisphere. Significant neo colonial trade distortions are present between the United Kingdom and her former dependencies. The findings suggest that deliberate trad e policy on behalf of the U.K. magnify trade volumes. These policy distortions are particularly prevalent for agricultural products Former British colonies (i.e., the economies in the Caribbean basin) ship about 15 60% of their exports to their former colonial ruler a sizeable percentage given the vast geographic distance between the Caribbean and the British Isles Not surpr isingly, the gravity model finds strong evidence of trade distortions resul ting from this historical legacy after controlling f or economics and geography The corresponding effect is significantly lower, however for former Iberian colonies and t he effect diminishes almost entirely for Portugals relationship vis vis Brazil. The gravity model reveals that regional linkages e .g ., regional trading agreements, significantly influence trade behavior after controlling for other factors In particular the smaller

PAGE 143

143 economies in the central region of the hemisphere tend to cooperate more fully than the larger, more self sufficient, peripheral ones. The evidence supports that the CARICOM and the CACM agreements strongly influence the trade relationship s of their respective member states. T his is not entirely surprising since smaller economies tend to have a narrower productive base and consequently a larger dependency on trade. For NAFTA and MERCOSUR on the other hand, the gravity model detects smaller effects on the trade behavior of their members after controlling for such factors as economic size, population, culture, and geogra phy The effect of the Andean Pact, consisting of medium sized economies, is somewhere in between, thus lending support to an inverse relationship between economic size and regional dependency. In the context of a future FTAA, historical linkages need t o be considered when considering hemispheric integration. The economies in the Caribbean basin are rather unique in that they are highly dependent upon exports to the United Kingdom for economic subsistence. Given the shorter geographic distance and cons equently lower transportation costs, the North American markets could through more accommodating trade policy be attractive substitute destinations for Caribbean and Latin American exports. In particular, the industrial economies of the United States an d Canada have presumably similar consumption patterns and demand structures as the U.K. and the rest of the European Union. However, the danger is that such a drastic change in trade patterns would be dominated by trade diversion rather than trade creation. In the static Vinerian tradition, trade diversion is welfare reducing as it decreases global productive efficiency; resources are diverted away from their most efficient use. It is therefore necessary to ensure via economic policies that the dy namic g ains from integration of specialization, economies of scale, and marketing efficiencies are realized. If sufficiently large, these gains can outweigh any efficiency reductions resulting from diverting trade.

PAGE 144

144 Taking into account regional tendencies a nd t he presence of neo colonial trade distortions hemispheric integration will be a c omplicated process. A s the polarization of the United States and Brazil increases, both taking on the dominant roles in the northern and southern cones respectively, a viab le option may be to look at the central region as a stepping stone to achieve wider integration. This can be realized by intensifying and expanding the trade relationships of the meridian economies with their northern and southern counterparts. Via the deepening of regional cooperation of the central region while at the same time escalating trade with, within, and among the e xtremities, wider integration can be achieved. Thus, an alternative approach of viewing the FTAA as enlarging and merging the nort hern and southern agreements ( i.e., NAFTA and MERCOSUR ) to in corporate the smaller economies (i.e., CARICOM CACM and the Andean Pact) would be to utilize the smaller agreements as building blocks to e xpand the integration process. The notion that these trade blocs have had significant roles in shaping the trade behavior of their member states supports this smaller to larger approach. Obviously, one cannot claim that the regional i ntegration of microstates is an end all panacea for achieving hemispheri c free trade. However, it might be a fruitful place to start, especially considering their relative openness to intra regi onal and inter regional cooperation. While the FTAA and member countries trade behavior are complex issues, a good starting point is to consider the impact of historical and regional linkages. As shown by this study the policy distortions resulting from history and regionalism are important factors In short, in order to move forward with free trade negotiations the hemisphere needs to consider its history; both its recent history of enacting regional trade agreements and its former imperial history.

PAGE 145

145 Interesting extensions of current research include using further disaggregated dataset s t o measure industry level effects Furt hermore, it would be appealing to include more years in the sample as to capture temporal changes and to assess whether trade patterns are stable over time. As indicated in Chapter 3, this study does not investigate the trade environment post the 9/11 ter rorist attacks. Given the data coverage, 1992 through 2001, an inherent characteristic of the data, which can be argued to be both a weakness and a strength of the current analysis, is that any changes in trade behavior post the 9/11 terrorist attacks are not addressed. For the years immediately following 2001 trade volumes may be significantly distorted as tra de patterns in all likelihood were altered while the world was coming to terms with new geo political conditions. It is plausible that a structur al break is present in the data after 9/11 This could be assessed if the time period included years for the decade following 2001. Clearly, that would be a fruitful extension of this study. However, that would be the focus of a different inquiry

PAGE 146

146 AP PENDIX A SUPPLEMENTAL TABLES Table A 1. F test for similarity of variances, MH specification Year AAee n A 'MMee n M k Calculated F statistic 1992 4982.053 1233 4924.449 1350 15 1.10888 1993 4774.809 1263 5081.01 7 1364 15 1.015787 1994 7651.581 1815 8530.616 2025 18 1.001775 1995 8125.984 1869 8037.056 2051 18 1.110478 1996 7728.951 1769 8011.09 1963 18 1.071673 1997 14755.79 2091 14433.85 2308 18 1.129318 1998 12382.19 2061 12489.97 2250 18 1.083083 1999 91 06.281 1883 10910.97 2033 18 1.108986 2000 8104.04 2021 9614.718 2204 18 1.087091 2001 10224.16 1867 11756.34 2006 18 1.069461 Testing the null hypothesis 22 012: H is tested against the alternative hypothesis, 22 12:aH AAee = Sums of squared errors from the agricultural trade estimation. nA = the number of observations for the manufactured goods trade estimation. 'MMee = Sums of squared errors from the manufactured goods trade estimation. nM = the number of observations for the manufactured goods trade estimation. k = the number of exogenous regressors. The F statistic is calculated as follows, 1,' 'AAMMAA AA nknk MM MMnk F nk ee ee for 1992 through 1998, as 'AAee > 'MMee and as 1,' 'MMAAMM MM nknk AA AAnk F nk ee ee for 1999 through 2001, as 'MMee > 'AAee In this case, df1 = 2 = F value at the 5% and 1% alpha level: 1.

PAGE 147

147 Table A 2. F tes t for similarity of variances, MTPLAB specification Year AAee n A 'MMee n M k Calculated F statistic 1992 4801.831 1233 3947.161 1350 16 1.333482 1993 4714.376 1263 3913.011 1364 16 1.302376 1994 7742.886 18 15 7244.241 2025 19 1.193808 1995 8297.081 1869 6852.27 2051 19 1.329973 1996 7933.907 1769 6669.346 1963 19 1.321484 1997 15091.22 2091 12256.89 2308 19 1.360191 1998 12796.51 2061 10411.27 2250 19 1.342862 1999 9126.574 1883 8777.188 2033 19 1.12348 2 2000 8267.162 2021 7662.014 2204 19 1.177608 2001 10248.69 1867 9559.277 2006 19 1.152761 Testing the null hypothesis 22 012: H against the alternative hypothesis, 22 12:aH AAee = Sum of square d errors from the agricultural trade data estimation. nA = the number of observations for the agricultural trade estimation. 'MMee = Sum of squared errors from the manufactured goods trade estimation. nM = the number of observations f or the manufactured goods trade estimation. k = the number of exogenous regressors. The F statistic is calculated as follows, 1,' 'AAMMAA AA nknk MM MMnk F nk ee ee for all years, as AAee > 'MMee In this case, df1 = a nd df2 = F value at the 5% and 1% alpha level: 1.

PAGE 148

148 Table A 3. F test for similarity of parameters, MH specification Year RRee 'URURee n A n M k (= r) Calculated F statistic 1992 11228.09 9906.502 1 233 1350 15 22.7057 1993 11397.69 9855.827 1263 1364 15 27.08527 1994 17892.43 16182.2 1815 2025 18 22.33498 1995 18344.44 16163.04 1869 2051 18 29.1218 1996 17592.18 15740.04 1769 1963 18 24.16174 1997 31674.19 29189.64 2091 2308 18 20.63159 1998 27 123.43 24872.16 2061 2250 18 21.49704 1999 22334.39 20017.25 1883 2033 18 24.95213 2000 19903.17 17718.76 2021 2204 18 28.69056 2001 24198.14 21980.5 1867 2006 18 21.50664 Testing the null hypothesis 0: H Estimated parameters are the same across sectors against the alternative hypothesis, :aH not 0H RRee = Sum of squared errors from the auxiliary restricted model obtained by stacking agricultural trade data and ma nufactured goods trade data. 'URURee =' AAee + 'MMee AAee = Sums of squared errors obtained from the agricultural trade estimation. nA = the number of observations for the ag ricultural goods trade estimation. 'MMee = Sums of squared errors obtained from the manufactured goods trade estimation. nM = the number of observations for the manufactured goods trade estimation. k = the number of exogenous regressor s. The F statistic is calculated as follows, ,2'' 2AMRRURUR rnnk URUR AMr F nnk eeee ee r is the number of restrictions. The degrees of freedom for the numerator equal both r and k whereas the degrees of freedom for the denominator equals ( nA + nM 2 k ). 1992 1993: critical F value at 1% alpha level = 2.04. 1994 2001: critical F value at the 1% alpha level = 1.88.

PAGE 149

149 Table A 4. F test for similarity of parameters MTPLAB specification Year RRee 'URURee n A n M k (= r) C alculated F statistic 1992 10539.86 8748.992 1233 1350 16 32.63596 1993 10688 8627.387 1263 1364 16 38.73773 1994 17071.45 14987.13 1815 2025 19 27.82952 1995 17730.02 15149.35 1869 2051 19 34.80484 1996 16925.23 14603.25 1769 1963 19 30.91376 1997 3 0456.5 27348.11 2091 2308 19 26.08795 1998 26196.07 23207.78 2061 2250 19 28.95803 1999 20794.03 17903.76 1883 2033 19 32.94937 2000 18771 15929.18 2021 2204 19 39.31453 2001 22853.05 19807.97 1867 2006 19 31.02923 Testing the null hypothesis 0: H Estimated parameters are the same across sectors against the alternative hypothesis, :aH not 0H RRee = Sum of squared errors from the auxiliary restricted model obtaine d by stacking agricultural trade data and manufactured goods trade data. 'URURee =' AAee + 'MMee AAee = Sums of squared errors obtained from the agricultural trade estimation. nA = the number of observations for the agricultural goods trade estimation. 'MMee = Sums of squared errors obtained from the manufactured goods trade estimation. nM = the number of observations for the manufactured goods trade estim ation. k = the number of exogenous regressors. The F statistic is calculated as follows, ,2'' 2AMRRURUR rnnk URUR AMr F nnk eeee ee r is the number of restrictions. The degrees of freedom for the numerator equal both r and k whereas the degrees of freedom for the denominator equals ( nA + nM 2 k ). 1992 1993: critical F value at 1% alpha level = 2.04. 1994 2001: critical F value at the 1% alpha level = 1.88.

PAGE 150

150 Table A 5. Estimated standard er rors of parameters for the population variables a ggregate trade G DP per capita a Absolute GDP b Population of exporter, SE( 3 ) Population of importer, SE( 4 ) Population of exporter, SE( 3 ) Population of importer, SE( 4 ) 199 2 0.061 0.072 0.101 0.133 1993 0.058 0.066 0.093 0.122 1994 0.040 0.029 0.074 0.077 1995 0.041 0.030 0.069 0.086 1996 0.043 0.032 0.069 0.084 1997 0.037 0.028 0.065 0.076 1998 0.039 0.033 0.066 0.079 1999 0.043 0.029 0.066 0.079 2000 0.040 0.029 0. 062 0.079 2001 0.043 0.030 0.067 0.080 The values in the table represent the white robust standard errors of the parameter estimates for the exporters population and the importers population (variables in natural logs). a) GDP per capita refers to t he MTPLAB specification gravity model as estimated in Chapter 4. The estimated parameters from the MTPLAB gravity model using GDP per capita of the exporter and GDP per capita of the importer as independent variables in the MTPLAB gravity model are presen ted Tables 4 2 through 43. b) Absolute GDP refers to estimates obtained for the MTPLAB specification gravity model when absolute GDP data instead of per capita income. The estimated parameters from the MTPLAB model using absolute GDP of the exporter a nd absolute GDP of the importer as independent variables are not presented; nor are standard errors of remaining variables.

PAGE 151

151 Table A 6. Estimated standard errors of parameters for the population variables a gricultural trade GDP per capita a Absolute GD P b Population of exporter, SE( 3 ) Population of importer, SE( 4 ) Population of exporter, SE( 3 ) Population of importer, SE( 4 ) 1992 0.037 0.044 0.060 0.076 1993 0.037 0.043 0.064 0.076 1994 0.034 0.024 0.053 0.064 1995 0.031 0.026 0.057 0.059 1996 0.036 0.030 0.059 0.063 1997 0.039 0.032 0.066 0.075 1998 0.036 0.030 0.058 0.066 1999 0.034 0.025 0.057 0.059 2000 0.029 0.025 0.049 0.058 2001 0.039 0.03 3 0.065 0.071 The values in the table represent the white robust standard errors of the parameter estimates for the exporters population and the importers population (variables in natural logs). a) GDP per capita refers to the MTPLAB specification gr avity model as estimated in Chapter 4. The estimated parameters from the MTPLAB gravity model using GDP per capita of the exporter and GDP per capita of the importer as independent variables in the MTPLAB gravity model are presented Tables 4 2 through 43. b) Absolute GDP refers to estimates obtained for the MTPLAB specification gravity model when absolute GDP data instead of per capita income. The estimated parameters from the MTPLAB model using absolute GDP of the exporter and absolute GDP of the importer as independent variables are not presented; nor are standard errors of remaining variables.

PAGE 152

152 Table A 7. Estimated standard errors of parameters for the population variables m anufactured goods trade GDP per capita a Absolute GDP b Population of e xporter, SE( 3 ) Population of importer, SE( 4 ) Population of exporter, SE( 3 ) Population of importer, SE( 4 ) 1992 0.033 0.038 0.056 0.065 1993 0.033 0.038 0.0 56 0.068 1994 0.026 0.023 0.049 0.053 1995 0.024 0.023 0.046 0.049 1996 0.026 0.025 0.046 0052 1997 0.033 0.026 0.056 0.062 1998 0.030 0.025 0.050 0.058 1999 0.032 0.024 0.052 0.060 2000 0.024 0.022 0.044 0.053 2001 0.034 0.030 0.056 0.065 The val ues in the table represent the white robust standard errors of the parameter estimates for the exporters population and the importers population (variables in natural logs). a) GDP per capita refers to the MTPLAB specification gravity model as estimated in Chapter 4. The estimated parameters from the MTPLAB gravity model using GDP per capita of the exporter and GDP per capita of the importer as independent variables in the MTPLAB gravity model are presented Tables 4 2 through 43. b) Absolute GDP refers to estimates obtained for the MTPLAB specification gravity model when absolute GDP data instead of per capita income. The estimated parameters from the MTPLAB model using absolute GDP of the exporter and absolute GDP of the importer as independent variables are not presented; nor are standard errors of remaining variables.

PAGE 153

153 APPENDIX B GRAPHICAL PRESENTATI ON OF HISTORICAL EFF ECTS The following appendix presents the estimated parameters for the three binary variables capturing the historical netw ork effects from both specifications of the model. The estimated parameters for British colonial trade effects, W3, Spanish colonial trade effects, W4, and Portuguese colonial trade effects, W5, are shown graphically over time. A similar graphical repres entation was provided in Sandberg, Seale, and Taylor (2006). As indicated in Chapter 3, variable W3, U.K. Colony is defined to equal 1 if one of the trading partners is the U.K. and the other a former British colony in the Western Hemisphere, 0 otherwi se; variable W4, Spain Colony is defined to equal 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere, 0 otherwise; and variable W5, Portugal Colony is defined to equal 1 if one of the trading par tners is Portugal and the other a former Portuguese colony in the Western Hemisphere, 0 otherwise. A separate graph is presented for each model specification and for each dataset Thus, there are a total of six graphical representations. Figures B 1 t hrough B 3 present the estimated binary history effects from the MH specification for aggregate trade, agricultural trade, and manufactu red goods trade respectively. Figures B 4 through B 6 present the corresponding estimated effects from the MTPLAB spec ification for aggregate trade, agricultural trade, and manufactured goods trade respectively. The numerical values can be found in Tables 4 2 through 47. A summary of the marginal effects is provided in Tables 4 8 and 49.

PAGE 154

154 Figure B 1. Estimated his torical effects: MH specification, aggregate trade Figure B 2. Estimated historical effects: MH specification, agricultural trade

PAGE 155

155 Figure B 3. Estimated historical effects: MH specification, manufactured goods trade Figure B 4. Estimated histo rical ef fects: MTPLAB specification, a ggregate trade

PAGE 156

156 Figure B 5. Estimated historical ef fects: MTPLAB specification, a gricultural trade Figure B 6. Estimated historical ef fects: MTPLAB specification, m anufactured goods trade

PAGE 157

157 APPENDIX C GRAPHICAL PRESENTATION OF REGI ONAL EFFECTS The following appendix presents the estimated parameters for the five binary variables capturing the regional network effects from both specifications of the model. The estimated parameters for the NAFTA effect, W6, the CARICOM effect, W7, the MERCOSUR effect, W8, the Andean Pact effect, W9, and the CACM effect, W10, are shown graphically over time. A similar graphical representation was provided in Sandberg, Seale, and Taylor (2006). As indicated in Chapter 3, variabl e W6, NAFTA is defined to equal 1 if both trading partners are members of NAFTA 0 otherwise; W7, CARICOM is defined to equal 1 if both trading partners are members of CARICOM 0 otherwise; W8, MERCOSUR is defined to equal 1 if both trading partne rs are members of MERCOSUR 0 otherwise; W9, Andean Pact is defined to equal 1 if both trading partners are members of the Andean Pact, 0 otherwise; and W10, CACM is defined to equal 1 if both trading partners are members of CACM 0 otherwise. A se parate graph is presented for each model specification and for each dataset Thus, there are a total of six graphical representations. Figures C 1 through C 3 present the estimated regional effects from the MH specification for aggregate trade, agricult ural trade, and manufactu red goods trade respectively. Figures C 4 through C 6 present the corresponding estimated effects from the MTPLAB specification for aggregate trade, agricultural trade, and manufactured goods trade respectively. The numerical v alues can be found in Tables 4 2 through 47. A summary of the marginal effects is provided in Tables 4 8 and 49.

PAGE 158

158 Figure C 1. Estimated regional effects: MH specification, aggregate trade Figure C 2. Estimated regional effects: MH specification, agricultural trade

PAGE 159

159 Figure C 3. Estimated regional effects: MH specification, manufactured goods trade Figure C 4. Estimated regional ef fects: MTPLAB specification, a ggregate trade

PAGE 160

160 Figure C 5. Estimated reigonal ef fects: MTPLAB specification, a gricultural trade Figure C 6. Estimated regional ef fects: MTPLAB specification, m anufactured goods trade

PAGE 161

161 APPENDIX D MEXICO CENTERED ESTIMATIONS To assess the effect that NAFTA has had on trade between the Mexico and the U.S. and between Mexico and Canada (and vice versa), the two gravity models are re estimated using Mexico centered dataset s The three original data sets are rearranged so that any transactions not involving Mexico are omitted. Thus, for the rearranged three data Mexico centered datasets all observations contain Mexico as either the exporter or the im porter. Again, one data set represents aggregate trade, one data set represents agricultural trade, and one dataset represents manufactured goods trade. Analogously to Equation s 3 8 and 41, the MH specification is the primary model, defined as 22 01 2 3ln ln ln ln1jt jt it it ijt itjt itjt itjt itjt YY YY X YY NN YYYY 45 6 1lnln lnM ij it jtmmijijt mDRemotenessRemotenessW ( D 1) where, again, Xijt represents the bilateral exports from country i to country j in time period t plus 1, Yi t is nominal GDP for country i in time period t Yj t is nominal GDP for country j in time period t Ni t is country i s population in time period t Nj t is country j s population in time period t and Dij is the bilateral distance in kilometers between the capitals of c ountry i and country j Remotenessit and Remotenessjt are defined as itY D Yjt Remoteness ij ji wt ( D 2) and

PAGE 162

162 jtY D Yit Remoteness ij ij wt (D 3) respectively. Wmijs are m binary variables in vector w corresponding to any qualitative relationshi ps between country i and country j and ijtv is a normally distributed error term. As explained previously, the fi rst bracketed term in Equation D 1 is the absolute value of the differences in GDP per capita, the second term represen ts combined economic size, and the third bracketed term is a dispersion index measuring the relative economic size of the two countries. The secondary model, or the MTPLAB specification, is defined as Equation 37 and 44, or ** 01 2 345ln lnlnlnlnjt it ijt it jtij it jtY Y X NNDNN 67 1ln ln ,M it jtmmijijt mRemotenessRemotenessWu (D 4) where Xijt *, Yi t, Yj t, Ni t, Nj t, Dij, Remotenessit, Remotenessjt, and Wmij are defined as above, and ijtu is a normally distributed disturbance. Since the dataset s are Mexico centered, modifications have to be made to vector w The only binary variables included in the following estimations are the dummy variables controlling for adjacency, sharing a common language, Spains post colonial influence, the effect exerted by NAFTA, and the EU being the importer. Thus, due to the structure of the data, the analysis only accounts for five binary effects. The other binary variables would naturally drop out of the estimation, since they would all equal zero for all years Each specifi cation of the model, Equations D 1 and D 4, is estimated three times for each year, i.e., once for each dataset. However, each model is estimated for eight years only covering the time period 1994 through 2001. Clearly, the NAFTA variable would equal 0 in 1992 and 1993 and statistical estimation

PAGE 163

163 would not be possible since this variable would cause perfect collinearity with the intercept term in two those years. Tables D2 through D 7 presents these Mexico centered estimations for 1994 through 2001. Table D 1 indexes these estimations for easier reference. Empirical estimation is performed using GAUSS econometric software (Aptech Systerms Inc. 2001). The evidence suggests that the MH model fits the data well with over all explanatory power. Judging from the calculated r squared values, the MH model explains 71 82 percent of the variations in Mexicos bilateral trade transactions. This is somewhat higher than when using the original comprehensive data ( see Tables 4 2 through 4 7) It seems that the gravity methodology is applicable when looking at a dataset centered around one particular country, at least empirically. In fact, the over all explanatory power of the model is increased when using the Mexico centered data. Looking at the empirical performance of the MH model, a couple of striking observations emerge. T he bilateral difference in resource endowments, as proxied by the absolute bilateral difference in per capita incomes, is not statistically significant. T he estimated parameters are not statistically dif ferent from zero at conventional levels in all but three instances for aggregate trade and agricultural trade. For manufactured goods trade, the estimated parameters are statistically zero for all years. The evidence suggests that traditional HO theory i s not significant in determining Mexicos bilateral trade transactions, at least not with these data. Interestingly, HOtheory is important in explaining over all trade volumes in the Western Hemisphere, with positive statistically significant parameters obtained from the comprehensive dataset, as was reported in Chapter 4. The remaining two Helpman (1987) variables, the bilateral sum of the trading partners respective GDPs and the dispersion index, perform well yielding statistically significant

PAGE 164

164 parameter estimates at conventional levels for all three datasets for all years. Combined economic size has a positive influence on Mexicos bilateral trade volumes, with marginally larger effects observed for manufactured goods trade. The estimated par ameters diminish in magnitude but not in statistical significance, for aggregate trade and even more so for agricultural trade. In particular, the confidence intervals of the parameters from the agricultural trade data and the manufactured goods trade da ta are not overlapping at conventional levels. The combined market effect is statistically larger for manufactured goods trade than for agricultural trade. The estimated parameters of the dispersion index suggest that Mexico trades more extensively with economies of similar size. The effect is stronger for aggregate trade and manufactured goods trade than for agriculture. As indicated previously, agriculture is unique and should be modeled separately. The bilateral distance between Mexico and its trad ing partners play a statistically significant role, as economic gravity would suggest. The estimated parameters are negative at conventional levels and the findings are robust across data sets. Distance seems to have a stronger effect on aggregate trade and manufactured goods versus agricultural products. T his provides support to the idea that Mexicos agricultural exports are shipped large distances and transportation costs would be less important as a determinant. The empirical performance of the r emoteness variables is just as ambiguous for Mexicos bilateral trade transactions as for the rest of the Western Hemisphere. The remoteness of the exporter exerts a positive, significant effect at conventional levels for aggre gate trade, but less so for agricultural trade The effect diminishes entirely when isolating trade in manufactur es The importer s remoteness, geographic adjacency and sharing a common language do not seem to exert significant influence s on Mexicos trade. Furthermore, Spains ne ocolonial trade

PAGE 165

165 distortion with Mexico is marginal and only significant for five years using aggregate trade and only one year when using manufactured goods trade. Mexicos agricultural trade does not seem to be affected by any neo imperial policy disto rtions. This should not come as a surprise when considering the marginal colonial export shares of Mexico, see Table 1 4. The EU importer bias is not present for agricultural exports, but there is a negative to ambiguous effect for aggregate trade and tr ade in manufactured products. The NAFTA binary effect is statistically significant and positive for all years using the agricultural data set. This is in contrast with the finding from using the comprehensive data in Chapter 4. When sorting away the noise provided by the full FTAA, focusing on Mexicos trade flows exclusively reveals a strong upward bias in trade volumes resulting from NAFTA. This effect diminishes for aggregate trade and manufactured goods trade. Thus, NAFTA seem to have had an impact on agricultural trade in particular and has definitely shaped the agricultural trade behavior of its members There is empirical support for the claim of American and Canadian farm communities that NAFTA has resulted in a n increase of agricultural products exported from M exico to the U.S. and Canada. This is of significance as major provisions of NAFTA are centered on the liberaliz ation of agricultural trade and the harmonization of agricultural policy. Using the methodology from Chapter 4, the m arginal effect of NAFTA on Mexicos agricultural trade volumes is a factor of roughly magnitude 7 times the benchmark trade behavior (e.g., the exponential of 2 equals 7.39) In this case, the benchmark trade behavior would be that between Mexico and its trading partners with which they do share a geographic border, do not share a common language, one of the trading partners is not Spain, the trading partner is not a NAFTA member, and the EU is not the importing party. Thus, examples of the benchmark trad e

PAGE 166

166 pattern would be the bilateral trade between Mexico and Brazil, Mexico and Barbad os, or Mexico and Australia, and so forth. Reassuringly, the MTPLAB model performs well empirically with high explanatory power and over all statistical significance. Th e gravity variables conform to a priori expectations and reveal nothing new vis vis the results obtained from the original comprehensive data. In fact, in most instances, constructed confidence intervals for estimated parameters overlap at conventional levels. Using the MTPLAB specification as a robustness check for the NAFTA variable, the evidence suggests that the effects are similar to those found using the MH model. For trade in agricultural products, the NAFTA variable is statistically significant and positive for all years (except for 1994) T he estimated parameters fall within the same confidence interval at conventional levels vis vis the parameters from the MH specification. This provides supports to the assertions above that NAFTA has significantly influenced agricultural trade flows of member states in general and Mexico in particular. For aggregate trade and manufactured goods trade, NAFTA trade flows behave in accordance with the benchmark trade volumes with estimated parameters statistically not different from zero; the e xponential of zero equals unity In summary, the NAFTA variable performs poorly for aggregate and manufactured goods trade. This conclusion holds true for both the Mexicocentered data and the comprehensive data. Ho wever, when focusing on Mexicos trade volumes exclusively the NAFTA variable becomes statistically significant in both the MH model and the MTPLAB model for trade in agricultural products Thus, when omitting any non Mexican trade flows, the evidence s uggests that NAFTA indeed has had a significant influence on agricultural trade.

PAGE 167

167 Table D 1. Regression index for Mexico centered estimations Table Empirical specification Dat a set D 2 MH specification Aggregate trade (UNCTAD total) D 3 MH specificati on Agricultural trade (UNCTAD cat. 1 & 2) D 4 MH specification Manufactured goods trade (UNCTAD cat. 5) D 5 MTPLAB specification Aggregate trade (UNCTAD total) D 6 MTPLAB specification Agricultural trade (UNCTAD cat. 1 & 2) D 7 MTPLAB specification Man ufactured goods trade (UNCTAD cat. 5) All observations feature Mexico as being either the exporting country or the importing country.

PAGE 168

168 Table D 2. MH specification: aggregate trade (UNCTAD total) Mexico centered data Dependent variable: ln(exports fro m country i to country j in U.S. dollars) 1994 1995 1996 1997 1998 1999 2000 2001 Constant 43.239 *** 47.321 *** 44.623 *** 43.084 *** 46.692 *** 40.166 *** 41.712 *** 41.412 *** (14.804) (13.273) (12.925) (12.232) (12.172) (12.548) (14.197) (12.314) Bilateral GDP difference a 0.001 0.349 0.354 0.293 0.287 0.171 0.099 0.353 (0.200) (0.202) (0.218) (0.242) (0.229) (0.280) (0.240) (0.260) Bilateral sum of GDPs b 2.176 *** 2.123 *** 2.222 *** 1.936 *** 2.097 *** 2.107 *** 2.058 *** 1.969 *** (0.217) (0.181) (0.192) (0.194) (0.220) (0.241) (0.266) (0.265) B ilateral similarity of GDPs c 1.352 *** 1.338 *** 1.602 *** 1.384 *** 1.467 *** 1.552 *** 1.411 *** 1.583 *** (0.111) (0.112) (0.221) (0.208) (0.208) (0.228) (0.215) (0.266) Distance d 1.622 *** 1.736 *** 1.928 ** 1.657 *** 1.881 *** 1.667 *** 1.654 *** 1.475 *** (0.370) (0.395) (0.475) (0.427) (0.448) (0.435) (0.446) (0.418) Remoteness exporter e 1.273 1.269 ** 1.578 ** 1.235 1.748 *** 1.472 ** 1.760 ** 1.696 *** (0.678) (0.625) (0.683) (0.668) (0.618) (0.652) (0 .718) (0.639) Remoteness importer e 0.845 1.547 0.866 1.636 1.290 0.591 0.611 0.735 (1.221) (1.234) (1.190) (1.104) (1.110) (1.045) (1.082) (0.898) Common border f 0.651 0.537 0.168 0.449 0.219 0.093 0.235 0.107 (0.684) (0.781) (0.653) (0.734) (0.7 33) (0.580) (0.732) (0.596) Common language g 0.383 0.317 0.675 0.274 0.046 0.491 0.246 0.739 (0.423) (0.474) (0.500) (0.430) (0.432) (0.397) (0.439) (0.376) Colonial linkage: Spain h 1.275 *** 1.001 0.539 0.879 1.242 *** 0.714 1.200 ** 0.529 (0.443) (0.553) (0.658) (0.473) (0.461) (0.461) (0.537) (0.411) NAFTA k 0.197 0.320 0.658 0.778 0.528 0.910 0.839 1.098 ** (0.495) (0.563) (0.660) (0.576) (0.539) (0.536) (0.521) (0.517) EU i mporter l 0.863 0.577 1.041 0.379 1.097 ** 1.077 ** 1.006 ** 0.88 5 ** (0.559) (0.568) (0.565) (0.481) (0.492) (0.492) (0.480) (0.405) R squared 0.819 0.809 0.768 0.749 0.757 0.745 0.715 0.719 F statistic 40.629 98.701 29.561 25.902 29.093 26.327 23.413 23.246 N 120 121 119 116 124 120 124 121 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms. *) Statistically significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level.

PAGE 169

169 Table D 2. Continued a) lnY Y jt it NN itjt b) ln YY itjt c ) 22 ln1 Y Y jt it YYYY itjtitjt d) Bilateral distance in kilometers between the capitals of trading partners, e) Remoteness it Y Yjt d ij ji wt Binary variables: f) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. g) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. h) E qual to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere (e.g., Mexico), 0 otherwise. k) NAFTA dummy: Equal to 1 if both trading partners are members of NAFTA, 0 otherwise. l) Equal to 1 if the impor ting country is a member of the European Union, 0 other wise.

PAGE 170

170 Table D 3. MH specification: agricultural trade (UNCTAD categories 1 & 2) Mexico centered data Dependent variable: ln(exports from country i to country j in U.S. dollars) 1994 1995 1996 1997 1998 1999 2000 2001 Constant 28.552 38.855 ** 43.7 79 *** 31.755 *** 37.958 *** 28.996 ** 29.012 ** 37.533 *** (18.753) (14.882) (13.689) (11.150) (12.187) (13.145) (14.333) (13.782) Bilateral GDP difference a 0.057 0.435 0.190 0.374 0.291 0.148 0.113 0.235 (0.279) (0.234) (0.189) (0.202) (0.181) (0. 194) (0.216) (0.194) Bilateral sum of GDPs b 1.713 *** 1.507 *** 1.699 *** 1.572 *** 1.593 *** 1.531 *** 1.495 *** 1.549 *** (0.287) (0.236) (0.261) (0.244) (0.250) (0.249) (0.267) (0.279) Bilateral similarity of GDPs c 1.069 *** 0.949 *** 1.016 *** 0.941 *** 0.873 ** 1.007 *** 1.026 *** 1.004 *** (0.106) (0.129) (0.101) (0.078) (0.077) (0.109) (0.093) (0.112) Distance d 1.347 *** 0.765 ** 1.109 *** 0.935 *** 1.120 *** 0.956 ** 1.216 *** 1.039 *** (0.353) (0.381) (0.300) (0.323) (0.321) (0.469) (0.394) (0.369) Remo teness exporter e 1.816 1.634 2.196 *** 0.936 1.764 ** 1.336 1.535 1.496 (0.999) (0.909) (0.751) (0.733) (0.780) (0.874) (0.939) (0.899) Remoteness importer e 0.538 0.746 0.582 0.654 0.660 0.112 0.304 0.955 (1.425) (1.076) (1.012) (0.697) (0.777) (0.8 75) (0.865) (0.775) Common border f 0.826 1.384 ** 0.479 1.231 ** 0.710 0.799 1.112 0.501 (0.578) (0.586) (0.399) (0.582) (0.527) (0.539) (0.579) (0.414) Common language g 0.849 1.238 ** 0.987 *** 1.196 *** 0.842 ** 1.292 ** 0.931 ** 1.111 ** (0.431) (0.502) (0.355) (0.451) (0.407) (0.557) (0.470) (0.445) Colonial linkage: Spain h 0.717 0.286 0.484 0.091 0.527 0.212 0.447 0.343 (0.529) (0.532) (0.511) (0.434) (0.430) (0.540) (0.472) (0.469) NAFTA k 1.629 ** 2.096 *** 2.098 *** 1.549 *** 1.846 *** 2.082 *** 1.725 *** 2.105 *** (0.656) (0.594) (0.529) (0.479) (0.511) (0.522) (0.554) (0.508) EU i mporter l 0.829 0.266 0.421 0.236 0.176 0.221 0.045 0.080 (0.645) (0.526) (0.512) (0.428) (0.467) (0.438) (0.431) (0.367) R squared 0.709 0.712 0.748 0.7 79 0.730 0.704 0.727 0.768 F statistic 20.941 21.394 24.746 29.187 23.905 20.178 23.737 28.639 N 115 116 112 111 118 114 119 116 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms. *) Statistical ly significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level.

PAGE 171

171 Table D 3. Continued a) lnY Y jt it NN itjt b) ln YY itjt c ) 22 ln1 Y Y jt it YYYY itjtitjt d) Bilateral distance in kilometers between the capitals of trading partners, e) Remoteness it Y Yjt d ij ji wt Binary variables: f) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. g) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. h) Equal to 1 if one of the trading partners is Spain and the other a former Spanish colony in the Western Hemisphere (e.g., Mexico), 0 otherwise. k) NAFTA dummy: Equal to 1 if both trading partners are members of NAFTA, 0 otherwise. l) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 172

172 Table D 4. MH specification: manufactured goods trade (UNCTAD category 5) Mexico centered data Depende nt variable: ln(exports from country i to country j in U.S. dollars) 1994 1995 1996 1997 1998 1999 2000 2001 Constant 50.289 *** 53.886 *** 44.216 *** 49.202 *** 45.113 *** 37.413 *** 43.417 ** 51.503 *** (16.491) (16.451) (14.306) (15.633) (14.413) ( 14.026) (16.041) (14.349) Bilateral GDP difference a 0.138 0.177 0.225 0.270 0.090 0.143 0.108 0.200 (0.244) (0.269) (0.206) (0.258) (0.271) (0.370) (0.239) (0.257) Bilateral sum of GDPs b 2.394 *** 2.323 *** 2.340 *** 2.107 *** 2.311 *** 2.319 *** 2.266 *** 2.133 *** (0.290) (0.248) (0.249) (0.276) (0.286) (0.306) (0.336) (0.325) Bilateral similarity of GDPs c 1.331 *** 1.384 *** 1.294 *** 1.238 *** 1.251 *** 1.436 *** 1.257 *** 1.249 *** (0.119) (0.140) (0.109) (0.128) (0.115) (0.221) (0.138) (0.124) Distance d 1.406 *** 1.568 *** 1.418 *** 1.521 *** 1.470 *** 1.222 *** 1.046 ** 0.850 ** (0.365) (0.451) (0.399) (0.402) (0.404) (0.434) (0.468) (0.411) Remoteness exporter e 0.057 0.075 0.097 0.262 0.575 0.032 0.498 0.693 (0.770) (0.804) (0.731) (0.746) (0.723) (0.733) (0.795) (0.705) Remoteness importer e 1.849 2.799 1.514 2.536 1.133 0.579 0.731 1.644 (1.273) (1.444) (1.213) (1.351) (1.102) (1.029) (1.076) (1.177) Common border f 0.901 0.898 0.133 0.124 0.238 0.034 0.340 0.023 (0.638) (0.769) (0.638) ( 0.712) (0.766) (0.656) (0.871) (0.627) Common language g 0.740 0.438 0.718 0.175 0.158 0.769 0.677 0.774 (0.409) (0.528) (0.507) (0.492) (0.469) (0.438) (0.535) (0.470) Colonial linkage: Spain h 0.614 0.501 0.140 0.824 0.881 0.156 0.203 0.053 (0. 529) (0.521) (0.512) (0.509) (0.503) (0.562) (0.566) (0.638) NAFTA k 0.081 0.213 0.901 1.232 0.801 1.197 1.369 ** 1.721 *** (0.565) (0.654) (0.685) (0.706) (0.617) (0.609) (0.637) (0.605) EU i mporter l 0.761 0.052 0.726 0.283 1.085 ** 1.042 ** 0.962 0.682 (0.557) (0.626) (0.524) (0.518) (0.522) (0.502) (0.493) (0.503) R squared 0.787 0.773 0.787 0.773 0.759 0.713 0.748 0.772 F statistic 33.061 30.638 32.020 28.969 28.828 22.119 27.426 30.254 N 119 120 116 114 122 119 123 119 Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms. *) Statistically signif icant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level.

PAGE 173

173 Tab le D 4. Continued a) lnY Y jt it NN itjt b) ln YY itjt c ) 22 ln1 Y Y jt it YYYY itjtitjt d) Bilateral distance in kilometers between the capitals of trading partners, e) Remoteness it Y Yjt d ij ji wt Binary variable s: f) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. g) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. h) E qual to 1 if one of the trading partners is Spain a nd the other a former Spanish colony in the Western Hemisphere (e.g., Mexico), 0 otherwise. k) NAFTA dummy: Equal to 1 if both trading partners are members of NAFTA, 0 otherwise. l) Equal to 1 if the impor ting country is a member of the European Union, 0 otherwise.

PAGE 174

174 Table D 5. MTPLAB specification: aggregate trade (UNCTAD total) Mexico centered data Dependent variable: ln(exports from country i to country j in U.S. dollars) 1994 1995 1996 1997 1998 1999 2000 2001 Constant 59.324 *** 57.308 *** 63.06 7 *** 63.525 *** 67.909 *** 62.270 *** 58.644 *** 65.569 *** (14.064) (14.072) (15.296) (14.453) (15.503) (14.948) (15.438) (14.778) GDP per capita exporter 1.619 *** 1.631 *** 1.932 *** 1.793 *** 1.712 *** 1.909 *** 1.701 *** 1.796 *** (0.164) (0.146) (0.209) (0.210) (0.177) (0.218) (0.240) (0.259) GDP per capita importer 1.279 *** 1.116 *** 1.259 *** 1.251 *** 1.278 *** 1.245 *** 1.268 *** 1.255 *** (0.161) (0.165) (0.165) (0.160) (0.233) (0.182) (0.184) (0.194) Population exporter 1.260 *** 1.365 *** 1.568 *** 1.47 0 *** 1.398 *** 1.547 *** 1.439 *** 1.567 *** (0.107) (0.080) (0.152) (0.166) (0.086) (0.166) (0.179) (0.225) Population importer 1.172 *** 1.018 *** 1.071 *** 0.967 *** 1.199 *** 1.051 *** 1.061 *** 1.169 *** (0.075) (0.074) (0.089) (0.074) (0.199) (0.092) (0.096) (0.105) Distance a 1.563 *** 1.561 *** 1.686 *** 1.544 *** 1.732 *** 1.531 *** 1.553 *** 1.354 *** (0.335) (0.295) (0.333) (0.298) (0.356) (0.336) (0.369) (0.352) Remoteness exporter b 1.894 *** 1.621 *** 2.104 *** 1.804 ** 2.288 *** 2.455 *** 2.477 *** 2.485 *** (0.710) (0.688) (0.790) (0.784) (0.701) (0.818) (0.832) (0.846) Remoteness importer b 0.799 1.120 0.466 1.228 1.170 0.018 0.052 0.032 (1.208) (1.240) (1.248) (1.157) (1.155) (1.157) (1.179) (1.021) Common border c 0.585 0.465 0.168 0.269 0.041 0.003 0.005 0.202 (0.582) (0.486) (0.591) (0.458) (0.510) (0.504) (0.438) (0.498) Common language d 0.706 0.546 1.023 *** 0.691 ** 0.405 0.796 ** 0.568 0.995 *** (0.367) (0.349) (0.379) (0.333) (0.328) (0.332) (0.374) (0.354) Colonial linkage: Spain e 0.977 *** 0.668 0.122 0.442 0.789 0.458 0.944 0.214 (0.416) (0.514) (0.673) (0.458) (0.424) (0.459) (0.552) (0.441) NAFTA f 0.349 0.037 0.328 0.081 0.073 0.066 0.148 0.094 (0.521) (0.486) (0.680) (0.605) (0.594) (0.602) (0.610) (0.728) EU i mporter g 0.886 0.502 1.050 0.715 1.012 1.436 ** 1.490 ** 1.310 ** (0.589) (0.657) (0.648) (0.552) (0.539) (0.606) (0.590) (0.557) R squared 0.829 0.842 0.816 0.819 0.765 0.802 0.755 0.747 F statistic 43.359 48.200 39.145 38.894 30.274 36.125 2 8.491 26.556 N 120 121 119 116 124 120 124 121

PAGE 175

175 Table D 5. Continued Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms. *) Statisticall y significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level. a) Bilateral distance in kilometers between the capitals of trading partners. b) Remoteness it Y Yjt d ij ji wt Binary variables: c) Equal to 1 if the two trading partne rs are contingent, i.e., they share a common border, 0 otherwise. d) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. e) Equal to 1 if one of the trading partners is Spain and the other a former Spanis h colony in th e Western Hemisphere (e.g., Mexico), 0 otherwise. f) NAFTA dummy: Equal to 1 if both trading partners are members of NAFTA, 0 otherwise. g) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 176

176 Table D 6. MTPLAB specification: agricultural trade (UNCTAD categories 1 & 2) Mexico centered data Dependent variable: ln(exports from country i to country j in U.S. dollars) 1994 1995 1996 1997 1998 1999 2000 2001 Constant 45.438 *** 41.110 *** 50.571 *** 31.921 *** 38.061 *** 36.069 ** 40.438 *** 45.626 *** (16.803) (16.177) (13.479) (11.594) (13.104) (15.265) (14.359) (14.155) GDP per capita exporter 1.234 *** 0.965 *** 1.122 *** 0.935 *** 0.888 *** 1.018 *** 1.091 *** 1.065 *** (0.214) (0.203) (0.181) (0.175) (0.187) (0.211) (0 .189) (0.194) GDP per capita importer 1.387 *** 1.002 *** 0.992 *** 0.855 *** 0.839 *** 0.887 *** 0.962 *** 0.824 *** (0.248) (0.157) (0.149) (0.126) (0.132) (0.155) (0.155) (0.117) Population exporter 0.928 *** 0.991 *** 0.974 *** 1.005 *** 0.938 *** 1.020 *** 1.01 4 *** 0.991 *** (0.115) (0.112) (0.097) (0.094) (0.085) (0.090) (0.088) (0.106) Population importer 0.937 *** 0.805 *** 0.889 *** 0.864 *** 0.809 *** 0.816 *** 0.862 *** 0.885 *** (0.079) (0.072) (0.076) (0.068) (0.065) (0.088) (0.083) (0.080) Distance a 1.336 *** 0.622 1.037 *** 0.817 ** 1.029 *** 0.848 1.136 *** 0.973 *** (0.401) (0.363) (0.320) (0.318) (0.328) (0.468) (0.391) (0.354) Remoteness exporter b 2.161 ** 1.198 2.302 *** 0.556 1.585 1.403 1.731 1.733 (1.050) (1.099) (0.856) (0.876) (0.993) (1.14 8) (1.103) (1.101) Remoteness importer b 0.158 0.381 0.507 0.221 0.413 0.197 0.031 0.626 (1.317) (1.157) (1.037) (0.816) (0.859) (1.020) (0.937) (0.812) Common border c 0.746 1.295 ** 0.519 1.220 ** 0.735 0.831 0.973 0.464 (0.583) (0.527) (0.452) (0. 541) (0.462) (0.600) (0.527) (0.388) Common language d 1.227 *** 1.357 *** 1.126 *** 1.302 *** 0.874 ** 1.426 *** 1.153 *** 1.214 *** (0.429) (0.452) (0.340) (0.414) (0.361) (0.499) (0.434) (0.400) Colonial linkage: Spain e 0.448 0.073 0.294 0.154 0.410 0.067 0.217 0.145 (0.460) (0.504) (0.496) (0.430) (0.404) (0.483) (0.438) (0.426) NAFTA f 1.064 1.837 *** 1.783 *** 1.385 ** 1.778 *** 1.702 *** 1.260 ** 1.801 *** (0.800) (0.668) (0.551) (0.529) (0.498) (0.565) (0.617) (0.528) EU i mporter g 1.141 0.507 0.419 0 .171 0.259 0.427 0.258 0.121 (0.585) (0.598) (0.561) (0.500) (0.519) (0.550) (0.511) (0.426) R squared 0.727 0.713 0.750 0.778 0.731 0.714 0.733 0.770 F statistic 22.711 21.273 26.762 28.602 23.745 21.014 24.203 28.748 N 115 116 112 111 118 114 119 116

PAGE 177

177 Table D 6. Continued Whites robust standard errors in parenthesis. All continuous variables are measured in natural logarithms. *) Statisticall y significant at the 10 % level. **) Statistically significant at the 5 % level. ***) Sta tistically significant at the 1 % level. a) Bilateral distance in kilometers between the capitals of trading partners. b) Remoteness it Y Yjt d ij ji wt Binary variables: c) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. d) Equal to 1 if the two trading partners share a common commercial language, 0 otherwise. e) Equal to 1 if one of the trading partners is Spain and the other a former Spanis h colony in the Western Hemisphere (e.g., Mexi co), 0 otherwise. f) NAFTA dummy: Equal to 1 if both trading partners are members of NAFTA, 0 otherwise. g) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 178

178 Table D 7. MTPLAB specification: manufactured goods trade ( UNCTAD category 5) Mexico centered data Dependent variable: ln(exports from country i to country j in U.S. dollars) 1994 1995 1996 1997 1998 1999 2000 2001 Constant 60.724 *** 65.609 *** 52.884 *** 67.703 *** 56.991 *** 58.322 *** 54.059 *** 68.466 *** (15.526) (17.354) (14.770) (15.614) (13.851) (15.853) (15.102) (16.037) GDP per capita exporter 1.671 *** 1.850 *** 1.775 *** 1.911 *** 1.834 *** 2.115 *** 1.787 *** 1.605 *** (0.207) (0.212) (0.151) (0.173) (0.184) (0.344) (0.189) (0.203) GDP per capita im porter 1.315 *** 0.980 *** 1.137 *** 1.203 *** 1.112 *** 1.139 *** 1.212 *** 1.343 *** (0.200) (0.196) (0.180) (0.196) (0.176) (0.186) (0.190) (0.216) Population exporter 1.311 *** 1.463 *** 1.437 *** 1.418 *** 1.404 *** 1.518 *** 1.416 *** 1.324 *** (0.128) (0.096) (0.072) (0.088) (0.072) (0.144) (0.089) (0.098) Population importer 1.207 *** 1.031 *** 0.996 *** 0.953 *** 0.955 *** 0.970 *** 0.939 *** 0.995 *** (0.099) (0.097) (0.091) (0.076) (0.071) (0.086) (0.089) (0.095) Distance a 1.429 *** 1.452 *** 1.445 *** 1.616 ** 1.517 *** 1.285 *** 1.092 *** 0.921 ** (0.343) (0.324) (0.289) (0.301) (0.316) (0.336) (0.379) (0.374) Remoteness exporter b 0.674 0.809 0.574 1.287 1.598 ** 1.738 1.606 1.530 (0.829) (0.936) (0.771) (0.784) (0.769) (0.934) (0.850) (0.806) Remoten ess importer b 1.693 2.364 1.213 2.275 0.866 0.064 0.027 1.702 (1.310) (1.459) (1.264) (1.363) (1.187) (1.092) (1.210) (1.530) Common border c 0.832 0.887 0.206 0.000 0.159 0.234 0.206 0.058 (0.509) (0.527) (0.402) (0.398) (0.401) (0.507) (0.436) (0. 449) Common language d 0.944 ** 0.657 0.788 ** 0.420 0.324 0.978 *** 0.884 ** 0.862 ** (0.376) (0.397) (0.334) (0.327) (0.305) (0.336) (0.386) (0.379) Colonial linkage: Spain e 0.370 0.183 0.022 0.523 0.742 ** 0.215 0.160 0.119 (0.476) (0.393) (0.413) (0 .337) (0.363) (0.414) (0.418) (0.521) NAFTA f 0.302 0.130 0.303 0.650 0.541 0.431 0.802 1.072 (0.535) (0.752) (0.562) (0.658) (0.537) (0.698) (0.496) (0.551) EU i mporter g 0.837 0.260 0.753 0.482 1.234 1.342 ** 1.587 *** 1.186 (0.688) (0.816) (0.607) (0.658) (0.632) (0.607) (0.600) (0.596) R squared 0.800 0.829 0.845 0.856 0.837 0.803 0.825 0.809 F statistic 35.278 43.292 46.926 50.028 46.739 35.970 43.204 37.520 N 119 120 116 114 122 119 123 119

PAGE 179

179 Table D 7. Continued Whites ro bust standard errors in parenthesis. All continuous variables are measured in natural logarithms. *) Statistically signific ant at the 10 % level. **) Statistically significant at the 5 % level. ***) Statistically significant at the 1 % level. a) Bilat eral distance in kilometers between the capitals of trading partners. b) Remoteness it Y Yjt d ij ji wt Binary variables: c) Equal to 1 if the two trading partners are contingent, i.e., they share a common border, 0 otherwise. d) Equal to 1 i f the two trading partners share a common commercial language, 0 otherwise. e) Equal to 1 if one of the trading partners is Spain and the other a former Spanis h colony in the Western Hemisphere (e.g., Mexico), 0 otherwise. f) NAFTA dummy: Equal to 1 if both trading partners are members of NAFTA, 0 otherwise. g) Equal to 1 if the importing country is a member of the European Union, 0 otherwise.

PAGE 180

180 APPENDIX E SPECIFICATION TESTIN G Preliminaries While both empirical models MTPLAB and MH, perform satis fact orily with robust findings across model s there are nevertheless subtle differences. Thus, it becomes interesting to address which econometric specification should be utilized. The following appendix applies three model discrimination tests to assess whi ch formulation of the gravity model is preferred for these particular sets of data. These discrimination tests are designated to appendix status due to the weak statistical foundation of such procedures (e.g., Seale 1994), and the possibility of reaching inconclusive results without presenting remedies (e.g., Kmenta 1986). Therefore, the results presented here are primarily for exposition purposes and should not be viewed as a rigorous discrimina tion between the two models. If one permits, it is more of a data mining exercise. T he two models under consideration are 22 01 2 3ln ln ln ln1jt jt it it ijt itjt itjt itjt itjt YY YY X YY NN YYYY 45 6 1lnln lnM ij it jtmmijijt mDRemotenessRemotenessW (E 1) which represents the MH specification and ** 01 2 345ln lnlnlnlnjt it ijt it jtijit jtY Y X NNDNN 67 1ln lnM it jtmmijijt mRemotenessRemotenessWu (E 2) which represent s the MTPLAB specification. It should be noted that Equation E 1 is the direct equivalent of Equations 37 and 41, and Equation E 2 is the equivalent of Equations 38 and 4-

PAGE 181

181 4. As in previous chapters, Xijt represents the bilateral exports from country i to country j in time period t plus 1, Yi t is nominal GDP for country i in time period t Yj t is nominal GDP for country j in time period t Ni t is country i s population in time period t Nj t is country j s population in time period t and Dij is the bilateral distance in kilometers between the capitals of country i and country j Remotenessit and Remotenessjt are defined as itY D Yjt Remoteness ij ji wt ( E 3) and jtY D Yit Remoteness ij ij wt ( E 4) respectively. Wmijs are m binary variables in vector w corresponding to any qualitative relationships between country i and country j and ijtu and ijt are assumed to be normally distributed error terms. By observation one can conclude that these are tw o nonnested models; the first three terms of the MH model cannot be linearly derived from the MTPLAB specification It is useful to reconsider these two models in slightly relabeled matrix format: y=Z 2 2~0, IIDvI ( E 5) and y=X 2 1~0, IIDuI ( E 6) where y is a vector representing ln (Xijt ) Z represents the regressors from the MH specification, X represents the regressors from the MTPLAB specification represents the parameters from the MH model represents the parameters for the MTPLAB model, where

PAGE 182

182 06111'(,...,,,...,) and 07111'(,...,,,...,) and v and u are normally distributed error terms. Equation E 5 thusly represents the MH specif ication in matrix form and Equation E 6 analogously represents the MTPLAB specification in matrix form. One should subsequently be able to test which of the two nonnested specifications is econometrically preferred. To do so, this study employs three te sts, namely the J test, the Cox test, and the Likelihood D iscrimination Criterion (LDC). Any statistical estimations needed for these tests are performed using GAUSS econometric software (Aptech Systerms Inc. 2001). The J Test The J test was initially proposed by Davidson and MacKinnon (1981).1 Since there is no null hypothesis, the J test is not a c lassical statistical test. Instead, two non nested hypotheses are postulated: H1: y=Z 2 2~0, IIDvI H2: y=X 2 1~0, IIDuI where, again, y is a vector representing ln (Xijt ) Z represents the regressors from the MH specification X represents the regressors from the MTPLAB specification and are the estimated parameters for each respective model, and u and v are normally distributed error terms. Hypothesis H1 refers to the MH specification and hypothesis H2 refers to the MTPLAB specification The mechanics of the J test are as foll ows: one first tests the validity of one hypothesis and then tests the validity of the other hypothesis and subsequently compares the results. To test the validity of the MH model, t he first step is to estimate the following auxiliary regression where the models are artificially nested 1 A more colloquial discussion can be found in Greene (2000) and Kmenta (1986)

PAGE 183

183 ^ y=X ( E 7) where Z are the fitted values from the MH model. Parameter is subsequently tested for statistical significance, i.e., one tests whether the estimated is statistically different from zero. The resulting test statistic follows asymptotically ^ ^~0,1 ()aN se ( E 8) where se is the standard error of the estimated If the estimated parameter is statistically zero, then hypothesis H1 is rejected, and if is different from zero, one cannot reject H1. In this particular instance, if one rejects H1, the MTPLAB model is preferred over the MH model; and if one cannot reject H1, the MH specification is preferred over the MTPLAB model. The second step of the J test is to test the validity of the MTPLAB model This is analogously achieved by artificially nesting the fitted values from the MTPLAB specification X into the MH model via the estimation of ^y=Z ( E 9) A similar test is carried out regarding hypothesis H2 using the asymptotically normal test statistic from Equation E 9. If the estimated param eter is statistically zero, hypothesis H2 is rejected, and if the estimated parameter is statistically different from zero, one fails to reject H2. If H2 is rejected, the MH specification is preferred over the MTPLAB model; and if one cannot reject H2, the MTPLAB specification is preferred over the MH mode l. While the test is relatively easy to implement, contradictory conclusions can be reached. As indicated by Kmenta (1986): Thus it is perfectly possible that both H1 and H2 are rejected, or

PAGE 184

184 that they both are accepted. If they are both rejected, ne ither model is useful in explaining the variation in the dependent variable. If they are both accepted, the data are apparently not rich enough to discriminate between the two hypotheses (p. 597). Since the test statistic is asymptotically normal, the t est tends to be more accurate as N It should be noted that for any gravity model, the number of observations is quite large. Table E 1 presents the results from the J test. The evidence suggests that the estimated test statistic, is statistically different from zero in both cases for almost all years Thus, Kmentas scenario materialized and the J test is inconclusive. The M H specification is preferred over the M TPLAB specification and the MTPLAB specification is preferred over the MH specification The exceptions are manufactured goods trade for 1992 and 1993. For 1992, the test suggests that both models are rejected and for 1993 there is a slight preference for the MTPLAB specification These contradictory results would suggest that the data cannot discriminate between the two models, at least not using the J test. The Cox Test Another procedure for discriminating between nonnested models is the Cox test. This test was introduced by Cox (1961) and was contextualized further via the subsequent work by Paseran (1974), Pesaran and Deaton (1978), Davidson and McKinnon (1981), and McAleer (1985).2 Similar to the J test, the Cost test is not a classical statistical test either as there is no null hypothesis. The following two nonnested hypotheses are once again postulated: H1: y=Z 2 2~0, IIDvI H2: y=X 2 1~0, IIDuI 2 Again, for a more colloquial overview, see Greene (2000) and Kmenta (1986).

PAGE 185

185 where the variables and parameters are defined above. The Cox test, which is more involved than the J test, is carried out in a series of steps. The regressors from the MTPLAB specification X has to be tested against the regressors from the MH specification, Z and vice versa. The residuals from t he MH specification, ez, and t he MTPLAB specification ex, need to be obtained via z^eY-Z ( E 10) and x^eY-X ( E 11) One then regresses the fitted values from the MTPLAB specification ^X on the varia bles from the MH specification, Z which yields ^ 1''x ^ ( E 12) and the following residuals are obtained zx x ^^eX ( E 13) These residuals are subsequently regressed on the variables from the MTPLAB s pecification X via ^ 1''zx xz ( E 14) and the following error terms are calculated ^ xzxzx xz eeX ( E 15) The fitted values from the MH specification, ^Z are regressed on the variables from the MTPLAB specification X yielding

PAGE 186

186 ^ 1''z ^ ( E 16) The following error terms are obtained, xz z ^^eZ ( E 17) which consequently are regressed on the variables from the MH specification, Z such that ^ 1''xz zx ( E 18) and the following error terms are calculated zxzxz zx^eeZ ( E 19) Two Cox tests are conducted regarding the underlying nonnested hypotheses. The first Cox test tests whether hypothe sis H2 can be rejected, i.e., testing the hypothesis that the M TPLAB specification is correctly specified and that the MH specification is not. The test statistic, qA, for the first Cox test is defined as 2' ln '' 2 '' ''zz xxzxzx A xx xzxxzx xxzxzxn n nn q nnn ee eeee eeee eeee ( E 20) where qA is asymptotically normally distributed such that ~0,1a AqN

PAGE 187

187 The Cox test should henceforth be more accurate as N If the calculated qA statistic falls in the rejection region, one can conclude that X is not th e correct set of variables, i.e., the MH specification is the preferred specification for the gravity model. The second Cox test is regarding the validity of hypothesis H1, i.e., testing whether the MH specification represents the correct set of regresso rs vis vis the MTPLAB specification Similarly, the test statistic for hypothesis H1, qB, is calculated as follows 2' ln 2 'xx xzxzzz B zxzzxz zz xzxz zzn n nn q n nn ee ee ee ee ee ee ee ( E 21) where asymptotic normality is implied, such that ~0,1a BqN If the calcul ated qB falls within the rejection region, one concludes that Z not is the correct set of variables and the MTPLAB specification is preferred. Table E 2 presents the calculated test statistics for the Cox test, qA and qB. At best, the Cox test provides ambiguous results. The test is inconclusive for aggregate trade, with 1992 and 1993 being the exception, where the MTPLAB specification appears to be correctly specified. For agricultural trade in 1996, 1997, and 1998, the Cox test suggests that the MH s pecification is preferred at conventional levels while the MTPLAB model is rejected. For manufactured goods trade, in six of the ten years, the test conveys that the MTPLAB specification is the proper

PAGE 188

188 model. It seems that, at least marginally, manufactur ed goods trade should be modeled using the MTPLAB specification Even though the Cox test performs somewhat better than the J test, the variation in the data is not plethoric enough for robust conclusions. The Likelihood Dominance Criterion Pollak and Wales (1991) introduced a novel approach to selecting among nonnested models called the Likelihood Dominance Criterion (LDC). The LDC methodology has been employed by agricultural economists in discriminating between different demand models (Green, Hass aon, and Johnson 1995; Lariviere, Larue, and Chalfant 2000; Skripnichenko and Chen 2002) and in technical efficiency applications (Mbaga 2003). However, as Seale (1994) pointed out in his criticism of the approach, it lacks foundations in statistical theo ry. Due to its construction, the LDC does not test two different hypotheses by rejecting or failing to reject one of them (or both). Rather, the test ranks the hypotheses based on appropriateness of the model specification given the information contained in the data. As such, the LDC is less likely to produce inconclusive results (Pollak and Wales 1991). In slightly relabeled form, the nonnested hypotheses are HMH: y=Z 2 2~0, IIDvI ( E 22) HMTPLAB: y=X 2 1~0, IIDuI ( E 23) where the variables and parameters are defined above. The number of parameters in the MH model is less than the number of parameters in the MTPLAB model, so the MH model is the first hy pothesis ( HMH ) and the MTPLAB model is the second hypothesis ( HMTPLAB). Let nMTPLAB be the number of regressors in the MTPLAB specification and let nM H be the number of regressors in the MH specification

PAGE 189

189 Following Pollack and Wales (1991), the LDC t est permits three possible cases. Let LMTPLAB represent the likelihood value from the MTPLAB specification and let LM H represent the likelihood value from the MH specification. 2(r) is the critical chisquare value for a specified significance level wit h degrees of freedom r where r corresponds to ( nMTPLAB +1) and ( nM H + 1) depending on the model considered. For any specified significance level, the LDC prefers the MH specification to the MTPLAB specification if 22 112MTPLAB MHnn MTPLABMHLL ( E 24) the LDC is indecisive between the MH specification and the MTPLAB specification if 22 22 11 1122MTPLABMH MTPLAB MHnn nn MTPLABMHLL ( E 25) and the LDC prefers the MTPLAB specification to the MH specification if 22 112MTPLABMHnn MTPLABMHLL ( E 26) The evidence fro m the LDC, presented in Table E 3, suggests that the MTPLAB specification is preferred for all years for both aggregate trade and manufactured goods trade. However, for agricultural products, the story is quite different. Inclusive of most years, except for 1992 and 1993, the MH model is preferred over the MTPLAB model. Based on the LDC, aggregate trade and manufactured goods trade should be estimated by the MTPLAB specification and the MH model should be used for agricultural trade. It seems that aggre gate trade patterns have more in common with manufactured goods trade than trade in agricultural commodities. More specifically, trade in minerals, metals, and fuels seems to follow the same trade behavior as manufacturers. This would help to explain why the aggregate trade data

PAGE 190

190 estimates are biased in favor of trade in manufactured goods over trade in primary products. However, one should be cautious of putting too much weight behind these tests. I n particular, the LDC has very little founding in statistical theory (Seale 1994) Instead, as Seale (1994) suggested economic theory should be applied in discriminating between two nonnested models. Nevertheless, a couple of interesting observations from the LDC emerges. Recall that the MH specification was initially derived for intra industry trade, where a country exports and imports goods in the same product category; i.e., there is no specialization as per HO based comparative advantage. Based on these results, one can make the initially perplexing argument that agricultural commodities are traded based on intra industry trade patterns (since the MH specification is preferred for the primary trade data set) and manufactured goods are traded based on inter industry trade (since the MTPLAB specificatio n is preferred for trade in manufactures). While this idea is somewhat contrary to economic intuition, there is a logical explanation behind these results upon further reflection. Consider, for example, the trade in fresh apples. During certain parts of the year the United States produces and exports fresh apples (i.e., the summer months) and during other parts of the year, the United States imports fresh apples (i.e., the winter months). As such, fresh apples would be intra industry traded goods. A s imilar argument can be applied to an extended array of agricultural products where seasonality creates intraindustry trade patterns. Henceforth, it is plausible to observe that agricultural commodities, or any product whose production is characterized by seasonality, would follow such a trade behavior. Conversely, manufactured goods trade tends to be based on resource endowments, where countries partially specialize in, and export, goods in which they have a comparative advantage. For instance, the abse nce of television production in the United States in favor of production under lower opportunity costs in

PAGE 191

191 the Pacific Rim countries and the domination of United States production in, say, pharmaceuticals, industrial supplies, and capital equi pment, illustrate this point. Perhaps, then, the findings regarding the preferred model specification are not that startling after all. This is, of course, if one admits the results obtained from these tests.

PAGE 192

192 Table E 1. Results from the J test J test statistic Test 1 a Test 2 b Alpha Standard error Alpha Standard error Year Aggregate trade (UNCTAD total): 1992 0.5367* 0.1232 0.9181* 0.0676 1993 0.6151* 0.1141 0.9017* 0.0695 1994 0.9811* 0.0991 0.9938* 0.0398 1995 0.9217* 0.0973 0.9850* 0.0411 1996 0 .8427* 0.1014 0.9727* 0.0445 1997 0.9499* 0.0964 0.9906* 0.0397 1998 0.9976* 0.0924 0.9927* 0.0497 1999 0.8564* 0.1033 0.9739* 0.0462 2000 0.9088* 0.0881 0.9829* 0.0386 2001 0.8546* 0.0965 0.9709* 0.0455 Agricultural trade (UNCTAD categories 1 & 2): 1992 0.7574* 0.2206 0.9386* 0.1233 1993 0.8089* 0.1599 0.8684* 0.1345 1994 1.1697* 0.1456 1.2530* 0.1941 1995 1.0908* 0.1321 1.1895* 0.2275 1996 1.0550* 0.1370 1.2092* 0.3572 1997 1.1108* 0.1303 1.3365* 0.2744 1998 1.0782* 0.1187 1.6638* 0.4614 1999 1.0807* 0.1770 1.0901* 0.1897 2000 1.1032* 0.1371 1.3389* 0.2734 2001 1.0994* 0.2510 1.1382* 0.2960 Manufactured goods trade (UNCTAD category 5): 1992 0.0161 0.2413 0.0016 0.0551 1993 0.2334 0.2143 0.9998* 0.0494 1994 1.6115* 0.2696 1.03 00* 0.0518 1995 1.4789* 0.3012 1.0198* 0.0524 1996 1.0971* 0.2499 1.0051* 0.0494 1997 1.5268* 0.2432 1.0288* 0.0484 1998 1.3865* 0.2063 1.0222* 0.0458 1999 1.2686* 0.2478 1.0095* 0.0442 2000 1.8570* 0.4227 1.0173* 0.0422 2001 1.1064* 0.1872 1.0055* 0.0450 *) The test statistic falls in the rejection region at the 5 % significance level. a) Tests the validity of the MH specification. If the test statistic falls in the rejection region i.e., is statistically different from zero one cannot reject the MH specification. b) Tests the validity of the MTPLAB specification If the test statistic falls in the rejection region i.e., is statistically different from zero one cannot reject the MT PLAB specification.

PAGE 193

193 Table E 2. Results from the Cox test Cox test statistic Test 1 a Test 2 b Year Aggregate trade ( UNCTAD total): 1992 0.48242 6.21095* 1993 0.72654 5.4065* 1994 53.0483* 45.0191* 1995 15.6573* 57.0426* 1996 5.69692* 67.5373* 1997 38.0979* 99.3346* 1998 58.3483* 20.9768* 1999 5.9003* 56.6154* 2000 17.3351* 116.07* 2001 6.73329* 71.9923* Agricultural trade (UNCTAD categories 1 & 2): 1992 2.68232* 17.7694* 1993 4.60829* 7.22989* 1994 12.508 9* 2.85513* 1995 30.1295* 2.20854* 1996 146.607* 1.17163 1997 33.0517* 0.919 1998 95.5448* 0.18593 1999 26.4639* 18.2791* 2000 32.2791* 2.31413* 2001 15.9328* 11.4249* Manufactured goods trade (UNCTAD category 5): 1992 0.012981 94.9009* 1993 0.09064 120.75* 1994 0.79079 94.016* 1995 0.9731 143.852* 1996 3.62583* 176.736* 1997 0.73062 121.741* 1998 2.05531* 104.393* 1999 4.19719* 278.274* 2000 0.03518 1209.44* 2001 26.3391* 192.205* *) The test statis tic falls in the rejection region at the 5% significance level. a) Tests whether the MTPLAB specification is correct and the MH specification is incorrect. If the test statistic falls in the rejection region, the MTPLAB specification is rejected. b) Test s whether the MH specification is correct and the MTPLAB specification is incorrect. If the test statistic falls in the rejection region, the MH specification is rejected.

PAGE 194

194 Table E 3. Results from the Likelihood Dominance Criterion Log likelihood val ues MTPLAB ( L S ) Helpman ( L H ) L S L H LDC conclusion a Year Aggregate trade (UNCTAD total): 1992 2605.1728 2683.7462 78.5734 MTPLAB preferred 1993 2594.9804 2661.26092 66.2805 MTPLAB preferred 1994 3944.533 4175.50541 230.972 MTPLAB pre ferred 1995 4025.3435 4240.47359 215.13 MTPLAB preferred 1996 3648.5521 3832.2648 183.713 MTPLAB preferred 1997 4452.9931 4686.90107 233.908 MTPLAB preferred 1998 4428.3008 4559.1281 130.827 MTPLAB preferred 1999 3799.5082 3970.56694 171.059 MTPLAB preferred 2000 4057.2021 4294.79917 237.597 MTPLAB preferred 2001 3584.6778 3754.88182 170.204 MTPLAB preferred Agricultural trade (UNCTAD categories 1 & 2): 1992 1454.7132 1477.4216 22.7084 MTPLAB preferred 1993 1463.3166 147 1.3541 8.0375 MTPLAB preferred 1994 2224.0504 2213.2803 10.77 MH preferred 1995 2327.4211 2307.944 19.477 MH preferred 1996 2211.9481 2188.7933 23.155 MH preferred 1997 3111.9435 3088.4385 23.505 MH preferred 1998 2912.2174 2878.2956 33 .922 MH preferred 1999 2427.5405 2425.4397 2.1008 MH preferred 2000 2434.0351 2413.8926 20.142 MH preferred 2001 2523.1287 2520.8865 2.2422 MH preferred Manufactured goods trade (UNCTAD category 5): 1992 1399.2499 1548.5647 149.315 MTPLAB preferred 1993 1400.7973 1578.9329 178.136 MTPLAB preferred 1994 2303.1149 2468.6082 165.493 MTPLAB preferred 1995 2262.5562 2426.1014 163.545 MTPLAB preferred 1996 2181.9678 2361.8769 179.909 MTPLAB preferred 1997 3080.8813 3269.5422 188.661 MTPLAB preferred 1998 2848.494 3053.2817 204.788 MTPLAB preferred 1999 2503.3219 2724.5202 221.198 MTPLAB preferred 2000 2475.1342 2725.3063 250.172 MTPLAB preferred 2001 2569.0988 2776.5952 207.496 MTPLAB preferred a) The LDC prefers the MTPLAB specification to the MH specification if ( LS LH) > 1.075. The LDC prefers the MH specification to the MTPLAB specification if (LS LH) < 0.6455 (for 1992 and 1993) and ( LS LH) < 0.6333 (for 1994 through 2001). The LDC is indecisive bet ween the MH specification and the MTPLAB specification if 0.6455 < ( LS LH) < 1.075 (for 1992 and 1993) and 0.6333 < ( LS LH) < 1.075 (for 1994 through 2001).

PAGE 195

195 LIST OF REFERENCES Abrams, Richard K. 1980. International Trade Flows Under Flexible Exch ange Rates Economic Review of the Federal Reserve Bank of Kansas City 65(3): 3 10. Aitken, Norman D. 1973. The Effect of the EEC and EFTA on European Trade: A Temporal Cross Section Analysis American Economic Review 63(5): 881 892. Aiken, Norman D ., and Robert S. Obutelewicz. 1976. A Cross sectional Study of EEC Trade with the Association of African Countries Review of Economics and Statistics 58(4): 425433. Alcaly, Roger E. 1967. Aggregation and Gravity Models: Some Empirical Evidence J ournal of Regional Science 7(1): 61 731. Anderson, James E. 1979. A Theoretical Foundation of the Gravity Equation. American Economic Review 69(1): 106116. Anderson, James. E., and Eric va n Wincoop. 2003. Gravity with Gravitas: A Solution to the B order Puzzle American Economic Review 93(1): 170 192. Anderson, Kym, and Hege Norheim. 1993. From Imperial to Regional Trade Preferences: Its Effect on Europe's Intra and Extra regional Trade Review of World Economics / Weltwirtschaftliches Archiv 129(1): 78102. Antonucci, Daniele, and Stefano Manzocchi. 2006. Does Turkey Have a Special Trade Relation with the EU? A Gravity Model Approach. Economic Systems 30(2): 157169. Aptech Systems, Inc. 2001. Gauss for Windows Kernel rev. 3.6.17, G UI rev. 3.6.7, Maple Valley, Washington. Beals, Ralph. E., Milred B. Levy, and Leon N. Moses. 1967. Rationality and Migration in Ghana Review of Economics and Statistics 49(4): 480486. Beckman, W. 1956. Distance and the Pattern of Intra European Tra de Review of Economics and Statistics 38(1): 31 40. Bergstrand, Jeffrey H. 1985. The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical Evidence Review of Economics and Statistics 67(3): 474481. Bergstrand, Jef frey H. 1989. The Generalized Gravity Equation, Monopolistic Competition and the Factor Proportions Theory in International Trade Review of Economics and Statistics 71(1): 143153. Bernal, Richard L. 1994. From NAFTA to Hemispheric Free Trade Colu mbia Journal of World Business 29(3): 22 31.

PAGE 196

196 Boisso, Dale, and Michael Ferrantino. 1997. Economic Distance, Cultural Distance, and Openness in International Trade: Empirical Puzzles Journal of Economic Integration 12(4): 456484. Bouzas, Roberto. 2005. The New Regionalism and the Free Trade Area of the Americas: A Less Benevolent View CEPAL Review 85: 718. Brada, Joself C., and Jose A. Mendez. 1983. Regional Economic Integration and the Volume of Intra Regional Trade: A Comparison of Deve loped and Developing Country Experience Kyklos 36(4): 589603. Breuss, Fritz, and Peter Egger. 1997. Use and Misuse of Gravity Equations in European Integration Research Working Paper 93/1997, Austrian Institute of Economic Research. Breuss, Fri tz, and Peter Egger. 1999. How Reliable Are Estimators of East West Trade Potentials Based on Cross Section Gravity Analyses? Empirica 26(2): 81 94. Brown, Drusilla K, Kozo Kiyota, and Robert M. Stern. 2005. Computational Analysis of the Free Trade A rea of the Americas (FTAA) North American Journal of Economics and Finance 16(2): 153185. Brysk, Alison, Craig Parsons, and Wayne Sandholtz. 2002. After Empire: National Identity and Post colonial Families of Nations. European Journal of International Relations 8(2): 267305. Campo Rivas, Jose Antonio, and Rafael Tiago Juk Benke. 2003. FTAA Negotiations: Short Overview Journal of International Economic Law 6(3): 661694. Carey, H. C. 1858. Principles of Social Science Volume 1. Philadel phia, Pennsylvania : J.B. Lippincott & Co. Caribbean Community Secretariat. 2001. http://www.caricom.org. Carrothers, Gerald A. P. 1956. An Historical Review of the Gravity and Potential Concepts of Human Interaction Journal of the American Institute of Planners 22(2): 94 102. Cavanaugh, Joseph A. 1950. Formulation, Analysis and Testing of the Interactance Hypothesis American Sociological Review 15(6): 763766. Central Intelligence Agency. 2007. https://www.cia.gov/library/publications/the world factbook/index.html Cheng, I. H., and Y. Y. Tsai. 2008. Estimated the Staged Effect of Regional Economic Integration on Trade Volumes Applied Economics 40(13): 383393.

PAGE 197

197 Coe, David T., and Alexander W. Hoffmaister. 1998. NorthS outh Trade: Is Africa Unusual? International Monetary Fund Working Paper WP/9894. Combes, Pierre Phillipe, Miren Lafourcade, and Thierry Mayer. 2005. The Trade creating Effects of Business a nd Social Networks: Evidence from France. Journal of International Economics 66(1): 129. Cox, D. R. 1961. Tests of Separate Families of Hypothesis in Jerzy Neyman, ed. Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics Berkel ey, California : University of Berkeley Press. Daio, Xinshen, and Agapi Somwaru. 2000. An Inquiry on General Equilibrium Effects of MERCOSUR An Intertemporal World Model. Journal of Policy Modeling 22(5): 557588. Daio, Xinshen, and Agapi Somwar u. 2001. A Dynamic Evaluation of the Effects of a Free Trade Area of the Americas An Intertemporal, Global General Equilibrium Model Journal of Economic Integration 16(1): 2147. Das, Gouranga Gopal, and Soamiely Andriamananjara. 2006. Hubs and S pokes Free Trade Agreements in the Presence of Technology Spillovers: An Application to the Western Hemisphere. Review of World Economics/Weltwirtschaftliches Archiv 142(1): 3366. Davidson, Russell and James G. MacKinnon. 1981. Several Tests for M odel Specification in the Presence of Alternative Hypothesis Econometrica 49(3): 781793. Deardorff, Alan V. 1998. Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World In Jeffrey A Frankel, ed. The Regionalization of the Worl d Economy Chicago Illinois : The University of Chicago Press. DeGroot, Henri L. F. et al. 2004. The Institutional Determinants of Bilateral Trade Patterns Kyklos 57(1): 103123. DiMauro, Francesca. 2000. The Impact of Economic Integration on FDI a nd Exports: A Gravity Approach. CEPS Working Paper No 156. Dodd, Stuart Carter. 1950. The Interactance Hypothesis: A Gravity Model Fitting Physical Masses and Human Groups American Sociological Review 15(2): 245 256. Egger, Peter. 2000. A Note on the Proper Econometric Specification of the Gravity Equation. Economics Letters 66(1): 25 31. Egger, Peter. 2002. An Econometric View of the Estimation of the Gravity Models and the Calculations of Trade Potentials World Economy 25(2): 297312.

PAGE 198

198 Eic hengreen, Barry. 2004. What Macroeconomic Measures are Needed for Free Trade to Flourish in the Western Hemisphere? Latin American Politics and Society 46(2): 127. Eichengreen, Barry, and Douglas A. Irwin. 1998. The Role of History in Bilateral Tra de Flows In Jeffrey A Frankel, ed. The Regionalization of the World Economy Chicago, Illinois : The University of Chicago Press. Evenett, Simon J, and Wolfgang Keller. 2002. On Theories Explaining the Success of the Gravity Equation. Journal of Pol itical Economy 110(2): 281316. Feenstra, Robert C., and Alan M. Taylor. 2008. International Trade 1st ed. New York, New York : Worth Publishers. Frankel, Jeffrey, Ernesto Stein, and Shang jin Wei. 1995. Trading Blocks and the Americas: The Natural an d Unnatural, and the Super Natural Journal of Development Economics 47 (1): 61 95. Frankel, Jeffrey A., Ernesto Stein, and Shang Jin Wei. 1997. Regional Trading Blocks in the World Economic System Washington, District of Columbia : Institute f or International Economics. Frankel, Jeffrey A, Ernesto Stein, and Shang Jin Wei. 1998. Continental Trading Blocks: Are They Natural or Supernatural? in Jeffrey A Frankel, ed. The Regionalization of the World Economy Chicago, Illinois : The University of Chi cago Press. Frankel, Jeffrey A, and ShangJin Wei. 1998. Regionalization of World Trade and Currencies: Economics and Politics in Jeffrey A Frankel, ed. The Regionalization of the World Economy Chicago, Illinois : The University of Chicago Press. Free Trade Area of the Americas Administrative Secretariat. 2002. http://www.ftaa alca.org/alca_e.asp. Furtan, W. H., and Blain M. van Melle. 2004. Canadas Agricultural Trade in North Americ a: Do National Borders Matter? Review of Agricultural Econom ics 26(3): 317 331. Garman, George, Janis Peterson, and Debora Gilliard. 1998. Economic Integration in the Americas: 1975 1992. Journal of Applied Business Research 14(3): 1 12. Garnaut, Ross. 1994. Open Regionalism: Its Analytical Basis and Relev ance to the International System. Journal of Asian Economics 5(2): 273 290. Geraci, Vincent J. and Wilfred Prewo. 1977. Bilateral Trade Flows and Transportation Cost The Review of Economics and Statistics 59(1): 67 74.

PAGE 199

199 Glejser, H., and A. Dramais. 1969. A Gravity Model of Interdependent Equations to Estimate Flow Creation and Diversion Journal of Regional Science 9 (5): 439449. Goodman, Leo A. 1963. Statisti cal Methods for the Preliminary Analysis of Transaction Flows Econometrica 31 (1/2): 197208. Gould, David M. 1994. Immigrant Links to the Home Country: Empirical Implications for the U.S. Bilateral Trade Flows Review of Economics and Statistic s 76(2): 302316. Gould, David M. 1998. Has NAFTA Changed North American Trade? Federal Reserve Bank of Dallas Economic Review First Quarter: 12 23. Grant, Jason H. and Dayton M. Lambert. 2008. Do Regional Trade Agreements Increas e Members Agricultural Trade? American Journal of Agricultural Economics 90(3): 765782. Graunt, John. 1665. Natural and Political Observations Made upon the Bills of Mortality London, United Kingdom: John Martyn, printer to the Royal Society. Green, Richard D, Zuhair A Hassaon, and S. R. Johnson. 1995. Selection of Nonnested Demand Models Canadian Journal of Agricultural Economics 43(3): 485499. Greene, William H. 2000. Econometric Analysis 4 th ed. Upper Saddle River, New Jersey : Prentice Hall Inc. Grier, Robin M. 1999. Colonial Legacies and Economic Growth. Public Choice 98(3 4): 317335. Hamilton, Carl B., and L. Alan Winters. 1992. Opening up International Trade with Eastern Europe Economic Policy 7(14): 78116. Hammer, Carl, and Fred Charles Ikle. 1957. Intercity Telephone and Airline Traffic Related to Distance and the Propensity to Interact Sociometry 20 (4): 306316. Hassan, M. Kabir. 2001. Is SAARC a Viable Economic Block? Evidence from the Gravity Mode l Journal of Asian Economics 12(2): 263 290. Head, Keith, and John Ries. 1998. Immigration and Trade Creation: Econometric Evidence From Canada. Canadian Journal of Economics 31(1): 47 62. Helliwell, John F. 1997. National Borders, Trade, and Migr ation NBER Working Paper No 6027. Helpman, Elhanan, and Paul R. Krugman. 1985. Market Structure and Foreign Trade: Increasing Returns, Imperfect Competition, and the International Economy Cambridge Massachusetts : MIT Press.

PAGE 200

200 Helpman, Elhanan. 1987. Imperfect Competition and International Trade: Evidence from Fourteen Industrial Countries Journal of the Japanese and International Economies 1(1): 62 81. Hertel, Thomas, David Hummels, Marcos Ivanic, and Roman Kee ney. 2007. How Confident Can W e be of CGE based Assessments of Free Trade Agreements? Economic Modeling 24(4): 611635. Hewett, Edward A. 1976. A Gravity Model of CEMA Trade in Josef C. Brada, ed. Quantitative and Analytical Studies in East West Economic Relations Bloomington, Indiana : International Development Research Center. Ikle, Fred Charles. 1954. Sociological Relationships of Traffic and Population and Distance. Traffic Quarterly 8(2): 123136. Inter American Development Bank. 2003. DATA Intal CD Rom. Version 4.0. Isard, Walter, and Merton J. Peck. 1954. Location Theory and International and Interregional Trade Theory Quarterly Journal of Economics 68 (1): 97 114. Isard, Walter. 1954. Location Theory and Trade Theory: Short Run Analysis Quarterly Journal of Economics 68 (2): 305320. Isard, Walter. 1956. Location and Space Economy: A General Theory Relating Industrial Location, Market Areas, Land Use, Trade, and Urban Structure Cambridge Massachusetts: MIT Press, and New York New York : John Wiley & S ons, Inc. Kehoe, Patrick J., and Timothy J. Kehoe. 1994. Capturing NAFTAs Impact with Applied General Equilibrium Models Federal Reserve Bank of Minneapolis Quarterly Review 18(2): 1734. Kerr, William A. 2002. NAFTA and Beyond: Challenges for Extending Free Trade in the Hemisphere. Estey Centre Journal of International Law and Trade Policy 3(2): 224238. Kennedy, Peter. 1989. Nonnested Hypothesis Tests: A Diagrammatic Exposition Australian Economic Papers 28(52): 160165. Kleiman, Ephraim. 1976. Trade and the Decline of Colonialism Economic Journal 86(343): 459480. Kleiman, Ephraim. 1977. Heirs to Colonial Trade Journal of Development Economics 4(2): 93103. Kmenta, Jan. 1986. Elements of Econometrics New York, New York : Macmillan Publishing Company.

PAGE 201

201 Kouparitsas, Michael A. 1997. A Dynamic Macroeconomic Analysis of NAFTA Economic Perspectives 21(1): 1435. Lariviere, Eric, Bruno Larue, and Jim Chalfant. 2000. Modeling the Demand for Alcoholic Beverages and Adv ertising Specifications Agricultural Economics 22 (2): 147 162. Leamer, Edward E., and Robert M. Stern. 1970. Quantitative International Economics Boston, Massachusetts : Allyn and Bacon Incorporated. Lee, David R. 1995. Western Hemisphere Econom ic Integration: Implications and Prospects for Agricultural Trade. American Journal of Agricultural Economics 77(5): 1274 1282. Linnemann, Hans. 1966. An Econometric Study of International Trade Flows Amsterdam, Holland: North Holland Publishing Compa ny. Livingstone, Ian. 1976. The Impact of Colonialism and Independence on Export Growth in Britain and France. Oxford Bulletin of Economics & Statistics 38(3): 211218. Mayer, Thierry and Soledad Zignago. 2006. CEPII Distance Measures. http://www .cepii.fr/anglaisgraph/bdd/distances.htm. Mbaga, Msafiri D, et al. 2003. Assessing Technical Efficiency of Quebec Dairy Farms Canadian Journal of Agricultural Economics 51(1): 121137. McAleer, Michael. 1985. Specification Tests for Separate Model s: A Survey In Maxell King and David Giles, eds. Specification Analysis in the Linear Model, London. London, United Kingdom: Routledge and Kegan Paul Book Ltd. McCallum, John. 1995. National Borders Matter: Canada U.S. Regional Trade Patterns Am erican Economic Review 85(3): 615623. Mittlehammer, Ron C, George G. Judge, and Douglas J. Miller. 2000. Econometric Analysis Cambridge, United Kingdom: Cambridge University Press. Molle, Willem. 2006. The Economics of European Integration: Theory, P ractice, Policy 5th ed. Aldershot, United Kingdom: Ashgate Publishing Ltd. Newton, Sir Isaac. 1686. Philosophi Naturalis Principia Mathematica. London, United Kingdom: S. Peppys the Royal Society Press. Nicho l ls, Shelton, et al. 2000. The Sate of and Prospects for the Deepening and Widening of Caribbean Integration. The World Economy 23(9): 11621166. Nilsson, Lars. 2000. Trade Integration and the EU Economic Membership Criteria European Journal of Political Economy 16(4): 807 827.

PAGE 202

202 Oguledo, Victor Iwuagwu, and Craig R. Macphee. 1994. Gravity Models: A Reformulation and an Application to Discriminatory Trade Arrangements Applied Economics 26(2): 107120. Ohlin, Bertil. 1933. Interregional and International Trade Cambridge, Massachuset ts : Harvard University Press. Paseran, M. H. 1974. On the General Problem of Model Selection. Review of Economic Studies 41(2): 153171. Pesaran M. H. and A. S. Deaton. 1978. Testing NonNested Nonlinear Regression Models Econometrica 46(3): 677694. Petty, Sir William. 1687. Two Essays in Political Arithmetick London, United Kingdom: Printed for J. Lloyd in the Middle Exchange. Podolny, Joel M, and Karen L. Page. 1998. Network Forms of Organization. Annual Review of Sociology 24(1): 54 76. Polak, Jacques J. 1996. Is APEC a Natural Regional Trading Bloc? A Critique of the Gravity Model of International Trade World Economy 19(5): 533 543. Pollak, Robert A, and Terence J. Wales. 1991. The Likelihood Dominance Criterion: A New Ap proach to Model Selection. Journal of Econometrics 47(23): 227242. Pyhnen, Pentti. 1963a. A Tentative Model for the Volume of Trade Between Countries Weltwirtschaftliches Archiv 90(1): 93 99. Pyhnen, Pentti. 1963b. Toward a General Theory of International Trade. Ekonomiska Samfundets Tidskrift 16(2): 6978. Prewo, Wilfred. 1978. Determinants of the Trade Pattern Among OECD Countries from 1958 to 1974. Jahrbucher fur Nationalokonomie und Statistik 193(4): 341358. Pulliainen, Kyosti. 1963. A World Trade Study: An Econometric Model of the Pattern of the Commodity Flows in International Trade in 19481960. Ekonomiska Samfundets Tidskirft 16(2): 7891. Rauch, James E. 1999. Networks versus Markets in International Trade Journal of International Economics 48(1): 7 35. Rauch, James E. 2001. Business and Social Networks in International Trade Journal of Economic Literature 39(4): 11771203. Ravenstein, E. G. 1885. The Laws of Migration. Journal of the Statistical Society of Lo ndon 48(2): 167235.

PAGE 203

203 Read, Robert. 2004. The Political Economy of International Integration: Small States in the Caribbean and the Free Trade Area of the Americas (FTAA) In Nikolaos Karagiannis and Michael Witter, eds. The Caribbean Economies in an Era of Free Trade Hants, United Kin gdom: Ashgate Publishing Ltd. Reilly, William J. 1929. Methods for the Study of Retail Relationships University of Texas Bulletin Business Research Monograph No. 4. Von Roozendaal, Gerda. 2006. Regional Policy in the Americas: The EU Experience as a Guide for North South Integration. Review of European Economic Policy 41(3): 159168. Ruiz, Jose Briceno. 2007. Strategic Regionalism and Regional Social Policy in the FTAA Process Global Social Policy 7(3): 294315. Rutherford, Thomas F, and Josefina Martinez. 2000. Welfare Effects of Regional Trade Integration of Central American and Caribbean Nations with NAFTA and MERCOSUR The World Economy 23(6): 799825. Salazar Xirinachs, Jose M., and Jose Ta vares de Araugo Jr. 1999. The Free Trade Area of the Americas: A Latin American Perspective. The World Economy 22(6): 783797. Samuelson, Paul A. 1954. The Transfer Problem and Transport Costs, II: Analysis of Trade Impediments Economic Journal 64(254): 264289. Samuelson, Paul A. 1983. Thnen at Two Hundred. Journal of Economic Literature 21(4): 14681488. Sandberg, H. Mikael, and T. L. Martin. 2001. Trade Relationships and Economic Development: The Case of the Southern African Development Community South African Journal of Economics 69(3): 405 435. Sandberg, H. Mikael, James L. Seale Jr, and Timothy Taylor. 2006. History, Regionalism, and CARICOM Trade: A Gravity Model Analysis Journal of Development Studies 42(5): 795811. Sap ir Andre. 1981. Trade Benefits under the EEC Generalized System of Preferences. European Economic Review 15(3): 339355. Savage, I. Richard, and Karl W. Deutsch. 1960. A Statistical Model of the Gross Analysis of Transaction Flows Econometrica 28(3): 551572. Sattinger, Michael. 1978. Trade Flows and Differences Between Countries Atlantic Economic Journal 6(2): 22 30.

PAGE 204

204 Schott, Jeffrey J, and Gary C. Hufbauer. 1999. Whither the Fr ee Trade Area of the Americas? The World Economy 22(6): 765782. Seale, James L, Jr. 1985. Fixed Effect CobbDouglas Production Functions for Floor Tile Firms, Fayoum and Kalyubiya Egypt, 1981 1983. Ph.D. Dissertation, Michigan State University. Seale, James L, Jr. 1990. Estimating Stochastic Frontier Sys tems with Unbalanced Panel Data: The Case of Floor Tile Manufactories in Egypt Journal of Applied Econometrics 5(1): 5974. Seale, James L. Jr. 199 4. Demand System Specification and Estimation Book review. Economics of Planning 27(1): 6366. Skr ipnichenko, Anitoliy and Kevin Chen. 2002. Estimation of an Effectively Globally Regular Demand System: An Application to United States Meat Consumption Empirical Economics 27(4): 601 606. Smith, Peter. H. 1999. Whither Hemispheric Integration. Bu siness Economics 34(3): 3846. Soloaga, Isidro, and L. Alan Winters. 2001. Regionalism in the Nineties: What Effects on Trade. North American Journal of Economics and Finance 12(1): 1 29. Spiegel, Henry William. 1991. The Growth of Economic Thought Durham, NC: Duke University Press. Stewart, John Q. 1941. An Inverse Distance Variation for Certain Social Influences Science 93 (2404): 8990. Stewart, John Q. 1942. A Measure of the Influence of a Population at a Distance Sociometry 5(1): 6371. Stewart, John Q. 1947. Empirical Mathematical Rules concerning the Distribution and Equilibrium of Population. Geographical Review 37(3): 461485. Stewart, John Q. 1948. Demographic Gravitation: Evidence and Applications Sociometry 11(1/2 ): 31 58. Stewart, John Q. 1950. The Development of Social Physics American Journal of Physics 18(5): 239253. Taplin, Grant B. 1697. Models of World Trade Int ernational Monetary Fund Staff Papers 14(3): 433453.

PAGE 205

205 Thoumi, Francisco, E. 1989a. Tr ade Flow and Economic Integration among the LDCs of the Caribbean Basin Social and Economic Studies 38 (2): 215233. Thoumi, Francisco, E. 1989b. Bilateral Trade Flow and Economic Integration in Latin America and the Caribbean. World Development 17(3): 421429. Tinbergen, Jan. 1951. Econometrics London, United Kingdom : George Allen & Unwin Ltd. Tinbergen, Jan. 1962. Shaping the World Economy: Suggestions for an International Economic Policy New York, New York : The Twentieth Century Fund, Inc. United Nations. 2004. United Nations Common Database. http://data.un.org. Vollrath, Thomas L ., Charles .B. Hallahan, and Mark J. Gehlhar. 2006. Consumer Demand and Cost Factors Shape the Global Trade Network in Comm odity and Manufactured Goods, Ca nadian Journal of Agricultural Economics 54(4): 497 511. Wackerly, Dennis D., William Mendenhall III, and Richard L. Scheaffer. 2002. Mathematical Statistics with Applications Pacific Grove, California: Duxbury/Wadsworth Group Wall, Howard J. 2003. NAFTA and the Geography of North American Trade. Federal Reserve Bank of St. Luis Review 85(2): 13 26. Wei, Shang Jin, and Jeffrey A. Frankel. 1997. Open versus Closed Trade Blocks In Takatoshi Ito and Anne O. Krueger, eds. Regionalism versus Multil ateral Trade Agreements Chicago, Il l inois : The University of Chicago Press. World Bank. 1998. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. World Bank. 1999. World Developme nt Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. World Bank. 2000. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. Wor ld Bank. 2001. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. World Bank. 2002. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. World Bank. 2003. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development.

PAGE 206

206 World Bank. 2004. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. World Bank. 2005. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. World Bank. 2006. World Development Indicators Washington, District of Columbia: The International Bank for Reconstruction and Development. Wrobel, Paulo S. 1998. A Free Trade Area of the Americas in 2005? International Affairs 74(3): 547561. Young, E. C. 1924. "The Movement of Farm Population. Cornell University Agricultural Experiment Station Bulletin 426. Zahniser, Steven S., et al. 2002. Regionalism in the Western Hemisphere and its Impact on U.S. Agricul tural Exports: A Gravity Model Analysis. American Journal of Agricultural Economics 84(3): 791 797. Zipf, George Kingsley. 1946. The P1P2/D Hypothesis: On the Intercity Movement of Persons American Sociological Review 11(6): 677 686.

PAGE 207

207 BIOGRAPHICAL SKETCH Harry Mikael Sandberg, now a permanent resident of the United States, was born in Visby, Sweden. Upon graduating from a Swedish high school in 1994, he decided to come to the United States for his post secondary education. In 1998, he earned a Bachelor of Science in Business Administration (B .S.B.A) degree with a major in e conomics from the University of Cent ral Florida The courses in the major instilled a great passion for the subject and convinced him to pursue a graduate degree in economics He decided to remain at the University of Central Flo rida and subsequently earned a Ma ster of Arts (M.A.) degree in applied e conomics two years later. The comp le tion of his masters thesis and his experiences as a classroom teaching assistant persuaded him to continue his studies at the doctoral level. He received offers of admission to doctoral programs with financial support from the University of Florida, University of Miami, University of Oregon, and Rice University. Upon visiting three of the campuses, the food and resource e conomics program at the University of Florida appeared as the best fit. He subsequently entered the program fully funded on a prestigious University of Florida Alumni Fellowship in the fall of 2000. While in d octoral candidacy, he was offered a full time faculty position in the Food and Reso urce Economics d epartment as a l ecturer. For the past six years, he has enjoyed his professional experiences at the University of Florida and he has plans to continue his academic career upon completion of the Doctor of Philosophy degree in 2010.