TRADE ADJUSTMENTS TO EXCHANGE RATES IN REGIONAL ECONOMIC INTEGRATION: ARGENTINA AND BRAZIL Except where reference is made to the work of others, the work described in this dissertation is my own or was done in collaboration with my advisory committee. This dissertation does not include proprietary or classified information. ________________________________________ Fernando Daniel Sedano Certificate of Approval: _________________________ __________________________ Dr. John Jackson, Co-Chair Dr. Henry Thompson, Co-Chair Professor Professor Economics Economics _________________________ __________________________ Dr. Mark Carpenter Dr. Keivan Deravi Professor Professor Mathematics and Statistics AUM Economics Department _______________________ Stephen L. McFarland Acting Dean Graduate School TRADE ADJUSTMENTS TO EXCHANGE RATES IN REGIONAL ECONOMIC INTEGRATION: ARGENTINA AND BRAZIL Fernando Daniel Sedano A Dissertation Submitted to the Graduate Faculty of Auburn University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy Auburn, Alabama December 16, 2005 iii TRADE ADJUSTMENTS TO EXCHANGE RATES IN REGIONAL ECO NOMIC INTEGRATION: ARGENTINA AND BRAZIL Fernando Daniel Sedano Permission is granted to Auburn University to make copies of this dissertation at its discretion, upon request of individuals or institutions and at their expense. The author reserves all publication rights. _____________________________ Signature of Author _____________________________ Date of Graduation iv VITA Fernando Daniel Sedano, son of Rodolfo Luis Sedano and Alicia Noem? (Leis) Sedano, was born August 22, 1972, in Buenos Aires, Argentina. He graduated from E.N.E.T. # 2, Ing. E. Mitre High School as mechanic technician in 1991. He attended Auburn University Montgomery, Alabama, and graduated summa cum laude with a Bachelor of Arts degree in International Studies in May 2000. He received a Master in Business Administration at Auburn University Montgomery in December, 2001. While working as a Research Specialist for Auburn University Montgomery, Fernando began the Doctor of Philosophy program at Auburn University in January 2003. In December 2004, he obtained a Master in Science in Economics at Auburn University. He married Juliana Piumatti, daughter of Roberto and Marilene (Moraes) Piumatti, on December 21, 2002. v DISSERTATION ABSTRACT TRADE ADJUSTMENTS TO EXCHANGE RATES IN REGIONAL ECONOMIC INTEGRATION: ARGENTINA AND BRAZIL Fernando Daniel Sedano Doctor of Philosophy, December 16, 2005 (M.S., Auburn University, 2004) (M.B.A., Auburn University Montgomery, 2001) (B.A., Auburn University Montgomery, 2000) 186 Typed Pages Directed by Dr. Henry Thompson and Dr. John Jackson Currency devaluation is one of the most commonly used economic policies when a country faces a trade imbalance. By making exports more competitive in world markets and imports more expensive in terms of local currency, devaluations should induce trade balance improvements at the aggregate and bilateral levels. This dynamic behavior known as the J-curve has been tested in numerous empirical papers that have generated mixed results. This literature has not formally addressed the role played by regional economic integration among the countries under examination. Since almost all countries vi are engaged in some sort of regional economic integration, this dissertation examines the trade effects of devaluations when countries are part of a regional integration agreement. First, Chapter 1 discusses the challenges raised by regionalization and introduces the case study by describing the Southern Cone Common Market (Mercosur) and the exchange rate regimes implemented in Argentina and Brazil. Chapter 2 investigates the effects devaluations on countries? trade balances and examines potential trade diversion effects. Chapter 3 looks at gravity models of trade to test for potential trade diversion effects of currency devaluations. This chapter also investigates whether these trade adjustments are a consequence of changing demand structures as proposed by Linder (1961). Chapter 4 presents the overall conclusions and policy implication issues. Each chapter has its own literature review, theory, and empirical sub-sections. All empirical work concentrates on Brazil and Argentina, the two major economies of Mercosur. Brazil devalued its currency in January 1999 and Argentina followed in January 2002. While both devaluations generated aggregate trade surpluses, unexpected adjustments emerged at the bilateral level. Although the empirical results are based on countries within a specific trading bloc, the economics behind them is subject to generalization. vii ACKNOWLEDGMENTS The author wishes to express his gratitude to Dr. Keivan Deravi for his technical guidance and personal support. Dr. Henry Thompson and Dr. John Jackson, Co-Chairs, provided valuable assistance on theoretical and econometric issues, and Leonardo Paz laboriously retrieved hard copy data in Argentina and transmitted it to Alabama. Thanks are owed to Almerisio Lopes who helped editing this work. Dr. John Veres provided the resources and time flexibility that made this whole process possible. Thanks are due to the author?s family, Rodolfo, Alicia, Diego, and Leandro Sedano for their unconditional support. The author?s unborn daughter, Gabriela Sedano, furnished the motivation needed to complete this dissertation in a timely manner. The author is especially grateful to his beautiful wife, Juliana Piumatti Sedano, for her invaluable emotional support during a long journey. Without her love and patience, this undertaking could not have been accomplished. viii Style manual or journal used: Journal of Applied Economics Computer Software used: Microsoft Word ix TABLE OF CONTENTS LIST OF FIGURES ........................................................................................................... xi LIST OF TABLES...........................................................................................................xiii CHAPTER 1: MERCOSUR: REGIONAL ECONOMIC INTEGRATION, EXCHANGE RATES, MACROECONOMICS, AND TRADE COMPOSITION ............................ 1 I. INTRODUCTION .........................................................................................................................1 II. STEPS TOWARDS REGIONAL ECONOMIC INTEGRATION ..........................................................2 III. DEVALUATION IN A REGIONAL TRADING BLOC: COMMON PROBLEMS ................................4 IV. CASE STUDY: MERCOSUR, ARGENTINA, AND BRAZIL..........................................................7 The Period of Fixed Exchange Regimes................................................................................................8 Brazil?s Currency Devaluation ............................................................................................................11 The End of Convertibility ....................................................................................................................13 Argentina?s Economic Recovery and Bilateral Trade Flows...............................................................15 Composition of Argentina?s Imports and Exports ...............................................................................16 Argentina?s Trade Regulations ............................................................................................................17 Customs Tariffs ...................................................................................................................................18 Taxes on Exports .................................................................................................................................19 APPENDIX I .............................................................................................................................20 APPENDIX II............................................................................................................................31 APPENDIX III...........................................................................................................................32 CHAPTER 2: TRADE ADJUSTMENTS TO CURRENCY DEVALUATION AND TRADE DIVERSION................................................................................................. 35 I. INTRODUCTION .......................................................................................................................35 II. LITERATURE REVIEW ............................................................................................................38 III. THEORIES ON CURRENCY DEVALUATION AND ITS EFFECT ON THE TRADE BALANCE .......47 The Absorption and Monetary Approaches.........................................................................................47 The Elasticities Approach and the J-curve...........................................................................................49 A J-curve in Argentina?.......................................................................................................................51 IV. THE MODEL .........................................................................................................................53 Model Adjustments..............................................................................................................................55 V. DATA DESCRIPTION AND STATIONARITY TESTS ..................................................................59 VI. MODEL RESULTS .................................................................................................................62 VII. CONCLUSIONS ....................................................................................................................64 APPENDIX I .............................................................................................................................66 APPENDIX II............................................................................................................................72 APPENDIX III........................................................................................................................... 73 x CHAPTER 3: TRADE DIVERSION IN THE CONTEXT OF GRAVITY MODELS: A TEST OF THE LINDER HYPOTHESIS................................................................... 75 I. INTRODUCTION .......................................................................................................................75 II. LITERATURE REVIEW ON GRAVITY MODELS........................................................................78 III. THEORETICAL FRAMEWORK OF GRAVITY MODELS ............................................................92 IV. THE LINDER HYPOTHESIS..................................................................................................106 V. ECONOMETRIC ISSUES ........................................................................................................112 Fixed Effects and a Time Series Cross Section Model ......................................................................116 VI. TRADE DIVERSION.............................................................................................................117 VII. THE LINDER HYPOTHESIS AND THE PESO DEVALUATION ...............................................127 VIII. CONCLUSIONS .................................................................................................................135 APPENDIX I ...........................................................................................................................137 APPENDIX II..........................................................................................................................142 APPENDIX III.........................................................................................................................148 CHAPTER 4: CONCLUDING REMARKS AND POLICY IMPLICATION ISSUES 160 I. CONCLUDING REMARKS.......................................................................................................160 II. POLICY IMPLICATIONS ........................................................................................................162 BIBLIOGRAPHY........................................................................................................... 164 xi LIST OF FIGURES FIGURE 1.1: ARGENTINA?S INTERNATIONAL RESERVES.................................................................................20 FIGURE 1.2: BRAZIL?S INTERNATIONAL RESERVES ........................................................................................20 FIGURE 1.3: ARGENTINA?S ANNUAL INFLATION RATE...................................................................................20 FIGURE 1.4: BRAZIL?S ANNUAL INFLATION RATE..........................................................................................20 FIGURE 1.5: ARGENTINA?S GDP ANNUAL GROWTH RATE.............................................................................20 FIGURE 1.6: BRAZIL?S GDP ANNUAL GROWTH RATE....................................................................................20 FIGURE 1.7: UNEMPLOYMENT RATE IN ARGENTINA ......................................................................................20 FIGURE 1.8: UNEMPLOYMENT RATE IN BRAZIL .............................................................................................20 FIGURE 1.9: FOREIGN DIRECT INVESTMENT...................................................................................................21 FIGURE 1.10: ARGENTINA?S BALANCE OF TRADE..........................................................................................21 FIGURE 1.11: BRAZIL?S BALANCE OF TRADE .................................................................................................21 FIGURE 1.12: ARGENTINA?S BALANCE OF TRADE WITH BRAZIL ....................................................................21 FIGURE 1.13: EXCHANGE RATES....................................................................................................................21 FIGURE 1.14: ARGENTINA?S EXTERNAL DEBT ...............................................................................................21 FIGURE 1.15: ARGENTINA?S IMPORTS FROM BRAZIL......................................................................................21 FIGURE 1.16: ARGENTINA?S EXPORTS TO BRAZIL..........................................................................................21 FIGURE 1.17: ARGENTINA?S IMPORTS FROM EU AND US...............................................................................22 FIGURE 1.18: ARGENTINA?S EXPORTS OF MANUFACTURES OF AGRICULTURAL ORIGIN................................22 FIGURE 1.19: ARGENTINA?S EXPORTS OF MANUFACTURES OF INDUSTRIAL ORIGIN......................................22 FIGURE 1.20: ARGENTINA?S EXPORTS OF PRIMARY PRODUCTS .....................................................................22 FIGURE 1.21: ARGENTINA?S EXPORTS OF FUELS AND ENERGY......................................................................22 FIGURE 1.22: ARGENTINA?S EXPORTS OF MANUFACTURES OF AGRICULTURAL ORIGIN (%SHARE) ..............22 FIGURE 1.23: ARGENTINA?S EXPORTS OF MANUFACTURES OF INDUSTRIAL ORIGIN (%SHARE) ....................22 FIGURE 1.24: ARGENTINA?S EXPORTS OF PRIMARY PRODUCTS (%SHARE)....................................................22 FIGURE 1.25: ARGENTINA?S EXPORTS OF FUELS AND ENERGY (% SHARE)....................................................23 FIGURE 1.26: ARGENTINA?S IMPORTS OF CAPITAL GOODS FROM BRAZIL (% SHARE) ...................................23 FIGURE 1.27: ARGENTINA?S IMPORTS OF INTERMEDIATE GOODS FROM BRAZIL (% SHARE) .........................23 FIGURE 1.28: ARGENTINA?S IMPORTS OF CONSUMER GOODS FROM BRAZIL (% SHARE) ...............................23 FIGURE 1.29: ARGENTINA?S IMPORTS OF CAPITAL GOODS BY COUNTRY OF ORIGIN .....................................23 FIGURE 1.30: ARGENTINA?S IMPORTS OF INTERMEDIATE GOODS BY COUNTRY OF ORIGIN ...........................23 FIGURE 1.31: ARGENTINA?S IMPORTS OF SPARE PARTS AND PIECES FOR CAPITAL GOODS BY COUNTRY OF ORIGIN .................................................................................................................................................23 FIGURE 1.32: ARGENTINA?S IMPORTS OF CONSUMER GOODS BY COUNTRY OF ORIGIN.................................23 FIGURE 1.33: ARGENTINA?S IMPORTS OF FUELS BY COUNTRY OF ORIGIN .....................................................24 FIGURE 1.34: ARGENTINA?S IMPORTS OF VEHICLES BY COUNTRY OF ORIGIN ...............................................24 FIGURE 2.1: ISLMBP WITH FIXED EXCHANGE RATES ...................................................................................56 FIGURE 2.2: ISLMBP WITH FLEXIBLE EXCHANGE RATES .............................................................................57 FIGURE 2.3: REAL EXCHANGE RATE (REALS /PESOS) .....................................................................................59 FIGURE 2.4: RELATIVE REAL GDPS (Y/Y*)...................................................................................................60 FIGURE 2.5: BILATERAL TRADE BALANCE MEASURED AS M/X .....................................................................60 FIGURE 2.6: RATIO OF IMPORTS FROM BRAZIL OVER NON-MERCOSUR PARTNERS (MB / MO) ......................60 FIGURE 2.7: ALMON PDL RESIDUAL PLOT ....................................................................................................72 FIGURE 3.1: TRADE IN MANUFACTURES FOR 2 COUNTRIES..........................................................................107 FIGURE 3.2: SHARES OF ARGENTINA?S IMPORTS..........................................................................................118 FIGURE 3.3: RESIDUALS FROM TSCS MODEL WITH COMMON INTERCEPT IN TABLE 3.1..............................142 FIGURE 3.4: RESIDUALS FROM TSCS MODEL WITH FIXED EFFECTS IN TABLE 3.1.......................................142 xii FIGURE 3.5: RESIDUALS FROM FE MODEL IN TABLE 3.1..............................................................................142 FIGURE 3.6: RESIDUALS FROM TSCS IV MODEL WITH COMMON INTERCEPT TABLE 3.2.............................143 FIGURE 3.7: RESIDUALS FROM TSCS IV MODEL WITH FE IN TABLE 3.2 .....................................................143 FIGURE 3.8: RESIDUALS FROM FE MODEL IN TABLE 3.2..............................................................................143 FIGURE 3.9: RESIDUALS FROM TSCS IV MODEL WITH COMMON INTERCEPT AND LAGGED IMPORTS IN TABLE 3.3 ......................................................................................................................................................144 FIGURE 3.10: RESIDUALS FROM TSCS IV MODEL WITH FE AND LAGGED IMPORTS TABLE 3.3...................144 FIGURE 3.11: RESIDUALS FROM IV MODEL WITH FE AND LAGGED IMPORTS TABLE 3.3.............................144 FIGURE 3.12: RESIDUALS FROM LINDER TSCS MODEL WITH COMMON INTERCEPT TABLE 3.4...................145 FIGURE 3.13: RESIDUALS FROM LINDER TSCS MODEL WITH FE TABLE 3.4................................................145 FIGURE 3.14: RESIDUALS FROM LINDER MODEL WITH FE TABLE 3.4..........................................................145 FIGURE 3.15: RESIDUALS LINDER TSCS IV MODEL WITH COMMON INTERCEPT TABLE 3.5........................146 FIGURE 3.16: RESIDUALS LINDER TSCS IV MODEL WITH FE TABLE 3.5 ....................................................146 FIGURE 3.17: RESIDUALS LINDER IV MODEL WITH FE TABLE 3.5...............................................................146 FIGURE 3.18: RESIDUALS LINDER TSCS IV MODEL WITH COMMON INTERCEPT AND LAGGED IMPORTS TABLE 3.6...........................................................................................................................................147 FIGURE 3.19: RESIDUALS LINDER TSCS IV MODEL WITH FE AND LAGGED IMPORTS IN TABLE 3.6 ...........147 FIGURE 3.20: RESIDUALS LINDER IV MODEL WITH FE AND LAGGED IMPORTS TABLE 3.6..........................147 xiii LIST OF TABLES TABLE 1.1: ARGENTINA?S AGGREGATE TRADE BALANCE (MILLIONS OF US$) .............................................25 TABLE 1.2: BRAZIL?S AGGREGATE TRADE BALANCE (MILLIONS OF US$) .....................................................25 TABLE 1.3: ARGENTINA?S EXPORTS BY TYPE OF GOOD .................................................................................26 TABLE 1.4: ARGENTINA?S IMPORTS BY TYPE OF GOOD (FOB) ......................................................................28 TABLE 1.5: INDICES OF PRICE AND QUANTITIES FOR ARGENTINA?S EXPORTS AND IMPORTS.........................29 TABLE 1.6: SHARE OF BRAZILIAN IMPORTS IN ARGENTINA ...........................................................................29 TABLE 1.7: ARGENTINA?S IMPORTS BY TYPE OF GOOD AND COUNTRY OF ORIGIN........................................30 TABLE 1.8: LIST OF TRADE AGREEMENTS......................................................................................................32 TABLE 2.1: AUGMENTED DICKEY-FULLER TESTS..........................................................................................62 TABLE 2.2: ALMON PDL RESULTS IN FIRST DIFFERENCES ............................................................................63 TABLE 2.3: ALMON PDL RESULTS WITH INSTRUMENT ..................................................................................74 TABLE 3.1: TRADE DIVERSION.....................................................................................................................121 TABLE 3.2: TRADE DIVERSION WITH IV.......................................................................................................124 TABLE 3.3: TRADE DIVERSION WITH IV AND LAGGED IMPORTS..................................................................126 TABLE 3.4: LINDER EFFECTS........................................................................................................................129 TABLE 3.5: LINDER EFFECTS WITH IV..........................................................................................................130 TABLE 3.6: LINDER EFFECTS WITH IV AND LAGGED IMPORTS.....................................................................131 TABLE 3.7: Q-STATISTICS TSCS COMMON INTERCEPT TABLE 3.1...............................................................142 TABLE 3.8: Q-STATISTICS TSCS FIXED EFFECTS TABLE 3.1........................................................................142 TABLE 3.9: Q-STATISTICS FE MODEL IN TABLE 3.1.....................................................................................142 TABLE 3.10: Q-STATISTICS TSCS IV MODEL WITH COMMON INTERCEPT IN TABLE 3.2..............................143 TABLE 3.11: Q-STATISTICS TSCS IV MODEL WITH FE IN TABLE 3.2...........................................................143 TABLE 3.12: Q-STATISTICS FE MODEL IN TABLE 3.2...................................................................................143 TABLE 3.13: Q-STATISTICS TSCS IV MODEL WITH COMMON INTERCEPT AND LAGGED IMPORTS IN TABLE 3.3 ......................................................................................................................................................144 TABLE 3.14: Q-STATISTICS TSCS IV MODEL FE AND LAGGED IMPORTS TABLE 3.3...................................144 TABLE 3.15: Q-STATISTICS IV MODEL FE AND LAGGED IMPORTS TABLE 3.3 .............................................144 TABLE 3.16: Q-STATISTICS LINDER TSCS MODEL WITH COMMON INTERCEPT TABLE 3.4..........................145 TABLE 3.17: Q-STATISTICS LINDER TSCS MODEL WITH FE TABLE 3.4.......................................................145 TABLE 3.18: Q-STATISTICS LINDER MODEL WITH FE TABLE 3.4.................................................................145 TABLE 3.19: Q-STATISTICS LINDER TSCS IV MODEL WITH COMMON INTERCEPT TABLE 3.5 .....................146 TABLE 3.20: Q-STATISTICS LINDER TSCS IV MODEL WITH FE TABLE 3.5..................................................146 TABLE 3.21: Q-STATISTICS LINDER IV MODEL WITH FE TABLE 3.5 ............................................................146 TABLE 3.22: Q-STATISTICS LINDER TSCS IV MODEL WITH COMMON INTERCEPT AND LAGGED IMPORTS TABLE 3.6...........................................................................................................................................147 TABLE 3.23: Q-STATISTICS LINDER TSCS IV MODEL WITH FE AND LAGGED IMPORTS IN TABLE 3.6.........147 TABLE 3.24: Q-STATISTICS LINDER IV MODEL WITH FE AND LAGGED IMPORTS TABLE 3.6 .......................147 1 CHAPTER 1: MERCOSUR: REGIONAL ECONOMIC INTEGRATION, EXCHANGE RATES, MACROECONOMICS, AND TRADE COMPOSITION I. Introduction Regional economic integration is rising. According to the World Trade Organization (WTO), close to 300 regional trading blocs will link countries around the world by the end of 2005. Of those 300 agreements, approximately two-thirds emerged after 1995. Many of these arrangements are among developed countries but others are formed by industrialized countries joining developing ones. This recent move toward regionalism raises a number of important issues for governments and policy makers around the world. At the center of those issues are the economic forces generated by interdependence among countries forming a regional agreement and how different policies should deal with those forces. Currency devaluation is one of the most commonly used economic policies when a country faces trade balance of payments deficits. By making exports more competitive in world markets and imports more expensive in terms of local currency, devaluation will raise the trade balance at the aggregate and bilateral levels given sufficient elasticities. Since regionalization is becoming the norm rather than the exception, the present dissertation provides a framework for the analysis of devaluation when countries are part 2 of a regional economic agreement. A survey describing regional economic integration and the macroeconomic behavior of the countries studied follows. II. Steps towards Regional Economic Integration The European Union (EU) is today the most deeply integrated of all trading blocs. Starting with the Treaty of Rome in 1957, European countries went through the different stages in the process of economic integration before launching the Euro in January 2000. The United States (US), Canada, and Mexico formed the North American Free Trade Agreement (NAFTA) in 1994, which allows for free trade in goods and services. In Asia, five countries established the Association of Southeast Asian Nations (ASEAN) in August 1967, which has grown to ten countries with the goal of creating the ASEAN Free Trade Area (AFTA) by 2005. According to the Inter-American Development Bank (IABD) (2002), some 15 integration agreements emerged recently in Africa and more than 30 agreements emerged since 1990 in Latin America. 1 There are four steps towards complete economic integration: free trade areas, customs unions, common markets, and monetary unions. 2 Starting with the least ambitious of the agreements, a free trade area is composed of countries that agree to gradually eliminate tariffs. Free movement of goods and services is the ultimate goal of a free trade area. A customs union is a free trade area that sets a common external trade policy. Common external tariff rates and quotas are applied to all goods and services entering the area from non-member countries. A customs union with free movement of factors of production becomes a common market. A monetary union is the last step 1 See Appendix III for a list of trade agreements. 2 See Thompson (2001) for a discussion of each step toward economic integration. 3 towards complete economic integration. Countries forming a monetary union have a single currency and common fiscal and monetary policies. Each step in the process of regional economic integration has its pros and cons. Among the positive aspects, regionalism reinforces the process of globalization by opening world markets, creating scale economies, and attracting foreign direct investment (FDI). It also helps small countries to have greater bargaining power when negotiating extra-regional agreements. According to IADB (2002), regionalism is also becoming a geopolitical tool by promoting peace, democracy, and cooperation in the development of regional infrastructure. On the negative side are the unforeseen economic forces that emerge in many regions and delay further integration. Among these forces are the uneven flows of FDI towards larger economies and the production and trade adjustments created by unilateral exchange rate movements (i.e., devaluations) in the absence of regulations such as currency bands or monetary unions. The uneven flow of FDI (usually biased towards larger economies) is one of the main barriers to deeper integration. With the reduction of tariffs among countries in a regional trade agreement, many firms decide to locate production at one site to supply all countries within the bloc. As a consequence, member countries compete for multinational firms by offering incentives such as tax discounts. Once FDI within a regional economic agreement starts to concentrate in one country, other members raise their tariffs or implement quotas, hindering the efficiency gains of integration. Uneven flows of FDI emerge also as a consequence of exchange rate adjustments. Devaluation in a member country continues to be the major barrier to deeper economic 4 ties. Next section examines the general problems that materialize when unilateral devaluations occur in a regional bloc and presents some empirical evidence. III. Devaluation in a Regional Trading Bloc: Common Problems Large swings in bilateral real exchange rates usually create problems among countries in a regional integration agreement. According to IADB (2002), the most common problems of devaluation are the relocation of FDI, protectionist measures enacted by the country that is losing competitiveness, trade adjustments, and exchange rate crises that have the potential to develop into recessions. 3 FDI relocation due to devaluation occurs more frequently among countries in a regional integration agreement. 4 According to IADB (2002), the factors that lead to relocation of FDI within a regional trading bloc depend on the type of FDI under consideration. When FDI is vertical or resource seeking, devaluation favors the depreciating country at the expense of all other potential hosts having similar factor endowments (regional trading partners). Specifically, lower production costs in the country devaluating its currency influence location criteria. In the presence of horizontal or market-seeking FDI within a regional trading block and in particular within a common market, firms are encouraged to produce in one single location (provided perhaps that scale economies are present) and from this location supply the entire market. Holding other factors constant, companies tend to choose the country with the lowest production costs. IADB (2002) finds that 1% depreciation in the real bilateral exchange rate 3 See IADB (2002) for a detailed analysis of these issues. 4 See Feenstra (1999), Pardo (2002), IADB (2002), and Eichengreen (1993). One of the most cited cases of FDI relocation is the move of the Hoover vacuum cleaning production facility from France to Scotland as a consequence of the 1992 European exchange rate mechanism crisis. 5 increases relative FDI inflows by 1.3% when both countries are members of a regional agreement. On the other hand, the impact of devaluations on FDI is found to be statistically insignificant for non-member countries. Appendix II presents evidence of companies moving from Argentina to Brazil following the devaluation of the Brazilian real in January 1999. Many large companies left Argentina and started production in Brazil, laying-off around 10,000 workers. More than a dozen car companies sent some 7,000 jobs (15% of the total industry) to the largest economy in Mercosur. According to this article, this corporate exodus was a consequence of the 35% depreciation of the real against the peso. Tariffs, quotas, or non-tariff barriers (NTBs) could arise among regional trading partners when a member country devalues its currency. These protectionist measures are among the major reasons for the failure of deeper integration within trading blocs. For example, when the United Kingdom (UK) devalued the sterling in 1992 French officials proposed protectionist measures. According to Eichengreen (1993), the creation of a monetary union in Europe is a consequence of the tensions generated by the sterling devaluation and the rise of protectionist ideas. Eichengreen (1997) suggests that these protectionist reactions depend on the degree of integration within the trading block. Deeper economic integration among member countries leads to more serious protectionist actions by countries losing competitiveness. Eichengreen compares the 1992 sterling devaluation to the 1994 Mexican devaluation pointing out that while protectionist reactions emerged in Europe (a monetary union), only a few specific complaints were made by the US or Canada (free trade area). 5 5 See introduction in Chapter 2 for a description of protectionist measures within Mercosur. 6 Regional trade agreements generally create trade diversion by substituting imports from more efficient non-member countries towards less efficient member countries. According to IADB (2002), regional trading blocs generate a demand for goods that are not internationally competitive and are known as regional goods. In general, devaluation depresses imports and therefore the demand for regional goods. Other member countries cannot redirect these regional products to markets outside the bloc due to their lack of competitiveness. IADB (2002) investigates whether exchange rate overvaluation leads to the same effect on exports to member and non-member countries. Results suggest that when 10% of a country?s exchange rate overvaluation is due to a member country?s devaluation, total exports decline by 14%. Further, when 10% of the overvaluation is due to non-member countries? devaluations, then exports decline by 3.5%. These results suggest that regional goods make up a significant portion of trade flows, especially when regional trading blocs are highly protected from the outside. Devaluation also generates exchange rate crises for regional trading partners. IADB (2002) states the exit of the Italian lira from the European exchange rate mechanism in 1992 led the UK to abandon its peg, which in turn exerted enormous pressure on the French franc. Similarly, the depreciation of the Thai baht in 1997 caused depreciations and economic contractions in Singapore, Malaysia, Indonesia, and the Philippines, all members of ASEAN. The report also mentions contagions such as when Mexico devalued its currency in 1994, affecting the stability of the Argentine peso, or when the 1997 Asian crisis affected the Russian ruble which in turn affected the Brazilian real. The paper examines the effects of devaluations on the currencies of regional and non-regional trading partners. Findings suggest that a ?10[%] overvaluation explained by 7 exchange rate movements within the [regional integration agreement] increases the probability of a crisis by 4 percentage points? (p. 182). On the other hand, when a 10% exchange rate overvaluation comes from a non-member country, the probability of a crisis increase by 1.7%. These findings suggest that when studying the effects of devaluation on bilateral trade balances with specific countries, the researcher should take into consideration the degree of economic integration between them. The effects of devaluation by a member country should be different from the effects generated by a similar devaluation in a non- member country. This dissertation investigates the effects of Argentina?s devaluation on the country?s bilateral trade balance with Brazil by taking into account the trade adjustments occurring with the US and EU (non-Mercosur members). 6 The following section presents a brief description of Mercosur, as well as the macroeconomic environment and composition of trade flows between Argentina and Brazil. IV. Case Study: Mercosur, Argentina, And Brazil The governments of Argentina, Brazil, Paraguay, and Uruguay signed the Treaty of Asunci?n and formed Mercosur on March 26, 1991. Mercosur?s ultimate goal is to create a common market with free movement of goods, services, and factors of production. While most of Mercosur?s intra-industry trade is tariff free, the steps toward a common market have been delayed on several occasions due to economic instability in the area. Each member maintains a list of a few sensitive products that are exempted from this zero rate and that are supposed to be gradually reduced by 2006. In January 6 Brazil, the US, and EU are Argentina?s major trading partners. Together, they explain almost 75% of the country?s trade during the period under study. 8 2001, Argentina and Brazil agreed to implement temporary tariffs or quotas to protect industries harmed by exchange rate fluctuations. A common external tariff (CET) set in 1995 contains many exemptions and can be suspended under some scenarios. Mercosur average external tariff is about 13.5%. 7 Since its implementation, Mercosur has been seeking extra-regional agreements. Chile became an associate member in 1996, but continues to maintain its own common external tariffs. Bolivia was also admitted to Mercosur as associate member in 1997. Mercosur and the European Union attempted to reach a free trade agreement but the high tariffs and subsidies for agricultural products in Europe have been a major barrier in the process. The same argument with regard to the US is delaying negotiations for a Free Trade Area of the Americas (FTAA). Trade between Argentina and Brazil has increased significantly since the creation of Mercosur. 8 Currently, Argentina absorbs 85% of Brazilian exports to Mercosur while 90% of Brazilian imports (from Mercosur) come from Argentina. Similarly, Brazil accounts for 85% of Argentine exports to Mercosur whereas 91% of what Argentina imports from member countries come from Brazil. This close trading relationship between Argentina and Brazil increased over the last fifteen years despite major exchange rate and macroeconomic instability. The Period of Fixed Exchange Regimes Argentina and Brazil experienced significant structural reforms during the 1990s that included trade liberalization, privatization of public enterprises, and deregulation of 7 See subsection on trade regulations for details. 8 Yeats (1998) offers a detailed description on Mercosur?s trade patterns. The paper shows that Mercosur has created substantial trade diversion since the most rapidly growing products traded within the bloc are generally products in which members do not have a comparative advantage. 9 markets. Currency boards known as the convertibility plan in Argentina and the Real Plan in Brazil became the cornerstones of all reforms. Both plans attempted to create a stable and market-friendly environment to attract foreign investments and generate sustainable economic growth. Argentina?s convertibility plan was implemented in April 1991 with the goal of stopping hyperinflation that almost reached an annual rate 5,000% in 1989. In fact, the plan became the ?Convertibility Law? with the peso and the US dollar ($) legally circulating at a one-to-one exchange rate. Peso holders could convert pesos into dollars without any restriction at this official rate. By law, the central bank was required to hold foreign reserves to fully cover its peso liabilities. 9 Reserves, consisting of gold and foreign currency or deposits and bonds payable in gold and foreign currency, had to be maintained at a level no less than 100% of the monetary base. Up to 30% of reserves could be held in bonds issued by the Argentine government. The autonomous creation of currency became legally impossible. Expansion was only possible when proper reserves existed to cover it, with a contraction occurring in the opposite case. Authorities could compensate for either of these situations by means of greater or lesser Central Bank holdings in public securities within the established 30% margin. The 1994 Brazilian Economic Stabilization Program, known as the Real Plan, became the most successful of all plans that previously attempted to solve Brazil's problems with chronic inflation. On July 1 st 1994, a new currency called the ?real? (R$) was created. The real was backed by the country?s international reserves at an exact ratio of one US dollar to each real emitted. Part of the country?s international reserves was to 9 See (Hanke, 2002). 10 be held in a special account at the central bank for this purpose. Exchange rate parity was to be held at a one-to-one rate with the US dollar for an indeterminate length of time. However, to avoid exchange rate rigidity, the Minister of Finance had the right to set the criteria to be used by the National Monetary Council to back the real as well as to make any changes to the parity policy. Both plans achieved price stability and economic growth during the first years of implementation. However, economic growth became vulnerable under external shocks. The so-called ?Tequila Crisis? in Mexico was the main cause of Argentina?s 1995 recession while the 1997 Asian crisis and 1998 Russian devaluation unveiled the weakness of the Real Plan. In Argentina, the requirement that all pesos in circulation had to be backed by the same amount of US dollars in reserves played a negative role. When all agents in the economy began changing pesos for dollars, the reduction of international reserves at the central bank meant a decrease in the monetary base. International reserves went from $17.8 billion in December 1994 to $12.4 billion in March 1995, and the monetary base went from 16 billion pesos to 10.8 billion pesos in the same time period. A $12 billion package from the World Bank saved Argentina?s banking system and put the economy back in the growth trend, peaking at an annual rate of 9.2% by the 4th quarter of 1996. By the 4th quarter of 1998, Argentina entered a 4-year recession that lasted until the 4 th quarter of 2002. Economic growth during the convertibility plan did not translate into a reduction in unemployment rates. In fact, the unemployment rate in Argentina doubled between 1991 and 2001 from less than 10% to almost 20%. In Brazil, the unemployment rate was 11 more stable ranging from annual rates of 8% to 12%. The highest unemployment figures in Brazil appeared in 1998, the year marking the collapse of the Real Plan. 10 During the 1992-1994 period, Argentina was receiving higher FDI than Brazil. It is only after the implementation of the Real Plan that FDI increased significantly in Brazil. An inflection point seems to appear in 1995, when Mercosur?s CET was implemented. By 2001, Argentina received only 8% of the amount of FDI in Brazil. The fixed exchange rate regimes generated aggregate trade deficits in both countries. Occasional trade surpluses emerged as economic contractions depressed import levels. In terms of bilateral trade, Argentina maintained a trade surplus with Brazil during the fixed exchange rate regimes. 11 Brazil?s Currency Devaluation In January 1999, the collapse of the Real Plan in Brazil and the subsequent devaluation shocked Mercosur?s economic stability. Brazil?s trade deficits coupled with the capital outflows that emerged after the Asian and Russian crisis were the major determinants for devaluation. The steady decrease in foreign reserves at the central bank starting in 1997 accelerated in 1998 as foreign investors were covering their losses from Asia and Russia. By January 1999, foreign reserves have dropped to $35 billion, almost half of the January 1997 level. On January 13 th 1999, the central bank announced that the real would be traded at a new and wider band of 1.20-1.32 reals to the dollar. Two days later the currency band was abandoned, thus making the real a free-floating currency. 10 The Brazilian Institute of Geography and Statistics (IBGE) changed the methodology used to calculate unemployment rates in 2001, therefore, figure 1.8 in Appendix I shows estimated rates for the period 1994- 2001. 11 See figure 1.10, 1.11, and 1.12 in Appendix I. 12 Following that decision, the real fell 64% against the US dollar between mid-January and the beginning of March 1999. 12 The devaluation of the real created inflation fears that never materialized. Amman and Baer (2002) state that the economic contraction of 1998 and part of 1999 led to low levels of capacity utilization, adding that responsible fiscal and monetary policies also contributed to price stability. Specifically, as the real was losing value in the first weeks after devaluation, the central bank dramatically tightened its monetary policy, causing interest rates to reach 43% by March 1999. The paper also mentions the fiscal surplus imposed by the International Monetary Fund (IMF) as a counter-inflationary measure. By the end of 1999, the devaluation of Brazil?s real stood at only 22% in real terms (Giambiagi and Averburg, 2000). Brazil?s economy grew for most of the post-devaluation period with the exception of the 4 th quarter of 2001 and the 1 st quarter of 2002. This contraction is usually attributed by the Argentine crisis and devaluation. The stable macroeconomic behavior and growth experienced by Brazil after the devaluation of the real kept the unemployment rate around 10% for most of the period. FDI increased significantly in 1999 and reached a maximum level of over 30 billion dollars in 2000. This unprecedented level of FDI started to decrease steadily before reaching less than 10 billion dollars in 2004. The aggregate trade balance went from deficit to a $2.6 billion surplus in 1999. By 2004, the surplus had reached an impressive $33 billion. Finally, the bilateral trade balance with Argentina stayed on the deficit side until 2003, where it became a surplus. 12 See Amman and Baer (2002) for details. 13 The End of Convertibility The convertibility plan became under pressure once Brazil devalued the real. Even though the government was committed to the currency board, the fiscal deficits continued and the external debt reached unprecedented levels. This led to a loss of confidence from investors that generated large capital outflows and increased the costs of accessing funds in the international markets. International reserves declined leading to a monetary contraction that severely affected credit markets. Argentina?s central bank lost around 40% of reserves in 2001 alone, reaching their lowest level since the inception of the convertibility plan. By then, three economic ministers have tried different policies with the result of further economic contraction. 13 The unemployment rate in Argentina reached an unprecedented 20% in October 2001. FDI decreased to its lowest level since the implementation of the convertibility plan. The aggregate balance of trade became a surplus in 2000 and increased to $7 billion in 2001. This trade surplus was a consequence of a reduction in imports and not a result of higher exports. Argentine exports to Brazil declined almost 30% in 1999 but lower imports maintained a trade surplus for Argentina even after the devaluation of the real. The economic recession and trade restrictions imposed by Argentina were the major contributors toward reduction of Brazilian imports. Three years after the economic recession started, decreasing international reserves raised doubts about the sustainability of the one-to-one rate between the peso and the US dollar. Once the IMF refused to send a 1.2 billion dollar package to sustain the currency board, the Argentine government imposed restrictions on bank withdrawals so as to stop 13 See Quispe-Agnioli and Kay (2002) for details on the collapse of the convertibility plan. 14 the accelerating run on deposits. Violent protests led economic minister, Domingo Cavallo and president, Fernando de la R?a to resign. Interim president Adolfo Rodr?guez Sa? announced that Argentina was halting payments on its $140 billion external debt, creating the largest sovereign default in history. 14 In January 2002, Argentina?s Congress repealed the Convertibility Law and set a rate of 1.4 pesos per dollar. Fears of further devaluation exerted pressure on the exchange rate. By February 2002, the peso started to float freely against the dollar. In fact, a managed floating system emerged since the government had the right to intervene the foreign exchange market when it deemed necessary. Argentina?s recession accelerated after the devaluation of the peso in January 2002. GDP contracted by an annual rate of 11% and the unemployment rate reached 23% in May 2002. The annual inflation rate in 2002 reached 41% as the price of imports (mainly intermediate goods) passed through domestic prices. The peso reached its lowest level in July 2002 by reaching a rate of 4 pesos per dollar. FDI in Argentina continued to decrease and remained around $2 billion per year in 2002 and 2003, representing only 15% of FDI received by Brazil. By December 2002, international reserves at the central bank dropped to $10.5 billion, the lowest level in ten years. Argentina?s aggregate trade surplus reached a record high in 2002, pushed by a 55% decrease in imports. This decrease in imports could be attributed to banking restrictions applied in December 2001, import payment restrictions, and overall exchange rate uncertainty after devaluation. 15 Exports decreased by almost 5% in 2002 as compared to 2001. Argentine exports to Brazil decreased by 22% and imports contracted 14 See Quispe-Agnioli and Kay (2002). 15 See sub-section on trade restrictions. 15 at a 52% rate in 2002. This led to the highest bilateral trade surplus for Argentina with its major trading partner in at least ten years. By 2003, Argentina experienced an economic recovery that spurred imports, especially those coming from Brazil. Argentina?s Economic Recovery and Bilateral Trade Flows Banking restrictions were lifted in December 2002 as the peso stabilized at a rate of 3 pesos per dollar. Restrictions on the transfer of funds from Argentine importers to foreign suppliers were relaxed. Argentina?s economy started recovering by the first quarter of 2003, after shrinking for seventeen consecutive quarters. GDP growth averaged 9% in 2003 and 2004 bringing the economic activity close to the peak reached in 1998. This growth was achieved with inflation rates running at 3% and 6% respectively for 2003 and 2004. By the fourth quarter of 2004, the unemployment rate fell to less than 12% and FDI almost doubled between 2002 and 2004. International reserves increased by 100% due to aggregate trade surpluses and central bank intervention aimed to keep the peso at a stable rate of 3 pesos per dollar. 16 By the second quarter of 2003, Argentina started to experience a trade deficit with Brazil. Imports from Brazil grew by 87% in 2003 and another 62% in 2004. Imports from Brazil reached record levels while imports from the US and EU grew at much lower rates, reaching only half of the 1998 level by the end of 2004. 17 After devaluation, Argentine exports to Brazil remained at levels below those of the currency board. The stable path of Argentine exports to Brazil and the unstable behavior of imports suggest a study of the composition of trade flows between both countries. 16 At this exchange rate, exports surged and the government could collect taxes on exports to maintain its fiscal surplus. 17 See figures 1.15, 1.16, and 1.17 in Appendix I. 16 Composition of Argentina?s Imports and Exports Table 1.3 in Appendix I presents Argentina?s exports by type of good for the 1992-2004 period. On average, manufactures of agricultural origin represented 34% of total exports during this period, followed by manufactures of industrial origin with 29%, primary products with 23%, and fuels and energy with 14% share of total. Residues and waste from the food industry is Argentina?s largest export followed by fats and oils, cereals, crude oil and carburant. Argentine exports have increased steadily since 1992 without changing its composition. 18 Exports of manufactures of agricultural origin increased almost three- fold, manufactures of industrial origin went from $3 billion to $9 billion, exports of primary products increased by 100%, and fuels and energy increased almost six times during the 1992-2004 period. Table 1.5 in Appendix I breaks down the value of total exports by price and quantity. Among imports, intermediate goods represent the largest category averaging a 35% of the total, followed by capital goods with 23%, spare parts and pieces for capital goods with 17%, and consumer goods with 16%. Vehicles and fuels amount to an average 4% of Argentina?s total imports. Imports of intermediate goods have been rising especially after the peso devaluation. These imports represented 28% of the total in 1994, peaking at 48% in 2002, and declining to 38% of total imports in 2004. Consumer goods declined from $3.7 billion 2001 to a low $1 billion in 2002. Brazilian exports to Argentina have been gaining market share relative to the US and EU. In the case of intermediate goods, Brazilian imports increased their share from 18 See figures 18-25 in Appendix I. 17 38% in 1995 to 50% in 2004 at the expense of US and EU products. 19 Imports of Brazilian capital goods went from 13% market share to 45%, fuels from 10% to 34%, consumer goods from 31% to 52%, and vehicles went from 20% to an impressive 88% share for the 1995-2004 period. Brazilian exports to Argentina have increased their share gradually since the inception of Mercosur CET in 1995, but growth accelerated after the peso devaluation. The share of Argentina?s imports of intermediate goods from Brazil jumped from 40% to 50% between 2001 and 2004. During this time period, capital goods increased their share from 27% to 45%, the share of spare parts and pieces for capital goods rose from 33% to 40%, and the share of vehicles increased from 58 to 88%. Argentina?s Trade Regulations On December 1 st 2001, the government imposed emergency exchange controls in order to stop the drain of dollars from the financial system and maintain the convertibility system. Under the new measures that were to be put in effect for 90 days, the general public could only withdraw $250 in cash per week from any bank account. Transfer of funds abroad were limited to $1,000 for the general public while companies had to obtain official clearance before transferring any amount over $1,000 abroad. Restrictions were imposed on external trade transactions to keep the peso from further depreciation. For instance, capital goods, high-tech goods, and telecommunications could be only paid six months or a year after the transaction. According to Argentina?s Central Bank Communiqu? ?A 3473? of February 9 th 2002, capital goods that represented f.o.b. value of $200,000 or less could be only paid after a 19 See Tables 1.6 and 1.7 where imports from Brazil, EU, and US are 100% of total imports. 18 180-day term, accepting anticipated payments only up to 30% of the f.o.b. value. For merchandise worth more than f.o.b. $200,000, 20% could be paid ahead of time and the rest was to be financed at a minimum of 360-day period. These rules help explain the 68.5% contraction in imports of capital goods in 2002 as compared to 2001. Similarly, depending on the consumer good, payments would not be accepted prior to anywhere from 90 to 360 days, a fact that might also explain their 72% contraction from 2001 to 2002. With regard to payments coming from exports, Argentina?s Central Bank Communiqu? ?A 3473? established a maximum of 15 days to exchange foreign currency into pesos. Some goods were exempted from these restrictions. According to ?resolution 61/02? of Argentina?s Ministry of Economy, health care products, critical intermediate goods, and primary products could be paid ahead of time. The importer was supposed to prove the acquisition of those imports within 90 days following payment. Most of the restrictions on payments for trading goods were lifted in January 2003 when the central bank stopped restrictions on foreign currency-denominated payments for imports and lifted the limits on foreign companies? ability to send dividend payments abroad. Customs Tariffs The Argentine Harmonized System was implemented on January 1 st 1992. This system complies with the WTO Customs Classification Code adopted in 1979. Ad- valorem duties are imposed on the cost of insurance and freight (c.i.f.) value of the imported merchandise. Tariffs range from zero to 30%, and the average applied tariff is about 13.5%. On January 1 st , 1995, Argentina adopted the Mercosur CET which reduced the average tariff to zero on certain goods not produced locally and established a 2% to 19 10% tariff rate on raw materials, intermediate industrial materials, and primary products, a 12% tariff on capital goods, informatics, and telecommunications goods, a 15 to 20% tariff on consumer durable and nondurable goods, and 22.5% on non-finished goods. Commercial importers and individuals are authorized to import automobiles equivalent in value to a maximum of 10% of the value of domestic automobile production during the previous year. On March 2000, Argentina and Brazil agreed to raise the CET on automobiles to 35%. Taxes on Exports Due to the economic crisis and the fiscal problems at the time of devaluation, Argentina began imposing taxes on exports, a policy that had not been used in this country since the 1980s. Exports of crude petroleum oils and oils from bituminous minerals are subject to a 20% tax. Primary products pay a 10% export tax. Comparatively, manufactures of industrial origin as well as of gas end electricity pay a 5% export tax. On March 5 th 2002, ?Resolution # 35/2002? imposed a 20% tax on exports of some agricultural products. These exports taxes remain in place at the time of the writing of this dissertation. 20 APPENDIX I Figure 1.1: Argentina?s International Reserves Argentina's International Reserves 0 5,000 10,000 15,000 20,000 25,000 30,000 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 M il lons of US $ Figure 1.2: Brazil?s International Reserves Brazil's International Reserves 0 10,000 20,000 30,000 40,000 50,000 60,000 70,000 199 4 199 5 199 6 199 7 199 8 199 9 200 0 200 1 200 2 200 3 200 4 Mil lio n s o f U S $ Figure 1.3: Argentina?s Annual Inflation Rate Annual Inflation Rates in Argentina -10% 0% 10% 20% 30% 40% 50% 60% 70% 80% J a n- 92 J a n- 93 J a n- 94 J a n- 95 J a n- 96 J a n- 97 J a n- 98 J a n- 99 J a n- 00 J a n- 01 J a n- 02 J a n- 03 J a n- 04 % C h a nge Figure 1.4: Brazil?s Annual Inflation Rate Annual Inflation Rates in Brazil 0% 5% 10% 15% 20% 25% 30% S ep- 95 S ep- 96 S ep- 97 S ep- 98 S ep- 99 S ep- 00 S ep- 01 S ep- 02 S ep- 03 S ep- 04 % C h ang e Figure 1.5: Argentina?s GDP Annual Growth Rate GDP Growth Rates for Argentina -20.00% -15.00% -10.00% -5.00% 0.00% 5.00% 10.00% 15.00% 19 94: 1 19 94: 4 19 95: 3 19 96: 2 19 97: 1 19 97: 4 19 98: 3 19 99: 2 20 00: 1 20 00: 4 20 01: 3 20 02: 2 20 03: 1 20 03: 4 20 04: 3 A nnu al % C h ang e Figure 1.6: Brazil?s GDP Annual Growth Rate GDP Growth Rates for Brazil -4.00% -2.00% 0.00% 2.00% 4.00% 6.00% 8.00% 10.00% 12.00% 19 94 : 1 19 94 : 4 19 95 : 3 19 96 : 2 19 97 : 1 19 97 : 4 19 98 : 3 19 99 : 2 20 00 : 1 20 00 : 4 20 01 : 3 20 02 : 2 20 03 : 1 20 03 : 4 20 04 : 3 A nnu al % C h a n g e Figure 1.7: Unemployment Rate in Argentina Argentina's Unemployment Rate 0% 5% 10% 15% 20% 25% Ma y - 9 1 Ma y - 9 2 Ma y - 9 3 Ma y - 9 4 Ma y - 9 5 Ma y - 9 6 Ma y - 9 7 Ma y - 9 8 Ma y - 9 9 Ma y - 0 0 Ma y - 0 1 Ma y - 0 2 Ma y - 0 3 Ma y - 0 4 % R a t e Figure 1.8: Unemployment Rate in Brazil Brazil's Unemployment Rate 0% 2% 4% 6% 8% 10% 12% 14% Ma y - 9 1 Ma y - 9 2 Ma y - 9 3 Ma y - 9 4 Ma y - 9 5 Ma y - 9 6 Ma y - 9 7 Ma y - 9 8 Ma y - 9 9 Ma y - 0 0 Ma y - 0 1 Ma y - 0 2 Ma y - 0 3 Ma y - 0 4 % Ra t e 21 Figure 1.9: Foreign Direct Investment Foreign Direct Investment 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 M il li ons of U S $ FDI in Brazil FDI in Argentina Figure 1.10: Argentina?s Balance of Trade Argentina's Trade Balance -10,000 -5,000 0 5,000 10,000 15,000 20,000 1 992 1 993 1 994 1 995 1 996 1 997 1 998 1 999 2 000 2 001 2 002 2 003 2 004 Mi l li ons of U S $ Figure 1.11: Brazil?s Balance of Trade Brazil's Trade Balance -10,000 -5,000 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 1 994 1 995 1 996 1 997 1 998 1 999 2 000 2 001 2 002 2 003 2 004 M illio n s o f U S $ Figure 1.12: Argentina?s Balance of Trade with Brazil Argentina's Bilateral Trade Balance with Brazil -800 -600 -400 -200 0 200 400 600 800 1000 1994: 3 1995: 1 1995: 3 1996: 1 1996: 3 1997: 1 1997: 3 1998: 1 1998: 3 1999: 1 1999: 3 2000: 1 2000: 3 2001: 1 2001: 3 2002: 1 2002: 3 2003: 1 2003: 3 2004: 1 2004: 3 Millio n s o f U S $ Figure 1.13: Exchange Rates Exchange Rates 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 1994: 3 1995: 2 1996: 1 1996: 4 1997: 3 1998: 2 1999: 1 1999: 4 2000: 3 2001: 2 2002: 1 2002: 4 2003: 3 2004: 2 R eal or P e s o per U S $ Pes o/$ R/$ Figure 1.14: Argentina?s External Debt Argentina's External Debt 0 20,000 40,000 60,000 80,000 100,000 120,000 1 994 1 995 1 996 1 997 1 998 1 999 2 000 2 001 2 002 2 003 2 004 M il lio n s o f U S $ Figure 1.15: Argentina?s Imports from Brazil Argentina's Imports from Brazil 0 500 1000 1500 2000 2500 19 94: 3 19 95: 2 19 96: 1 19 96: 4 19 97: 3 19 98: 2 19 99: 1 19 99: 4 20 00: 3 20 01: 2 20 02: 1 20 02: 4 20 03: 3 20 04: 2 M ill ions o f U S $ Figure 1.16: Argentina?s Exports to Brazil Argentina's Exports to Brazil 0 500 1000 1500 2000 2500 19 94: 3 19 95: 2 19 96: 1 19 96: 4 19 97: 3 19 98: 2 19 99: 1 19 99: 4 20 00: 3 20 01: 2 20 02: 1 20 02: 4 20 03: 3 20 04: 2 M i l l i ons of U S $ 22 Figure 1.17: Argentina?s Imports from EU and US Argentina's Imports from EU and US 0 500 1000 1500 2000 2500 1 994 : 3 1 995 : 2 1 996 : 1 1 996 : 4 1 997 : 3 1 998 : 2 1 999 : 1 1 999 : 4 2 000 : 3 2 001 : 2 2 002 : 1 2 002 : 4 2 003 : 3 2 004 : 2 M il li o ns of US $ US Imports EU Impor ts Figure 1.18: Argentina?s Exports of Manufactures of Agricultural Origin Argentina's Exports of Manufactures of Agricultural Origin 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 M il li ons of US $ Figure 1.19: Argentina?s Exports of Manufactures of Industrial Origin Argentina's Exports of Manufactures of Industrial Origin 0.0 1,000.0 2,000.0 3,000.0 4,000.0 5,000.0 6,000.0 7,000.0 8,000.0 9,000.0 10,000.0 199 2 199 3 199 4 199 5 199 6 199 7 199 8 199 9 200 0 200 1 200 2 200 3 200 4 M illions of U S $ Figure 1.20: Argentina?s Exports of Primary Products Argentina's Exports of Primary Products 0.0 1,000.0 2,000.0 3,000.0 4,000.0 5,000.0 6,000.0 7,000.0 8,000.0 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 M illi o n s o f U S $ Figure 1.21: Argentina?s Exports of Fuels and Energy Argentina's Exports of Fuels and Energy 0.0 1,000.0 2,000.0 3,000.0 4,000.0 5,000.0 6,000.0 7,000.0 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 M i ll ions of US $ Figure 1.22: Argentina?s Exports of Manufactures of Agricultural Origin (%Share) Argentina's Exports of Manufactures of Agricultural Origin 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 S har e of T o t a l Figure 1.23: Argentina?s Exports of Manufactures of Industrial Origin (%Share) Argentina's Exports of Manufactures of Industrial Origin 0% 5% 10% 15% 20% 25% 30% 35% 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 S har e of T o tal Figure 1.24: Argentina?s Exports of Primary Products (%Share) Argentina's Exports of Primary Products 0% 5% 10% 15% 20% 25% 30% 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 M i ll ions of US $ 23 Figure 1.25: Argentina?s Exports of Fuels and Energy (% Share) Argentina's Exports of Fuels and Energy 0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 1 992 1 993 1 994 1 995 1 996 1 997 1 998 1 999 2 000 2 001 2 002 2 003 2 004 S har e of T o t a l Figure 1.26: Argentina?s Imports of Capital Goods from Brazil (% Share) Argentina's Imports of Capital Goods from Brazil 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 S h ar e of T o t a l Figure 1.27: Argentina?s Imports of Intermediate Goods from Brazil (% Share) Argentina's Imports of Intermediate Goods from Brazil 0% 10% 20% 30% 40% 50% 60% 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 S har e of T o t a l Figure 1.28: Argentina?s Imports of Consumer Goods from Brazil (% Share) Argentina's Imports of Consumer Goods from Brazil 0% 10% 20% 30% 40% 50% 60% 1 995 1 996 1 997 1 998 1 999 2 000 2 001 2 002 2 003 2 004 S h a r e of T o t a l Figure 1.29: Argentina?s Imports of Capital Goods by Country of Origin Imports of Capital Goods 0 500 1000 1500 2000 2500 3000 1 995 1 996 1 997 1 998 1 999 2 000 2 001 2 002 2 003 2 004 M il li o ns of US $ Brazil US EU Figure 1.30: Argentina?s Imports of Intermediate Goods by Country of Origin Imports of Intermediate Goods 0 500 1000 1500 2000 2500 3000 3500 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 M i llio n s o f U S $ Brazil US EU Figure 1.31: Argentina?s Imports of Spare Parts and Pieces for Capital Goods by Country of Origin Imports of Spare Parts and Pieces for Capital Goods 0 500 1000 1500 2000 2500 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 M illio n s o f U S $ Brazil US EU Figure 1.32: Argentina?s Imports of Consumer Goods by Country of Origin Imports of Consumer Goods 0 200 400 600 800 1000 1200 1400 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 M ill ion s o f U S $ Brazil US EU 24 Figure 1.33: Argentina?s Imports of Fuels by Country of Origin Imports of Fuels 0 50 100 150 200 250 300 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 M il li ons of US $ Brazil US EU Figure 1.34: Argentina?s Imports of Vehicles by Country of Origin Imports of Vehicles 0 200 400 600 800 1000 1200 1 995 1 996 1 997 1 998 1 999 2 000 2 001 2 002 2 003 2 004 M ill ion s of U S $ Brazil US EU 25 Table 1.1: Argentina?s Aggregate Trade Balance (Millions of US$) Period Exports Imports BOT 1992 12,399 13,795 -1,396 1993 13,269 15,633 -2,364 1994 16,023 20,162 -4,139 1995 21,162 18,804 2,357 1996 24,043 22,283 1,760 1997 26,431 28,553 -2,123 1998 26,434 29,531 -3,097 1999 23,309 24,103 -795 2000 26,341 23,889 2,452 2001 26,543 19,158 7,385 2002 25,709 8,473 17,236 2003 29,566 13,118 16,448 2004 34,453 21,185 13,267 Table 1.2: Brazil?s Aggregate Trade balance (millions of US$) Period Exports Imports BOT 1994 43,545 33,079 10,466 1995 46,506 49,972 -3,466 1996 47,747 53,346 -5,599 1997 52,994 59,747 -6,753 1998 51,140 57,763 -6,624 1999 48,011 49,295 -1,283 2000 55,086 55,839 -753 2001 58,223 55,572 2,650 2002 60,362 47,237 13,125 2003 73,084 48,291 24,793 2004 96,475 62,809 33,666 26 Table 1.3: Argentina?s Exports by Type of Good Categories 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Primary Products 3,500.2 3,270.9 3,735.3 4,815.8 5,817.1 5,704.7 6,603.3 5,144.4 5,345.6 6,052.1 5,289.5 6,459.9 6,827.6 Live Animals 8.7 13.2 51.0 97.8 44.6 35.2 19.3 17.9 15.9 17.5 8.2 8.8 11.3 Unprocessed Fish and Shell Fish 321.4 427.2 439.3 498.1 609.2 613.7 525.9 505.4 590.4 708.7 481.6 621.7 475.6 Honey 51.8 50.2 53.8 70.4 90.6 108.4 89.3 96.1 87.4 71.5 114.7 159.9 120.1 Unprocessed Vegetables, Legumes 168.2 185.5 259.2 268.4 270.5 352.1 460.6 270.2 210.2 233.5 184.9 187.4 198.7 Fresh Fruits 286.1 215.4 243.8 417.0 475.5 504.6 492.0 459.2 416.0 505.9 391.3 473.0 537.6 Cereals 1,547.7 1,453.6 1,332.7 1,862.6 2,560.1 3,006.7 3,042.1 2,063.1 2,419.1 2,447.8 2,134.6 2,306.7 2,703.7 Oil seeds and Fruits 790.1 696.5 951.8 884.6 963.7 338.7 1,052.1 869.7 1,016.8 1,401.1 1,294.6 1,992.5 1,829.8 Unprocessed tobacco 142.7 117.0 88.8 100.8 145.9 186.4 130.3 166.1 120.8 162.1 148.0 151.1 185.1 Raw Wool 41.2 49.1 74.6 86.2 64.7 61.3 39.7 38.7 43.2 30.4 35.4 35.2 41.8 Cotton fiber 76.6 25.7 176.3 432.8 497.0 332.3 224.3 177.9 53.3 73.1 12.1 2.3 10.8 Copper material and concentrates 0.0 0.0 0.0 0.0 0.0 68.5 438.6 412.6 307.0 346.7 437.2 467.3 642.4 Other Primary Products 65.7 37.5 64.0 97.1 95.3 96.8 89.1 67.5 65.5 53.8 46.9 53.8 70.7 Manufactures of Agricultural Origin 4,863.7 4,970.5 5,857.7 7,528.6 8,493.5 9,104.6 8,762.0 8,193.2 7,863.5 7,460.1 8,167.9 9,990.9 11,932.0 Meat 767.2 748.2 918.1 1,229.1 1,073.6 1,024.8 830.0 830.1 791.2 364.9 579.4 735.4 1,229.4 Processed Fish and Shellfish 236.6 279.3 285.8 416.2 394.9 416.5 385.8 296.5 242.7 237.7 236.2 253.8 322.3 Dairy Products 35.2 75.8 135.3 260.1 280.5 291.4 315.3 376.8 320.9 284.0 302.4 270.9 525.7 Other Products of Animal Origin 9.8 12.4 17.3 16.4 21.8 20.7 15.7 11.8 14.7 12.4 13.9 21.4 29.4 Dry or Frozen Fruits 23.7 21.9 32.0 27.8 33.4 31.3 31.6 34.4 35.2 31.0 42.3 54.6 63.6 Tea, Yerba Mate, Spices, etc. 46.9 62.3 61.0 67.3 64.6 79.2 84.4 65.0 64.7 67.0 60.8 53.4 61.0 Milled Products 51.4 59.3 87.8 90.2 166.0 203.5 165.1 131.4 158.3 145.1 115.9 86.1 92.9 Fats and Oils 1,109.1 1,078.6 1,533.6 2,097.1 1,890.5 2,225.0 2,733.7 2,332.1 1,678.1 1,636.6 2,095.4 2,831.5 3,168.0 Sugar and Confectionery items 65.4 43.3 58.7 122.0 144.5 133.7 135.9 106.1 137.9 120.6 159.2 131.2 146.6 Processed Legumes and Vegetables 260.4 166.4 160.1 321.2 400.1 391.5 319.0 340.7 308.4 325.7 291.1 365.9 447.6 Beverages, Alcoholic Liq. and Vinegar 64.0 64.2 79.8 165.2 153.1 197.4 231.3 205.9 215.2 209.8 173.6 216.1 273.1 Residues and waste from Food Industry 1,459.3 1,451.0 1,348.5 1,254.3 2,366.7 2,404.0 2,005.9 2,049.7 2,431.1 2,627.7 2,797.8 3,500.3 3,843.3 Tanning and Dyeing Extracts 40.3 44.2 43.2 39.6 41.5 49.8 46.2 39.0 39.8 39.7 34.9 33.8 35.6 Fats and Leather 475.1 617.8 762.8 937.0 889.3 980.1 812.4 779.8 835.7 819.5 700.6 727.3 840.9 Processed Wool 92.1 95.8 113.2 115.5 121.1 116.2 69.5 70.5 89.7 100.9 109.0 126.9 135.8 27 Table 1.3: Argentina?s Exports by Type of Good (continued) Other MAO 127.2 150.0 220.5 369.6 451.9 539.5 580.2 523.4 499.9 437.5 455.4 582.2 716.8 Manufactures of Industrial Origin 2,823.4 3,678.9 4,646.8 6,504.1 6,465.7 8,334.6 8,624.3 6,965.6 8,230.0 8,305.6 7,634.5 7,703.2 9,522.0 Chemicals and Related Products 533.4 558.8 727.5 972.5 980.0 1,176.1 1,370.0 1,373.1 1,386.6 1,432.1 1,349.1 1,558.9 2,017.5 Artificial Plastic Materials 148.0 133.0 180.6 340.7 339.9 349.2 380.0 369.3 518.6 628.7 642.7 695.9 937.8 Rubber and its manufactures 39.8 54.7 82.0 128.8 129.5 137.5 161.8 149.7 166.3 151.1 168.1 160.0 190.9 Leather goods 78.8 118.3 156.6 138.0 146.6 118.1 80.5 55.8 52.9 78.4 62.5 66.3 102.9 Paper, Cardboard, Printing and Publications 127.3 149.6 202.3 413.6 377.7 394.0 407.9 344.2 427.3 357.8 334.5 388.9 488.2 Textiles and Garments 121.5 164.9 210.1 383.8 304.5 334.7 320.5 278.3 304.7 263.9 227.6 210.4 271.5 Footwear and its Components 51.6 92.3 86.8 102.4 72.7 105.0 68.3 35.7 27.7 17.7 12.4 17.7 20.0 Stone, Gypsum and Ceramic Manufactures 71.2 78.8 70.9 109.8 106.7 120.2 113.7 96.1 96.7 91.3 94.7 102.0 123.6 Precious Stones and Metals and their Manuf. 4.2 52.0 251.6 23.1 4.9 3.7 29.6 113.0 102.3 103.9 118.0 115.8 145.5 Base metals and their manufactures 643.6 702.5 759.7 1,214.4 1190.3 1330.7 1234.5 1079.1 1412.1 1444.8 1601.9 1545.6 1670.5 Machines and Devices, Electric Material 518.4 754.8 866.5 983.0 961.5 1,230.4 1,109.6 1,054.8 1,102.4 1,125.1 942.2 861.0 1,052.1 Transport Material 404.8 719.4 918.2 1,307.8 1,641.9 2,786.4 3,102.5 1,751.9 1,957.0 1,982.2 1,615.1 1,432.2 2,068.1 Vehicles for air, maritime and river transportation n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. 318.9 231.5 105.4 199.4 39.3 Other MIO 80.8 99.8 134.0 386.2 209.5 248.6 245.4 264.6 356.5 397.1 360.3 349.1 394.1 Fuels and Energy 1,211.6 1,348.6 1,783.5 2,313.2 3,266.4 3,286.9 2,444.1 3,005.4 4,901.9 4,724.9 4,617.5 5,411.7 6,171.0 Oil crude 348.8 527.4 1,125.6 1,591.9 2,320.0 2,191.4 1,462.7 1,589.6 2,808.8 2,363.3 2,235.3 2,298.6 2,314.6 Carburant 760.0 712.5 539.3 463.6 696.1 842.0 696.6 983.3 1,368.3 1,426.9 1,557.8 2,016.6 2,389.1 Lubricants from Fats and Oils 0.0 0.0 0.0 107.9 58.3 48.1 56.8 43.4 53.7 69.2 55.5 89.9 106.1 Oil gas and other hydrocarbures 44.3 65.9 71.8 75.7 109.6 128.9 162.1 278.9 451.8 609.8 628.1 872.1 1,131.1 Electric energy 1.9 1.1 0.9 7.8 13.0 11.3 2.5 27.9 148.2 159.4 67.2 36.7 93.3 Others 56.6 41.7 45.9 66.3 69.4 65.2 63.4 82.3 71.1 96.3 73.6 97.9 136.8 Total 12,398.9 13,268.9 16,023.3 21,161.7 24,042.7 26,430.8 26,433.7 23,308.6 26,341.0 26,542.7 25,709.4 29,565.8 34,452.6 28 Table 1.4: Argentina?s Imports by Type of Good (FOB) 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Capital Goods 2,900 3,889 5,696 4,509 5,348 7,387 8,154 6,515 5,728 3,981 1,248 2,426 5,209 Intermediate Goods 4,313 4,629 5,724 6,644 7,744 9,310 9,260 7,760 7,850 6,878 4,062 5,873 8,078 Fuels 468 422 598 781 841 897 785 676 970 791 457 518 919 Spare Parts and Pieces for Capital Goods 2,429 2,637 3,201 3,171 3,855 5,215 5,217 3,979 4,218 3,224 1,454 2,133 3,443 Consumer Goods 2,898 3,206 3,561 2,917 3,311 4,206 4,516 4,227 4,323 3,758 1,073 1,660 2,354 Vehicles 746 803 1,338 745 1,157 1,514 1,576 926 778 511 170 497 1,169 Others 41 48 45 39 27 26 24 21 22 15 10 12 13 Total 13,795 15,633 20,162 18,804 22,283 28,553 29,531 24,103 23,889 19,158 8,473 13,118 21,185 29 Table 1.5: Indices of Price and Quantities for Argentina?s Exports and Imports Period Value Price Quantity Value Price Quantity 1992 93.3 99.8 93.5 88.6 102.7 86.3 1993 100.0 100.0 100.0 100.0 100.0 100.0 1994 120.7 102.9 117.4 128.6 101.4 126.9 1995 159.8 108.8 146.9 119.9 106.9 112.2 1996 181.5 115.9 156.6 141.6 105.6 134.0 1997 201.5 111.9 180.1 181.4 103.2 175.8 1998 201.6 100.3 201.0 187.1 97.9 191.1 1999 177.7 89.1 199.5 152.0 92.4 164.6 2000 200.8 98.0 204.9 150.6 92.4 163.1 2001 202.3 94.7 213.7 121.1 89.9 134.7 2002 196.0 91.0 215.3 53.6 86.7 61.5 2003 224.0 99.7 224.7 84.2 87.0 94.7 2004 226.6 109.0 214.0 133.0 93.7 142.0 Exports Imports Table 1.6: Share of Brazilian Imports in Argentina Period Capital Goods Intermediate Goods Fuels Spare Parts and Pieces for Capital Goods Consumer Goods Vehicles Others 1995 13% 38% 10% 37% 31% 20% 9% 1996 19% 36% 17% 39% 33% 29% 24% 1997 22% 37% 17% 32% 35% 50% 23% 1998 24% 37% 9% 30% 35% 48% 24% 1999 22% 37% 26% 28% 37% 45% 23% 2000 32% 40% 32% 31% 46% 66% 22% 2001 27% 40% 66% 33% 50% 58% 17% 2002 21% 45% 31% 34% 42% 83% 33% 2003 46% 49% 19% 39% 50% 88% 56% 2004 45% 50% 34% 40% 52% 88% 41% *Argentina?s imports from Brazil, EU, and US represent 100% of imports. 30 Table 1.7: Argentina?s Imports by Type of Good and Country of Origin Period Brazil US EU Brazil US EU Brazil US EU Brazil US EU Brazil US EU Brazil US EU 1995 465.4 1588.4 1568.4 2004.3 1478.4 1827.2 39.9 85.2 261.4 988.6 546.2 1156 542.8 448.5 743.8 132.4 56.8 457.7 1996 804.6 1674.8 1811.2 2269.3 1844 2111.8 51 71.4 175.5 1234 584.5 1349.4 665.4 501.7 819.2 273.9 58.1 628.8 1997 1305.3 2321.8 2240.7 2689.2 2033 2522.7 33.3 70.2 94.9 1370.4 931.2 1929.2 878 643.5 993.5 635.5 91.8 535.6 1998 1527 2374 2419.3 2660 1915.3 2581.6 13.9 64.4 77.5 1246.1 1052.6 1820.9 973.8 658.7 1162.7 671.5 89.3 624.4 1999 1132.4 1936.8 2024.8 2202.4 1560.3 2158.6 36.9 39.8 66.2 885.9 807.2 1430.5 997.1 618 1044.9 342.4 31.5 390.7 2000 1385.3 1621.9 1296.9 2425.1 1688.7 1929.3 58.6 58.7 68.7 926.9 801.7 1309.3 1256.7 536 965.5 421.8 36.8 184 2001 811.2 1116.5 1090.4 2120.3 1521.8 1614.1 134 32.2 35.9 758.7 677.9 893.9 1219.7 406.6 818.9 232.6 24.1 141.7 2002 214.9 484.8 331 1449.3 837.6 925.3 25.5 33.5 23.4 371.9 287.7 430.4 334.9 157.7 298.4 121.1 2.7 22.3 2003 855.8 532.7 466.2 2223.4 1095.5 1219 17 35.4 37 593.7 383.4 564 602.9 208.9 384.2 413.2 7.8 47.3 2004 1745.8 1077.7 1020.5 3075.6 1485.8 1603 61.7 53.2 65.8 942.7 523.5 912.8 817 276.5 481.1 959.8 14.7 114.1 Spare Parts and Pieces for Capital Goods Consumer Goods VehiclesCapital Goods Intermediate Goods Fuels 31 APPENDIX II 32 APPENDIX III Table 1.8: List of Trade Agreements Acronym Trade Agreement Countries AFTA ASEAN Free Trade Area Brunei Darussalam Cambodia Indonesia Laos Malaysia Myanmar Philippines Singapore Thailand Vietnam ASEAN Association of South East Asian Nations Brunei Darussalam Cambodia Indonesia Laos Malaysia Myanmar Philippines Singapore Thailand Vietnam BAFTA Baltic Free-Trade Area Estonia Latvia Lithuania BANGKOK Bangkok Agreement Bangladesh China India Republic of Korea Laos Sri Lanka CAN Andean Community Bolivia Colombia Ecuador Peru Venezuela CARICOM Caribbean Community and Common Market Antigua & Barbuda Bahamas Barbados Belize Dominica Grenada Guyana Haiti Jamaica Monserrat Trinidad & Tobago St. Kitts & Nevis St. Lucia St. Vincent & the Grenadines Surinam CACM Central American Common Market Costa Rica El Salvador Guatemala Honduras Nicaragua Bulgaria Croatia Romania CEFTA Central European Free Trade Agreement CEMAC Economic and Monetary Community of Central Africa Cameroon Central African Republic Chad Congo Equatorial Guinea Gabon Australia New Zealand CER Closer Trade Relations Trade Agreement CIS Commonwealth of Independent States Azerbaijan Armenia Belarus Georgia Moldova Kazakhstan Russian Federation Ukraine Uzbekistan Tajikistan Kyrgyz Republic COMESA Common Market for Eastern and Southern Africa Angola Burundi Comoros Democratic Republic of Congo Djibouti Egypt Eritrea Ethiopia Kenya Madagascar Malawi Mauritius Namibia Rwanda Seychelles Sudan Swaziland Uganda Zambia Zimbabwe EAC East African Cooperation Kenya Tanzania Uganda EAEC Eurasian Economic Community Belarus Kazakhstan Kyrgyz Republic Russian Federation Tajikistan EC European Communities Austria Belgium Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Malta Netherlands Poland Portugal Slovak Republic Slovenia Spain Sweden United Kingdom ECO Economic Cooperation Organization Afghanistan Azerbaijan Iran Kazakhstan Kyrgyz Republic Pakistan Tajikistan Turkey Turkmenistan Uzbekistan 33 EEA European Economic Area EC Iceland Liechtenstein Norway EFTA European Free Trade Association Iceland Liechtenstein Norway Switzerland GCC Gulf Cooperation Council Bahrain Kuwait Oman Qatar Saudi Arabia United Arab Emirates GSTP General System of Trade Preferences among Developing Countries Algeria Argentina Bangladesh Benin Bolivia Brazil Cameroon Chile Colombia Cuba Democratic People's Republic of Korea Ecuador Egypt Ghana Guinea Guyana India Indonesia Islamic Republic of Iran Iraq Libya Malaysia Mexico Morocco Mozambique Myanmar Nicaragua Nigeria Pakistan Peru Philippines Republic of Korea Romania Singapore Sri Lanka Sudan Thailand Trinidad and Tobago Tunisia United Republic of Tanzania Venezuela Vietnam Yugoslavia Zimbabwe LAIA Latin American Integration Association Argentina Bolivia Brazil Chile Colombia Cuba Ecuador Mexico Paraguay Peru Uruguay Venezuela MERCOSUR Southern Common Market Argentina Brazil Paraguay Uruguay MSG Melanesian Spearhead Group Fiji Papua New Guinea Solomon Islands Vanuatu Canada Mexico United States NAFTA North American Free Trade Agreement OCT Overseas Countries and Territories Greenland New Caledonia French Polynesia French Southern and Antarctic Territories Wallis and Futuna Islands Mayotte Saint Pierre and Miquelon Aruba Netherlands Antilles Anguilla Cayman Islands Falkland Islands South Georgia and South Sandwich Islands Montserrat Pitcairn Saint Helena Ascension Island Tristan da Cunha Turks and Caicos Islands British Antarctic Territory British Indian Ocean Territory British Virgin Islands PATCRA Agreement on Trade and Commercial Relations between the Government of Australia and the Government of Papua New Guinea Australia, Papua New Guinea PTN Protocol relating to Trade Negotiations among Developing Countries Bangladesh Brazil Chile Egypt Israel Mexico Pakistan Paraguay Peru Philippines Republic of Korea Romania Tunisia Turkey Uruguay Yugoslavia SADC Southern African Development Community Angola Botswana Lesotho Malawi Mauritius Mozambique Namibia South Africa Swaziland Tanzania Zambia Zimbabwe 34 SAPTA South Asian Preferential Trade Arrangement Bangladesh Bhutan India Maldives Nepal Pakistan Sri Lanka SPARTECA South Pacific Regional Trade and Economic Cooperation Agreement Australia New Zealand Cook Islands Fiji Kiribati Marshall Islands Micronesia Nauru Niue Papua New Guinea Solomon Islands Tonga Tuvalu Vanuatu Western Samoa TRIPARTITE Tripartite Agreement Egypt India Yugoslavia WAEMU West African Economic and Monetary Union Benin Burkina Faso C?te d'Ivoire Guinea Bissau Mali Niger Senegal Togo Source: WTO 35 CHAPTER 2: TRADE ADJUSTMENTS TO CURRENCY DEVALUATION AND TRADE DIVERSION I. Introduction In December 2001, with passage of the Law of Public Emergency and Reform of the Exchange Rate Regime, the ?convertibility plan? was abolished and Argentina abandoned the one-to-one parity between the dollar and peso. By the end of January 2002, the peso started to float freely against all major currencies. Argentina?s exchange rate regime reform and devaluation was intended to improve the country?s balance of payments. Economic theory suggests that devaluation should lead to a higher trade balance. According to Magee (1973), real devaluations lower the trade balance in the short run but raise it in the long run. This behavior is known as the J-curve since plotting the trade balance over time generates a curve with the shape of the letter J. Since Magee?s seminal paper, a number of empirical studies have found mixed evidence on the J-curve. While papers finding J-curves use Magee?s theoretical framework as a justification, studies finding no evidence of a J-curve do not highlight the reasons for the lack of empirical support. 20 Argentina has been experiencing an aggregate (with the rest of the world) trade surplus since 1999, which indeed widened after the peso devaluation of 2002. However, 20 See literature review and Appendix I in Chapter 2. 36 two issues emerge in the analysis of Argentina?s trade balance adjustments to devaluation. First, when studying a country that belongs to a common market (Mercosur), trade balance adjustments to exchange rate movements may differ between member and non-member countries primarily due to the regionalization forces detailed in Chapter 1. Second, as pointed out by Bahmani-Oskooee and Brooks (1999), aggregate trade data may not be an accurate gauge since a country?s trade balance may improve against some trading partners (Argentina vis-?-vis Europe and the United States) and deteriorate against others (Argentina vis-?-vis Brazil). These two issues suggest that the analysis of bilateral trade flows might uncover trade patterns not apparent in aggregate data. Bilateral trade studies should lead to more precise policy implications. After the peso devaluation of January 2002, Argentina?s trade balance with Brazil deteriorated at a fast pace and became a deficit in the second quarter of 2003. In this case, Argentina?s currency devaluation seems to have worked opposite to the theoretical prediction. On the other hand, Argentina?s trade deficit with non-Mercosur trading partners US and EU became a surplus after devaluation. Argentina?s post-devaluation trade deficit with Brazil is primarily attributed to a surge in imports. This increase in Argentina?s imports from Brazil has become one of the main economic topics in both countries. The effects of the peso devaluation on the trade balance with Brazil have been widely covered by the local media. Government officials have already implemented tariffs and non-tariff barriers (NTBs) to correct the current ?asymmetries? in trade flows between the two countries. For example, in July 2004, Argentina?s government imposed temporary import licenses on Brazilian home appliances. Also, a 21% tariff on imports of Brazilian television sets was imposed in July 37 2004. While the NTBs on home appliances are still in place, the 21% tariff imposed on imports of Brazilian televisions from the free trade zone of Manaus was eliminated. On March 2005, Brazilians rice producers complained to their government that Mercosur members are exporting rice to Brazil at dumping prices. A similar argument was made by Brazil?s wine and flour producers. Those sectors plan to retaliate against Argentina?s import tariffs. 21 This chapter examines the statistical relationship between Argentina?s currency devaluation and its bilateral trade balance with Brazil using quarterly data for the 1994:3- 2004:4 period. An analysis of the short-run dynamics of trade balance adjustments is outlined. This analysis augments the conventional J-curve dynamics by including a measure of trade diversion in the balance of trade model. Real exchange rate devaluations lead to changes in a country?s income and purchasing power relative to its trading partners. The peso devaluation in January 2002 could have lowered Argentina?s income and accelerated the country?s diversion of more expensive imports from the US and EU to cheaper substitutes from Brazil (a Mercosur member). 22 When analyzed in the context of regional economic integration, the dynamics of trade balance adjustments should investigate trade diversion effects from non-member to member countries. Section II presents a review of the literature on the J-curve phenomenon. This review describes models and findings from previous research in the area starting with Magee?s (1973) seminal paper. Section III provides an overview of relevant theoretical issues regarding the impact of currency devaluations on trade balances. Section IV 21 See Chapter 1 for a general discussion on tariffs, quotas, and NTB?s within a regional economic integration agreement. 22 Chapter 3 investigates the Linder hypothesis and trade diversion issues. 38 covers the empirical model and section V describes and prepares the data for model estimation. Section VI outlines results and section VII presents the conclusions and future research direction. II. Literature Review When the United States trade balance deteriorated from a surplus of $2.2 billion in 1970 to a deficit of $2.7 billion in 1971, the government devalued the dollar to correct such a deficit. The first year after the dollar devaluation (1972), the trade deficit reached $6.8 billion. Magee (1973) explains this phenomenon in terms of adjustment lags and analyzes the currency contract, pass-through, and quantity adjustments periods. 23 This paper analyzes in detail the short-term dynamics of devaluations on the trade balance when imports and exports are measured in home currency as well as in foreign currency. He argues that there may or may not be a J-curve in the short run and concludes that the long-run impact of devaluation on the trade balance is favorable. Since Magee?s seminal paper, the J-curve theory has been tested in different countries. The most cited papers in the literature are presented below. 24 Himarios (1985) uses annual data from 1956 to 1972 for Costa Rica, Ecuador, Finland, France, Iceland, Israel, the Philippines, Spain, Sri Lanka, and the United Kingdom. The study adds lagged values of exchange rates to a model investigating the trade dynamics of currency devaluations. Himarios suggests that the real exchange rate and not the nominal exchange rate is what affect trade flows. Domestic and foreign government expenditures as well as a variable accounting for the opportunity cost of 23 See section III in this chapter for a detailed discussion of the currency contract, pass-through, and quantity adjustments periods. 24 See Appendix I in this chapter for a table presenting the most frequently cited papers on J-curve. 39 money are part of the estimated model. Himarios (1985) finds that devaluation improves the trade balance in nine out of ten countries under study. Bahmani-Oskooee (1985) analyzes the effect of devaluation in India, Greece, Korea, and Thailand using quarterly data for the 1973-1980 period. The paper is the first of a series of papers that test the presence of a J-curve by using an Almon lag structure on the exchange rate variable. A maximum of 12 lags are used to test for the J-curve phenomenon. With the exception of Thailand, the author finds the presence of a J-curve in Greece, Korea, and India, but with different time adjustments. The study also shows that devaluation in the long run deteriorates the trade balance for all countries except for Thailand. By using a dummy variable that represents a sudden change in the exchange rate for Korea, the author shows that sudden one-time shocks in the exchange rate affect the trade balance in a different way than small daily adjustments. The absorption and monetary approach are also tested with insignificant results in most cases. Felmingham (1988) uses an unrestricted distributed lag model to test for the presence of a J-curve in Australia for the 1965:1-1985:2 period. The terms of trade, Australian GNP, and the US GNP (proxy for world income) are the model?s explanatory variables whereas the import-export ratio is the dependent variable. Model estimations are based on two different periods: the fixed exchange rate period (1965-1973) and the managed or floating exchange rate period (1974-1983). Findings suggest evidence of a ?delayed J-curve? since it takes more than eight quarters to improve the trade balance after a change in the terms of trade. Felmingham finds no evidence of a J-curve for the floating period. It is worth noting that changes in the terms of trade not always reflect 40 changes in exchange rates; therefore, results may not properly reflect the impact of exchange rate adjustments on the trade balance. Bahmani-Oskooee (1989) improves his 1985 paper by changing the way the real exchange variable is measured. The author adds a measure of foreign price level in the real exchange rate variable and defines the exchange rate as the number of units of domestic currency per unit of foreign currency. The paper shows that once these changes are in place, the real exchange variable should have negative coefficients followed by positive ones if a J-curve is present. The author finds an ?inverse J-curve? when re- estimating the model in Bahmani-Oskooee (1985). Results suggest that devaluations first improve and then lower the trade balance. There are no changes in his long-run results that devaluation improved the trade balance only in Thailand. Himarios (1989) examines the effectiveness of devaluation on trade balance adjustments by looking at two different samples (1953-73 and 1975-84) involving 27 countries and 60 devaluations. His evidence indicates that devaluations have been a successful tool in inducing trade balance adjustments. Specifically, he finds that nominal devaluations resulted in significant real devaluations that last for at least three years, and this significant real devaluation increased exports relative to imports. Brissimis and Leventankis (1989) use an Almon lag structure to test the elasticities and monetary approaches to the balance of trade in Greece for the 1975-1984 period. The paper examines the impact of a non-sustained 10% devaluation of the Greek currency (drachma) on the country?s trade balance. With the use of instrumental variables (IV), they find similar long-run results to Bahmani-Oskooee (1989). In the short run, they find evidence of a J-Curve in Greece with an initial deterioration that 41 lasted one quarter. Contrary to Brissimis and Leventankis (1989), Karadeloglou (1990) finds evidence of an ?inverse J-curve? for Greece during the 1974-1983 period. The results of this paper are based on a macroeconometric model of the Greek economy that includes consumption, private investment, imports, exports, inventory changes, prices, wages, and other macroeconomic variables. Bahmani-Oskooee and Pourheydarian (1991) use an Almon lag structure on the real exchange rate variable. They test for the presence of a J-curve in Australia during the 1977?1988 period using quarterly data. They find evidence of a ?delayed J-curve? for Australia and conclude that depreciation leads to trade balance improvements. The authors suggest that measuring the trade balance as an export-import ratio (X/M) does not affect the model?s results. Bahmani-Oskooee and Malixi (1992) use a similar model and test for the dynamics of the J-curve in 13 less developed countries (LDCs). Using quarterly data from 1973:1 to 1985:4, they find support for the J-curve in Brazil, Greece, Korea, and India. They also report, in line with Magee (1973), shapes such as the N-, M-, and I- curves, concluding that the short-run effects may not follow a standard pattern. Bahmani-Oskooee and Malixi find that in the long run devaluations have a positive impact on trade balance. Bahmani-Oskooee and Alse (1994) improve Himarios (1989) by using stationary variables and measuring the trade balance as the import-export (M/X). The study discounts the results in Bahmani-Oskooee (1985) and Himarios (1989) arguing that these papers used non-stationary variables. Bahmani-Oskooee and Alse use Engle-Granger?s cointegration technique on quarterly data from 1971 to 1990 for 19 developed and 22 42 LDCs. In the long run, devaluation improves the trade balance for Costa Rica, Brazil, and Turkey and has a negative effect in Ireland. No long-run effects are found in other countries. Using an error-correction model (ECM), the authors report the presence of a J- curve effect in Costa Rica, Ireland, the Netherlands, and Turkey. Buluswar, Thompson, and Upadhyaya (1996) compare absorption, elasticity, and monetary models and conclude that the monetary model performs better in India. This paper uses quarterly data from 1960 to 1990 and imposes an Almon lag structure on the real exchange rate. They find no evidence of a J-curve and conclude that devaluations have had no significant long-run effects on the trade balance. Upadhyaya and Dhakal (1997) test the effectiveness of devaluation on the trade balance of eight developing countries in Asia, Europe, Africa, and Latin America. They use a distributed lag model on the dependent variable and the real exchange rate. Their findings show that devaluation does not improve the trade balance in the long run. Specifically, only in Mexico does devaluation improve the trade balance in the long run, with the opposite applying to Cyprus, Greece, and Morocco. In Colombia, Guatemala, Singapore, and Thailand, devaluation is neutral in the long run. Gupta?Kapoor and Ramakrishnan (1999) use an ECM for Japan during the 1975:1-1996:4 period. The trade balance is measured as the import-export ratio. The model is estimated with variables in nominal terms. Using Johansen?s likelihood ratio cointegration test, the authors find evidence of a long-run relationship between the trade balance variable and the exchange rate. The authors find evidence of the J-curve phenomenon. Results remained unchanged even when real variables were used in the estimation. 43 Most of the recent papers examine the J-curve phenomenon with bilateral trade models. Rose and Yellen (1989) examine bilateral trade patterns for the US using quarterly data for the 1963-1988 period. They estimate a log-linear model with the trade balance measured as US net exports with the foreign country. Real GNP in the US and the foreign country along with the real exchange rate are the model?s independent variables. The authors use stationary variables and test for cointegration among variables. Even though they find no evidence of cointegration among variables, they estimate a model in first differences with different lag structures on independent variables. They also correct for simultaneity bias and measurement errors by using IV. The paper fails to find a J-curve and concludes that the real exchange rate does not affect the trade balance with the exception to Germany and Italy in which lagged coefficients of the real exchange rate are statistically significant. Marwah and Klein (1996) use an Almon distributed lag model for the US and Canada to show evidence of J-curves with France, Germany, the UK, and Japan on a bilateral basis. The authors use the export-import ratio as the dependent variable. The model is estimated using ordinary least square (OLS) and IV. A polynomial distributed lag is also used in the estimations. Marwah and Klein find that the timing and shape of J- curves are similar for both countries, but the initial deterioration of the trade balance is deeper in Canada. The paper covers the 1977:1-1992:1 period using quarterly data. Bahmani-Oskooee and Brooks (1999) improve Rose and Yellen (1989) by using the US import-export ratio with its trading partners. They suggest that this ratio captures real and nominal movements in the trade balance. The paper also objects to the use of non-stationary data in Marwah and Klein (1996). Estimations are based on the 44 Autoregressive Distributed Lag (ARDL) model introduced by Peasaran and Shin (1995) and Peasaran, Shin, and Smith (2001). The paper examines the 1973:1-1996:2 period and finds no statistical evidence of a J-curve effect in the US. Model results show that a real depreciation has a positive effect on the bilateral trade balances between the US and its 6 major trading partners in the long run. Wilson and Tat (2001) examine the relationship between the balance of trade and the real exchange rate between Singapore and the United States. Using quarterly data from 1970 to 1996 and an ARDL model, they find that the real exchange rate does not have a significant impact on the bilateral trade between both countries. They also find no evidence of a J-curve effect. Similarly, Wilson (2001) studies bilateral trade flows between Singapore, Malaysia, and Korea with both the US and Japan. In order to test for the presence of a J-curve, he uses a Vector Autoregressive (VAR) model and finds evidence of a J-curve only in the Korean case. Baharumshah (2001) studies the effect of exchange rates on bilateral trade balances for Malaysia and Thailand with the US and Japan. With the use of an unrestricted VAR model, he finds that depreciation of Malaysia and Thailand currencies causes trade balances to improve with both the US and Japan. Further, he finds that the improvement in Malaysia-US trade balance place the same quarter that devaluation occurred. Results also suggest that the real effects of devaluation on the trade balance are distributed over a period of eight to nine quarters. The author finds evidence of stable long-run relationships between trade and exchange rates as well as between trade and incomes (domestic and foreign). The paper uses quarterly data from 1980:1 to 1996:4. 45 Bahmani-Oskooee and Kanitpong (2001) use an ARDL model to investigate the presence of a J-curve phenomenon for Thailand with Germany, the UK, the US, Japan, and Singapore. They use quarterly data from 1984 to 1997 to find a J-curve in the US and Japan. Chen (2001) estimates two models by imposing an Almon lag structure on the real exchange rate. The first model estimates real exports as a function of foreign income, real imports, and real exchange rate. The second model estimates the trade balance as a function of foreign income and real exchange rates. Using quarterly data from 1981:1 to 1998:1, the paper studies bilateral trade flows for Taiwan with the US and Japan. Chen finds that real income affects real exports in both cases and that real exchange rates and real imports do not affect Taiwan?s exports to the US, but they do affect Taiwan?s exports to Japan. Chen also finds that real exchange rates have significant effects on the trade balance with US and Japan, but income does not. Lal and Lowinger (2002) use quarterly data between 1980 and 1998 for seven East Asian countries. This paper examines the determinants of trade balances using Johansen?s cointegration technique, error correction model, and impulse-response function. Among other findings, their investigation confirms the existence of a J-curve effect and results show that there are significant differences in the duration and extent of the J-curve effect across countries. Bahmani-Oskooee and Goswami (2003) use an ARDL approach to cointegration and error correcting modeling to test for the presence of a J-curve in Japan with 9 major trading partners. The paper utilizes quarterly data from 1973 to 1998 and finds support 46 for the J-curve only with Germany and Italy. They conclude that currency depreciation improves the trade balance in the long run. Arora, Bahmani-Oskooee, and Goswami (2003) investigate the occurrence of a J- curve phenomenon in India?s trade with Australia, France, Germany, Italy, Japan, the UK, and the US with an ARDL approach to cointegration and error correcting modeling. Defining the J-curve as a short-run deterioration followed by long-run improvements, they find a J-curve in India?s trade with Australia, Germany, Italy, and Japan. Hacker and Abdulnasser Hatemi-J (2003) test for the J-curve in five North European countries (Belgium, Denmark, the Netherlands, Norway, and Sweden) using generalized impulse-response functions. The results provide empirical support for the J- curve. Each country has an impulse-response function generated from a vector error- correction model suggesting that after depreciation there will be a dip in the export- import ratio within the first half-year. The long-run export-import ratio appears to be higher than the low point of this early dip in almost all cases. Bahmani-Oskooee and Ratha (2004) expand Bahmani-Oskooee and Brooks (1999) by adding 12 US industrial trading partners. The study employs an ARDL model to investigate whether a J-curve is present. The results show that a J-curve is present only in the case of the Netherlands. Due to this lack of support for the theory, the authors redefined the J-curve phenomenon as a short-run deterioration of the trade balance with long-run improvements. Based on this version, results are supportive of a J-curve for Austria, Denmark, France, Germany, Ireland, Italy, Japan, New Zealand, Sweden, and Switzerland. In other words, a real devaluation of the dollar has a positive long-run impact on the trade balance. Finally, Bahmani-Oskooee and Ratha (2004b) broaden the 47 scope of the previous paper by investigating the J-curve phenomenon between the US and 13 developing countries. With the use of quarterly data from 1975:1 to 2000:2, results from an ARDL model support the new definition of the J-curve in seven out of the thirteen countries studied. Empirical findings are ambiguous. Note that none of the cited papers investigates whether the countries involved are part of a regional economic integration agreement. The focus of this chapter is to study the short-term dynamic of bilateral trade flows between two countries in of a common market that have intentionally devalued their currencies in order to induce trade balance improvements. At a time when talks about the Free Trade Area of the Americas (FTAA) seem to advance as slowly as the potential trade agreements between the European Union and Mercosur, examining trade patterns between the two major economies of Mercosur may uncover interesting policy implications. III. Theories on Currency Devaluation and its Effect on the Trade Balance Most of the models explaining a country?s current account originated during the fixed exchange rates era (1950s and 1960s). 25 Consequently, the literature focuses on the effect of currency devaluation on the trade balance. Three main approaches emerged from the models of current account: the absorption, elasticity, and monetary approaches. The Absorption and Monetary Approaches The absorption approach focuses on home and foreign incomes, and states that an increase in home income relative to the income of trading partners should lower the trade balance. Domestic export revenue (X) is the value of domestic goods bought by 25 See Krueger (1983). 48 foreigners and is a function of trading partners? real income (Y*). As foreign income Y* increases, some of that income would be spent on domestic goods and domestic exports will expand with the consequent improvement on the trade balance. Domestic imports (M) are the value of foreign goods bought by domestic residents and are a function of domestic real income (Y). A rise in domestic income is associated with higher imports since part of the additional income will be spent on foreign goods. Higher domestic real income leads to deterioration of the trade balance while higher foreign real income leads to trade balance improvements. A country?s real GDP is commonly used as a measure of real income. The monetary approach to the exchange rate states that devaluation decreases the real supply of money creating an excess demand that leads to hoarding, which in turn generates trade balance improvements. The monetary approach can be explained by analyzing the determinants of the price level in a small open economy. This scenario fits the case of the two countries in question. The price level on the economy is a weighted average of the prices of exportable goods and imported goods: P = ?P x + (1- ?) P m (2.1) where P x is the price of exportable goods, P m is the price of imports, ? is the share of exportable goods, and (1- ?) is the share of imports in the economy. It is assumed that this small open economy is a price taker, so to calculate domestic prices we convert world prices with the use of the nominal exchange rate e. Therefore, the domestic price of imports and domestic price of exports can be written as: P m = e P m * (2.2) P x = e P x * (2.3) 49 where P m * is the world price of imports and P x * is the foreign currency price of exports. Substituting (2.2) and (2.3) into (2.1), we obtain: P = ?eP x * + (1- ?)eP m * = e (?P x * + (1- ?) P m * ) P = eP * (2.4) where P * = (?P x * + (1- ?) P m * ) is the world price of a basket of goods consumed domestically. As equation (2.4) implies, an increase in the nominal exchange rate e will raise the domestic price level P. As a consequence, the real value of money balances declines as shown by equation (2.5) where Ms is the money supply and L is real money demand or real money balances: Ms/P = L (Y, i) (2.5) A decrease in real money balances leads to hoarding while domestic spending or absorption decreases bringing about an improvement in the trade balance. The monetary approach predicts trade balance improvements in the short term. However, the payment surplus will generate additions to the money supply over time, taking the economy to the pre-devaluation equilibrium (Rivera-Batiz, 2002). In short, the monetary approach to devaluation states that currency devaluation improves the trade balance temporarily and in the long run leaves the real money supply and the balance of trade unchanged. The Elasticities Approach and the J-curve The elasticities approach focuses on the dynamics that generate the post- devaluation time-path of the trade balance. The response of trade flows to changes in exchange rates takes time because consumers are slow to change habits and, more importantly, because changes in production possibilities and supply require long-term 50 investment decisions. 26 According to the elasticities approach, devaluation lowers the foreign currency price of exports and raises the domestic price of imports. As a consequence, quantities adjust and the trade balance improves due to import substitution, assuming that the Marshall-Lerner condition holds. 27 However, the increase in the price of imports may offset the decrease in quantity and lower the balance of trade. In this context, the outcome depends on price elasticities of demand for domestic exports and imports. The theoretical justification of the J-curve phenomenon is as follows. The ?contract period? hypothesis states that at the time of devaluation many contracts are already signed and many goods are in transit. 28 Krueger (1983) argues that the completion of these transactions dominate the short-run trade balance behavior. During the contract period, the trade balance should deteriorate due to fixed quantities and higher domestic prices for imports. Also, consumers and producers do not adjust instantaneously to changes in relative prices generated by real devaluations. The ?pass-through? period refers to slow quantity adjustments by producers and consumers to any price changes. Magee (1973) explains that quantities do not change during the pass-through period because of two reasons. First, supply might be perfectly inelastic for some time because exporters cannot suddenly adjust their output and sales abroad. Second, domestic demand of foreign goods might also be perfectly inelastic because it takes time for importers to substitute goods and change the flow of orders. 26 See Krugman (1989). 27 According to the Marshall-Lerner condition, if the sum of imports and exports elasticities exceeds unity, a nominal devaluation has a positive effect on the trade balance. 28 See Magee (1973) for a detailed analysis on the contract and pass through periods. 51 Finally, the ?quantity adjustment? period calls for trade balance improvements in the long run when demand and supply for imports and exports become price elastic (Marshall-Lerner condition). Foreign importers have enough time to adjust their purchases to a lower foreign price of exports. Similarly, domestic importers adjust import quantities due to the increase in the domestic currency price of imports. Clearly, the quantity adjustment period calls for trade balance improvements due to an increase in the quantity of exports and decrease in the quantity of imports. Based on the absorption, monetary, and elasticities approaches, the trade balance model and the expected sign for the different coefficients can be expressed as follows: BOT t = ? 0 + ? 1 tY )(+ + ? 2 tY )(? * + ? 3 tM )(+ + ? 4 tM )(? * + ? 5 tE )(+ (2.6) where BOT t is the trade balance at time t, Y t and Y t * are home and foreign incomes measured by real GDP, M t and M t * are home and foreign real money supply, and E t is the real exchange rate at time t. The expected positive sign for E t reflects the long-run expectations of improved trade balances after real exchange rate depreciations. It is assumed that the real exchange rate variable uses the nominal exchange rate as the amount of local currency per unit of foreign currency. Using this definition, an increase in E t implies a depreciation of the local currency. A J-curve in Argentina? An important issue not often addressed in the literature is that the post- devaluation time-path of the trade balance depends on the currency used to measure imports and exports. This becomes important in the case of Argentina and Brazil because the trade balance is measured in US dollars. Magee argues that when the trade balance is 52 measured in foreign currencies, it is not affected during the contract period, but it could be negatively affected during the pass-through period. As a demonstration, consider the particular case of Argentina?s bilateral trade balance with Brazil. The trade balance equation is presented in equation (2.7): BOT = Px*X ? Pm*M (2.7) where Px is the price of exports, X is the quantity of exports, Pm is the price of imports, and M is the quantity of imports. During the currency contract period, the quantities of exports (X) and the quantities of imports (M) are fixed. When the price of exports and imports are measured in US dollars (foreign currency), devaluation does not affect prices. Assume that an importer from Argentina signed a contract to buy 100 television sets from Brazil at a price of $1 each in November 2001. Clearly, the value of imports before the peso devaluation is $100. Assume that delivery and payments take place in March 2002 when the dollar was worth over 2 pesos. At this point, the Argentine importer has to pay the agreed $100 to the Brazilian exporter. In dollars, the value of imports continues to be $100. It is only when measuring the price of imports in local currency (pesos) that the value of imports increases during the contract period. A similar example could be used to demonstrate that the value of exports will not change during the contract period. Therefore, in line with Magee (1973), no initial deterioration of Argentina?s trade balance with Brazil should be expected. Now, consider the pass-through period, where the quantities are still fixed due to the short-term inelastic supply and demand for tradeables. During this period, the price of imports in dollars do not change and the price of exports measured in dollars decrease only if Argentine exporters are willing to allow devaluation to affect their prices. If 53 Argentine exporters do not pass through exchange rate changes onto the prices of their products, then the balance of trade will be again unaffected during this period. If they do charge Brazilian importers of Argentine goods a price reflecting the peso devaluation, the value of exports will decrease and the trade balance will deteriorate. As pointed out by Magee, a successful pass-through implies a deterioration of the trade balance during this short period. Finally, during the quantity-adjustment period, we should expect an improvement in the trade balance. Specifically, the price of Argentine exports decreases and the quantity of exports increases. If the increase in quantities offsets the decrease in the price exports, then the value of exports increases. In equation (2.7), the dollar price of imports remains fixed and the quantity of imports decreases. Consequently, the value of imports decreases. Then, the overall result of the quantity-adjustment period depends on the elasticities of supply and demand for imports and exports or the Marshall-Lerner condition. 29 Specifically, higher elasticities lead to larger trade balance improvements in the presence of devaluation. IV. The Model Testing for the presence of a J-curve calls for examination of the contemporaneous and lagged effects of exchange rate movements on the trade balance. In general, a linear model could be used as follows: y t = ? + ? = p i 0 ? i X t-i + ?Z t + ? t, (2.8) 29 See Magee (1973) for details. 54 where Z t is a covariate or a simple regressor without lagged coefficients, and X t-i is a regressor with lagged coefficients. However, when the lag length (p) is long, multicollinearity becomes a problem. In the present case, although I do not find multicollinearity problems, a large number of lags (p) would use up degrees of freedom and the model would not generate consistent estimators. 30 Also, when dealing with trade data, one should impose constraints in the parameters ? i according to the notion that exchange rate effects on trade balance may peak after several quarters, then show diminishing effects, and finally disappear at a specific lag. Almon (1965) shows that a smooth pattern of lag weights could be approximated by a polynomial of low order. The Almon PDL is considered a finite distributed lag model in which a change on an independent variable has an effect on the dependent variable that is distributed over several periods. The Almon PDL model is proven to be successful in capturing the lagged effects of exchange rate movements on the trade balance (see Bahmani-Oskooee (1985) and (1989), and Buluswar, Thompson, Upadhaya (1996)). Almon introduces the PDL model and its application by trying to predict quarterly capital expenditures in manufacturing industries for current and past appropriations for the period 1953-1961. With the use of an Almon PDL, one solves eventual multicollinearity problems and more importantly, the problem of inconsistent estimates due to low degrees of freedom. ? i = ? 0 * + ? = d j 0 ? j * i j + ? t, for j = 0,?, d and i = 0,?, p where d ? p (2.9) Equation (2.9) indicates that the coefficients ? i s in (2.8) lay on a polynomial curve, which demonstrates that the effects of these lagged coefficients could first increase 30 This is particularly important given that only 42 observations are available. See section V for data description. 55 and then decrease to finally disappear at some specific lag. The degree of the polynomial is generally not known, but can be determined empirically with the use of t-statistics. The degree of the polynomial (j) and number of lags (i) are determined based on Akaike Information Criteria (AIC), which is a measure that accounts for the trade-off between minimizing the sum of squares error (SSE) and limiting the number of regressors in a regression. Model Adjustments Equation (2.6) is adjusted before it is applied to the bilateral trade balance between Argentina and Brazil. First, the monetary variables (M and M*) should be excluded when using real GDP because real income accounts for changes in income due to real money supply fluctuations (see for example, Bahmani-Oskooee and Artatrana (2004)). Second, M and M* should be excluded because Brazil and Argentina had fixed exchange rates during most of the period under study, which, in turn, restrained all possibilities for implementing an autonomous monetary policy. The first argument in favor of excluding monetary variables relates to the relationship between the monetary and the absorption approach. According to the monetary approach, a decrease in real money balances leads to hoarding while lower domestic spending improves the trade balance. Therefore, any changes in real money balances will affect the trade balance through absorption, phenomenon that is already captured by the relative income variable. The Mundell-Fleming (ISLMBP) model is used here to help explain the second argument. According to the ISLMBP model, in a small open economy (SOE) that takes the world interest rate and has a fixed exchange rate regime coupled with perfect capital 56 mobility, an increase in the money supply has no effect on income. This was the case of the convertibility plan established in Argentina and Brazil?s Real Plan. Under these plans, the only manner to increase the money supply was through an increase in foreign reserves. Figure 2.1 shows the ISLMBP model in a SOE with fixed exchange rates where i is the interest rate, Y is income, IS is the investment-savings curve (goods market equilibrium), LM and LM? are liquidity-money curves (money market equilibrium), and BP is the balance of payments. An increase in the money supply implies a rightward shift on the LM curve. This shift lowers home interest rates, which in turn leads to outflows of foreign funds (foreign bonds are a bargain). The government uses foreign reserves to maintain the exchange rate at the fixed level and this decrease in reserves shifts LM left to the original position, leaving income (Y) unchanged. In this model of fixed exchange rates, there is no effect of money supply on income and therefore no effect through the absorption approach on the trade balance. Figure 2.1: ISLMBP with Fixed Exchange Rates Y i LM LM? IS Y e Y? e i e i? e BP BP? 57 When a SOE with fixed exchange rates devalues its currency, domestic goods become more competitive and expenditures switch towards domestic goods. This has a positive and direct effect on aggregate expenditures shifting the IS curve rightward and increasing home interest rates that generate capital inflows. The central bank accommodates these capital inflows by accumulating foreign reserves. As a consequence, the LM curve shifts out increasing income. This is exactly the experience of Argentina and Brazil after both devaluations. These adjustments are similar to those of a SOE under flexible exchange rates and are displayed in Figure 2.2. Figure 2.2: ISLMBP with Flexible Exchange Rates In the ISLMBP model with flexible exchange rates, an increase in the money supply shifts LM rightward to LM?. This causes a decrease in interest rates that in turn leads to outflows of foreign funds. Capital outflows are offset by depreciation of home currency (decrease in demand for home currency) shifting the IS curve out due to increased consumption of domestic goods. This new equilibrium finds the level of interest rates unchanged and a higher income. An increase in money supply under a Y i LM LM? IS Y e Y? e Y?? e i e i? e BP BP? IS? 58 flexible exchange rate only affects income and changes in income affect the trade balance through the absorption approach. Once the monetary variables are excluded, the model uses relative incomes to test for the absorption approach. In order to preserve degrees of freedom, the ratio of Y/Y* (instead of Y and Y* separately) is included in equation (2.6). 31 A priori, the sign on the estimated parameter of Y/Y* is positive, indicating that income growth in Argentina relative to Brazil should increase the value of imports relative to exports. Furthermore, two more variables accounting for the trade diversion phenomenon and Argentina?s deep recession in 2002 are added to the model. In order to measure trade diversion, the ratio of imports from Brazil over imports from the Unites States and Europe (Mb/Mo) is used, where Mb is imports from Brazil and Mo is imports from outside Mercosur countries (the US and EU). It is expected, a priori, a positive relationship between (Mb/Mo) and BOT indicating trade diversion from non-member countries to Brazil. The switch from a fixed to a free-floating exchange rate led to higher interest and inflation rates that helped deepen Argentina?s recession. In order to account and correct for this structural break the model includes an interaction dummy (DY) that is defined as the change in relative incomes with respect to the same time period last year multiplied by a dummy that takes the value of 1 after the first quarter of 2002. The Almon PDL model used to test for the J-curve phenomenon in this chapter is as follows: BOT t = ? 0 + ? 1 Y t /Y t * + ? 2 DY + ? 3 Mb/Mo + ? = n i 1 b i E t-i + ? t (2.10) 31 Models using Y and Y* separately give similar results than models using Y/Y*. 59 V. Data Description and Stationarity Tests Consistent with Bahmani-Oskooee and Alse (1994), Gupta-Kapoor and Ramakrishnan (1999), Bahmani-Oskooee and Brooks (1999), Baharumshah (2001), Lal and Lowinger (2002), trade balance is defined as the ratio of Argentine imports over exports. The value of imports and exports are measured in US dollars ($). Real GDP for both countries (Y, Y*) are indices where 1994:3 is 100. The real exchange rate is calculated as E = eP/P*, where E is the real exchange rate, e is the nominal exchange rate (measured as reals per pesos), P is Argentina?s price level measured by the consumer price index (CPI), and P* is the price level in Brazil (CPI). 32 The trade diversion proxy is measured as the ratio of imports from Brazil over imports from the US and Europe. 33 Figure 2.3 plots the real exchange rate for the time period under study, Figure 2.4 shows relative incomes, Figure 2.5 introduces the trade balance, and Figure 2.6 shows the trade diversion proxy. All figures plot the series in levels. Figure 2.3: Real Exchange Rate (reals /pesos) PLOTREXCH REXCH 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1. 0 1. 1 1. 2 1. 3 1. 4 DATE 1 9 9 4 : 3 1 9 9 4 : 4 1 9 9 5 : 1 1 9 9 5 : 2 1 9 9 5 : 3 1 9 9 5 : 4 1 9 9 6 : 1 1 9 9 6 : 2 1 9 9 6 : 3 1 9 9 6 : 4 1 9 9 7 : 1 1 9 9 7 : 2 1 9 9 7 : 3 1 9 9 7 : 4 1 9 9 8 : 1 1 9 9 8 : 2 1 9 9 8 : 3 1 9 9 8 : 4 1 9 9 9 : 1 1 9 9 9 : 2 1 9 9 9 : 3 1 9 9 9 : 4 2 0 0 0 : 1 2 0 0 0 : 2 2 0 0 0 : 3 2 0 0 0 : 4 2 0 0 1 : 1 2 0 0 1 : 2 2 0 0 1 : 3 2 0 0 1 : 4 2 0 0 2 : 1 2 0 0 2 : 2 2 0 0 2 : 3 2 0 0 2 : 4 2 0 0 3 : 1 2 0 0 3 : 2 2 0 0 3 : 3 2 0 0 3 : 4 2 0 0 4 : 1 2 0 0 4 : 2 2 0 0 4 : 3 32 An increase in the real exchange rate variable is evidence of an appreciation of the peso. This is a consequence of measuring the nominal exchange rate as reals per pesos. 33 It is worth noting that Brazil, Europe, and the US account for almost 75% of Argentina?s trade during the period under study. 60 Figure 2.4: Relative Real GDPs (Y/Y*) PLOTratioq ratioq 0. 94 0. 95 0. 96 0. 97 0. 98 0. 99 1. 00 1. 01 1. 02 1. 03 DATE 1 9 9 4 : 3 1 9 9 4 : 4 1 9 9 5 : 1 1 9 9 5 : 2 1 9 9 5 : 3 1 9 9 5 : 4 1 9 9 6 : 1 1 9 9 6 : 2 1 9 9 6 : 3 1 9 9 6 : 4 1 9 9 7 : 1 1 9 9 7 : 2 1 9 9 7 : 3 1 9 9 7 : 4 1 9 9 8 : 1 1 9 9 8 : 2 1 9 9 8 : 3 1 9 9 8 : 4 1 9 9 9 : 1 1 9 9 9 : 2 1 9 9 9 : 3 1 9 9 9 : 4 2 0 0 0 : 1 2 0 0 0 : 2 2 0 0 0 : 3 2 0 0 0 : 4 2 0 0 1 : 1 2 0 0 1 : 2 2 0 0 1 : 3 2 0 0 1 : 4 2 0 0 2 : 1 2 0 0 2 : 2 2 0 0 2 : 3 2 0 0 2 : 4 2 0 0 3 : 1 2 0 0 3 : 2 2 0 0 3 : 3 2 0 0 3 : 4 2 0 0 4 : 1 2 0 0 4 : 2 2 0 0 4 : 3 Figure 2.5: Bilateral Trade Balance measured as M/X PLOTBOT BOT 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1. 0 1. 1 1. 2 1. 3 1. 4 1. 5 1. 6 DATE 1 9 9 4 : 3 1 9 9 4 : 4 1 9 9 5 : 1 1 9 9 5 : 2 1 9 9 5 : 3 1 9 9 5 : 4 1 9 9 6 : 1 1 9 9 6 : 2 1 9 9 6 : 3 1 9 9 6 : 4 1 9 9 7 : 1 1 9 9 7 : 2 1 9 9 7 : 3 1 9 9 7 : 4 1 9 9 8 : 1 1 9 9 8 : 2 1 9 9 8 : 3 1 9 9 8 : 4 1 9 9 9 : 1 1 9 9 9 : 2 1 9 9 9 : 3 1 9 9 9 : 4 2 0 0 0 : 1 2 0 0 0 : 2 2 0 0 0 : 3 2 0 0 0 : 4 2 0 0 1 : 1 2 0 0 1 : 2 2 0 0 1 : 3 2 0 0 1 : 4 2 0 0 2 : 1 2 0 0 2 : 2 2 0 0 2 : 3 2 0 0 2 : 4 2 0 0 3 : 1 2 0 0 3 : 2 2 0 0 3 : 3 2 0 0 3 : 4 2 0 0 4 : 1 2 0 0 4 : 2 2 0 0 4 : 3 Figure 2.6: Ratio of Imports from Brazil over Non-Mercosur Partners (Mb / Mo) PLOTratiom ratiom 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1. 0 1. 1 DATE 1 9 9 4 : 3 1 9 9 4 : 4 1 9 9 5 : 1 1 9 9 5 : 2 1 9 9 5 : 3 1 9 9 5 : 4 1 9 9 6 : 1 1 9 9 6 : 2 1 9 9 6 : 3 1 9 9 6 : 4 1 9 9 7 : 1 1 9 9 7 : 2 1 9 9 7 : 3 1 9 9 7 : 4 1 9 9 8 : 1 1 9 9 8 : 2 1 9 9 8 : 3 1 9 9 8 : 4 1 9 9 9 : 1 1 9 9 9 : 2 1 9 9 9 : 3 1 9 9 9 : 4 2 0 0 0 : 1 2 0 0 0 : 2 2 0 0 0 : 3 2 0 0 0 : 4 2 0 0 1 : 1 2 0 0 1 : 2 2 0 0 1 : 3 2 0 0 1 : 4 2 0 0 2 : 1 2 0 0 2 : 2 2 0 0 2 : 3 2 0 0 2 : 4 2 0 0 3 : 1 2 0 0 3 : 2 2 0 0 3 : 3 2 0 0 3 : 4 2 0 0 4 : 1 2 0 0 4 : 2 2 0 0 4 : 3 61 A series is said to be stationary in the mean if its mean does not depend upon time. Similarly, a series is stationary in variance if the variance does not depend upon time or level. Differencing or detrending makes a series stationary in the mean and variance stabilizing transformations make the variance of a series to become constant. A logarithmic transformation is a variance stabilizing transformation that helps overcome the problem of nonstationarity in the variance. Also, when using a logarithmic transformation one can interpret the coefficients as elasticities. Therefore, all variables are transformed to a logarithmic scale before performing stationarity tests. Following Nelson and Plosser (1982), augmented Dickey-Fuller (ADF) tests are performed in each series in order to test for stationarity. 34 ?y t = ? 0 + ? 2 t + ?y t-1 + ? = p i 2 ? i ? y t-i+1 + ? t (2.11) where y t is the variable of interest, t is a time trend, and ? is the first difference operator. In equation (2.11) the null hypothesis for the ADF test is ? = 0. Failure to reject the null hypothesis indicates that the series is not stationary. 35 Table 2.2 presents ADF results for the logarithmic series and for the differenced series. Based on the ADF tests none of the variables are stationary in logarithms. However, all series are difference stationary. 36 The lag length is determined using the AIC. 37 34 To test for the robustness of ADF results, Perron?s (1989) procedure was performed to find out whether the series are trend or difference stationary. None of the variables was found to be trend stationary at the 5% level. 35 Dickey and Fuller (1979) provide the t-statistics for the ADF test. For a sample size of 50 (the closest to our study), the t-statistics are 3.18, 3.50, and 4.15 at the 10%, 5%, and 1% confidence levels. 36 For Y/Y*, stationarity is achieved by first differencing relative annual growth rates in real GDP (first difference of the change in the logarithm of relative incomes with respect to same period last year). 37 AIC chooses ADF?s with 1 lag or none due to low degrees of freedom. 62 Table 2.1: Augmented Dickey-Fuller Tests Variable Logarithmic Level First Differences Number of Lags E 1.08 5.20*** 0 Y/Y* -0.53 3.51** 1 BOT 1.93 6.84*** 0 (Mb/Mo) 2.31 5.69*** 1 Note: Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. VI. Model Results The Almon PDL model in first differences is as follows: ?BOT t = ? 0 + ? 1 ?Y t /Y t * + ? 2 DY + ? 3 ?Mb/Mo + ? = n i 1 b i ?E t-i + ? t (2.12) Equation (2.12) estimates the trade balance in first differences as a function of the ratio of Argentina?s real GDP over Brazil?s real GDP (?(Y/Y*)) in seasonal first differences, an interaction dummy (DY) between relative incomes (?(Y/Y*)) and Argentina?s devaluation, the first difference of the ratio of Brazilian imports over US and EU imports (?(Mb/Mo)), and the real exchange rate in first differences (?E) and its lags. The optimal lag length of the real exchange rate and the degree of the polynomial are determined using AIC. AIC determined the lag length and the degree of polynomial to be 5 and 3, respectively. Table 2.3 presents the model results and Figure 2.7 (in Appendix II) shows the model?s residuals. As is indicated in Table 2.3, changes in relative incomes do not affect the trade balance. The coefficient for DY is positive, which suggests that after the modification of the exchange rate regime GDP growth in Argentina relative to Brazil?s income growth leads to higher imports relative to exports. Therefore, the absorption approach holds for the post-devaluation period. 63 Table 2.2: Almon PDL Results in First Differences Variable Coefficient t-value Constant 0.001 0.09 ?(Y/Y*) -4.344 1.53 DY 4.843 2.56** ?(Mb/Mo) 0.407 2.49** ?E t 0.449 4.38*** ?E t-1 -0.294 3.50** ?E t-2 -0.367 4.47*** ?E t-3 -0.161 1.97* ?E t-4 -0.065 0.81 ?E t-5 -0.470 5.03*** Adjusted R 2 =0.79 Degrees of Freedom = 28 Note: Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. Results in Table 2.3 indicate the presence of an ?inverse J-curve.? An inverse J- curve calls for initial improvements followed by deterioration of the trade balance. The positive and significant coefficient of a contemporaneous change in the real exchange rate E, followed by the negative and significant coefficients for lags 1, 2, 3, and 5 are indicative of an ?inverse J-curve.? Bahmani-Oskooee (1989) finds inverse J-curves for Greece, India, Korea, and Thailand from 1973 to 1980. Karadeloglou (1990) also finds an inverse J-curve for Greece for the 1974-1983 period. The temporary improvement on Argentina?s bilateral trade balance with Brazil could be attributed to initial exchange rate uncertainty, the banking restrictions imposed on December 3 rd 2001, and to the foreign exchange (capital controls) restrictions set by the Argentine government in 2002. Known as the ?corralito? (little fence), the banking restrictions allowed deposits to be transferred within the financial system, but it prohibited deposits to be converted into cash or to be transferred outside the financial system beyond a certain limit. The freezing of bank accounts and the capital controls 64 deterred import payments to foreign exporters. In December 2001 alone imports from Brazil decreased by almost 55%. Through the ?A 3827? Communiqu? (making reference to Resolution # 668/2002 of the Ministry of Economy) the Argentine Central Bank announced the end of banking restrictions in Argentina as of December 2 nd 2002. Argentina?s imports from Brazil started to pick up in 2003, once banking restrictions and extensive capital controls were lifted. 38 The positive and significant coefficient for ?Mb/Mo means that imports from Brazil can be also explained by the reduction in imports from the US and EU. 39 This suggests the presence of import diversion from non-Mercosur trading partners to Brazil. The later deterioration of the trade balance seems to emerge when the economy (and the exchange rate) started to stabilize and recover early in 2003. With a floating exchange rate managed by the government at 3 pesos per dollar, imports from the US and EU became too expensive. 40 VII. Conclusions This chapter studies the short-term dynamic adjustments in the bilateral trade balance between Argentina and Brazil after the peso devaluation in January 2002. An ?inverse J-curve? emerges from estimates indicating that the peso devaluation was not effective in improving Argentina?s trade balance with Brazil. Trade diversion contributes significantly to Argentina?s unexpected trade deficit with Brazil. This supports the theoretical belief that the reduction in Argentina?s purchasing power due to the peso 38 See Levi Yeyati et al. for a discussion on the ?corralito? and its effect on the Argentine stock market. 39 See Appendix IV for theoretical simultaneity concerns. 40 Chapter 3 deals with the issue of trade diversion by through a Linder effect. 65 devaluation has triggered a process of trade diversion from the more expensive non- member products to the more affordable Brazilian products. Results suggest that trade adjustments to exchange rates are distributed over time. The estimates suggest exchange rates affect trade at a lag length of up to five quarters. Results also support the idea that regionalization should be taken into account when examining the impact of devaluation on bilateral trade balances. Devaluations affect bilateral trade flows in a manner that favors regional trading partners at the expense of non-member countries. Perhaps, economic forces linked to regionalization such as the uneven flow of FDI targeting Mercosur?s largest economy and the migration of firms from Argentina to Brazil after devaluation of the real explain the ?inverse J-curve.? The substitution of imports from the US and EU for Brazilian imports may be an income effect as proposed by Linder. Chapter 3 examines these issues in detail. The results presented here have policy implications. For countries within a regional trade agreement, a devaluation intended to raise the trade balance may have contrasting effects depending on timing. When analyzing the case of Argentina and Brazil, it becomes clear that the country that devalued first (Brazil) has somehow managed to improve its trade balance with its partner (Argentina) even after the latter devalued three years later. The results in this chapter could mean that the member country that devalues first gains a first-mover advantage in exporting. 66 APPENDIX I J-Curve Literature Review Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Type Magee (1973) X/M Exchange rate, domestic real income, foreign real income US Monthly (1969?1973) Develops the ideas of currency contract, pass-through, and quantity adjustments. There may or may not be a J-Curve in the short-run, but the long-run impact of devaluation on trade-balance is favorable. Rest of the world Junz and Rhomberg (1973) Time series OLS (i) Market shares in manufacturing exports (ii) Manufacturing exports Price variable: relative price of exports Austria, Belgium, Luxemburg, Canada, Denmark, France, Germany, Italy, Japan, Netherlands, Norway, Sweden, Switzerland, UK, US Annual (1953?1969) They find lags of up to five years in the effects of exchange rate changes on market shares of countries in world trade due to: lags in the recognition of the devaluation, in the decision to change real variables, in delivery time, in the replacement of inventories and materials, and in production. Rest of the world Himarios (1985) Time series OLS The first specification involves: [(X-M)/GNP]. The second equation involves trade balance measured in foreign currency (B t ) ?g i , ?g R , ?M i , ?M R , ?G i , ?G R , and ?ER i , where gi and gr are income growth rates, M i and M R are money supplies (M1), G i and G R are the ratio of government expenditures to output, and ER i is the country's exchange rate. The second equation involves real exchange rates Costa Rica, Ecuador, Finland, France, Iceland, Israel, Philippines, Spain, Lanka, UK Annual (1956-1972) Devaluation improves the trade balance in nine out of ten cases. Rest of the world Bahmani- Oskooee (1985, 1989a) Time series Almon lag structure imposed on real exchange rate Index of (X-M) . Base year = 1975 GNP (Y t ), world income (YW t ), domestic high powered money (M t ), world high power money (MW t ), effective exchange rate deflated by wholesale prices. All variables are expressed in index forms with base year 1975 Greece, India, Korea, and Thailand Quarterly (1973?1980) Finds evidence of an inverse J-curve for Greece, India, and Korea. The long-run impact is favorable only in the case of Thailand. Rest of the world 67 Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Type Felmingham (1988) Unrestricted distributed lag model M/X Terms of trade (px/pm), domestic income (y), and proxy for world income (y f ) Australia Quarterly (1965:1? 1985:2) No strong evidence of J-curve during the fixed exchange rate era (1965-1974). No resemblance of an Australian J-curve during the era of managed or free-floating exchange rates (1974-1983). Rest of the world Himarios (1989) Time series OLS Real trade balance Domestic and foreign real income (Y, Y*), domestic and foreign real government expenditures (G, G*), domestic and foreign real money balances (M, M*), interest rate (i), and a proxy for expectations (anticipated devaluation) 27 countries and 60 devaluation episodes (1953-73 and 1975-84) Indicates that devaluations have been a successful tool in inducing trade balance adjustments. Specifically, he finds that nominal devaluations resulted in significant real devaluations that last for at least three years, and this significant real devaluation increased exports relative to imports for the same time period. Rest of the world Brissimis and Leventankis (1989) Time series Almon lag structure. IV method Petroleum and non- petroleum exports and imports Exchange rate, export and imports weighted effective exchange rate, balance on invisibles and capital flows in drachmas, etc. Greece Quarterly (1975-1984) Evidence of a J-Curve for Greece. The initial deterioration lasts one quarter. Rest of the world Karadeloglou (1990) Time Series. Simulations of the Macro- econometric (MYKL) of the Greek Economy Consumption, private investment, imports, exports, inventory changes, prices, wages, etc. Demographic variables, government expenditures, foreign demand, foreign export prices, monetary variables, exchange rate Greece 1974-1983 Evidence of an inverse J-curve. Rest of the world Bahmani- Oskooee and Pourheydarian (1991) Time series Almon lag structure on the real exchange rate (X-M) t in real terms. GNP (Y t ), world income (YW t ), domestic high powered money (M t ), world high-power money, real effective exchange rate E*PW/P, where P is domestic price level, PW is the world price level, and E is the effective exchange rate Australia Quarterly (1977?1988) Depreciation leads to trade balance improvements. Evidence of a delayed J-curve for Australia. Measuring the trade balance as the ratio of X/M does not affect the results of the model. Rest of the world 68 Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Type Bahmani- Oskooee and Malixi (1992) Time series Almon lag structure (X/M)t in real terms (X-M) t in real terms. GNP (Y t ), world income (YW t ), domestic high powered money (M t ), world high-power money, real effective exchange rate E*PW/P, where P is domestic price level, PW is the world price level, and E is the effective exchange rate Brazil, Dominican Republic, Egypt, Greece, India, Korea, Mexico, Pakistan, Peru, Philippines, Portugal, Thailand, Turkey Quarterly (1973 Q1 ?1985 Q4) They find support for the J-curve for Brazil, Greece, Korea, and India. They also report, in line with Magee (1973), shapes such as the N-, the M-, and the I-Curves, concluding that the short-run effects may not follow a standard pattern, though the long-run effects are favorable in most cases. Rest of the world Bahmani- Oskooee and Alse (1994) Time series. Engle-Granger cointegration technique M/X Real effective exchange rate 19 developed and 22 less developed countries Quarterly (1971-1990) Finds J-curves for Costa Rica, Ireland, Netherlands, and Turkey. Concludes that levels used by Himarios and Bahmani-Oskooee (1985) were not stationary. Use ratio of imports to exports for trade balance. Rest of the World Buluswar, Thompson, and Upadhyaya (1996) Time series Almon lag structure on the real exchange rate X - M Y is an index of industrial production in India, Y* is a proxy for rest of the world income, M is India's M1, M* is a proxy for rest of the world M1 (major trading partners), and E is the real exchange rate India Quarterly (1960-1990) The paper compares absorption, elasticity, and monetary models and concludes that the monetary model performs better in India. They find no evidence of a J-curve and they conclude that devaluations have had no significant long-run effect on the BOT. Rest of the World Upadhyaya and Dhakal (1997) Distributed lag model with lags on the dependent variable and the real exchange rate X - M Real exchange rate Colombia, Cyprus, Greece, Guatemala, Mexico, Morocco, Singapore, Thailand Annual (1967- 1992) for Colombia, Cyprus, Guatemala, and Mexico, (1964- 1992) for Greece and Morocco, (1962-1992) for Singapore, and (1962- 1992) for Thailand Their findings show that devaluation, in general, does not improve the trade balance in the long run. Specifically, only in Mexico does devaluation improve the trade balance in the long run, while in Cyprus, Greece, and Morocco it does not. In Colombia, Guatemala, Singapore, and Thailand, devaluation is neutral in the long run. Rest of the World 69 Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Type Gupta? Kapoor and Ramakrishnan (1999) Error Correction Model (ECM) Ln (M/X) Y, Y* are nominal domestic and foreign incomes respectively, and the nominal effective exchange rate Japan Quarterly (1975 Q1? 1996 Q4) Finds evidence in favor of J-Curve. Rest of the world Akbostanci (2004) Cointegration and error correction modeling. Impulse response function X-M Y, Y* are nominal domestic and foreign incomes respectively, and the real exchange rate (q) Turkey Quarterly (1987:1- 2000:4) In the long run, a real depreciation of the Turkish lira improves the country?s trade balance. In the long run, domestic a foreign income have no effects on the trade balance. Rest of the world Rose and Yellen (1989) Time series: cointegration approach X-M in domestic currency Y USt , Y Jt is the US and country j's real GDP. REX jt is the bilateral real exchange rate between $ and j's currency US with Germany, Italy, Canada, France, Japan, UK Quarterly (1963-1988) Finds no indication of J-curve and no significant effect of the exchange rate on the trade balance. Bilateral Marwah and Klein (1996) Almon distributed lag X/M The ratio of world trade to the country?s GNP, and bilateral real exchange rate US and Canada with France, Germany, Japan, and UK Quarterly (1977:1 ? 1992:1) Evidence of J-curve effects for both the Us and Canada. The shapes of the J-curves are quite similar in both countries, but they seem to peak sooner and become more negative at the beginning of the adjustment process. Bilateral Sukar and Zoubi (1996) Almon distributed lag. GLS estimates a reported X in real terms Foreign real income and real exchange rate US with Canada and Japan Quarterly (1975:3- 1993:4) Real income and real exchange rates are important determinants of bilateral trade flows. Bilateral Tongzon and Felmingham (1998) Cointegration X/M Domestic and foreign real income (Y, Y*), domestic and foreign real cash balances (h, h*), and the real exchange rate (q) US, Japan, Singapore, and Australia Quarterly (1977:3- 1994:3) Real exchange rates have limited or no effect on bilateral trade balances in Australia, Japan, and US. Singapore?s trade with US and Japan is the exception. Real cash balances and real income affect bilateral trade flows in the short-run. Bilateral Bahmani- Oskooee and Brooks (1999) Time series ARDL approach: cointegration and error correction modeling M/X Y USt , Y jt is the US and country j's real GDP. REX jt is the bilateral real exchange rate between $ and j's currency. US with six major trading partners: Canada, France, Germany, Italy, Japan, UK Quarterly (1973 Q1 ?1996 Q2) They improve Rose and Yellen by using M/X instead of X-M. They find no evidence of a J-curve. In the long-run, a $ depreciation leads to trade balance improvements. Bilateral 70 Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Type Wilson and Tat (2001) Time series: ARDL approach X ? M Real bilateral exchange rate (q), and real domestic and foreign income (Y, Y*) measured with the manufacturing production index and industrial production index respectively Singapore with the US Quarterly (1970-1996) They find that the real exchange rate does not have a significant impact on bilateral trade between these two countries. They also find no evidence of a J-curve effect. Bilateral Wilson (2001) VAR specification (X-M)/CPI Domestic income, foreign income and real bilateral exchange rate Singapore, Malaysia, and Korea. Trading partners for these countries are the US and Japan Quarterly (1970?1996) Only in the case of Korea they find evidence of a J-curve. Bilateral Baharumshah (2001) Time series. Unrestricted VAR model Ln(X/M) Domestic income, foreign income and the real effective (rather than bilateral) exchange rate Malaysia and Thailand. Trading partners for these 2 countries are the US and Japan. Quarterly (1980?1996) No evidence of the J-curve. Bilateral Bahmani- Oskooee and Kanitpong (2001) Time series ARDL approach: cointegration and error correction modeling Ln(X/M) Domestic income, foreign income and the real bilateral exchange rate. Improves Baharumshah (2001) by using the real bilateral exchange rate Thailand with 5 partners: Germany, Japan, Singapore, UK, US Quarterly (1984?1997) Supports J-curve between Thailand and the US and Thailand with Japan. Bilateral Chen (2001) Almon PDL structure imposed on the real exchange rate i) Ln(REX) where REX is real exports ii) Ln(RTB) where RTB i real trade balance i) Foreign income in logs (ln RY*), lo of real imports (ln RIM), and log of real exchange rate (ln RER). ii) log of foreign income (ln RY*), and log of real exchange rate (ln RER) Taiwan with the US and Japan Quarterly (1981:1 -1998:1) Real income affects real exports in both cases. Real exchange rates and real imports do not affect Taiwan?s exports to the US, but they do affect Taiwan?s exports to Japan. Real exchange rate has significant effects on trade balance with US and Japan, but income does not. Bilateral Lal and Lowinger (2002) Time series Johansen?s cointegration and error- correction modeling and impulse response function. Ln(M/X) Domestic income, world income and real effective exchange rate Indonesia, Japan, Korea, Malaysia, Philippines, Singapore, Thailand Quarterly (1980?1998) They confirm the J-curve and show that there are significant differences in the duration and extend of the J-curve effect across countries. Bilateral 71 Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Type Bahmani- Oskooee and Goswami (2003) Time series. ARDL approach to cointegration and error correcting modeling Ln(X/M) Domestic income, trading partner?s income and real bilateral exchange rate Japan with Australia, Canada, France, Germany, Italy, Netherlands, Switzerland, UK, US Quarterly (1973?1998) Evidence supports the J-curve only with Germany and Italy. They conclude that the long-run effects of currency depreciation are to improve the trade balance. Bilateral Arora, Bahmani- Oskooee and Goswami (2003) Time series. ARDL approach to cointegration and error correcting modeling Ln(X/M) Domestic income, trading partner?s income and real bilateral exchange rate India with: Australia, France, Germany, Italy, Japan, UK, US Quarterly (1977?1998) A new concept of the J-curve appears with Australia, Germany, Italy, and Japan. Bilateral Hacker and Abdulnasser Hatemi-J. (2003) Time series Impulse response function Ln(X/M) Domestic income, trading partner?s income and real bilateral exchange rate Belgium, Denmark, Netherlands, Norway, Sweden Quarterly and Monthly (1977?2000) Supportive of the J-Curve. Bilateral Bahmani- Oskooee and Ratha (2004a) Time series. ARDL approach to cointegration and error correcting modeling Ln(X/M) Domestic income, trading partner?s income and real bilateral exchange rate US with 18 trading partners: Australia, Austria, Belgium, Canada, Denmark, Finland, Germany, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Spain, Sweden, Switzerland, UK Quarterly (1975 Q1 ? 2000 Q4) Results support a new definition of the J-curve in 11 out of 18 cases. Bilateral Bahmani- Oskooee and Ratha (2004b) Time series. ARDL approach to cointegration and error correcting modeling Ln(X/M) Domestic income, trading partner?s income and real bilateral exchange rate USA with 13 developing countries: Argentina, Chile, Ecuador, India, Indonesia, Israel, Korea, Malaysia, M?xico, Nigeria, Pakistan, Singapore, South Africa Quarterly (1975 Q1 ? 2000 Q2) Results support a new definition of the J-curve in 7 out of 13 cases. Bilateral 72 APPENDIX II Figure 2.7: Almon PDL Residual Plot res -0.14 -0.13 -0.12 -0.11 -0.10 -0.09 -0.08 -0.07 -0.06 -0.05 -0.04 -0.03 -0.02 -0.01 0. 00 0. 01 0. 02 0. 03 0. 04 0. 05 0. 06 0. 07 0. 08 0. 09 0. 10 0. 11 0. 12 0. 13 0. 14 0. 15 0. 16 0. 17 0. 18 0. 19 0. 20 0. 21 0. 22 time 0 1020304050 73 APPENDIX III Investigation of Theoretical Simultaneity Issues: Theoretical simultaneity concerns may arise because imports from Brazil (Mb) enter the model as the numerator of the dependent variable (Mb/Xb) as well as in the numerator of an independent variable (Mb/Mo). An instrumental variables regression is used to estimate imports from Brazil (Mb) as follows: Mb = ? 0 + ? 1 Y t + ? 2 e t + ? 3 M t-1 + ? 4 P + ? t (2.13) In equation 2.13, Y refers to Argentina?s real GDP, e is the nominal exchange rate (reals/pesos), M is Argentina?s monetary base, and P is the price level in Argentina. Estimates for equation 2.13 follow: Mb = 1190.40 + 42.45 Y ? 121.18 e t + 0.02 M t-1 ? 0.0001 P (0.80) (11.64) (1.55) (3.82) (3.20) The predicted values for Brazilian imports from this instrumental variables regression ( ^ Mb ) are used to create an instrument for the trade diversion proxy as ( ^ Mb / Mo). After taking a logarithmic transformation, the first difference of the series is stationary and used in the Almon PDL model (equation 2.12) which is re-estimated as follows: ?BOT t = ? 0 + ? 1 ?Y t /Y t * + ? 2 DY + ? 3 ? ^ Mb / Mo + ? = n i 1 b i ?E t-i (2.14) As shown in Table 2.4, results are similar to the originals presented in Table 2.3. The re-estimated trade diversion proxy shows a positive and significant coefficient, although it is smaller in magnitude. This suggests that results from the original Almon PDL model are subject to simultaneity problems. One can conclude that part of Argentina?s bilateral trade deficit with Brazil is due to a decrease in imports from US and 74 Europe. In other words, import flows are being diverted from non-member countries to Brazil. Table 2.3: Almon PDL Results with Instrument Variable Coefficient t-value Constant 0.009 0.56 ?(Y/Y*) -3.383 1.18 DY 4.807 2.46** ?( ^ Mb / Mo) 0.123 1.97* ?E t 0.443 4.14*** ?E t-1 -0.288 3.28*** ?E t-2 -0.362 4.24*** ?E t-3 -0.160 1.88* ?E t-4 -0.063 0.75 ?E t-5 -0.451 4.56*** Adjusted R 2 =0.77 Degrees of Freedom = 28 Note: Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. 75 CHAPTER 3: TRADE DIVERSION IN THE CONTEXT OF GRAVITY MODELS: A TEST OF THE LINDER HYPOTHESIS I. Introduction Starting with Ricardo?s trade theory of comparative advantage, trade economists have developed frameworks to explain trade flows. The factor abundance theory (HOS theory) developed by Heckscher (1919), Ohlin (1933), and refined by Samuelson in the 1950s, is based on the principle of comparative advantage and suggests that trade is the result of differences in relative factor endowments between nations. Countries relatively well endowed with labor tend to export goods that use labor intensively. This factor proportion model suggests that trade patterns are mainly a supply-side phenomenon. The HOS model was challenged by Leontief (1953) in a paper that studies trade patterns for the United States in 1947. Leontief found that US exports were on average labor intensive and that US imports were capital intensive in their factor contents. Since the US was believed to be a capital abundant country, this finding seemed to contradict factor proportion theory and became known as Leontief paradox. The evidence regarding the HOS model is mixed and a set of alternative theories has been developed. 41 Linder (1961) proposes a demand-side approach that argues that HOS theory only applies to trade in primary products. Linder suggests that countries produce manufactures for their own consumption and that any excess supply is exported. He 41 See Bharadwaj (1962), Bowen, Leamer, and Skeivaskas (1995), and Deardorff (1984). 76 argues that countries interested in buying this excess supply must have similar demand patterns. This principle of overlapping demands and production capacity explains why most world trade takes place between countries with similar endowments. The Linder hypothesis suggests that per capita income is the most important determinant of a country?s demand structure and argues that similarity in per capita incomes increases the amount of trade between countries. It is worth noting that the Linder theory applies only to trade in manufactures and does not question the validity of the HOS theory for trade in resource based products. During the 1980s and 1990s, trade theorists developed a new approach to explain world trade patterns. Krugman (1990) refers to a ?new trade theory? that supplements other theories such as HOS or Linder. This new trade theory is based on increasing returns to scale, product differentiation, and imperfect competition. New trade theory played a major role in the development of a theoretical framework for the gravity model of trade, the most widely used model in the research examining the determinants of trade flows. 42 Gravity models have successfully explained bilateral trade flows since the 1960s. Eichengreen and Irwin (1995) refer to the gravity model as the ?workhorse? for empirical investigation due to its success explaining trade flows. The gravity model of trade proposes that trade depends upon economic size and geographic or economic distance between countries. The dependent variable becomes the sum of exports and imports, usually measured in US dollars. The gravity model was developed by Tinbergen (1962) 42 See Section III for a discussion on the theoretical framework of the gravity model of trade. 77 and Linnemann (1966). 43 Tinbergen (1962) proposes that the amount of trade between two countries is directly related to their economic sizes (measured by GNPs or GDPs) and inversely related to their distance. Linnemann (1966) adds countries? populations as a measure of size. The GDP of the exporting country measures productive capacity, while that of the importing country measures absorptive capacity. The geographic distance between the two countries has been commonly used as a proxy for transportation costs. Per capita income has been also widely used because it accounts for both measures of country size GDP and populations. Besides testing for size and distance, the gravity model has been used to examine the presence of Linder effects. 44 Gravity models became a tool for testing the effects of regional economic integration, common language, monetary unions, exchange rate variability, and adjacency on trade flows as well as trade creation and trade diversion effects. This chapter provides an analysis of the trade patterns between member and non- members countries by using a gravity type model. Section II presents a literature review that shows the development of gravity models since the 1960s. Theoretical underpinnings of the model are presented in section III. Section IV presents the Linder hypothesis and a review of the papers investigating its empirical success. A discussion of different econometric techniques is introduced in Section V. Section VI inspects the trade diversion effects of Argentina?s devaluation and Section VII tests for the presence of Linder effects that could explain the diversion from non-Mercosur countries to Brazil. Section VIII presents the chapter?s conclusions. 43 See literature review section. 44 See Kennedy and McHugh (1983), Hoftyzer (1984), Hanink (1988 and 1990), Greytak and Tuchinda (1990), and McPherson, Redfearn, and Tieslaw (2000 and 2001). 78 II. Literature Review on Gravity Models Tinbergen (1962) introduces the gravity equation to investigate standard patterns of trade that would prevail in the absence of trade restrictions. Any difference between actual and theoretical trade flows (actual vs. predicted values) is used as evidence of a preferential or discriminatory treatment of a country?s exports in world markets. The main factors determining trade flows between two countries are GNPs and the geographic distance between them. The amount of exports a country is able to supply depends on its economic size, the size of the importing country, and on transportation costs. GNPs explain economic size and distance is a proxy for transportation costs. In its simplest form, the gravity model introduced by Tinbergen is: ln E ij = ? 0 + ? 1 ln Y i + ? 2 ln Y j + ? 3 ln D ij + ? (3.1) where E ij is the value of exports from country i to country j, Y i is country i's GNP, Y j is country j?s GNP, and D ij is the distance between countries i and j. The model is cross- section with 18 countries in 1958. Dummy variables accounting for neighboring countries and regional integration are also part of the model. Results indicate that economic size and distance are the main factors explaining trade flows and that deviations from theoretical trade flows are considerable. A positive coefficient for the dummy representing the British Commonwealth preference implies higher trade among member countries. Tinbergen repeats the previous exercise by using export data on 42 countries (70% of world trade) in 1959. He also estimates a model by adding a proxy for export commodity concentration. The correlation coefficient (R 2 ) for all regressions is 0.81. Significant deviations between actual and expected trade flows suggest the presence of 79 discriminatory trade barriers. Tinbergen presents the reasons for those deviations and suggests that further research is needed. Among the reasons for positive deviations, the author cites preferential treatment of a country?s exports, utilization of previously accumulated foreign exchange in the case of imports, or a net inflow of capital. Negative deviations are based on discriminatory treatment of exports, import restrictions, or net outflows of capital. Linnemann (1966) explores the gravity model in detail by analyzing the factors that explain trade flows: i) factors indicating total potential supply of country A (exporting country); ii) factors affecting total potential demand of country B (importing country); and iii) factors representing the ?resistance? to trade flows from A to B. Linnemann?s trade flow equation follows: X ij = ? 0 542 631 ??? ??? ijji ijji DNN PYY (3.2) where X ij are trade flows from country i to country j, Y i and Y j are GNPs and have an expected positive effect, N i and N j are populations that are expected to have a negative coefficient, P ij is a preferential trade factor (British, French, and Portuguese colonies), and D ij is the distance between countries. GNPs and populations are factors affecting potential supply and demand, while P ij and D ij account for resistance to trade. Linnemann investigates the link between his proposed equation and economic theory by using a model similar to Walras? model of general equilibrium prices. 45 Linnemann estimates world trade flows by applying (3.2) in a log-linear form to 80 45 See Nicholson (1998) for a presentation of Walras? model. A more detailed presentation of Linnemann?s model is introduced in the theoretical section of this chapter. 80 countries. The model is applied to different sets of data and is adjusted by adding a variable that measures the commodity composition of trade. Findings suggest a positive relationship between GNP and trade flows, a negative relationship between trade and population, a negative relationship between natural trade barriers (distance) and trade flows, and a substantial effect of preferential trade arrangements. By isolating and quantifying these effects, Linnemann (1966) improves Tinbergen (1962). Linnemann acknowledges the existence of ?possible econometric shortcomings? and suggests further research in this area. Aitken (1973) tries to separate the major forces that shaped trade flows in Europe during the 1951-1967 period. Aitken follows Tinbergen and Linneman by using a cross- section model with dummy variables examining the impact of the European Economic Community (EEC) and the European Free Trade Association (EFTA) on trade flows. Yearly regressions are performed for the pre- and post-integration periods to examine the forces that were in place before the formation of the EEC. Projection estimates are generated based on a base year equation to investigate trade creation and trade diversion effects. 46 Results are consistent with customs union theory. EEC has generated gross trade creation effects that are greater than those generated by EFTA ($9.2 billion and $1.3 billion respectively). Findings suggest that EEC had a net external trade creation effect on EFTA through 1964 that was offset by a growing net trade diversion effect from 1965 46 Aitken (1973) defines gross trade creation (GTC) as the total increase in trade among members of a trading community due to integration, regardless of whether the additional trade replaces domestic production or whether it replaces non-members exports. Trade diversion (TD) is the substitution of imports from non-member countries (lower costs imports) for imports from member countries (higher cost imports). Finally, external trade creation (ETC) refers to integration-caused increases in trade between a trading community and countries outside the agreement. ETC minus TD yields the net effect of a trading bloc on the outside world. 81 to 1967. Aitken also suggests that 1958 is the last year for which it is safe to assume that European trade flows were not affected by the EEC. Thoumi (1989) uses a gravity model to analyze intra-Latin American and Caribbean trade in 1971, 1975, and 1979. The paper uses GDP of the exporting country to account for productive capacity, the GDP of the importing country to capture its absorptive capacity, physical distance and country adjacency (border) as proxies for transportation costs, income per capita, bilateral exchange rates, and dummies capturing economic integration effects. The author applies the gravity equation to aggregate trade data as well as to three product categories: total goods traded except fuels, manufactures, and natural resource based products. Thoumi finds that exporters? GNP and distance are the most influential factors affecting trade patterns. Results also suggest that there is a tendency for richer countries to import more natural resource based products than manufactures from poor countries. In general, the author suggests that integration systems among countries that are not too distant, have similar sizes and development levels, and follow similar policies are more likely to succeed than other integration agreements. Frankel, Stein, and Wei (1993) investigate the effects of trading blocs on trade flows. They estimate a gravity model using cross-sectional data including a large number of developing and industrial countries. The paper presents estimates every five years starting in 1965. The authors find that the EEC became a significant trade-creating force in the 1980s, peaking in 1985 and declining thereafter. Frankel, Stein, and Wei find that if two countries are members of the EEC, trade becomes 70% higher than it would have been otherwise (1990 estimates). They also find no trade creating effects for EFTA. 82 Bayoumi and Eichengreen (1995) examine the effects of preferential trade agreements in Europe since the 1950s. The paper?s goal is to find whether regionalism creates trade diversion by using the EEC and EFTA as case studies. The paper estimates the gravity equation in differences rather than in levels to correct for the heterogeneity across countries. 47 The authors argue that the problem of omitting third-country effects is solved by including the real exchange rate between European countries and the US. The dependent variable of the gravity equation is bilateral trade between 21 developed countries. Real incomes, populations, distance, and the real exchange rate between European countries and the US are the model?s independent variables. The sample data is divided into three overlapping periods: formation of the EEC and EFTA (1956-73), the entry of the United Kingdom, Ireland, and Denmark in the EEC (1965-80), and EEC expansion to include Greece, Portugal, and Spain (1975-92). Five dummy variables measure trade within the EEC, trade within EFTA, trade between EEC and EFTA, trade between EEC and other industrial countries, and trade between EFTA and other industrial countries. Results resemble those of Aitken (1973) and suggest that the formation of EEC and EFTA had a significant effect on European trade flows that cannot be attributed to economic factors or even unobservable characteristics. Bayoumi and Eichengreen find that EFTA was trade-creating, while EEC generated trade creation and trade diversion. Eichengreen and Irwin (1995) add a historical perspective to the gravity model and examine the effect of regional trade agreements on trade. They suggest that the standard gravity model neglects the effect of historic ties on trade patterns and therefore suffers from an omitted variable problem. The paper presents evidence that demonstrate 47 See section on econometric issues. 83 that the coefficients on traditional variables (incomes, populations, and distance) are distorted when a lagged dependent variable is added to the equation. The authors make this point by analyzing the evolution of trade between 1949 and 1965. They find a significant effect on lagged trade variables. While the paper suggests interpreting these lagged coefficients with caution, the results are robust to instrumental variables replacing lagged trade values. Specifically, Eichengreen and Irwin find that in the absence of lagged trade variables, the trade-creating effects of the European Payments Union (EPU) as well as the importance of the Dillon Round in the early 1960s are exaggerated. 48 They conclude that one should always include lagged variables in the gravity equation. Frankel and Wei (1997) estimate an augmented gravity model using data for 63 countries for four years between 1970 and 1992. The dependent variable is the value of exports from country i to country j rather than the value of exports plus imports. Besides the standard explanatory variables included in gravity models, the authors use distance between trading partners and dummies for contiguous borders, common language, and regional groupings. Results from this augmented gravity model show that affinity variables such as common language or adjacency are significant and that intraregional trade biases exist. Frankel and Wei show that Western European countries are estimated to have traded 17% more than when these estimates are obtained with a standard gravity model. Similarly, Western Hemisphere and ASEAN countries are estimated to have traded 40% and 145% more than what a model without dummies would have estimated. 48 Following the establishment of the European Economic Community in 1957, large-scale negotiations were held between September 1960 and May 1961 under Article XXIV:6 of the General Agreement. These negotiations were supplemented by a round of tariff negotiations, proposed by Douglas Dillon, Under Secretary of State of the United States. The Dillon Round yielded modest results: only 4,400 tariff concessions were exchanged, and agriculture and certain sensitive products were not covered. 84 Among other findings, results also suggest that increased trade in ASEAN and EEC did not occur at the expense of third countries. Finally, Frankel and Wei examine the extent to which currency blocs and currency stability follow regional trading blocs and trade flows between countries. Their findings suggest evidence of a currency bloc in Europe that follows the mark and a dollar bloc in the Pacific. The authors also find evidence suggesting that exchange rate volatility hinders trade. Frankel, Stein, and Wei (1997) estimate a gravity model that resembles the one in Frankel and Wei (1997) but adds per capita income levels as an explanatory variable. Western European countries traded 36% more that what the standard gravity model would have predicted between 1970 and 1992. The authors also examined the extent to which intraregional trade was higher due to higher openness than average and they study trends in intraregional trade over time. Results show that trade increased over time as a consequence of trade-creating and trade-diverting effects. The coefficient for the per capita income variable is positive, suggesting that richer countries trade more. Frankel, Stein, and Wei add transportation costs and imperfect competition to their model. They claim that regional preferential agreements are welfare-improving, but conclude that the extent of preferences among regional partners has probably exceeded optimal levels. Frankel (1997) provides a comprehensive investigation of the gravity model. The dependent variable is the logarithm of the total value of merchandise traded (exports plus imports) between two countries. Frankel estimates the gravity model with 65 countries every five years from 1965 to 1985 and then in 1987, 1990, 1992, and 1994. The model?s independent variables are GNPs, per capita incomes, distance, and dummies accounting for adjacency between a pair of countries, common language, and preferential 85 trade agreements. Results show that trade increases with a country?s GNP but less than proportionally. According to Frankel, this suggests that smaller countries tend to be more open to trade than larger ones. The coefficients on per capita income are highly significant and indicate that richer countries trade more than poor ones. The coefficients on the distance variable are sensitive to the inclusion of the common border dummy. When the common border dummy appears in the equation, the coefficient for distance ranges from -0.5 to -0.7. In other words, increasing the distance by 1% reduces trade by 0.6%. Results also suggest that two countries sharing a common border trade 82% more than two similar countries not sharing borders. Results in Frankel (1997) also show that two countries sharing linguistic or colonial links trade 55% more than they would otherwise. When examining the effects of trading blocs, Frankel finds that members of the ASEAN and the Australian-New Zealand Closer Economic Relationship (CER) have increased trade by almost fivefold. The Andean Pact and Mercosur have increased trade by more than two times. He also claims that the EEC has increased intra-trade by 65% after 1985. Frankel also shows that there are no factor endowment effects and finds significant historical-political effects and bilateral FDI effects on trade. During the 1990s a number of researchers started to raise some questions about the econometric properties of the gravity model of trade. They argue that the standard cross-sectional ordinary least square (OLS) method in gravity regressions generates biased results because it cannot properly account for heterogeneity in trade flows between 86 countries. 49 Specifically, it is claimed that gravity models of trade overestimate the effects regional integration as well as the effects of time invariant variables such as distance, common language, or adjacency. Misspecification issues and omitted variable problems are cited as reasons for the biased results. M?ty?s (1997) shows that all gravity models of international trade are misspecified from an econometric point of view. The paper starts by presenting the standard gravity model with the addition of country and time effects. These country and time effects are treated as unknown fixed parameters. The author states that cross- sectional studies restrict the model by assuming no time effects and time series models restrict local specific effects. M?ty?s claims that the gravity models used up to that time did not take into account the time, local, and target country (importing country) effects. The study shows that imposing these restrictions leads to incorrect inferences due to the misinterpretation of the coefficients on dummies accounting for trading blocs, common border, or common language. He suggests that models explaining trade should take into account these fixed effects. M?ty?s, K?nya, and Harris (1997) study the volume of exports in the APEC countries for the 1982-94 period. They present results for the restricted model and for the fixed effects model showing that most country-specific parameters (fixed effects model) are statistically significant. Dell?Ariccia (1999) analyzes the effects of exchange rate volatility on bilateral trade flows in Western Europe. The argument is that exchange rate volatility could have negative effects on trade and investment. The paper argues that the European Monetary 49 See M?ty?s (1997), Bayoumi and Eichengreen (1997), Cheng (1999), Pakko and Wall (2001), Glick and Rose (2001), Wall (1999, 2000, 2002, and 2003), Egger (2002), Millimet and Osang (2004), and Cheng and Wall (2005). 87 System (EMS) and later the European Monetary Union (EMU) had the objective to control exchange rate movements and misalignments in Europe. The author uses a panel data approach with different measures and techniques that focus on solving potential simultaneous causality problems. Results from a Hausman test suggest that the OLS regression generates biased results indicating the existence of simultaneity bias. This bias is due to the existence of unobserved country-pair specific effects and is addressed with the use of instrumental variables and a fixed effects model. A fixed effects model is preferred over a random effects model and results are similar to OLS estimates. The sample data covers the 1975-1994 period for the fifteen countries forming the EU and Switzerland. Results suggest that exchange rate volatility decreases international trade and these results are robust for different specifications. The coefficients on the standard gravity variables are in line with expectations. M?ty?s, K?nya, and Harris (2000) follow M?ty?s, K?nya, and Harris (1997) in analyzing trade patterns among the 12 original APEC members between 1978 and 1997. The dependent variable is exports from country i to country j and the explanatory variables are GDPs, populations, foreign currency reserves, real exchange rates, and distance. Local, target, and time specific effects are also added. The authors follow the econometric analysis presented in M?ty?s (1997) and then estimate four different models. Model A is a fully restricted model that assumes no local or target country effects, and no time effects. Model B includes local effects, Model C adds target effects to the previous one, and Model D is a fully unrestricted model. They conclude, based on F-tests, that Model D is the preferred specification claiming that this specification is superior in terms of statistics and economics. The paper identifies countries with strong propensity to 88 import and export. APEC members trying to increase exports should look at Singapore and New Zealand as potential markets. The authors claim that policy implications could be wrong in the absence of specific effects. Results also suggest that foreign GDP effects were underestimated in previous studies, that the effect of population on trade could be positive, and that the effect of real exchange rates is significant. Rose (2000) uses a gravity model to estimate the separate effects on trade of exchange rate volatility and common currencies. A large cross-country panel data set includes the EU countries as well as other 92 countries that have some sort of common currency arrangement. The augmented gravity model explains bilateral trade as a function of GDP, income per capita, distance, and a series of dummies accounting for common language, regional trade agreement, colonies, common nations, a common currency dummy, and a variable explaining exchange rate volatility. 50 Rose finds that two countries with a common currency trade three times as much as countries not sharing a common currency. This common currency effect is larger than the effect of reducing exchange rate volatility to zero but keeping separate currencies. The author performs a sensitivity analysis that suggests robust results. Soloaga and Winters (2001) investigate the effect of regional preferential trade agreements (PTAs) on trade. The paper applies a gravity model to 1980-1996 annual non-fuel imports data for 58 countries representing 70% of world trade. The authors modify the usual gravity equation by adding dummy variables that identify separate effects of PTAs on intra-bloc trade, members? total imports, and their total exports. They also test the significance of changes in the estimated coefficients before and after the 50 Exchange rate volatility is measured as the standard deviation of the first difference of the monthly natural logarithm of the bilateral nominal exchange rate. 89 formation of trading blocs. Results show no indication that increasing regionalism during the 1990s raised intra-bloc trade significantly. The paper presents evidence of trade diversion taking place in the EU and EFTA. Soloaga and Winters also suggest that trade liberalization efforts in Latin America had a positive impact on bloc members? imports. Pakko and Wall (2001) proposes a gravity equation that uses trading pair-specific fixed effects to control time invariant or fixed geographic, cultural, and historical factors instead of controlling these factors through the use of specific dummy variables. They argue that Rose (2000) has an estimation bias problem that leads to unprecedented findings. The authors claim that the fixed effects model avoids the estimation bias that may arise due to misspecification or omitted variables. Misspecification could arise with the creation of the variable distance that is supposed to reflect relative costs of trading. Omitted variable problems arise because it becomes impossible to include enough variables to account for all the important fixed factors (time invariant factors). Further, Pakko and Wall suggest that the fixed effects model not only controls variables such as language, common nation, colony, and distance, but also accounts for factors that are usually not included in gravity models. The paper shows that by using the data from Rose (2000) the fixed effects model results in much weaker evidence. Rose (2000) found that countries sharing a common currency trade three times more as they would with different currencies. Pakko and Wall find that having a common currency has no significant effect on trade flows between trading partners and conclude that one should be cautious in drawing conclusions when models are not robust. Martinez-Zarzoso and Nowak-Leman (2002) use an adjusted gravity equation to study the role of economic and geographical distance on Mercosur plus Chile exports to 90 15 EU countries. The authors use a panel data approach for annual exports disaggregated by sectors for the 1988-1996 period. Using a log-linear model, they estimate a gravity equation with sector specific exports for different countries at different time periods as the dependent variable. The independent variables are the differences in per capita income between countries (economic distance), distance between countries scaled by infrastructure, and the bilateral real exchange rate. The economic distance variable accounts for Linder and HOS effects. When trading partners have contrasting per capita income, higher economic distance might deter trade (Linder effect). When higher economic distance leads to higher trade, then HOS effects are present. The infrastructure variable is an index capturing information on roads, paved roads, railroads, and telephones. The geographical distance is scaled by using this infrastructure index. The authors utilize a fixed effects model allowing for country-pair specific effects and time specific effects. This paper finds that products that are highly sensitive to economic distance and not sensitive to geographical distance are the best candidates for future trade with EU. The authors find evidence of Linder effects in some industries and of HOS effects in other industries. Specifically, the Linder hypothesis applies to telecommunications, iron and steel, metals, industrial machinery, and animal feed. Sectors with a dominant HOS effect are furniture, footwear, beverages, meat and fish (products in which Mercosur has a comparative advantage). Martinez-Zarzoso and Nowak-Leman (2003) follow the previous paper by applying a gravity model to investigate Mercosur-EU trade patterns and trade potential. A sample of 20 countries consisting of the four members of Mercosur, plus Chile, and the 15 countries forming the EU is used. The fixed effects model is preferred over the 91 random effects model. Exporter and importer incomes have a positive effect on trade flows. Results also show that exporter?s population has a negative effect on exports and importer?s population has a positive effect. Findings also suggest that for Mercosur-EU trade flows, only exporter infrastructure has a positive effect on trade. Preferential trade agreements also increase trade flows. Potential trade estimates show that Mercosur was exporting below its potential levels in 1996, but results are varied for previous years. Cheng and Wall (2005) study the various fixed effects specifications and evaluate them in terms of their econometric appropriateness. First, they show that standard pooled-cross-section methods used in gravity models have an estimation bias problem due to omitted or misspecified variables. The paper shows that a two-way fixed effects model solves this by using country-pair and period dummies that explain bilateral trade patterns. This two-way fixed effects model replaces M?ty?s? (1997) three-way model due to its bilateral nature. Country-specific dummies capture factors such as distance, common border, common language, history, culture, and others that are constant over time. Cheng and Wall show that alternative fixed effects models such as M?ty?s (1997), Glick and Rose (2001), and Bayoumi and Eichengreen (1997) are special cases of their proposed two-way model. They claim that the restrictions applied to obtain these alternative models are not supported statistically. Finally, the paper investigates the effect of integration on trade patterns by adding dummies accounting for preferential agreements and controlling country-pair heterogeneity using the two-way fixed effect model proposed. Results indicate that unless heterogeneity is accounted for properly, gravity models of bilateral trade can overestimate the effects of integration on trade flows. 92 III. Theoretical Framework of Gravity Models The first theoretical foundation of the gravity model was developed by Linnemann (1966). Chapter 3 of Linnemann?s book attempts to reconcile the trade flow equation in (3.2) with economic theory. The author suggests that trade flows between individual countries could be derived from a Walrasian type general model. Linnemann starts with a 3-country model that accounts for transportation costs and production. The demand equations for the product of country 1 are a system composed of domestic demand and the demand for the product in countries 2 and 3. X D 11 = D 11 (Y 1 , N 1 , p 1 , p 2 , p 3 , t 21 , t 31 ), (3.3) X D 12 = D 12 (Y 2 , N 2 , p 1 , p 2 , p 3 , t 12 , t 32 ), (3.4) X D 13 = D 13 (Y 3 , N 3 , p 1 , p 2 , p 3 , t 13 , t 23 ), (3.5) where, X D ij is the demand for the product of country i in country j, Y i is national product or income in country i, N i is the population in country i capturing the notion of optimum size in a production unit, p i is the price of a product unit of country i in country i, and t ij are transport costs between countries i and j for a product unit of country i. Equation (3.3) shows domestic demand for the product of country 1 while (3.4) and (3.5) represent foreign demand. The supply equation shows that total supply depends on production capacity K i and on price p i . X S 1 = S 1 (K 1 , p 1 ) (3.6) Linnemann?s model is a short-run model in which production capacity and income are given. He assumes a constant capital-output ratio in the short-run (income and production capacity are given) and rewrites equation (3.6) as follows: X S 1 = S? 1 (Y 1 , p 1 ) (3.7) 93 Equality of supply and demand is given as: X S 1 = X D 11 + X D 12 + X D 13 (3.8) Equation (3.8) is an equilibrium equation reached through the interaction of supply and demand. It is neither a supply nor a demand function and therefore it does not contribute to explain trade flows X ij as proposed in equation (3.2). Consequently, X S i and p i are excluded so that we have X ij as a function of other X ij , t ij , Y i , and N i . In order to eliminate the X ij as an explanatory variable, Linnemann reduces this 3-country model to a bilateral one, in which X ij is a function of t ij , Y i , and N i . Before defining this bilateral trade model, Linnemann defines the foreign supply for product i as follows: X SF i = X S i - X D ii (3.9) where X D ii is the domestic demand for product i, X S i is domestic supply, and X SF i is foreign supply. For small countries that have no power to affect neither world prices nor third countries? trade resistances (t ij ), it is sensible to eliminate them from the model. Therefore, Linnemann?s bilateral model of trade flows is as follows: X D 12 = D? 12 (Y 2 , N 2 , p 1 , t 12 ) (3.10) X SF 12 = S F 12 (Y 1 , N 1 , p 1 ) (3.11) X SF 12 = X D 12 (3.12) This bilateral equilibrium model of X 12 depends on economic sizes (Y i and N i ) and trade resistances (t ij ) between the two countries under consideration as in the standard gravity model presented in equation (3.2). Anderson (1979) uses a linear expenditure system with homothetic and uniform preferences for a country?s goods and with products that are differentiated by place of origin. Using a pure Cobb-Douglas expenditure system model, Anderson presents the 94 simplest possible gravity equation assuming that each country specializes in the production of one good with no tariffs or transportation costs. The fraction of income spent on the product produced by country i is denoted by b i and is the same across countries due to identical Cobb-Douglas preferences. Income in country j is denoted by Y j and assuming prices are equal to unity, imports of good i by country j M ij equals: M ij = b i Y j (3.13) Income should equal sales Y i = b i (?Y j ) and substituting into (3.13), Anderson finds the ?simplest form of gravity model?: M ij = Y i Y j /?Y j (3.14) Anderson also uses a trade share expenditure system to derive a gravity-type equation by appending to the previous Cobb-Douglas expenditure system a traded-non- traded goods split. He presents a weakly separable utility function as u = u(g(traded goods), nontraded goods) where individual demand for traded goods are determined by homothetic preferences and are maximized subject to a budget constraint. Since preferences are identical across countries, expenditure shares are the same everywhere. For any country j, ? i is the expenditure on country i tradeable good divided by total expenditure on tradeables in country j. Let ? j be the share of expenditures on all traded goods in country j?s total expenditure, ? j = F(Y j ,N j ). Anderson defines the demand of imports of country i?s good on country j as follows: M ij = ? i ? j Y j (3.15) The share of national expenditures on tradeables is a function of income and population and the share of total tradeable goods expenditures accounted by each good is a function of transport costs. The balance of trade equation for country i implies that 95 Y i ? i = (?Y j ? j )? i . Solving for ? i and substituting this equation into (3.15), Anderson finds a deterministic form of gravity equation without the distance term and with a scale term added: M ij = ?? ji ij jjii M YY?? (3.16) Krugman?s (1979) seminal paper develops a model that explains trade flows as a function of economies of scale instead of factor endowments or technology. He assumes that scale economies are internal to firms with a market structure that follows a Chamberlinian monopolistic competitive market where firms have some monopoly power but entry drives monopoly profits to zero. 51 It is assumed a one factor (labor) economy that is able to produce many goods. All consumers share the following utility function U: U = ? = n i i cv 1 )( , v? > 0, v? < 0 (3.17) where c i is the consumption of good i. All goods share the same cost function and the amount of labor needed for production is a linear function of output: l i = ? + ?x i, ?, ? >0 (3.18) where ? are fixed costs, l i is the labor used to produce good i, and x i is the quantity produced of good i. Through conventional consumer?s utility maximization subject to a budget constraint, Krugman derives the demand function faced by a firm. Then, 51 Chamberlin (1933, 1962) introduces the tangency solution in his theory of monopolistic competition in which entry and exit of firms lead to a zero profit situation. At the tangency solution, an individual?s demand curve is tangent to a falling portion of the average cost curve in which scale economies prevail. 96 following a profit maximizing behavior and Chamberlin?s tangency solution, he finds the firm?s supply curve and pricing scheme. Krugman examines the effects of population growth, trade, and factor mobility with conventional comparative statics. Population growth and trade increase the production of each good as well as the number of goods. It is also shown that the volume of trade is maximized when economies of scale between two countries equal in size, a result that is in line with the Linder hypothesis. Further, in the presence of increasing returns and trade impediments, factor mobility seems to create a process of agglomeration in which all workers will concentrate in one country or the other. The author suggests that the more populous the region, the more variety of goods and the higher real wages which would induce immigration. Krugman proposes that trade may be a channel to extend markets and therefore to allow the utilization of scale economies. The paper shows that trade need not be a result of international differences in technology or factor endowments. Krugman argues that trade is a way of extending a domestic market allowing for scale economies. He concludes that economies of scale are ?underemphasized? in formal trade theory. Krugman (1980) extends the previous paper by examining the effects of transportation costs and the effects of larger domestic markets on wages as well as the effects of home market size on trade. The author follows the same approach as in Krugman (1979) with scale economies where firms can differentiate their products at no cost and equilibrium is reached by Chamberlinian monopolistic competition. Krugman shows that transportation costs have no effects on pricing policies and also no effects on output and the number of firms. He also finds that countries with larger domestic 97 markets tend to have higher wage rates. Finally, Krugman argues that countries with larger domestic markets on specific goods will tend to export those goods. He also suggests that findings are in line with Linder (1961). Krugman (1981) develops a model that formalizes previous work such as Balassa (1967), Grubel (1970), and Kravis (1971). The paper attempts to explain the fact that much of the world trade is between countries with similar factor endowments, trade between similar countries is mainly intra-industry, and that the growth on intra-industry trade has not caused serious income distribution problems. Krugman proposes that the usual forces of comparative advantage operate on ?groups of products? giving rise to inter-industry specialization and trade. On the other hand, scale economies lead each country to produce only a subset of goods within a group, leading to intra-industry trade. Krugman argues that similar countries have an incentive to trade, that this trade will be mainly on goods that use similar productive factors, and that this intra-industry trade will not generate income distribution problems that usually arise with inter-industry trade. Helpman and Krugman (1985) present the theoretical framework supporting that bilateral trade flows depend on the product of GDPs. 52 In a nutshell, demand for variety drives consumer expenditures and monopolistic competitive firms produce differentiated products. The authors argued that under the usual assumptions, the Heckscher-Ohlin theory does not have the property that bilateral trade depends on the product of GDPs. Empirical gravity models find a significant effect for the product of incomes, suggesting that a model of trade with differentiated products is preferred. 52 See Helpman and Krugman (1985), section 1.5. 98 Bergstrand (1985) uses a general equilibrium model of trade where consumers across countries maximize the same utility function with constant elasticity of substitution (CES) subject to an income constraint: 1 1 1 1 1 1 ? ? ? ? ? = ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? = ? j j j j j j j j j j jj N jk k kjj XXU ? ? ? ? ? ? ? ? ? ? , j = 1, ?, N (3.19) where X kj is aggregate demand in country j for k?s goods, X jj are domestically produced goods, ? j is the CES between domestic and importable goods, and ? j is the CES among importable goods. Expenditures in country j are constrained by income as follows: Y j = kj N k kj XP ? =1 _ , j = 1,?, N (3.20) where _ kj P = P kj T kj C kj /E kj and P kj is the k-currency price of k?s product sold in the j th market, T kj is one plus j?s tariff rate on k?s product, C kj are transport costs to ship k?s product to j, and E kj is the spot value of j?s currency in terms of k?s currency. Maximizing utility in equation (3.19) subject to (3.20) generates N(N + 1) first order conditions that lead to N(N ? 1) bilateral aggregate import demands and N domestic demand equations. In terms of supply, Bergstrand uses equation (3.21) in which firms in each country i maximize the profit function: ? i = ? = N k 1 P ik X ik - W i R i , i = 1, ? , N (3.21) where R i is the amount of the single, fixed (internationally immobile) resource to produce different goods and W i is the price of this resource (i.e., wage in the case of labor). R is 99 allocated in each country according to a constant elasticity of transformation (CET) function as follows: i i i i i i i i i i ii N ik k iki XXR ? ? ? ? ? ? ? ? ? ? + + + + ? = + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? = ? 1 1 1 1 1 1 , j = 1, ?, N (3.22) where ? i is i?s CET between production at home and foreign markets and ? i is i?s CET for production among exports markets. Substituting (3.22) into (3.21) and maximizing this equation, Bergstrand finds N(N ? 1) bilateral exports supply equations and N domestic supply equations. 53 Bergstrand derives a general equilibrium model of trade by equating supply and demand equations. To find a general equilibrium consistent with the gravity specification, he assumes countries are small open economies taking prices and foreign incomes as given. He also assumes identical utility and production functions across countries. Assuming also perfect substitutability of goods across nations, perfect commodity arbitrage, zero tariffs, and zero transport costs, the general equilibrium equation is simplified to the general form of the gravity equation: PX ij = (1/2)Y i 1/2 Y j 1/2 (3.23) In the last section of his paper, Bergstrand estimates a generalized gravity model for 1965, 1966, 1975, and 1976. 54 He finds that price and exchange rates have significant statistical effects on trade flows. Bergstrand suggests that if trade flows are differentiated by origin, the typical gravity equation omits prices and exchange rates. He also finds that 53 See equations 8 and 9 in Bergstrand (1985). 54 See Appendix III for a list of the variables used by Bergstrand. 100 exporter?s income increases trade flows, which implies that the elasticity of substitution among importables ? j is greater than 1. The negative coefficient on country i?s GDP deflator supports this conclusion and suggests that the elasticity of substitution between domestic and imported goods is less than 1. Finally, the negative coefficient for i?s export unit value index implies that ?the elasticity of transformation among export markets exceeds that between production for domestic and foreign markets? (Bergstrand, p. 480). Overall, these elasticities imply that in terms of production and spending, countries are more flexible when it comes to substitute among export and import markets than when they substitute between a domestic and a foreign market. Bergstrand (1989) extends the previous paper by incorporating factor endowments differences (H-O theory) and non-homothetic preferences (the Linder hypothesis) to the model. Consumers maximize a Cobb-Douglas-CES-Stone-Geary utility function subject to an income constraint. National income, income per capita, and prices explain bilateral trade flows. According to Bergstrand, the utility function of consumer l in country j (U jl ) is: U jl = ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?? == A An A N n H h Ahnjl X /1 11 x ? ? ? ? == ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?? 1 _ /1 11 B N n H h Bhnjl XX B Bn B - ? < ? A , ? B < 1; 0 < ? <1 (3.24) where X Ahnjl (X Bhnjl ) is the amount of manufactured (non-manufactured) goods produced by industry A?s (B?s) firm h in country n demanded by consumer-worker l in country j, and BX _ is the minimum consumption of good B (necessity). Expenditures are constrained by consumer?s nominal income (Y jl ): 101 Y jl = ??? ===BAa N n H h an ,11 (P anj T anj /E nj )X ahnjl (3.25) where T anj is one plus the tariff rate on industry a (a = A, B) exports from country n to j, E nj is exchange rate between n and j, and P anj is the free on board (f.o.b.) price of firm h?s output of industry a exported from country n to j. Maximizing (3.24) subject to (3.25) leads to a set of bilateral import demand functions that are aggregated since consumers in country j are identical. Bergstrand derives country j?s demand curve for the output of A produced by firm g in country i as: P Aij = ijAijjjAgij ETyYX AAAA 1/11/1/1/1 )1()( ??? ? ???? ? x () A A n A h njAnjAhjn ETP ? ? /1 1 / ? ? ? ? ? ? ? ? ?? , i =1,?,N (3.26) where, ? A = 1/(1 ? ? A ), Y j is j?s nominal GDP and y j is j?s GDP per capita. Bergstrand assumes that similar demand exists for industry B?s output. On the supply side, each firm in each of the two industries produces a differentiated product in a Chamberlinian monopolistic competition market using labor (L) and capital (K). Bergstrand?s profit function assumes a linear technology function shared by all firms, a fixed supply of labor and capital in each country, and assumes that each firm?s output is distributed among domestic and foreign markets according to a CET function: ? agi = ?P ain X agin ? (W i ? La + R i ? Ka ) - W i ? La () a a n aginain XC ? ? /1 ? ? ? ? ? ? ? - R i ? Ka () a a n aginain XC ? ? /1 ? ? ? ? ? ? ? , g = 1,?,H ai ; a = A, B; i = 1,?,N (3.27) 102 Maximizing the profit function, Bergstrand finds equations for the marginal costs of exporting industries A and B. 55 After making the appropriate substitutions, solving for reduced form equations, and summing up all firms in industry A or B in country i, the author presents a generalized gravity equation that explains trade flows as a function of GDP, per capita income, distance, tariffs, price levels, and the exchange rate. Bergstrand applies the derived gravity equation to the same data of his previous paper to test the H-O model. Results show that between 40% and 80% of the variation across countries in one digit SITC trade flows are explained by the model. Coefficients on exporter and importer?s income are positive as expected, and coefficients for exporter?s per capita income suggest that chemicals, raw materials, manufactures, machinery and transport equipment, and food products are usually capital intensive in production whereas beverages, tobacco,and miscellaneous manufactures are labor intensive. Bergstrand further notes that the coefficient on importer?s per capita income suggests that manufactures tend to be luxuries and raw materials necessities. Finally, he suggests further research for the effects of prices on trade flows. Bergstrand (1990) extends previous theoretical work by examining how average levels and inequality of GDP, GDP per capita, tariff rates, and capital-labor ratios affect the share of intra-industry trade. The paper provides a theoretical framework for such a model and then presents an empirical analysis for 14 developed countries. These theoretical foundations are similar to Bergstrand (1989) with minor differences. By maximizing constrained utility functions, a bilateral import demand function is as follows: 55 See Bergstrand (1989) for details on marginal costs functions and on the final derived gravity equation. 103 X hij = ?Y j (1-y j -1 ) (P hij D ij T ij /E ij ) -? (P j ) -1 (3.28) where X hij is the aggregate demand in country j for output of country i?s firm h, Y j is j?s national income, y j is j?s income per capita, P hij is the f.o.b. i-currency price of firm h?s output of X sold in j, D ij is the transport cost to ship X from i to j, T ij is one plus the tariff rate on imports of X from i to j, E ij is the exchange rate, P j is an import price index for good X in country j, and ? is the elasticity of substitution in consumption. On the supply side, a Chamberlinian monopolistic competition characterized by profit maximization and zero economic profits yields the following mark-up pricing function: P hij = [(1- ? -1 ) -1 (X hij /X hi ) 1/? ](W i ? LX + R i ? KX ) (3.29) where W i and R i are the wage and rental rate in country i, X hij is the output of firm h in country i that is exported to j, X hi is total output of firm h, ? is the elasticity of transformation of output among domestic and foreign markets, and (W i ? LX + R i ? KX ) are marginal costs. Firm?s output is determined by: X hi = (?-1)[(W i ? LX + R i ? KX )/(W i ? LX + R i ? KX )] (3.0) where, (W i ? LX + R i ? KX ) are fixed costs. Then, substituting firm?s output in (3.30) and demand for imports (3.28) in the mark-up price equation (3.29) and then solving for the equilibrium price and quantity, one can determine the value of bilateral trade flows of firm h from country i to j. Multiplying the value of the flow of firm h by the number of firms, Bergstrand finds the gravity equation. 56 Using a Grubel-Lloyd index that measures the share of intra-industry trade between i and j, Bergstrand presents eight propositions based on comparative statics. 56 See Bergstrand (1990) to find the derived gravity equation. 104 These propositions are empirically tested for each of the two-digit SITC category 7 for each possible bilateral trade flow among 14 major developed countries in 1976. Empirical results are all in line with the theoretical framework. In general, Bergstrand?s model reveals that more similar per capita income between two countries leads to higher intra-industry trade. In terms of supply, the author proposes that the more inequality among capital-labor ratios, the lower intra-industry trade (Heckscher-Ohlin-Samuelson). Regarding demand, the greater the inequality between per capita incomes, the lower the share of intra-industry trade due to differences in tastes (Linder). Harrigan (1994) proposes an econometric approach to test the monopolistic competitive model of intra-industry trade summarized in Helpman and Krugman (1985) using 1983 disaggregated data from OECD countries. He suggests using gross trade volumes to examine the contribution of scale economies to trade because using a Grubel- Lloyd index of intra-industry trade is subject to aggregation bias. Harrigan distinguishes between a model of monopolistic competition and what he called the ?Armington-HOV? model explaining that aggregate intra-industry trade with taste for variety and technological differences make it possible for foreign varieties to be produced at home. He indicates that if the monopolistic competitive model explains gross trade flows, industries with high gross trade flows should be described as having large scale economies. If that is not the case, then the Armington-HOV model is right and high gross trade is determined by substitution between domestic and foreign production. Harrigan uses two different equations and four distinct proxies for scale economies to test whether trade is explained better by any of the previous models. The level of aggregation is at the 3-digit ISIC, which consists of 26 industry categories. 105 Results strongly support both models in the sense that the elasticity of imports with respect to a country?s output is one. Harrigan further finds some evidence that higher volumes of gross trade are associated with scale economies but this is sensitive to the choice of proxy variables. He concludes that scale economies and product differentiation by location of production are important causes of trade patterns. Deardorff (1995) demonstrates that the gravity equation can be also derived from the HOS theory. Deardorff uses two scenarios: one in which he assumes frictionless trade with no barriers to trade and homothetic products and another in which he introduces impediments to trade and product differentiation. In the first case, consumers are indifferent when choosing among goods from different countries, including their own country. With homothetic preferences, Deardorff derives what he calls a ?simple frictionless gravity equation? where trade depends on incomes and a factor of proportionality. 57 He shows that this result holds even assuming arbitrary preferences. In the case of impeded trade, Deardorff derives equations for bilateral trade using Cobb- Douglas and CES preferences. In the first case, Deardorff derives a simple frictionless gravity equation for trade measured as c.i.f. and then presents the standard gravity equation (adds transport costs) with trade valued as f.o.b. Results based on CES show similar findings but trade is lower for distant countries. His findings show that gravity models could be easily derived from standard HOS trade theory and consequently it is not appropriate to conclude that the empirical success of gravity models suggest failure of the factor proportions or any other theory. 57 See Frankel (1998), page 13. 106 IV. The Linder Hypothesis Linder (1961) proposes a supplement to HOS theory by letting trade be determined by demand rather than supply. He argues that the more similar the demand structures between two countries, the more intensive their bilateral trade in manufactures. Linder develops this principle due to an apparent lack of empirical success of the factors proportion theory explaining patterns of trade. Linder emphasizes that factor proportions works well when primary products are being traded, but fails to explain trade in manufactures. Linder?s basic proposition is that the production of non-primary exportable products is determined by internal demand. He claims that goods will not be produced at a comparative advantage unless there is a domestic market for those products. Therefore, countries produce manufactures for their own consumption and exports are surpluses or excess supply. He concludes that countries interested in these surpluses should have similar demand patterns to the exporting country. Determining which factors affect a country?s demand structure should predict which countries will trade more intensively. Linder suggests that the average level of income is the single and most important determinant of a country?s demand structure. Considering consumer goods, Linder argues that higher average income leads to replacement of less sophisticated consumer goods toward better quality goods. He suggests that higher income induces qualitative changes in demand besides the usual increase in quantities. In terms of capital goods, Linder argues that because per capita income is determined (among other factors) by the capital stock, higher income countries demand more sophisticated capital equipment than lower income ones. Therefore, he suggests that differences in per capita income are an 107 obstacle to trade. When income differences are of a certain magnitude, goods produced at comparative advantage in one country are not demanded in the other and vice versa. Figure 3.1: Trade in Manufactures for 2 Countries Linder hypothetical reasoning is summarized in Figure 3.1. The horizontal axis measures per capita income and the vertical axis measures the quality of each product demanded in ordinal numbers. The positive relationship between per capita income and product quality is represented by line OP. Consumer and capital goods of different qualities are demanded in a specific country because a limited degree of quality could result from a single product and also due to unequal income distribution. A given specific per capita income may cover a range of qualitative degrees around the average degree on OP. Therefore, for country I in Figure 3.1, the products demanded have a degree of quality that ranges from a to e, with b as the average quality. Similarly for country II, the quality range is c-g with f as the average. Both countries share the y O P q o I II g f e d c b a 108 qualitative range c-e. These overlapping demands relate to products that have a quality within this range c-e and these products are the candidates for trade. Country I will not demand products whose quality is higher than e, and country II will not demand products with a quality below c. Linder mentions that factors such as distance, languages, and cultural and political affinity act as ?trade-braking functions? that could dissolve the effect of similar per capita incomes on actual trade. 58 He tests his theory using 32 countries and trade data for 1958. Linder uses a graphical approach by plotting the average marginal propensity to import of all countries with respect to a specific country. These diagrams seem to indicate that Linder hypothesis is correct, but he suggests further empirical research. One of the first attempts to test the Linder hypothesis is Hirsch and Lev (1973). This paper follows Linneman (1966) and estimates a gravity model to explain commodity flows in five industries for Denmark, the Netherlands, Israel, and Switzerland. The five industries comprise processed food, textiles, clothing and footwear, processed chemicals, machinery, electrical machinery, appliances, and professional equipment. The independent variables are GNP, distance, a dummy for preferential trade, and a variable indicating per capita income differential. This income differential variable is specified as a ratio, with the numerator being the smallest per capita income of the two countries. Other specifications are tested with no significant changes in results. Hirsch and Lev find results that are consistent with the Linder hypothesis. Most coefficients for the income differential variable are negative and significant, implying that the greater the difference in per capita income, the lower the volume of trade between two countries. 58 See Linder (1961), page 108. 109 Thursby and Thursby (1987) use bilateral trade flows between 17 countries to examine the Linder hypothesis and the effects of exchange rate variability using a gravity model and running separate regressions for different countries. The period under study is 1974-1982. Following standard demand and supply analysis, the authors estimate a gravity model that shows overwhelming support for the Linder hypothesis and the theoretical proposition that exchange rate risk affects bilateral trade flows. The coefficient on the Linder variable is negative and significant with the exception of 2 countries. Similarly, most of the countries with proper specified equations show a statistically significant negative coefficient for the exchange rate risk variable. They also conclude that using nominal or real exchange rates does not make a difference in terms of results. Other papers such as Hoftyzer (1975) or Greytak and McHugh (1977) empirically test the Linder hypothesis and suggest adding a distance variable. They show that distance plays a significant role explaining trade flows of manufactures. They conclude that Linder?s empirical test for his hypothesis overestimates the effects of income differential on trade flows. Similarly, Qureshi, French, and Sailors (1980) claim that the main problem concerning empirical verifications of the Linder hypothesis is the inability to statistically account for the separate influence of distance and the Linder effect on the intensity of trade. In order to solve the distance problem, they propose an alternative empirical method that tests the theory in terms of changes in propensities to trade against changes in income differences between two points in time. This model fails to find evidence of a Linder effect. Kennedy and McHugh (1980) follow Greytak and McHugh (1980) and find no evidence of a Linder effect when distance is accounted for. 110 Hanink (1988) adds to Linder?s model of bilateral trade flows by incorporating the hierarchical flow of goods that is common in regional trade. This hierarchical trade is determined by population size and the largest country has the widest variety of goods and the smallest country has the smallest variety. Hanink calls this model the ?extended Linder model? and he performs empirical tests to examine its validity explaining actual trade flows. Results indicate that trade flows are positively related to market homogeneity (the Linder hypothesis), negatively related to distance, and positively related to variety across goods. The dependent variable is trade intensity measured as per capita imports of country i from country j. The explanatory variables are the absolute difference in per capita GNP, the distance between economic centers, and the variability across goods measured by absolute difference in populations. Hanink concludes that his extended model should be further tested to examine its validity in ?the increasingly complex international economy? (Hanink, p. 333). Chow, Kellman, and Shachmurove (1999) investigate whether the Linder hypothesis explains trade flows between four Newly Industrialized Countries (NICs) in East Asian and their major OECD trading partners. The authors examine the behavior of exports from these four NICs to OECD markets during the 1965-1990 period. They emphasize the fact that per capita income differentials between these NICs and those of OECD have been constantly decreasing during the period under study. Linder?s theory is tested by using disaggregated data on manufactured exports. With the use of an improved measure of trade intensity (trade complementary index) as the dependent variable, the authors use OLS for twelve country-partner pairs (4 NICs and 3 OECD 111 countries). Results support the Linder hypothesis and suggest that tastes significantly affect trade flows in those countries. McPherson, Redfearn, and Tieslau (2000) examine the Linder hypothesis using a random effects Tobit approach. Findings support the Linder hypothesis for all but one of the 19 OECD countries under consideration. Results provide strong evidence that countries with similar income levels trade more. The paper uses a panel data approach to capture both time-invariant and time-variant effects. They also compare the results of the Tobit random effects model to a simple random effects model and conclude that a simple model leads to misleading results. The data includes 161 potential trading partners of each of the OECD countries covering the 1990-1995 period. The dependent variable is the dollar value of exports from OECD country j to potential trading partner i at time t. The explanatory variables are real GDP of trading partner i, real exchange rates, and the absolute differences in per capita income between trading partners (the Linder effect). Besides finding support for the Linder hypothesis, the paper finds that the relative size of a trading partner?s economy has a positive effect on trade. In terms of exchange rates, all but three of the OECD countries show the expected positive and significant effect on exports. 59 McPherson, Redfearn, and Tieslau (2001) examine the Linder hypothesis in six East African developing countries. In five out of six countries, the Linder hypothesis holds. The paper uses a fixed effect panel data model that captures the time-invariant country-specific effects. It is one of the first attempts to test the Linder hypothesis in developing countries. Results show that Ethiopia, Kenya, Rwanda, Sudan, and Uganda 59 See McPherson, Redfearn, and Tieslaw (2001) for a description of the exchange rate variable. 112 trade more intensively with countries that have similar per capita incomes. As in their previous paper, the authors enforce the idea that a censored Tobit model should be used when studying bilateral trade flows between one country and a large number of partners. The authors argue that using imports as a dependent variable is appropriate to test for Linder effects since most imports in developing countries tend to be manufactures. Country specific effects are for the most part statistically significant at the 95% confidence level. The authors conclude that the factor proportions theory is inadequate when investigating trade flows in developing countries and suggest the appropriateness of the Linder hypothesis in such a context. Other papers such as Martinez-Zarzoso and Nowak-Leman (2002 and 2003) have found a Linder effect for trade patterns between Mercosur and EU. 60 V. Econometric Issues The gravity model of trade has been specified in many different forms. The standard gravity model with country income, population, and distance as explanatory variables was augmented by adding income per capita, real exchange rates, and dummy variables that account for common borders, common language, preferential trade agreements, and so on. Starting with M?ty?s (1997), recent literature has focused on estimating gravity models of trade using different fixed effects models. This section presents the different specifications of the gravity model going from the standard model to the various fixed effects models. The standard gravity model introduced by Tinbergen (1962) in log linear form is presented in (3.1) of this chapter. Adding population as suggested by Linnemann (1966) 60 This paper?s findings are discussed in Section II. 113 and using as dependent variables the product of income and population gives (3.31) as in Frankel (1997): log T ij = ? 0 + ? 1 log (GDP i * GDP j ) + ? 2 log (pop i * pop j ) + ? 3 log Dist ij + ? (3.31) where T ij is trade between country i and country j (exports plus imports), GDP i and GDP j are real gross domestic products, pop i and pop j are populations, and Dist ij is the geographic distance between the two countries. Modifications of (3.31) use income per capita as another explanatory variable. The product of GDPs is expected to be positively related to trade since large countries should trade more than small ones. The coefficient for distance should be negative given that proximity reduces transportation and information costs. Populations are expected to be negatively related to trade since larger countries tend to be relatively less open to trade as a percentage of GDP. Adding dummy variables to the standard gravity specification leads to the first version of the augmented gravity model of trade: log T ij = ? 0 + ? 1 log (GDP i *GDP j ) + ? 2 log (pop i * pop j ) + ? 3 log Dist ij + ? 4 Adj + ? 5 Lang + ? ? i PTA + ? (3.2) where Adj is a dummy that stands for adjacency or common border, Lang accounts for common language effects, and PTAs are dummies capturing the effects of preferential trade agreements. Equation (3.32) has been used repeatedly in the literature (see Frankel (1997), Frankel, Stein, and Wei (1997), Bayoumi and Eichengreen (1995), Chow, Kellman, and Shachmurove (1999), Del?Ariccia (1999), and Pakko and Wall (2001)). However, other papers such as Hirsch and Lev (1973), Bergstrand (1985), Thoumi (1989), Bergstrand (1989), M?ty?s, K?nya, and Harris (1997), Soloaga and Winters (1999), M?ty?s, K?nya, and Harris (2000), Martinez-Zarzoso and Nowak-Lehmann 114 (2002), and Cheng and Wall (2005) have specified the gravity equation by separating GDPs and populations. This is appropriate when trying to explain a country?s imports or exports since it captures a country?s productive and absorptive capacities separately. log X ij = ? 0 + ? 1 log GDP i + ? 2 log GDP j + ? 3 log pop i + ? 4 log pop j + ? 5 log Dist ij + ? 6 Adj + ? 7 Lang + ? ? i PTA + ? (3.3) where X ij is the value of exports of country i to j. Most of the literature has used some sort of the standard or augmented gravity models presented above. M?ty?s (1997) examines the econometric properties of the equation. He suggests that all gravity type models used to measure the effect of trading blocs on trade patterns are misspecified and lead to incorrect interpretation and improper economic inference since they do not account for the local country, target country, and time effects. According to M?ty?s, the correct econometric specification is as follows: log E ijt = ? i + ? j + ? t + ? 1 log GDP it + ? 2 log GDP jt + ? 3 log Dist ij + ?+ ? ijt (3.34) where E ijt is exports from country i to j at time t, ? i is the local country effect, ? j is the target country effect, and ? t is the time (business cycle) effect. According to M?ty?s, ? i , ? j , and ? t are fixed unknown parameters. M?ty?s claims that all other forms of gravity equations are restrictions of (3.34). For N countries, M?ty?s presents this panel data approach in vector form as follows: y = D N ? + D J ? + D T ? + Z ? + ? (3.5) where, y is a vector of observations on the dependent variable, Z is the matrix of observations of the explanatory variables in (3.34), and D N , D J , and D T are dummy 115 variable matrices. 61 M?ty?s claims that when dummies accounting for common language or common borders are added to the model, equation (3.35) becomes: y = D? + Z ? + ? (3.6) where ? is the parameter vector of the previously mentioned dummies. The author claims that it is easy to show using simple matrix algebra that the column vectors on D can be expressed as a linear combination of the column vectors of the matrices D N , D J , and D T . If the parameters ?, ?, and ? in (3.35) are significant, then ? should be significant in (3.36) due to misspecification of the model. Therefore, he concludes that any inference based on ? is misleading. The panel data fixed effects model proposed by M?ty?s (1997) dominates the recent empirical literature. Dell?Ariccia (1999), M?ty?s, K?nya, and Harris (2000), Pakko and Wall (2001), Martinez-Zarzoso and Nowak-Lehmann (2002), and Cheng and Wall (2005) use the fixed effects model in the gravity equation. Different fixed effects models claim to solve the problem of country-pair heterogeneity not addressed when using the classical OLS gravity approach. Cheng and Wall (2005) claim that the model introduced by M?ty?s (1997) is a three-way fixed effects that becomes a two-way fixed effects model when dealing with country pairs. Specifically, Cheng and Wall propose the following equation: log E ijt = ? 0 + ? ij + ? t + ??Z ijt + ? ijt (3.7) where, Z ijt is a row vector of gravity variables (GDP, population, per capita income, differences in per capita income, real exchange rate, etc), ? 0 accounts for that portion of the intercept that is common to all years and all country pairs, ? ij is the part of the 61 See M?ty?s (1997) for details on dummy matrices. 116 intercept that is specific to each country pair and common to all years, and ? t is specific to year t and common to all country pairs. The dependent variable is defined as in (3.34). In this equation, country pair effects are allowed to differ depending on the direction of trade flows (? ij ? ? ji ). Cheng and Wall (2005) compares different fixed effects specifications and a pooled cross-section model to the general fixed effects in (3.37). Cheng and Wall (2005) show that the standard pooled-cross-section model suffers from estimation bias due to omitted or misspecified variables. 62 This issue is solved by using fixed effects to capture those factors that remain constant over time such as distance, common border, common language, historical links, or cultural affinities. The authors also showed that alternative fixed effects are special cases of their general equation presented in (3.37) that introduce restrictions with no statistical validity. Fixed Effects and a Time Series Cross Section Model According to Greene (2003), in the presence of relatively small cross-sectional units and relatively large time periods, it is reasonable to specify a common conditional mean function across groups (countries) with heterogeneity taking the form of different variances rather than different intercepts. A gravity model estimating trade for a small number of countries over a long period of time requires a special type of panel data that takes into account the time series properties of the data set. Greene (2003) refers to this model as a time-series-cross-section (TSCS) that is specified as: y it = ??X it + ? it (3.38) 62 See literature review section for more on Cheng and Wall (2005). 117 where it is assumed that ? 1 = ? 2 =?= ? n . The model allows for E[? it 2 ] = ? ii (groupwise heteroscedasticity), Cov[? it ,? jt ] = ? ij (cross group correlation), and ? it = ? i ? i,t-1 + u it (within group autocorrelation). 63 A groupwise heteroscedastic model assumes variances differ across countries, a cross group correlation model allows for correlation among units (countries), and within group autocorrelation models permits different ?s for each country. TSCS models can be estimated by using the econometric software Limdep, which allows for different estimations producing up to nine sets of results that depend upon specification. In terms of the disturbance covariance, the TSCS command allows for a model specification with no correlation or heteroscedasticity (S0), a model with groupwise heteroscedasticity (S1), and a model with cross group correlation and groupwise heteroscedasticity (S2). Regarding autocorrelation, a model with no autocorrelation (R0), a model with autocorrelation and same ? for all groups (R1), and one with autocorrelation allowing for different ?s (R2) can be specified. All nine combinations of these models can be estimated. Model selection for models not accounting for autocorrelation (R0 with S0, S1, and S2) can be based on likelihood ratio tests. Since the autocorrelation models are not estimated by maximum likelihood, performing likelihood ratio tests between two models specified as R1 or R2 is not appropriate. VI. Trade Diversion The J-curve analysis presented in Chapter 2 suggests that trade could have been diverted from non-Mercosur members (EU and US) to Brazil. Results imply that 63 See Greene (2003) for a discussion on models and estimators Chapter 13. 118 Argentina?s trade deficit with Brazil since 2003 could be partially attributed to a surge in imports from Brazil that replaced American and European products. 64 These trade adjustments emerged after Argentina?s devaluation as suggested by an inverse J-curve. Trade diversion is trade that emerges between two or more countries within a regional trade agreement that replaces trade that otherwise would have taken place with more efficient non-members. 65 Arguably, devaluations intended to converge exchange rates of countries in a regional trading bloc may also divert trade from non-member to member countries. Figure 3.2: Shares of Argentina?s Imports Shares of Argentina's Imports 0.0000 0.0500 0.1000 0.1500 0.2000 0.2500 0.3000 0.3500 0.4000 1994: 3 1995: 2 1996: 1 1996: 4 1997: 3 1998: 2 1999: 1 1999: 4 2000: 3 2001: 2 2002: 1 2002: 4 2003: 3 2004: 2 Brazil US EU Figure 3.2 shows market shares of Brazilian, American, and European exports to Argentina. Note that Argentina?s imports from Europe were almost 35% of total imports in 1994 and stayed above 30% until the end of 1999. The European import share reached their lowest share at 20% for the period following the peso devaluation. Similarly, American imports maintained a share of 23% for the period preceding Argentina?s devaluation and then declined to 15% in 2004. On the other hand, Brazilian exports to Argentina maintained an average 23% share during the 1994-1999 period. Brazil?s 64 See model results in Chapter 2. The trade diversion variable was significant at the 5% level. 65 See Frankel (1997) for examples of papers measuring trade diversion and trade creation. 119 devaluation seems to have started a process of import diversion in Argentina favoring Brazilian goods. The share of Brazilian imports in Argentina grew to over 25% between 1999 and 2002. This growth trend accelerated after Argentina?s devaluation leading to a share that exceeded 35%. In this section, a gravity model examines trade diversion using a TSCS approach and a fixed effects (FE) model. 66 The TSCS approach captures heterogeneity among country-pairs with different specifications of the covariance matrix either with a common constant term as proposed by Greene (2003) or with country-specific time invariant effects (different intercepts for each country pair). The FE model captures the country- specific time invariant effects with the use of country-pair dummies and corrects for heteroscedasticity and autocorrelation (common ?). The goal is to estimate the impact of the peso devaluation on Argentina?s imports. If imports have been diverted from non- members to Brazil after devaluation, Argentina?s trade deficit with Brazil and the inverse J-curve found in Chapter 2 are a reasonable outcome of this policy. The data comprises three cross section units (three country-pairs) and 42 time series observations for each country-pair (126 total observations). Imports of country i from country j are the model?s dependent variable as in Frankel and Wei (1997). Equation (3.39) presents the gravity model testing for trade diversion effects. When heterogeneity is captured with different covariance specifications and a common intercept, the dummies Europe and Brazil are excluded from the equation. log M ijt = ? 0 + ? 1 Europe + ? 2 Brazil + ? 1 log Y it + ? 2 log Y jt + ? 3 log P it + ? 4 log P jt + ? 5 log REX ijt + ? 6 DR + ? 7 DB + ? 8 DNM + ? ijt (3.39) 66 See Section V in this chapter for details on the econometrics of these models. 120 where M ijt is Argentina?s imports from country j at time t in US dollars, Europe and Brazil are dummies capturing country-pair fixed effects, Y it is Argentina?s GDP at time t (index), Y jt is trading partner?s GDP at time t (index), P it is Argentina?s population at time t, and P jt is trading partner?s population at time t. 67 REX ijt is the bilateral real exchange rate between pairs of countries. Imports, incomes, populations, and real exchange rates are transformed to natural logarithms. DR is a dummy variable accounting for the effects of the temporary restrictions on payments for imports, the initial exchange rate uncertainty, and the banking restrictions implemented by the Argentine government in 2002. Finally, DB and DNM are multiplicative dummy variables that separate the effects of Argentina?s devaluation between member-country Brazil (DB) and non-members US and EU (DNM). These dummies test for structural or regime changes in Argentina. The sign of the coefficients for these dummies should reveal if trade has been diverted from the US and EU toward Brazil after the peso devaluation. Trade diversion is implied in the case that ? 7 is positive and ? 8 is either negative or insignificant and also in the case that ? 7 is positive or insignificant and ? 8 is negative. Even if ? 7 and ? 8 have the same sign, comparing the magnitude of the coefficients could reveal trade diversion effects as in Frankel (1997). 68 When TSCS estimates are robust across all nine possible combinations of Ss and Rs, the S2-R2 model with cross group correlation, groupwise heteroscedasticity, and different ?s for each country-pair is selected since it is the least restrictive of models. Table 3.1 presents the results from (3.39). Coefficients for all variables except dummies are read as elasticities. 67 Annual population estimates from the International Database of the US Census Bureau were interpolated to obtain quarterly estimates. 68 Frankel (1997) calls trade diversion the fact that after the 1995 Mexican devaluation, Mexican imports from US declined less than non-NAFTA imports from EU and Japan. 121 Table 3.1: Trade Diversion Variable TSCS with Common Intercept TSCS with Country-Pair Effects FE Model Intercept -12.761 (0.591) 24.822 (1.042) 15.701 (0.545) Europe -1.195*** (5.495) -1.405*** (3.827) Brazil 2.411*** (7.894) 2.770*** (5.044) Argentina?s GDP 2.785*** (9.464) 2.974*** (11.065) 2.836*** (13.053) Trading Partner?s GDP -1.860*** (6.031) -0.597 (1.496) -0.438 (0.734) Argentina?s Population 0.675 (0.517) -6.970*** (5.894) -7.257*** (4.613) Trading Partner?s Population 0.205*** (3.400) 4.766*** (7.115) 5.487*** (4.713) Real Exchange Rate 0.051 (0.473) 0.106 (1.517) 0.045 (0.412) Restrictions -0.367*** (2.680) -0.300*** (2.851) -0.392*** (3.109) Dev. Effect for Brazil 0.227* (1.797) 0.200** (2.008) 0.178 (1.563) Dev. Effect for Non-members -0.312** (2.462) -0.193** (2.117) -0.241* (1.887) Rho 0.167* (1.867) Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. The first column in Table 3.1 presents the results of a TSCS S2-R2 model capturing heterogeneity across country-pairs with different covariance specifications and common intercept. The second column shows estimates from a TSCS S2-R2 model capturing time invariant effects for each country-pair with the use of different intercepts. The last column in Table 3.1 presents the results of a fixed effects model that does not allow for different autocorrelation coefficients as in the TSCS S2-R2 model. Results indicate the presence of trade diversion. The model with common intercept shows that the peso devaluation in Argentina increased Brazilian imports while decreasing imports from non-Mercosur members US and EU. Specifically, a devalued 122 and floating peso increased imports from Brazil by 25% (e 0.227 -1 = 0.254) and reduced imports from non-members by 27% (e -0.312 -1 = -0.268). Estimates from the TSCS S2-R2 with country pair specific effects show also a positive impact of Argentina?s devaluation in the country?s demand for Brazilian products and a negative impact for goods imported from the US and EU. The effects are smaller in magnitude since this model captures time invariant effects that are country specific. After devaluation, imports from Brazil increased 22% and imports from non-member countries decreased by 17%. TSCS estimations are robust with regard to the different specifications of the covariance matrix. The FE model also generates results consistent with the hypothesis of trade diversion. This model estimates that Brazilian imports remained at pre-devaluation levels but it shows a 21% decrease in imports from non-member countries. Time invariant country-specific effects are important. The coefficient for Europe is negative, suggesting that factors such as distance, lack of common border, or not being part of Mercosur depressed Argentina?s imports from Europe for the time period under study. The coefficient for Brazil shows that Mercosur membership, common borders, and short geographic distances increase Brazilian exports to Argentina. Argentina?s GDP is positively related to the country?s imports showing the economy?s absorptive capacity. This positive coefficient for Argentina?s GDP is robust and does not change significantly in magnitude across all models in Table 3.1. An increase in Argentina?s GDP of 1% leads to a 2.8% increase in imports. Trading partners? GDP has no significant effect in Argentine imports (only the model specified with a common intercept shows a negative sign, significant at the 10% level). Both models capturing time invariant effects show that a larger population in Argentina leads to a 123 reduction in the country?s imports. As the country experiences population growth, it becomes more self-sufficient with a wider variety of goods being produced and less need for imports. Similarly, the larger the population of trading partners, the greater Argentina?s imports, which supports the idea that larger populations lead to more diversified economies and greater possibilities for exports. This is in line with Krugman?s Chamberlinian competitive model with scale economies. Once devaluation effects are captured by dummies, the real exchange rate has no significant effect on imports. The initial exchange rate uncertainty and the restrictions on payments for imports and capital mobility imposed by the Argentine government in 2002 reduced the amount of the country?s imports. This negative effect is robust across all models in Table 3.1. Appendix II shows models? residuals and Q-statistics suggesting non-white noise errors. The model in (3.39) treated Argentina?s real GDP as exogenous when in fact there are good reasons to believe that this variable is affected by the level of imports. Since Argentina?s imports are the model?s dependent variable, the significant effect of Argentina?s income on imports could be spurious and therefore biased due to simultaneity problems. Even though results from Ramsey?s Reset test show no specification bias, the gravity model is estimated by using an instrumental variables (IV) approach to replace Argentina?s real GDP in equation (3.39). 69 Results from this IV estimation are presented in Table 3.2. 69 The instrument is the predicted values of a regression that estimates Argentina?s real GDP as a function of a lagged dependent variable, Brazil?s real GDP, an economic activity estimator calculated by INDEC, time, and a dummy for 2002 capturing the deep contraction in Argentina?s GDP during that year. 124 Table 3.2: Trade Diversion with IV Variable TSCS with Common Intercept TSCS with Country-Pair Effects FE Model Intercept -42.735** (2.221) 2.784 (0.130) -23.324 (0.933) Europe -1.166*** (4.812) -1.568*** (4.621) Brazil 2.367*** (6.804) 2.943*** (5.750) Argentina?s GDP 3.233*** (12.700) 3.248*** (13.214) 3.201*** (15.555) Trading Partner?s GDP -1.866*** (5.705) -0.578 (1.419) -0.894* (1.717) Argentina?s Population 2.374** (2.054) -5.680*** (4.987) -5.473*** (3.935) Trading Partner?s Population 0.119* (1.670) 4.674*** (6.190) 5.920*** (5.475) Real Exchange Rate 0.045 (0.445) 0.105 (1.421) -0.003 (0.032) Restrictions -0.373*** (3.285) -0.308*** (3.347) -0.441*** (4.087) Dev. Effect for Brazil 0.047 (0.441) 0.071 (0.824) 0.007 (0.066) Dev. Effect for Non-members -0.432*** (3.885) -0.324*** (3.846) -0.398*** (3.579) Rho 0.289*** (3.334) Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. Using an instrument instead of Argentina?s GDP as an explanatory variable does not affect results significantly. Estimates in Table 3.2 show the occurrence of trade diversion favoring Brazilian products, since imports from non-Mercosur countries declined significantly and imports from Brazil stayed at pre-devaluation levels. The decline in non-member imports after the devaluation of the peso ranges from 28% for the TSCS model with country-pair fixed effects to 35% for the TSCS model with common intercept. The instrument for Argentina?s GDP is positively related to imports, showing again Argentina?s economy absorptive capacity. Trading partners? GDP now appears to be inversely related to imports in two out of three models. Population and restrictions 125 effects do not change from those in Table 3.1. The real exchange rate continues to have no effect on Argentina?s imports. A visual inspection of the models? residuals and Q-statistics suggest autocorrelation or misspecification problems. 70 An Omitted variable problem could arise since the previous models do not capture potential habit formation effects that are common in demand equations. 71 Therefore, a dynamic model using a lagged dependent variable is estimated as in Eichengreen and Irwin (1995) and Harris and M?ty?s (1998). It is shown in Appendix II that the TSCS model capturing time invariant effects specific to country-pairs and the FE model generate white noise residuals. This suggests that equation (3.39) should be estimated with a lagged dependent variable. Inferences are based on estimates appearing in the second and third column of Table 3.3. Lagged imports exert a positive and highly significant effect on current import flows. Both models capturing time invariant fixed effects generate estimates that are similar. Argentina?s GDP has a positive effect on imports at the 1% level, thus suggesting that a 1% increase in economic activity leads to 2.3% to 2.4% growth in imports from its major trading partners. Economic growth in Brazil, the US, and EU does not lead to higher Argentinean imports. The dummy variable Brazil remains positive, the coefficient on Argentina?s population continues to be negative, and the coefficient on trading partners? population stays positive. Again, the initial instability 70 See Appendix II. 71 Pollak (1970) defines a habit such that (i) past consumption influences current preferences and hence, current demand and (ii) a higher level of past consumption of a good implies, ceteris paribus, a higher level of present consumption of that good. 126 brought by devaluation coupled with the capital restrictions of 2002 have depressed Argentine imports. The real exchange rate still has no effect on imports. Table 3.3: Trade Diversion with IV and Lagged Imports Variable TSCS with Common Intercept TSCS with Country-Pair Effects FE Model Intercept -2.029 (0.154) 22.934 (1.198) 2.014 (0.097) Europe -0.637*** (2.813) -0.958*** (3.486) Brazil 1.344*** (3.972) 1.820*** (4.341) Argentina?s GDP 2.124*** (10.129) 2.344*** (10.828) 2.446*** (11.860) Trading Partner?s GDP -1.035*** (4.279) -0.218 (0.590) -0.486 (1.119) Argentina?s Population -0.036 (0.046) -4.558*** (4.794) -4.439*** (4.004) Trading Partner?s Population 0.056 (1.286) 2.615*** (3.599) 3.636*** (4.102) Real Exchange Rate -0.032 (0.474) 0.056 (0.899) -0.022 (0.284) Restrictions -0.165** (1.978) -0.146* (1.817) -0.235** (2.559) Dev. Effect for Brazil 0.156** (2.118) 0.175** (2.368) 0.102 (1.282) Dev. Effect for Non-members -0.120 (1.537) -0.090 (1.190) -0.181* (1.923) Lagged Imports 0.519*** (9.909) 0.399*** (7.168) 0.370*** (6.991) Rho 0.118 (1.311) Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. Estimates in Table 3.3 also show the presence of trade diversion. While the TSCS model capturing time invariant effects shows that imports from Brazil increased and non- Mercosur imports remained at pre-devaluation levels, the FE model shows that imports from non-members declined and those from Brazil remained at levels that prevailed during the fixed exchange rate regime. Since TSCS estimates are not robust, inferences are based on the FE model. According to FE estimates, devaluation in Argentina 127 depressed the country?s imports from the US and EU by 17%. Trade diversion impacts of devaluation are still implied since the 17% decline in non-members imports is compared to a lack of decline on Brazilian imports. As in Frankel (1997), one can conclude that after the devaluation of the peso Brazilian goods gained market share in Argentina relative to the US and EU. Based on the results from the different estimated models, trade diversion favoring imports from Brazil emerged as a consequence of the peso devaluation. However, further examination is required since dummy variables suggest the presence of structural changes but do not uncover the specific economic factors behind these changes. The next section presents a TSCS model that examines the determinants of Argentine imports and tests for the presence of a Linder effect. VII. The Linder Hypothesis and the Peso Devaluation Since devaluation reduces a country?s purchasing in international markets, trade adjustments could be demand driven. These theoretical demand-side adjustments and the composition of Argentine imports (mainly manufactures) suggest an investigation of Linder effects. 72 The Linder hypothesis suggests that trade in manufactures depends on demand structures which are mainly determined by per capita incomes. 73 The greater the difference between two countries? per capita incomes, the lower the amount of trade in manufactures between them. This theory has been tested empirically using gravity models of trade. The difference in per capita incomes between countries becomes the variable of interest. The hypothesis is that Argentina?s devaluation depressed the 72 See Chapter 1 for a description on the composition of imports. 73 See section on Linder hypothesis for details. 128 country?s per capita income generating changes in its demand structure. A negative coefficient for the Linder variable indicates the presence of a Linder effect. The Linder variable is created by first converting per capita incomes to US dollars and then taking the difference between countries. For example, Argentina?s real GDP in pesos is divided by the country?s population and then converted to US dollars by using the nominal exchange rate. Equation (3.40) shows the gravity model testing for Linder effects: log M ijt = ? 0 + ? 1 Europe+ ? 2 Brazil + ? 1 log Y it + ? 2 log Y jt + ? 3 log P it + ? 4 log P jt + ? 5 log REX ijt + ? 6 DR + ? 7 D + ? 8 log LINDER + ? ijt (3.40) where most of the variables are defined as in (3.39), D is a dummy that accounts for structural changes after devaluation, and LINDER is the difference in per capita income between countries. Equation (3.40) is estimated following the approach of Section VI. All models in Table 3.4 show the presence of a Linder effect. The TSCS model with common intercept indicates that a 1% increase in the difference between Argentina?s per capita income and that of its major trading partners reduce imports by 0.30%. When dummies are used to capture the country-specific time invariant effects, a 1% increase in the difference between per capita incomes leads to a reduction of 0.13% in imports. As in Section VI, membership in a regional integration agreement, and time invariant factors such as shorter distances, common borders, or cultural affinities increase trade. This is implied by a positive coefficient on the dummy Brazil, and a negative coefficient for Europe. Economic growth in Argentina leads to higher imports as previously shown with other models. Trading partners? GDP and the real exchange rate do not impact 129 significantly the level of Argentinean imports. The initial devaluation effects and the restrictions in capital mobility depressed imports. The negative coefficient on the dummy capturing any structural break after 2003 indicates that the newly devalued and floating peso has reduced Argentine imports. 74 Results also indicate that population growth in Argentina lower imports while population growth abroad leads to higher imports. None of these models generate white noise residuals. Table 3.4: Linder Effects Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. Numbers in parenthesis are t-statistics. 74 Only the FE model shows no significant structural break once restrictions were lifted and exchange rate instability diminished. Variable TSCS with Common Intercept TSCS with Country-Pair Effects FE Model Intercept 58.957*** (2.666) 0.279 (0.011) 5.956 (0.187) Europe -1.569*** (6.747) -1.941*** (4.770) Brazil 2.724*** (7.859) 3.398*** (5.362) Argentina?s GDP 2.216*** (8.768) 2.768*** (10.691) 2.641*** (11.080) Trading Partner?s GDP 0.665 (1.580) -0.831* (1.880) -0.366 (0.558) Argentina?s Population -4.057*** (2.992) -6.417*** (5.297) -8.354*** (4.717) Trading Partner?s Population 0.428*** (5.521) 5.706*** (7.891) 7.065*** (5.428) Real Exchange Rate 0.010 (0.092) 0.020 (0.173) 0.050 (0.331) Restrictions -0.536*** (4.122) -0.501*** (3.564) -0.431** (2.512) Devaluation -0.323*** (3.078) -0.265** (2.206) -0.112 (0.729) Linder -0.296*** (7.565) -0.127** (2.213) -0.128* (1.867) Rho 0.365 (4.325) 130 Equation (3.40) is estimated using an IV approach that corrects for potential endogeneity in Argentina?s GDP. For the most part, results are similar to those in Table 3.4 and are consistent with the Linder hypothesis. Only the TSCS S2-R2 model shows no significant Linder effects. However, this may arise due to over parameterization of the model. The fact that six out of the nine combinations of Ss and Rs show a significant Linder effect suggests over parameterization. 75 Table 3.5 summarizes the results. Again, none of these models generate white noise residuals. Table 3.5: Linder Effects with IV Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. Numbers in parenthesis are t-statistics. 75 See Appendix I for Limdep?s output with all nine models. Variable TSCS with Common Intercept TSCS with Country- Pair Effects FE Model Intercept 21.748 (0.996) -19.853 (0.852) -31.133 (1.127) Europe -1.560*** (6.414) -2.111*** (5.524) Brazil 2.836*** (7.708) 3.661*** (6.062) Argentina?s GDP 2.732*** (11.037) 3.178*** (12.470) 3.093*** (13.213) Trading Partner?s GDP 0.047 (0.107) -0.897** (2.039) -0.770 (1.344) Argentina?s Population -1.875 (1.394) -5.466*** (4.690) -6.837*** (4.318) Trading Partner?s Population 0.377*** (3.645) 5.778*** (7.573) 7.591*** (6.155) Real Exchange Rate 0.037 (0.302) 0.046 (0.407) 0.028 (0.207) Restrictions -0.487*** (3.929) -0.425*** (3.232) -0.437*** (3.009) Devaluation -0.383*** (3.702) -0.314*** (2.841) -0.247* (1.866) Linder -0.223*** (5.074) -0.070 (1.217) -0.101* (1.629) Rho 0.467*** (5.830) 131 The absorptive capacity of Argentina?s economy, the country-specific time invariant effects, population impacts, the effect of capital mobility restrictions with the exchange rate uncertainty of 2002, and the structural break caused by the new exchange rate regime closely resemble estimates in Table 3.4. Results indicate that increasing the difference in per capita income between countries by 1% reduces imports by 0.22% in the TSCS model with a common intercept, and by 0.10% in the FE model. The lack of white noise residuals suggests potential omitted variable problems. In order to account for any habit formation effect, a lagged dependent variable is added to this IV model. Table 3.6: Linder Effects with IV and Lagged Imports Variable TSCS with Common Intercept TSCS with Country- Pair Effects FE Model Intercept 37.807** (2.291) 3.157 (0.150) -3.117 (0.139) Europe -0.904*** (3.857) -1.221*** (4.178) Brazil 1.537*** (4.318) 1.987*** (4.366) Argentina?s GDP 1.955*** (9.597) 2.223*** (10.083) 2.325*** (10.654) Trading Partner?s GDP 0.116 (0.324) -0.551 (1.342) -0.460 (0.991) Argentina?s Population -2.638** (2.559) -3.976*** (3.999) -4.781*** (3.985) Trading Partner?s Population 0.170*** (2.957) 3.235*** (4.320) 4.265*** (4.514) Real Exchange Rate -0.024 (0.356) -0.021 (0.225) -0.066 (0.647) Restrictions -0.173** (2.120) -0.195* (1.841) -0.267** (2.250) Devaluation -0.060 (0.851) -0.074 (0.811) -0.099 (0.948) Linder -0.140*** (4.479) -0.070 (1.518) -0.108** (2.383) Lagged Imports 0.517*** (10.093) 0.459*** (8.271) 0.426*** (7.885) Rho 0.158* (1.766) Significance at the 10%, 5%, and 1% level are denoted by *, **, and *** respectively. Numbers in parenthesis are t-statistics. 132 Results in Table 3.6 show that regardless of model specification, factors such as longer distances, absence of preferential agreements, or lack of common borders reduce the amount of trade between Argentina and EU. As expected, the magnitude of the coefficients is smaller when lagged imports are used as an explanatory variable. On the other hand, Mercosur and time invariant factors such as Brazil?s common borders and relatively short distance with Argentina exert a positive impact on trade. The instrument replacing Argentina?s GDP is positive at the 1% level for different model specifications showing again the country?s absorptive power. Models capturing time invariant effects suggest that a 1% growth in Argentina?s economic activity leads to an increase of 2.2% in the country?s imports from Brazil, the US, and EU. Trading partner?s GDP has no influence on Argentine imports. Argentina?s population has a negative effect on imports and trading partner?s population is positively related to the level of Brazilian, European, and American exports to Argentina. Real exchange rates seem to have no impact on Argentine imports. Restrictions on import payments and capital mobility as well as the exchange rate uncertainty brought by devaluation worked as expected by reducing the level of the country?s imports. The coefficient for the dummy variable capturing any structural break for the pre- and post- devaluation period is insignificant. This is evidence in favor of the Linder hypothesis since any structural change in the demand for imports is explained by the differences in incomes per capita rather than a dummy variable. Lagged imports are positive and significant at the 1% level suggesting that habit formation effects are important when modeling demand for imports in Argentina. 133 Results suggest evidence of a Linder effect even after capturing country-pair time invariant effects and any structural break generated by devaluation or capital restrictions. Estimates support Linder?s proposition that the greater the difference between two countries? per capita incomes the lower the amount of trade between them. The coefficient for LINDER is negative at the 1% level when the TSCS model is estimated with a common intercept. This suggests that a 1% increase in the difference between Argentina?s per capita income and that of its major trading partners leads to 0.14% decrease in imports. These estimates are robust to different covariance specifications. Over parameterization may exist since the TSCS S2-R2 model capturing time invariant effects shows no significant Linder effects while all other eight combinations of Ss and Rs show a negative coefficient for LINDER. 76 Estimates from the FE model in the third column of Table 3.6 support Linder?s theory. At the 5% significance level, results indicate that a 1% increase in the difference between Argentina?s per capita income and that of its major trading partners leads to a 0.11% decrease in the country?s imports. There are (at least) four ways to model (parameterize) the pair-wise cross-country differential trade effects: (1) dummy variables (fixed effects), (2) different variances (random effects), (3) different autoregressive processes (different ?s), and (4) cross trading partner income differences (Linder effects). The above TSCS with common intercept approach and the FE results suggest that using up to any three of these parameterizations in a given specification produces very reasonable results. However, the TSCS model capturing country-specific effects indicate that when all four parameterizations are used the Linder effect is only marginally significant, if at all. To 76 See Appendix I in this chapter. 134 reinforce this notion, if Greene?s philosophy that problems such as the current one should be modeled with a common mean (Greene, p. 320) so that the fixed effects dummies can be dropped, then the TSCS model with common intercept in Table 3.6 confirms that the Linder effect is statistically significant in all specifications. In summary, modeling heterogeneity across countries with different covariance specifications and a common intercept shows significant Linder effects; in the FE model, which allows for groupwise heteroscedasticity and corrects for autocorrelation with a common ? for all trading country-pairs, the Linder effect is significant; and the TSCS estimates capturing countries? fixed effects shows in eight out of the nine possible specifications a significant Linder effect. Only in the TSCS S2-R2 model capturing time invariant effects and therefore incorporating all four parameterizations of the pair-wise cross-country differential trade effects, was the Linder pair-wise cross-country differential trade effects effect found to be statistically insignificant at traditional levels. Thus, it is reasonable to conclude that the Linder effect is in fact a statistically significant determinant of Argentinean imports, and any apparent insignificance is due to an over parameterization of pair-wise cross-country differential trade effects. Results suggest that the Linder hypothesis explains part of the trade diversion favoring Brazilian products. In sum, Argentina?s devaluation depressed the country?s per capita income generating changes in its demand structure, which seems now more closely related to its Mercosur partner Brazil. 135 VIII. Conclusions This chapter provides evidence of trade diversion effects after devaluation. Argentina?s imports have been diverted favoring Brazilian goods to the detriment of imports from non-Mercosur trading partners, and this diversion can be explained by Linder effects. Empirical results suggest that trade diversion may occur as a consequence of exchange rate adjustments and not necessarily as the direct result of the formation of a trading bloc. A currency devaluation that sets a country?s exchange rate more in line with regional trading partners can create trade diversion. Estimates support the idea that devaluation depressed Argentina?s per capita income to levels closer to Brazil?s and more distant from the US and EU. In sum, findings suggest that Argentina?s trade deficit with Brazil is a consequence of an import diversion process after a devaluation that generated drastic changes in relative per capita incomes. First, a gravity model separates the effects of Argentina?s devaluation on imports from Brazil and from non-Mercosur US and EU with dummy variables. Estimates from different specified models indicate trade diversion in favor of Brazilian products. Some estimates suggest a significant increase in Argentina?s imports from Brazil coupled with a decrease in non-member imports. Other estimates show a decline in non-member imports with no effects on Brazilian goods. In this case, trade diversion favoring Brazil is implied since lower imports from the US and EU have increased Brazil?s market share. Results are robust to the use of an instrumental variable approach and when adding a lagged dependent variable. Examination of model residuals suggests a dynamic IV model is more appropriate. 136 Finally, a gravity model focuses on the Linder hypothesis as a possible explanation of this trade diversion. Two main reasons make the case under study appropriate for a test of the Linder hypothesis. First, Argentina?s imports are composed mainly of manufactures as shown in Chapter 1. Second, the peso devaluation has depressed Argentina?s per capita income to levels that are much closer to Brazil?s. In other words, the peso devaluation lowered Argentina?s buying power in international markets, making the country?s demand structure more similar to that of Brazil. This section presents a gravity model of trade that tests the presence of a Linder effect that could explain the trade adjustments under examination. A Linder effect is evident suggesting that devaluation depressed Argentina?s per capita income affecting its import demand structure. Results suggest that every 1% increase in the difference between Argentina?s per capita income and those of its major trading partners leads to 0.11% decrease in imports. The trade diversion effects of devaluation are a consequence of a change in Argentina?s import demand structure. This becomes a reasonable explanation for the inverse J-curve found in Chapter 2 and for Argentina?s post-devaluation trade deficit with Brazil. 137 APPENDIX I Output for TSCS Model with Country-Pair Fixed Effects in Table 3.5 Constant: intercept; DEU: EU?s fixed effects; DB: Brazil?s fixed effects; IV: instrument for Argentina?s GDP; LRGDPTP: trading partners? GDP; LPA: Argentina?s population; LPTP: trading partners? populations; LREXCH: real exchange rates; DR: dummy for restrictions and initial exchange rate instability; D: dummy testing for structural break; LLINDER: Linder effects. +-------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Homoskedastic Regression (S0) | | Nonautocorrelated disturbances (R0) | | Pooled OLS residual variance (SS/nT) .0139| | Test statistics for homoscedasticity: | | Deg.Fr. = 2 C*(.95) = 5.99 C*(.99) = 9.21 | | Lagrange multiplier statistic = .4062 | | Log-likelihood function = 88.038328 | +-------------------------------------------+ +---------+--------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+--------------+--------+---------+ Constant -51.48850340 -2.007 .0447 DEU -2.214822839 -7.482 .0000 DB 3.522114733 7.859 .0000 IV 3.436856862 17.074 .0000 LRGDPTP -1.487776687 -2.798 .0051 LPA -5.581821569 -4.074 .0000 LPTP 7.640157113 8.202 .0000 LREXCH -.9338096301E-01 -.790 .4294 DR -.5436679333 -4.046 .0001 D -.3305538788 -2.819 .0048 LLINDER -.1647341573 -3.235 .0012 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. Regression (S1) | | Nonautocorrelated disturbances (R0) | | Test statistics for homoscedasticity: | | Deg.Fr. = 2 C*(.95) = 5.99 C*(.99) = 9.21| | Wald statistic = .7921 | | Likelihood ratio statistic = .6698 | | Test statistics against the correlation | | Lagrange multiplier statistic = 18.9285 | | Log-likelihood function = 88.373244 | +------------------------------------------+ +---------+--------------+--------+--------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z]| +---------+--------------+--------+--------+ Constant -54.66486465 -2.072 .0383 DEU -2.203326600 -7.481 .0000 DB 3.496215899 7.913 .0000 IV 3.420367714 17.127 .0000 LRGDPTP -1.589000719 -2.918 .0035 LPA -5.313355075 -3.861 .0001 LPTP 7.589312657 8.235 .0000 LREXCH -.9136991987E-01 -.756 .4497 DR -.5379784140 -3.967 .0001 D -.3365355759 -2.834 .0046 LLINDER -.1600420348 -3.059 .0022 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. and Correlated (S2) | | Nonautocorrelated disturbances (R0) | | Test statistics against the correlation | | Deg.Fr.= 3 C*(.95) = 7.81 C*(.99) = 11.34| | Test statistics against the correlation | | Likelihood ratio statistic = 32.3206 | | Log-likelihood function = 104.533556 | +------------------------------------------+ +---------+--------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+--------------+--------+---------+ Constant -65.22837805 -2.815 .0049 DEU -2.047727145 -9.245 .0000 DB 3.153428153 10.204 .0000 IV 3.531696643 14.800 .0000 LRGDPTP -1.895733396 -3.917 .0001 LPA -3.945576954 -3.396 .0007 LPTP 6.962724639 10.510 .0000 LREXCH -.1621153580 -1.453 .1462 DR -.6135926526 -4.434 .0000 D -.3905205474 -3.395 .0007 LLINDER -.1744948964 -3.287 .0010 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Homoskedastic Regression (S0) | | Common autocorrelation (R1) | | Autocorrelation coeff. r = .30048 | | Pooled OLS residual variance (SS/nT).0121| | Corrected residual var.= (s2/(1-r2) .0133| | Test statistics for homoscedasticity: | | Deg.Fr.= 2 C*(.95) = 5.99 C*(.99) = 9.21 | | Lagrange multiplier statistic = .2224 | | Log-likelihood function = 96.824261 | +------------------------------------------+ +---------+--------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+--------------+--------+---------+ Constant -29.30068623 -1.111 .2665 DEU -2.049038471 -5.665 .0000 DB 3.534286741 6.200 .0000 IV 3.106245792 14.079 .0000 LRGDPTP -.8516051351 -1.564 .1178 LPA -6.664164690 -4.479 .0000 LPTP 7.361089326 6.311 .0000 LREXCH .3816237421E-01 .301 .7632 DR -.4257695960 -3.071 .0021 D -.2380300682 -1.897 .0578 LLINDER -.1046321027 -1.788 .0738 138 +-----------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. Regression (S1)| | Common autocorrelation (R1)| | Autocorrelation coeff. r = .30048 | | Test statistics for homoscedasticity: | | Deg.Fr. = 2 C*(.95) = 5.99 C*(.99) =9.21| | Wald statistic = .3073 | | Likelihood ratio statistic = .2781 | | Test statistics against the correlation | | Lagrange multiplier statistic = 18.9790 | | Log-likelihood function = 96.963303 | +-----------------------------------------+ +---------+--------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+--------------+--------+---------+ Constant -30.45959809 -1.161 .2455 DEU -2.073470579 -5.821 .0000 DB 3.580267343 6.370 .0000 IV 3.112024229 14.134 .0000 LRGDPTP -.8472722888 -1.568 .1168 LPA -6.697070184 -4.567 .0000 LPTP 7.446380250 6.484 .0000 LREXCH .3097006680E-01 .245 .8063 DR -.4320181646 -3.117 .0018 D -.2404074324 -1.909 .0562 LLINDER -.1022020250 -1.764 .0778 +-------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. and Correlated (S2) | | Common autocorrelation (R1) | | Autocorrelation coeff. r = .30048 | | Test statistics against the correlation | | Deg.Fr. = 3 C*(.95) = 7.81 C*(.99) = 11.34| | Test statistics against the correlation | | Likelihood ratio statistic = 25.5855 | | Log-likelihood function = 109.756048 | +-------------------------------------------+ +---------+--------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+--------------+--------+---------+ Constant -22.16508109 -.902 .3669 DEU -1.785021674 -7.093 .0000 DB 3.121988981 8.316 .0000 IV 3.195241630 11.519 .0000 LRGDPTP -.7236082904 -1.492 .1357 LPA -6.158654349 -4.618 .0000 LPTP 6.491619773 8.317 .0000 LREXCH .2506924037E-01 .213 .8315 DR -.4461429597 -3.090 .0020 D -.2320266335 -1.806 .0709 LLINDER -.1111659187 -2.102 .0356 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Homoskedastic Regression (S0) | | Group specific autocorrelation (R2) | | Autocorrelation coefficients: | | .635 .080 .186 | | Pooled OLS residual variance (SS/nT).0105| | Test statistics for homoscedasticity: | | Deg.Fr. = 2 C*(.95) = 5.99 C*(.99) = 9.21| | Lagrange multiplier statistic = .2523 | | Log-likelihood function = 105.381853 | +------------------------------------------+ +---------+--------------+--------+--------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z]| +---------+--------------+--------+--------+ Constant -22.58413029 -.898 .3694 DEU -1.698841609 -5.065 .0000 DB 3.095530181 5.940 .0000 IV 3.142792803 15.453 .0000 LRGDPTP -.9900509139 -2.041 .0412 LPA -5.850580030 -4.641 .0000 LPTP 6.282933591 5.829 .0000 LREXCH .1224329178 .963 .3355 DR -.3426319367 2.584 .0098 D -.2334961153 -2.012 .0442 LLINDER -.4738260781E-01 -.738 .4606 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. Regression (S1) | | Group specific autocorrelation (R2) | | Autocorrelation coefficients: | | .635 .080 .186 | | Test statistics for homoscedasticity: | | Deg.Fr. = 2 C*(.95) = 5.99 C*(.99) = 9.21| | Wald statistic = .3903 | | Likelihood ratio statistic = .3218 | | Test statistics against the correlation | | Lagrange multiplier statistic = 18.6963 | | Log-likelihood function = 105.542770 | +------------------------------------------+ +---------+--------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+--------------+--------+---------+ Constant -20.86407858 -.852 .3940 DEU -1.711805614 -5.094 .0000 DB 3.107752920 5.923 .0000 IV 3.141020155 15.330 .0000 LRGDPTP -.9253603152 -1.957 .0503 LPA -6.009467756 -4.766 .0000 LPTP 6.325636412 5.840 .0000 LREXCH .1243844649 1.000 .3173 DR -.3413836872 -2.609 .0091 D -.2287966047 -1.999 .0456 LLINDER -.5549234217E-01 -.887 .3751 139 +-----------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. and Correlated (S2)| | Group specific autocorrelation (R2)| | Autocorrelation coefficients: | | .635 .080 .186 | | Test statistics against the correlation | | Deg.Fr.= 3 C*(.95) = 7.81 C*(.99)= 11.34| | Test statistics against the correlation | | Likelihood ratio statistic = 25.0810 | | Log-likelihood function = 118.083280 | +-----------------------------------------+ +---------+-------------+--------+--------+ |Variable |Coefficient |b/St.Er.|P[|Z|>z]| +---------+-------------+--------+--------+ Constant -19.85336430 -.852 .3941 DEU -1.560062126 -6.414 .0000 DB 2.836483244 7.708 .0000 IV 3.177866078 12.470 .0000 LRGDPTP -.8973043201 -2.039 .0415 LPA -5.465667793 -4.690 .0000 LPTP 5.778095087 7.573 .0000 LREXCH .4624001189E-01 .407 .6839 DR -.4249076912 -3.232 .0012 D -.3136158694 -2.841 .0045 LLINDER -.7002352755E-01 -1.217 .2236 Output for TSCS Model with Country-Pair Fixed Effects in Table 3.6 Constant: intercept; DEU: EU?s fixed effects; DB: Brazil?s fixed effects; IV: instrument for Argentina?s GDP; LRGDPTP: trading partners? GDP; LPA: Argentina?s population; LPTP: trading partners? populations; LREXCH: real exchange rates; DR: dummy for restrictions and initial exchange rate instability; D: dummy testing for structural break; LLINDER: Linder effects; LLM: lagged imports. +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Homoskedastic Regression (S0) | | Nonautocorrelated disturbances (R0) | | Pooled OLS residual variance (SS/nT).0085| | Test statistics for homoscedasticity: | | Deg.Fr.= 2 C*(.95) = 5.99 C*(.99) = 9.21 | | Lagrange multiplier statistic = 3.8898 | | Log-likelihood function = 118.587125 | +------------------------------------------+ +---------+--------------+--------+--------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z]| +---------+--------------+--------+--------+ Constant -4.328539165 -.209 .8346 DEU -1.186839605 -4.587 .0000 DB 1.880526469 4.747 .0000 IV 2.299379938 11.328 .0000 LRGDPTP -.5670874650 -1.324 .1854 LPA -4.491064980 -4.169 .0000 LPTP 4.093817757 4.929 .0000 LREXCH -.7888716608E-01 -.854 .3929 DR -.2599494720 -2.369 .0179 D -.8929186880E-01 -.934 .3501 LLINDER -.1146430845 -2.853 .0043 LLM .4491282638 8.862 .0000 +-----------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. Regression (S1)| | Nonautocorrelated disturbances (R0)| | Test statistics for homoscedasticity: | | Deg.Fr.= 2 C*(.95) = 5.99 C*(.99) = 9.21| | Wald statistic = 11.1897 | | Likelihood ratio statistic = 5.7117 | | Test statistics against the correlation | | Lagrange multiplier statistic = 10.0028 | | Log-likelihood function = 121.442980 | +-----------------------------------------+ +---------+------------+--------+---------+ |Variable | Coefficient|b/St.Er.|P[|Z|>z] | +---------+------------+--------+---------+ Constant -13.43228469 -.628 .5303 DEU -1.199427084 -4.921 .0000 DB 1.864988943 5.043 .0000 IV 2.223133585 11.505 .0000 LRGDPTP -.7492502318 -1.718 .0857 LPA -3.894243767 -3.722 .0002 LPTP 4.079513688 5.255 .0000 LREXCH -.1182971194 -1.244 .2135 DR -.2827611220 -2.592 .0095 D -.1181493207 -1.212 .2254 LLINDER -.1093501724 -2.682 .0073 LLM .4701993929 9.549 .0000 140 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. and Correlated (S2) | | Nonautocorrelated disturbances (R0) | | Test statistics against the correlation | | Deg.Fr.= 3 C*(.95) = 7.81 C*(.99) = 11.34| | Test statistics against the correlation | | Likelihood ratio statistic = 11.0771 | | Log-likelihood function = 126.981512 | +------------------------------------------+ +---------+-------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+-------------+--------+---------+ Constant -6.485590308 -.310 .7568 DEU -1.039327388 -4.798 .0000 DB 1.640163100 5.107 .0000 IV 2.158760077 9.728 .0000 LRGDPTP -.6818866959 -1.676 .0938 LPA -3.743287432 -3.777 .0002 LPTP 3.568682539 5.213 .0000 LREXCH -.8206692881E-01 -.907 .3643 DR -.2309253188 -2.108 .0351 D -.6742807191E-01 -.688 .4916 LLINDER -.9512367887E-01 -2.436 .0149 LLM .4971602796 9.137 .0000 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Homoskedastic Regression (S0) | | Common autocorrelation (R1) | | Autocorrelation coeff. r = .11594 | | Pooled OLS residual variance (SS/nT).0084| | Corrected residual var.= (s2/(1-r2).0085 | | Test statistics for homoscedasticity: | | Deg.Fr.= 2 C*(.95) = 5.99 C*(.99) = 9.21 | | Lagrange multiplier statistic = 4.0128 | | Log-likelihood function = 119.295641 | +------------------------------------------+ +---------+-------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z]| +---------+-------------+--------+---------+ Constant -3.513507828 -.165 .8690 DEU -1.211780853 -4.370 .0000 DB 1.966239592 4.561 .0000 IV 2.321325946 11.190 .0000 LRGDPTP -.4773424870 -1.081 .2798 LPA -4.712384387 -4.150 .0000 LPTP 4.228882511 4.720 .0000 LREXCH -.6717585366E-01 -.691 .4893 DR -.2685893260 -2.384 .0171 D -.1009231434 -1.019 .3084 LLINDER -.1095868300 -2.544 .0110 LLM .4259261602 8.273 .0000 +-----------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. Regression (S1)| | Common autocorrelation (R1)| | Autocorrelation coeff. r = .11594 | | Test statistics for homoscedasticity: | | Deg.Fr.= 2 C*(.95) = 5.99 C*(.99) = 9.21| | Wald statistic = 10.3369 | | Likelihood ratio statistic = 5.5414 | | Test statistics against the correlation | | Lagrange multiplier statistic = 8.4777 | | Log-likelihood function = 122.066340 | +-----------------------------------------+ +---------+------------+--------+---------+ |Variable | Coefficient|b/St.Er.|P[|Z|>z] | +---------+------------+--------+---------+ Constant -9.385832925 -.432 .6654 DEU -1.213516720 -4.675 .0000 DB 1.958273975 4.872 .0000 IV 2.241131371 11.324 .0000 LRGDPTP -.5659917893 -1.275 .2024 LPA -4.316529526 -3.978 .0001 LPTP 4.202782322 5.036 .0000 LREXCH -.9944158334E-01 -1.007 .3139 DR -.2815969539 -2.522 .0117 D -.1166766043 -1.159 .2463 LLINDER -.1003391581 -2.346 .0190 LLM .4514300650 9.021 .0000 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. and Correlated (S2) | | Common autocorrelation (R1) | | Autocorrelation coeff. r = .11594 | | Test statistics against the correlation | | Deg.Fr.= 3 C*(.95) = 7.81 C*(.99) = 11.34| | Test statistics against the correlation | | Likelihood ratio statistic = 9.2445 | | Log-likelihood function = 126.688598 | +------------------------------------------+ +---------+-------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+-------------+--------+---------+ Constant -.7858523221 -.037 .9708 DEU -1.054469711 -4.566 .0000 DB 1.748903148 4.993 .0000 IV 2.206686846 9.755 .0000 LRGDPTP -.4598841120 -1.100 .2711 LPA -4.300398013 -4.136 .0000 LPTP 3.714137359 5.040 .0000 LREXCH -.5262695206E-01 -.554 .5799 DR -.2244885564 -1.980 .0477 D -.5779015481E-01 -.563 .5732 LLINDER -.8592608030E-01 -2.115 .0344 LLM .4697544127 8.504 .0000 141 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Homoskedastic Regression (S0) | | Group specific autocorrelation (R2) | | Autocorrelation coefficients: | | .337 -.181 .191 | | Pooled OLS residual variance (SS/nT).0078| | Test statistics for homoscedasticity: | | Deg.Fr.= 2 C*(.95) = 5.99 C*(.99) = 9.21 | | Lagrange multiplier statistic = 2.6841 | | Log-likelihood function = 124.156140 | +------------------------------------------+ +---------+-------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+-------------+--------+---------+ Constant 2.453876151 .119 .9054 DEU -1.073684103 -3.859 .0001 DB 1.796759495 4.164 .0000 IV 2.346478696 11.748 .0000 LRGDPTP -.5092769136 -1.203 .2292 LPA -4.600952542 -4.234 .0000 LPTP 3.817546954 4.244 .0000 LREXCH -.1227996339E-01 -.126 .8998 DR -.2110490463 -1.973 .0486 D -.8062070258E-01 -.878 .3798 LLINDER -.8876857445E-01 -1.901 .0573 LLM .4148341393 8.162 .0000 +-----------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. Regression (S1)| | Group specific autocorrelation (R2)| | Autocorrelation coefficients: | | .337 -.181 .191 | | Test statistics for homoscedasticity: | | Deg.Fr.= 2 C*(.95) = 5.99 C*(.99) = 9.21| | Wald statistic = 6.6433 | | Likelihood ratio statistic = 3.8838 | | Test statistics against the correlation | | Lagrange multiplier statistic = 9.8574 | | Log-likelihood function = 126.098059 | +-----------------------------------------+ +---------+------------+--------+---------+ |Variable |Coefficient |b/St.Er.|P[|Z|>z] | +---------+------------+--------+---------+ Constant -3.419443371 -.159 .8739 DEU -1.048838745 -3.961 .0001 DB 1.732512011 4.249 .0000 IV 2.279375245 11.780 .0000 LRGDPTP -.6528403904 -1.497 .1344 LPA -4.077125930 -3.879 .0001 LPTP 3.689520309 4.338 .0000 LREXCH -.4618038487E-01 -.458 .6467 DR -.2306228000 -2.147 .0318 D -.1000224029 -1.068 .2854 LLINDER -.8177022979E-01 -1.689 .0911 LLM .4371750965 8.716 .0000 +------------------------------------------+ | Groupwise Regression Models | | Estimator = 2 Step GLS | | Groupwise Het. and Correlated (S2) | | Group specific autocorrelation (R2) | | Autocorrelation coefficients: | | .337 -.181 .191 | | Test statistics against the correlation | | Deg.Fr.= 3 C*(.95) = 7.81 C*(.99) = 11.34| | Test statistics against the correlation | | Likelihood ratio statistic = 10.4309 | | Log-likelihood function = 131.313516 | +------------------------------------------+ +---------+-------------+--------+---------+ |Variable | Coefficient |b/St.Er.|P[|Z|>z] | +---------+-------------+--------+---------+ Constant 3.157003063 .150 .8809 DEU -.9038215936 -3.857 .0001 DB 1.536627635 4.318 .0000 IV 2.223201630 10.083 .0000 LRGDPTP -.5512611447 -1.342 .1796 LPA -3.975726923 -3.999 .0001 LPTP 3.235175268 4.320 .0000 LREXCH -.2136747710E-01 -.225 .8219 DR -.1954216562 -1.841 .0656 D -.7415923703E-01 -.811 .4173 LLINDER -.7037746598E-01 -1.518 .1290 LLM .4589651116 8.271 .0000 142 APPENDIX II Figure 3.3: Residuals from TSCS Model with Common Intercept in Table 3.1 Residuals TSCS Common Intercept -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 04182123 Observations R e s idual s Table 3.7: Q-statistics TSCS Common Intercept Table 3.1 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 72.50 6 <.0001 12 82.68 12 <.0001 18 89.07 18 <.0001 24 151.20 24 <.0001 Figure 3.4: Residuals from TSCS Model with Fixed Effects in Table 3.1 Residuals OLS TSCS Fixed-Effects -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 41 82 123 Observations R e s idu al s Table 3.8: Q-statistics TSCS Fixed Effects Table 3.1 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 21.52 6 0.0015 12 44.88 12 <.0001 18 52.36 18 <.0001 24 66.49 24 <.0001 Figure 3.5: Residuals from FE Model in Table 3.1 Residuals Fixed-Effects -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 41 82 123 Observations R es id u a ls Table 3.9: Q-statistics FE Model in Table 3.1 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 17.34 6 0.0081 12 35.16 12 0.0004 18 40.40 18 0.0018 24 56.76 24 0.0002 143 Figure 3.6: Residuals from TSCS IV Model with Common Intercept Table 3.2 Residuals IV -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 04182123 Observations Re s id u a ls Table 3.10: Q-statistics TSCS IV Model with Common Intercept in Table 3.2 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 142.38 6 <.0001 12 167.54 12 <.0001 18 189.79 18 <.0001 24 284.61 24 <.0001 Figure 3.7: Residuals from TSCS IV Model with FE in Table 3.2 Residuals IV TSCS FE -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 04182123 Observations R e s idua ls Table 3.11: Q-statistics TSCS IV Model with FE in Table 3.2 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 22.08 6 0.0012 12 26.11 12 0.0103 18 33.84 18 0.0132 24 36.20 24 0.0520 Figure 3.8: Residuals from FE Model in Table 3.2 Residuals IV FE -0.3 -0.2 -0.1 0 0.1 0.2 0.3 04182123 Observations R e s idual s Table 3.12: Q-statistics FE Model in Table 3.2 Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 18.20 6 0.0058 12 23.54 12 0.0235 18 30.33 18 0.0343 24 35.19 24 0.0656 144 Figure 3.9: Residuals from TSCS IV Model with Common Intercept and Lagged Imports in Table 3.3 Residuals IV with Lagged Imports -0.3 -0.2 -0.1 0 0.1 0.2 0.3 04182123 Observations R e s idual s Table 3.13: Q-statistics TSCS IV Model with Common Intercept and Lagged Imports in Table 3.3 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 13.16 6 0.0406 12 20.41 12 0.0597 18 25.66 18 0.1077 24 40.83 24 0.0174 Figure 3.10: Residuals from TSCS IV Model with FE and Lagged Imports Table 3.3 Residuals IV with Lagged Imports TSCS FE -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0 41 82 123 Observations R es id u a ls Table 3.14: Q-statistics TSCS IV Model FE and Lagged Imports Table 3.3 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 4.21 6 0.6480 12 7.71 12 0.8073 18 8.37 18 0.9727 24 10.24 24 0.9935 Figure 3.11: Residuals from IV Model with FE and Lagged Imports Table 3.3 Residuals IV with Lagged Imports FE -0.3 -0.2 -0.1 0 0.1 0.2 0.3 04182123 Observations R es idua ls Table 3.15: Q-statistics IV Model FE and Lagged Imports Table 3.3 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 5.16 6 0.5229 12 7.95 12 0.7888 18 8.86 18 0.9630 24 10.48 24 0.9923 145 Figure 3.12: Residuals from Linder TSCS Model with Common Intercept Table 3.4 Residuals Linder Effects OLS -0.6 -0.4 -0.2 0 0.2 0.4 0.6 04182123 Observations R es idual s Table 3.16: Q-statistics Linder TSCS Model with Common Intercept Table 3.4 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 102.29 6 <.0001 12 114.21 12 <.0001 18 121.63 18 <.0001 24 168.47 24 <.0001 Figure 3.13: Residuals from Linder TSCS Model with FE Table 3.4 Residuals Linder Effects TSCS FE -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 04182123 Observations R e s id ual s Table 3.17: Q-statistics Linder TSCS Model with FE Table 3.4 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 38.83 6 <.0001 12 54.38 12 <.0001 18 62.07 18 <.0001 24 79.28 24 <.0001 Figure 3.14: Residuals from Linder Model with FE Table 3.4 Residuals Linder Effects FE -0.6 -0.4 -0.2 0 0.2 0.4 0 41 82 123 Observations R e s idu al s Table 3.18: Q-statistics Linder Model with FE Table 3.4 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 35.02 6 <.0001 12 45.59 12 <.0001 18 50.07 18 <.0001 24 83.58 24 <.0001 146 Figure 3.15: Residuals Linder TSCS IV Model with Common Intercept Table 3.5 Residuals Linder Effects IV -0.4 -0.2 0 0.2 0.4 0.6 04182123 Observations R e s idual s Table 3.19: Q-statistics Linder TSCS IV Model with Common Intercept Table 3.5 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 126.74 6 <.0001 12 127.76 12 <.0001 18 142.93 18 <.0001 24 191.35 24 <.0001 Figure 3.16: Residuals Linder TSCS IV Model with FE Table 3.5 Residuals Linder IV TSCS FE -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 04182123 Observations R e s id ual s Table 3.20: Q-statistics Linder TSCS IV Model with FE Table 3.5 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 53.84 6 <.0001 12 55.28 12 <.0001 18 58.45 18 <.0001 24 64.42 24 <.0001 Figure 3.17: Residuals Linder IV Model with FE Table 3.5 Residuals Linder IV FE -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 41 82 123 Observations R e s idu al s Table 3.21: Q-statistics Linder IV Model with FE Table 3.5 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 51.06 6 <.0001 12 53.73 12 <.0001 18 65.15 18 <.0001 24 91.39 24 <.0001 147 Figure 3.18: Residuals Linder TSCS IV Model with Common Intercept and Lagged Imports Table 3.6 Residuals Linder IV and Lagged Imports -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0 41 82 123 Observations R e s idu al s Table 3.22: Q-statistics Linder TSCS IV Model with Common Intercept and Lagged Imports Table 3.6 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 17.53 6 0.0075 12 22.45 12 0.0328 18 30.65 18 0.0316 24 52.57 24 0.0007 Figure 3.19: Residuals Linder TSCS IV Model with FE and Lagged Imports in Table 3.6 Residuals Linder IV and Lagged Imports TSCS FE -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 04182123 Observations R e s id ual s Table 3.23: Q-statistics Linder TSCS IV Model with FE and Lagged Imports in Table 3.6 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 8.71 6 0.1903 12 14.47 12 0.2715 18 20.39 18 0.3112 24 28.49 24 0.2399 Figure 3.20: Residuals Linder IV Model with FE and Lagged Imports Table 3.6 Residuals Linder IV and Lagged Imports FE -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 04182123 Observations R e s idual s Table 3.24: Q-statistics Linder IV Model with FE and Lagged Imports Table 3.6 To Lag Chi-Square Degrees of Freedom Prob. > Chi-Square 6 7.23 6 0.2998 12 11.44 12 0.4915 18 16.43 18 0.5625 24 23.00 24 0.5200 148 APPENDIX III Literature Review for Gravity Models Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Theoretical and Empirical Linder (1961) Graphical approach by plotting countries? average propensity to import from other countries 32 countries 1958 Proposes an alternative to the HOS theory by claiming that trade is determined by demand rather than supply. He argues that the more similar the demand structure between two countries, the more intensive their bilateral trade in manufactures. Linder suggests that the average level of income is the single and most important determinant of a country?s demand structure. Therefore, he suggests that differences in per capita income are an obstacle to trade. Empirical Tinbergen (1962) OLS Cross- Section E ij : value of exports from country i to country j Y i : country?s i GNP; Y j : country?s j GNP; D ij is the distance between countries i and j; and dummies for Adjacency and Regional Integration 18 countries, 42 countries 1958 and 1959 Economic size and distance are the main factors explaining trade flows. The only dummy variable that is positive and significantly different from zero is the one representing the Commonwealth preference. Significant deviations between actual and expected trade flows suggest the presence of discriminatory trade barriers. The reasons for these deviations are preferential treatment of a country?s exports, utilization of previously accumulated foreign exchange in the case of imports, or a net inflow of capital. Negative deviations are based on discriminatory treatment of exports, import restrictions, or net outflows of capital. Theoretical and Empirical Linnemann (1966) OLS Cross- Section Xij: the trade flows from country i to country j in logs. Y i and Y j are GNPs; N i and N j are populations; P ij is a preferential trade factor; and D ij is the distance between countries 80 1958-1960 Findings suggest a positive relationship between GNP and trade flows, a negative relationship between trade and population, a negative relationship between natural trade barriers (distance) and trade flows, a substantial effect of preferential trade arrangements, and a trade regulating effect of trading partners? commodity composition of exports and imports. The paper derives the gravity equation from a Walrasian type general model. 149 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Empirical Aitken (1973) OLS Cross- Section: yearly regressions Xij: log of the dollar value of exports from country i to country j Y i and Y j are GNPs; N i and N j are populations; P ij is a preferential trade factor (EEC and EFTA); D ij is the distance between countries; Aij is a dummy for adjacency Belgium- Luxembourg, France, Germany, Italy, Netherlands, Austria, Denmark, Norway, Portugal, Sweden, Switzerland, UK 1951-1967 EEC generated gross trade creation effects that are greater than those generated by EFTA. Findings suggest that EEC had a net external trade creation effect on EFTA through 1964 that was offset by a growing net trade diversion effect from 1965 to 1967. Results imply that 1958 is the last year for which it is safe to assume that European trade flows were not affected by the EEC. Empirical. Tests the Linder Hypothesis Hirsch and Lev (1973). OLS Cross- Section Xijk: log of exports of commodity k from country i to country j GNPs, distance, a dummy for preferential trade, and a variable indicating per capita income differential Denmark, Netherlands, Israel, and Switzerland 1966 Hirsch and Lev find results that are consistent with Linder?s hypothesis. Most coefficients for the income differential variable are negative and significant, implying that the greater the difference in per capita income, the lower the volume of trade between two countries. Theoretical Anderson (1979) Uses expenditures systems with identical homothetic Cobb-Douglas preferences across countries and with products that are differentiated by place of origin. Using a pure Cobb-Douglas expenditure system model, Anderson presents the simplest possible gravity equation assuming that each country specializes in the production of one good, and no tariffs or transportation costs exist. The paper explains the multiplicative form of the equation; permits an interpretation of the distance in the equation; implies that the usual estimator of the gravity equation may be biased. Theoretical Krugman (1979) Explains trade flows as a function of economies of scale instead of factor endowments or technology. He assumes that scale economies are internal to firms with a market structure that follows a Chamberlinian monopolistic competitive market where firms have some monopoly power but entry drives monopoly profits to zero. The paper shows that trade need not be a result of international differences in technology or factor endowments. Krugman argues that trade is a way of extending a domestic market allowing for scale economies. 150 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Theoretical Krugman (1980) Shows that transportation costs have no effects on firms? pricing policies and also no effects on output and number of firms. He also finds that countries with larger domestic markets tend to have higher wage rates. Finally, Krugman argues that countries with higher domestic markets on specific goods will tend to export those goods. Overall, findings lead to implications for the pattern of trade that are in line with Linder (1961). Theoretical Krugman (1981) This paper proposes that the usual forces of comparative advantage operate on ?groups of products? giving rise to inter-industry specialization and trade. On the other hand, scale economies lead each country to produce only a subset of goods within a group, originating intra-industry trade. Krugman argues that the gains from larger markets when countries are similar have outweighed any income distribution problem. He concludes that similar countries have an incentive to trade, that this trade will be mainly on goods that use similar production factors, and that this intra-industry trade will not generate income distribution problems that usually arise with inter- industry trade. Theoretical Helpman and Krugman (1985) Demand for variety drives consumer expenditures and monopolistic competitive firms produce differentiated products. The authors argued that the Heckscher-Ohlin theory does not have the property that bilateral trade depends on the product of GDPs. Since all empirical gravity models estimated a significant effect for the product of incomes, this approach suggests that a model of trade with differentiated products is preferred. 151 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Theoretical and Empirical using exchange rates as explanatory variable Bergstrand (1985) OLS Cross- Section PXij: dollar value of exports from country i to country j in logs Nominal GDPs, Distance, Adjacency dummy, Preferential trade agreement dummy, Exchange Rate, GDP deflators Canada, US, Japan, Belgium- Luxembourg, Denmark, France, West Germany, Italy, Netherlands, UK, Austria, Norway, Spain, Sweden, Switzerland 1965, 1966, 1975, and 1976 Uses a general equilibrium model of trade where consumers maximize a utility function with constant elasticity of substitution (CES) that is subject to an income constraint. Producers maximize profits based on a constant elasticity of transformation (CET) production function. Bergstrand finds that price and exchange rates have significant statistical effects on trade flows. Bergstrand suggests that if trade flows are differentiated by origin, the typical gravity equation omits prices and exchange rates. He also finds that the elasticity of substitution among importables is greater than 1, the elasticity of substitution between domestic and imported goods is less than 1, and that ?the elasticity of transformation among exports markets exceeds that between production for domestic and foreign markets.? Empirical. Tests the Linder Hypothesis. Uses exchange rates as explanatory variable Thursby and Thursby (1987) OLS: Cross sections. Regression for each country against al others PQij: dollar value of exports from country i to country j in logs Import price of i's exports to j, index of import prices of exports from other countries, CPIs, GNPs, export price of i's exports to j, index of net export prices of i's exports to other countries, variable reflecting tastes in j for i's export good (Linder), spot price of i's currency in terms of j, tariffs, transport costs, and a factor reflecting hedging by importers Canada, US, Japan, Belgium, Denmark, France, Finland, Germany, Greece, Italy, Netherlands, UK, Austria, Norway, Sweden, Switzerland, South Africa 1974-1982 The gravity model shows overwhelming support for Linder?s hypothesis and the theoretical belief that exchange rate risk affects bilateral trade flows. The coefficient on the Linder variable is negative and significant with the exception of 2 countries. Similarly, most of the countries with proper specified equations show a statistically significant negative coefficient for the exchange rate risk variable. 152 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Empirical. Tests the Linder Hypothesis Hanink (1988) Spatial effects model Trade intensity measured as per capita imports of country i from country j. Absolute difference in per capita GNPs, the distance between economic centers, and the variability across goods that is measured by absolute difference in populations Australia, Argentina, Canada, US, Japan, Belgium-Lux., Denmark, France, Finland, West Germany, Greece, Italy, Israel, Ireland, Mexico, Netherlands, New Zealand, Portugal, Singapore, South Korea, UK, Austria, Norway, Sweden, Switzerland, South Africa, and Spain 1984 Incorporating the hierarchical flow of goods that is common in regional trade. This hierarchical trade is determined by population size. Trade flows are positively related to market homogeneity (Linder hypothesis), negatively related to distance, and positively related to variety across goods. Theoretical and Empirical. Uses exchange rate as explanatory variable Bergstrand (1989) OLS Cross- Section: yearly regressions PXaij: the value of exports in industry a from country i to j National income, income per capita, distance, adjacency, preferential trade agreement dummies (EFTA and EEC), appreciation of importer's currency, wholesale price indices for each country, and prices Canada, US, Japan, Belgium- Luxembourg, Denmark, France, Germany, Italy, Netherlands, UK, Austria, Norway, Portugal, Spain, Sweden, Switzerland 1965, 1966, 1975, 1976 Extends Bergstrand (1985) by incorporating factor endowments differences (H-O theory) and non- homothetic preferences (Linder hypothesis) to the model. Consumers maximize a Cobb-Douglas-CES- Stone-Geary utility function subject to an income constraint. Demand curves based on this utility function use national income, income per capita, and prices to explain bilateral trade flows. Results show that between 40 to 80% of the variation across countries in one digit SITC trade flows are explained by the model. Coefficients on exporter and importer?s income are positive as expected, and coefficients for exporter?s per capita income suggest that chemicals, raw materials, manufactures, machinery and transport equipment, and food products are usually capital intensive in production whereas beverages and tobacco with miscellaneous manufactures are labor intensive. Bergstrand further notes that the coefficient on importer?s per capita income suggests that manufactures tend to be luxuries and raw materials necessities. 153 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Empirical using exchange rates as explanatory variable. Thoumi (1989) OLS Cross- Section: yearly regressions PXij: dollar value of exports from country i to country j in logs GDP of the exporting country; the GDP of the importing country; physical distance and country adjacency; income per capita, bilateral exchange rates, and dummies capturing economic integration effects Mexico, all Central American countries except Belize; all South American countries except Suriname; Jamaica, Haiti, the Dominic Republic, Barbados, Trinidad & Tobago 1971, 1975, and 1979 Exporters? GNP and distance are the most influential factors affecting trade patterns. Results also suggest that there is a tendency for richer countries to import more natural resource-based products than manufactures from poor countries. In general, the author suggests that integration systems among countries that are not too distant have similar sizes and development levels, and follow similar policies are more likely to succeed than other integration agreements. Theoretical and Empirical. Tests for Linder hypothesis Bergstrand (1990) OLS Cross- Section Grubel-Lloyd intra-industry trade index GDPs, GDPs per capita, tariffs rates, dummy for adjacency, inequality of GDPs, inequality of GDPs per capita, and capital-labor ratios 14 developed countries 1976 Extends previous theoretical work by examining how average levels and inequality of GDPs, GDPs per capita, tariffs rates, and capital-labor ratios affect the share of intra-industry trade. The paper provides a theoretical framework for such a model and then presents an empirical analysis for 14 developed countries. These theoretical foundations are similar to Bergstrand (1989) with minor differences. Bergstrand?s model reveals that the more similar per capita income within two countries, the more intra-industry trade. Specifically, in terms of supply, the author proposes that the more inequality among countries? capital-labor ratios, the lower intra- industry trade (Heckscher-Ohlin-Samuelson). Regarding demand, the greater the inequality between per capita incomes, the lower the share of intra-industry trade due to differences in tastes (Linder). Empirical Frankel, Stein, and Wei (1993) OLS Cross- Section: yearly regressions The value of exports plus imports in log form GNPs, GNPs per capita, distance, dummy for adjacency, preferential trade agreement dummies (East Asia, European Community, and NAFTA) 63 countries 1965-1990 every 5 years EEC became a significant trade-creating force in the 1980s, peaking in 1985 and declining thereafter. If two countries are members of the EEC, trade becomes 70% higher than it would have been otherwise (based on 1990 estimates). No trade creating effects for EFTA were found. 154 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Theoretical and Empirical Harrigan (1994) OLS Cross- Section Log of the value of imports divided by adjusted GNP Log of exporting country output; 4-firm concentration ratio, Herfindahl index, minimum efficient plant scale, and price cost margin variables Australia, Belgium, Canada, France, Germany, Italy, Netherlands, Austria, Finland, Norway, Sweden, Switzerland, UK 1983 He proposes an econometric approach to test the monopolistic competitive model of intra-industry trade summarized in Helpman and Krugman (1985). The author indicates that if the monopolistic competitive model explains gross trade flows, then industries with high gross trade flows should be described as having large scale economies. If that is not the case, then the Armington-HOV model is right and high gross trade is determined by substitution between domestic and foreign production. Results strongly support both models in the sense that the elasticity of imports with respect to a country?s output is one. Harrigan further finds some evidence that higher volumes of gross trade are associated with scale economies but this is sensitive to the choice of proxy variables. He concludes that scale economies and product differentiation by location of production are important causes of trade patterns. Empirical using exchange rates as explanatory variable Bayoumi and Eichengreen (1995) Gravity equation in differences rather than in levels Bilateral trade flows between countries in US dollars Real incomes, populations, distance, and the real exchange rate between European countries with the US. Five dummy variables measure trade within the EEC, trade within EFTA, trade between EEC and EFTA, trade between EEC and other industrial countries, and trade between EFTA and other industrial countries Australia, Canada, US, Japan, Belgium- Luxembourg, Denmark, Finland, France, Germany, Greece, Italy, Ireland, Netherlands, UK, Austria, Norway, New Zealand, Portugal, Spain, Sweden, Switzerland 1956-1992 divided in 3 periods The formation of EEC and EFTA had a significant effect on European trade flows that cannot be attributed to economic factors or even unobservable characteristics. Bayoumi and Eichengreen find that EFTA was trade creating, while EEC generated trade creation and trade diversion. Empirical adding lagged variables and exchange rates Eichengreen and Irwin (1995) OLS in logs, OLS scaled, Tobit Bilateral trade between countries i and j The product of the 2 countries national income, the product of the 2 countries per capita income, distance, lagged trade (dependent), and dummy for adjacency 38 countries 1928, 1938, 1949, 1954, 1964 They find a significant effect on lagged trade variables. Results are robust to instrumental variables replacing lagged trade values. Specifically, they find that in the absence of lagged trade variables, the trade-creating effects of the European Payments Union (EPU) as well as the importance of the Dillon Round in early 1960s are exaggerated. They conclude that one should always include lagged variables in the gravity equation. 155 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Theoretical Deardorff (1995) Deardorff uses two scenarios: one in which he assumes frictionless trade with no barriers to trade and homothetic products and another in which he introduces impediments to trade and product differentiation. With homothetic preferences, Deardorff derives what he called a ?simple frictionless gravity equation? where trade depends on incomes and a factor of proportionality. He shows that this result holds even when assuming arbitrary preferences. In the case of impeded trade, Deardorff derives equations for bilateral trade using Cobb-Douglas and CES preferences. His findings show that gravity models could be easily derived from standard HOS trade theory and consequently, it is not appropriate to conclude that the empirical success of gravity models suggest failure of the factor proportions or any other theory. Empirical using exchange rates as explanatory variable Frankel and Wei (1997) OLS Cross- Section: yearly regressions Value of exports from country i to country j Real incomes, populations, distance between trading partners, and dummies for contiguous borders, common language, and regional groupings 60 countries Four yearly regressions between 1970-1992 Affinity variables such as common language or adjacency are significant and intraregional trade biases exist. European countries are estimated to have traded 17% more than when these estimates are obtained with a standard gravity model. Similarly, Western Hemisphere and ASEAN countries are estimated to have traded 40% and 145% more than what a model without dummies would have estimated. Results also suggest that increased trade in ASEAN and EEC did not occur at the expense of third countries. The paper presents evidence of a currency bloc in Europe that follows the mark and a dollar bloc in the Pacific. The authors also find evidence suggesting that exchange rate volatility hinders trade. Empirical using exchange rates as explanatory variable Frankel, Stein, and Wei (1997) OLS Cross- Section: yearly regressions Value of exports from country i to country j Real incomes, populations, per capita incomes, distance between trading partners, and dummies for contiguous borders, common language, and regional groupings 60 countries Four yearly regressions between 1970-1992 Western European countries traded 36% more that what the standard gravity model would have predicted between 1970 and 1992. Results show that trade increased over time and suggest that this growth is a consequence of trade creating as well as trade diverting effects. The coefficient for the per capita income variable is positive, suggesting that richer countries trade more. Findings also suggest that regional preferential agreements are welfare improving, but the authors conclude that the extent of preferences among regional partners has probably exceeded optimal levels. 156 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Empirical using exchange rates as explanatory variable Frankel (1997) OLS Cross- Section: yearly regressions Total value of merchandise traded (exports plus imports) between two countries GNPs, per capita incomes, distance and dummies accounting for adjacency between a pair of countries, common language, and preferential trade agreements 65 countries Every five years from 1965 to 1985 and then in 1987, 1990, 1992, and 1994 Trade increases with a country?s GNP but less than proportionally. This suggests that smaller countries tend to be more open to trade than larger ones. The coefficients on per capita income are highly significant and indicate that richer countries trade more than poor ones. The coefficients on the distance variable are sensitive to the inclusion of the common border dummy. When the common border dummy appears in the equation, increasing the distance by 1% reduces trade by 0.6%. Results also suggest that two countries sharing a common border trade 82% more than two similar countries not sharing borders. Empirical. Addresses econometric issues M?ty?s (1997) Standard gravity model with country and time effects as unknown fixed parameters EXPij: exports from country i to country j Countries GDPs, populations, foreign currency reserves, and real exchange rates. Local, target, and time specific effects are also added Australia, Canada, India, Japan, Korea, Malaysia, New Zealand, Philippines, Singapore, Thailand, US 1982-1994 M?ty?s claims that the gravity models used up to that time did not take into account the time, local, and target country (importing country) effects. The study shows that imposing these restrictions leads to incorrect inferences due to the misinterpretation of the coefficients on dummies accounting for trading blocs, common border, or common language. He suggests that models explaining trade should take into account these fixed effects. Empirical. Tests the Linder Hypothesis Chow, Kellman, and Shachmurove (1999) OLS Trade intensity (trade complementary index) Log of per capita income ratio, a relative price variable that takes the log of the ratios of exchange rates and wholesale price indices Taiwan, Korea, Hong Kong, Singapore with US, EC, and Japan 1965-1990 Linder theory is tested by using disaggregated data on manufactured exports. Results support the Linder hypothesis and suggest that tastes significantly affect trade flows in these countries. Empirical using exchange rates as explanatory variable. Addresses econometric issues Dell?Ariccia (1999) Panel data Log of exports plus imports between countries i and j GDPs, populations, distance, exchange rate volatility, a variable accounting for "third country" volatility, dummies for common border, common language, and EU Belgium- Luxembourg, Finland, France, Greece, Germany, Ireland, Italy, Netherlands, Austria, Denmark, Spain, Portugal, Sweden, Switzerland, UK 1975-1994 Results from a Hausman test show that OLS regression generates biased results suggesting the existence of simultaneity bias. Specifically, this bias is due to the existence of unobserved country-pair specific effects. This simultaneity bias is addressed with the use of instrumental variables and a fixed effects model. A fixed effects model is preferred over a random effects model and results are similar to OLS estimates. Results suggest that exchange rate volatility decreases international trade and these results are robust for different specifications. The coefficients on the standard gravity variables are also as expected. 157 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Empirical using exchange rates as explanatory variable. Addresses econometric issues M?ty?s, K?nya, and Harris (2000) Panel Data: Fixed Effects. Four models are estimated Exports from country i to country j Countries GDPs, populations, foreign currency reserves, real exchange rates, and distance. Local, target, and time specific effects are also added Australia, Indonesia, Japan, Korea, Malaysia, New Zealand, Philippines, Singapore, Thailand, US, and the European Economic Area 1978-1997 APEC members trying to increase exports should look at Singapore and New Zealand as potential markets. They claim that policy implications could be wrong in the absence of specific effects. Results also suggest that foreign GDP effects were underestimated in previous studies, that the effect of population on trade could be positive, and that the effect of real exchange rates is significant. Empirical. Tests the Linder Hypothesis. Uses exchange rates as explanatory variable McPherson, Redfearn, Tieslau (2000) Panel Data: Fixed Effects, Tobit Approach Dollar value of exports from OECD country j to potential trading partner i GDP of trading partner i, real exchange rates, and the absolute differences in per capita income between trading partners Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Spain, Sweden, Switzerland, UK, US, and 161 potential trading partners 1990-1995 at annual intervals Supports the Linder hypothesis for all but one of the 19 OECD countries under consideration. The relative size of a trading partner?s economy has a positive effect on trade. In terms of exchange rates, all but three of the OECD countries show the expected positive and significant effect on exports. Empirical using exchange rates as explanatory variable Rose (2000) Cross- country panel data, yearly regressions Bilateral trade GDPs, incomes per capita, distances, and a series of dummies accounting for common language, regional trade agreement, colonies, common nations, a common currency dummy, and a variable explaining exchange rate volatility EU countries as well as other 92 countries 1970, 1975, 1980, 1985, 1990 Rose finds that two countries with a common currency trade three times as much as countries not sharing a common currency. This common currency effect is larger than the effect of reducing exchange rate volatility to zero but keeping separate currencies. The author performs a sensitivity analysis that suggests robust results. Empirical. Addresses econometric issues Soloaga and Winters (2001) 17 separate yearly regressions and then, pooled data estimation Value of imports of country i from country j GDPs, populations, distance, weighted distance, land area, dummies for common border, island, cultural affinities, and preferential trade agreement 58 countries representing 70% of world trade 1980-1996 annual non- fuel imports Results show no indication that increasing regionalism during the 1990s raised intra-bloc trade significantly. The paper presents evidence of trade diversion taking place in the EU and EFTA. Soloaga and Winters also suggest that trade liberalization efforts in Latin America had a positive impact on bloc members? imports. 158 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Empirical using exchange rates as explanatory variable. Addresses econometric issues Pakko and Wall (2001) Panel Data: Fixed Effects Xij: log of total real trade of country i with country j GDPs, per capita GDPs, dummies for preferential agreement, common currency, and a vector of time dummies EU countries as well as other 92 countries 1970, 1975, 1980, 1985, 1990 The authors claim that the fixed effects model avoids the estimation bias that may arise due to misspecification or omitted variables. Misspecification could arise with the creation of the variable distance that is supposed to reflect relative costs of trading. Omitted variables problems arise because it becomes impossible to include enough variables to account for all the important fixed factors (time invariant factors). Further, Pakko and Wall suggest that the fixed effects model not only controls for variables such as language, common nation, colony, and distance, but also, it accounts for any other factors that are usually not included in gravity models. The paper shows that using the same data of Rose (2000), the fixed effects model results in much weaker evidence. Empirical. Tests the Linder Hypothesis. Uses exchange rates as explanatory variable McPherson, Redfearn, Tieslau (2001) Panel Data: Fixed Effects, Tobit Approach Value of imports from trading partners GDP of trading partner i, real exchange rates, and the absolute differences in per capita income between trading partners Ethiopia, Kenya, Rwanda, Sudan, Tanzania, Uganda and their trading partners 1984-1992 Annual In five out of six countries, the Linder hypothesis holds. In particular, results show that Ethiopia, Kenya, Rwanda, Sudan, and Uganda trade more intensively with countries that have per capita incomes similar to them. The authors conclude that factor proportions theory is inadequate when investigating trade flows in developing countries and they suggest the appropriateness of Linder hypothesis in such a context. Empirical Evenett and Keller (2002) OLS Cross- Section Grubel-Lloyd intra-industry trade index GDPs per capita, GDPs, Capital per Worker, # of industries traded 58 countries accounting for 67% of world imports and 79% of world GDP 1985 Specialization and trade have a positive relationship with the share of intraindustry trade on total trade. Evenett and Keller suggest that a model with IRS and product differentiation explains the North-North trade. They also find that trade volumes increase when there are larger differences in factor endowments suggesting that factor abundance is important explaining North- South trade. 159 Type Author (Year) Method Dependent Variable Independent Variables Countries Time Period (frequency) Results Empirical using exchange rates as explanatory variable Martinez- Zarzoso and Nowak- Leman (2002) Panel Data: fixed effects Sector specific exports for different countries at different time periods Differences in per capita income between countries (economic distance), distance between countries scaled by infrastructure, and the bilateral real exchange rate Mercosur plus Chile exports to 15 EU countries 1988-1996 Products that are highly sensitive to economic distance and not sensitive to geographical distance are the best candidates for future trade with EU. Results suggest that the Linder hypothesis applies to telecommunications, iron and steel, metals, industrial machinery, and animal feed. Sectors with a dominant HOS effect are furniture, footwear, beverages, meat and fish (products in which Mercosur has a comparative advantage). Results also indicate that some industries have a high and significant geographical distance effect. Empirical using exchange rates as explanatory variable Martinez- Zarzoso and Nowak- Leman (2003) Panel Data: fixed effects Value of exports from country i to country j GDPs, difference in GDPs per capita, populations, distance, infrastructure, real exchange rate, EU dummy, Mercosur dummy Mercosur plus Chile exports to 15 EU countries 1988-1996 Exporter and importer incomes have a positive effect on trade flows. Results also show that exporter?s population has a negative effect on exports and importer?s population has a positive effect. The findings also suggest that for Mercosur-EU trade flows, only exporter infrastructure has a positive effect on trade. Preferential trade agreements are also shown as variables increasing trade flows. Potential trade estimates show that Mercosur was exporting below its potential levels in 1996 (each country member), but in previous years, results are varied. Empirical using exchange rates as explanatory variable. Addresses econometric issues Cheng and Wall (2005) Panel data: fixed effects Xij: exports from country i to country j in logs GDPs, populations, distance, country pair dummies Argentina, Australia, Austria, Belgium-Lux., Brazil, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong, Ireland, Israel, Italy, Japan, Korea, Mexico, Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sweden, Switzerland, UK, Uruguay, US 1982, 1987, 1992, 1997. The authors show that standard pooled-cross-section methods used in gravity models have an estimation bias problem due to omitted or misspecified variables. The paper shows that a two-way fixed effects model solve this econometric issues by using country-pair and period dummies that explain bilateral trade patterns. Country- specific dummies capture factors such as distance, common border, common language, history, culture, and others that are constant over time. Cheng and Wall show that alternative fixed effects models such as M?ty?s (1997), Glick and Rose (2001), and Bayoumi and Eichengreen (1997) are special cases of their proposed two-way model. They claim that the restrictions applied to obtain these alternative models are not supported statistically. Results indicate that unless heterogeneity is accounted for properly, gravity models of bilateral trade can overestimate the effects of integration on trade flows. 160 CHAPTER 4: CONCLUDING REMARKS AND POLICY IMPLICATION ISSUES I. Concluding Remarks This dissertation examines the trade adjustments to exchange rate policies in a world of increasing regional economic integration. It provides a framework for the analysis of currency devaluations and their effect on bilateral trade balances within the context of a regional trading bloc. The empirical investigation is based on Argentina and Brazil, two countries that have devalued their currencies to correct trade imbalances. Argentina experienced a deterioration of its trade balance with Brazil after the peso devaluation leading to a lively political discussion between the two governments that blocked further advancements to a more meaningful Mercosur economic integration. Chapter 1 introduces the problems and challenges faced by countries entering regional trading blocs. It highlights the effects of devaluation in a country party to a regional trade agreement. The most common problems of devaluation in this context are the relocation of foreign direct investment (FDI), protectionist measures enacted by the country that is losing competitiveness, trade adjustments, and exchange rate crises that have the potential to develop into recessions. There is evidence of these issues in Mercosur. Starting with Brazil?s devaluation in January 1999, a large number of firms relocated from Argentina to Brazil. This led the Argentine government to adopt protectionist measures such as tariffs and quotas. Chapter 1 also describes the 161 composition of trade in Argentina as well as the macroeconomic performance of both countries for the period under study. Chapter 2 studies dynamic adjustments of trade balances to devaluations or the J- curve phenomenon. The proposed model augments conventional J-curve models by adding a trade diversion variable that captures the effects of regional economic integration. Results from an Almon polynomial distributed lag (PDL) model provide evidence of an ?inverse J-curve? and a significant trade diversion effect. Argentina?s trade balance with Brazil initially improved and then became a deficit as the exchange rate stabilized and capital restrictions were relaxed or lifted. Potential simultaneity concerns between the variable capturing trade diversion effects and the trade balance variable are addressed by estimating the Almon PDL model with an instrumental variables (IV) approach. Results are robust to model specification. Chapter 3 investigates the reasons for the inverse J-curve and the potential trade diversion effects found in Chapter 2. A gravity model of trade examines trade diversion effects of devaluations and the Linder hypothesis. Estimates from different specifications of the time-series cross-section model described in Greene (2003) and from a fixed effects model provide evidence of trade diversion. Findings imply that trade was diverted from the US and EU to Brazil as a consequence of devaluation rather than a direct consequence of regionalization. Models testing the Linder hypothesis show that a change in Argentina?s demand structure might explain diversion effects. As devaluation depressed Argentina?s per capita income, the country?s demand structure has become more similar to Brazil?s, diverging from the US and EU demand structures. In other 162 words, changes in demand structures suggested by Linder explain the increased demand for Brazilian manufactures in Argentina at the expense of non-Mercosur countries. The inverse J-curve in Chapter 2 should not be surprising considering the effects of regionalization and changing demand structures. First, if these two countries were not Mercosur members, the relocation of production capabilities from Argentina to Brazil after the devaluation of the real would have been of a lesser magnitude. Increased production capabilities through an FDI surge in Brazil expanded the country?s potential for exports. By the time Argentina devalued its currency, imports from the US and EU became too expensive and were substituted for goods produced in Brazil, the only Mercosur country with capacity to supply foreign markets with manufactures. In summary, this dissertation contributes to the existing literature by examining trade adjustments to exchange rates in the context of regional economic integration. The analysis of the dynamic effects of currency devaluation on trade balances also tests for trade diversion effects. A gravity model investigates whether trade diversion effects emerged as a consequence of devaluation rather than being a direct consequence of a preferential trade agreement. By testing the Linder hypothesis for the first time in Mercosur countries, this dissertation contributes to the literature examining the presence of Linder effects in developing countries. An up-to-date literature review on gravity models is also provided. II. Policy Implications The empirical results in this dissertation support the notion that currency links play a major role in explaining trade flows. The fact that Mercosur has not become a 163 fully working common market is due for the most part to the divergent or contradicting exchange rate policies implemented by member countries. Consequently, countries entering regional trade agreements should set either a common currency or exchange rate convergence criteria for better results. This is in line with Frankel (1997) who says that ?European leaders believe that currency links are not just a desirable supplement to a successful common market, but are actually a necessary component of it? (p. 135). Also, IADB (2002) argues that trade agreements seem to fail in the presence of ?exchange rate disagreements.? Results indicate that in the absence of specific rules regulating exchange rates among countries in a regional trading bloc, devaluation by a country member may induce adjustments that are irreversible when other members do not follow in a timely fashion. Argentina?s devaluation came too late to redirect FDI flows from Brazil to Argentina. This is similar to a first mover advantage where the country that devalues first derives most of the benefits in terms production and export capabilities. Once countries in a regional trading bloc start experiencing uneven benefits, the whole issue of regionalization becomes questionable, at least over an intermediate time period. Evidently, monetary agreements are essential for successful regional economic integration between countries. 164 BIBLIOGRAPHY Abrams, R. K. (1980). International Trade Flows under Flexible Exchange Rates. Economic Review, Federal Reserve Bank of Kansas City, March: 3-10. Aitken, N. D. (1973). The Effect of the EEC and EFTA on European Trade: A Temporal Cross- Section Analysis.? American Economic Review, 63, 881-92. Akbostanci, E. (2004) Dynamics of the Trade Balance: The Turkish J-curve. Emerging Markets Finance and Trade, 40:5, 57-73. Almon, S. (1965) The Distributed Lag Between Capital Appropriations and Expectations, Econometrica, 33, 178-96. Amman, E. and Baer W. (2002). Anchors Away: The Costs and Benefits of Brazil?s Devaluation. World Development, 31(6): 1033-1046. Anderson, J. E. (1979) A Theoretical Foundation for the Gravity Equation. American Economic Review, 69, 1, 106-116. Arndt, H. W. and Dorrance, G. (1987). The J-Curve, The Australian Economic Review. 1st. Quarter 1987. Arora, S., Bahmani-Oskooee, M., and Goswami, G. G. (2003). Bilateral J-curve between India and her Trading Partners. Applied Economics, 35, 1037-41. Backus, D. K., Kehoe, P. J., and Kydland, F. E. (1994). Dynamics of the Trade Balance and the Terms of Trade: The J-curve? The American Economic Review, 84, 84-103. Baharumshah, A. Z. (2001). The effect of Exchange Rate on Bilateral Trade Balance: New Evidence from Malaysia and Thailand, Asian Economic Journal, 15, 291-312. Bahmani-Oskooee, M. (1985). Devaluation and J-curve: some evidence from LDCs, The Review of Economics and Statistics, August, 49-58. Bahmani-Oskooee, M. (1989). Devaluation and the J-curve: Some Evidence from LDC?s: Errata, The Review of Economics and Statistics, 71, 553-4. Bahmani-Oskooee, M. and Alse, J. (1994). Short-Run versus Long?Run Effects of Devaluation: Error Correction Modeling and Cointegration, Eastern Economic Journal, 20(4), 453-64. Bahmani-Oskooee, M. and Brooks, T. J. (1999). Bilateral J-curve between US and her Trading Partners, Weltwirtschaftliches Archiv, 135, 156-65. 165 Bahmani-Oskooee, M. and Goswami, G. G. (2003). A Disaggregated Approach to Test the J- curve Phenomenon: Japan vs. Her major Trading Partners. Journal of Economics and Finance, 27, 102?13. Bahmani-Oskooee, M. and Malixi, M. (1992). More Evidence on the J-curve from LDCs. Journal of Policy Modeling, 14, 641?53. Bahmani-Oskooee, M. and Pourheydarian, M. (1991). The Australian J-curve: A Reexamination. International Economic Journal, 5, 49?58. Bahmani-Oskooee, M. and Ratha, A. (2004). The J-curve Dynamics of US Bilateral Trade, Journal of Economics and Finance, 28(1), 32-38. Bahmani-Oskooee, M. and Ratha, A. (2004b). Dynamics of US Trade with Developing Countries. Journal of Developing Areas, forthcoming. Balassa, B. (1967). Trade Liberalization among Industrial Countries: Objectives and Alternatives. New York: McGraw-Hill. Banco Central de la Rep?blica Argentina (BCRA), Estad?sticas. Banco Central do Brasil (BCB), Economia e Finan?as, Series Temporais. Bayoumi, T. and Eichengreen, B. (1997). Is Regionalism Simply a Diversion? Evidence from the Evolution of the EC and EFTA, in Takatoshi Ito and Anne O. Krueger, eds., Regionalism versus Multilateral Trade Arrangements. Chicago: University of Chicago Press, 141-64. Bayoumi, T. and Eichengreen, B. (1995). Is Regionalism Simply a Diversion? Evidence from the Evolution of the EC and EFTA. Center for Economic Policy Research, Working Paper 1294. November. Bergstrand, J. H. (1985). The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical Evidence. The Review of Economics and Statistics. 67: 474- 481. Bergstrand, J. H. (1989). The Generalized Gravity Equation, Monopolistic Competition, and the Factor-Proportions Theory in International Trade. The Review of Economics and Statistics. 71: February 143-153. Bergstrand, J. H. (1990). The Heckscher-Ohlin-Samuelson Model, The Linder Hypothesis and the Determinants of Bilateral Intra-Industry Trade. The Economic Journal. 100: December 1216-1229. Bergstrand, J. H. and Scott, L. B. (2001). ?The Growth of World Trade: Tariffs, Transport Costs, and Income Similarity.? Journal of International Economics. 53: 1-27. Bharadwaj, R. (1962). Factor Proportions and the Structure of Indo-U.S. Trade, Indian Economic Journal, 10 (October), 105-16. 166 Bowen, H. P., Leamer, E.E., and Sveikavskas, L. (1995). Multicountry, Multifactor Tests of the Factor Abundance Theory. International Trade: Structure, Trade and Growth, 2 (Ed.) J. P. Neary, Edward Elgar, Brookfield, Vermont, 269-87. Brissimis, S. N. and Leventakis, J. A. (1989). The Effectiveness of Devaluation: A General Equilibrium Assessment with Reference to Greece. Journal of Policy Modeling, 11(2), 247?71. Buluswar, M. D., Thompson, H., and Upadhyaya, K. P. (1996). Devaluation and the Trade Balance in India: Stationarity and Cointegration, Applied Economics, 28, 429-432. Chen, H.C. (2001). Taiwan Exports and Trade Imbalance against US and Japan: An Empirical Analysis. Applied Economics, 33, 1283-1287. Cheng, I-H. (1999). The Political Economy of Economic Integration. Ph.D. Dissertation, Birkbeck College, University of London, July. Cheng, I-H. and Wall, H. (2005). Controlling for Heterogeneity in Gravity Models of Trade and Integration. Federal Reserve Bank of St. Louis Review, January/February, 87(1), 49-64 Chow, P., Kellman, M., and Shachmurove, Y. (1999). A Test of the Linder Hypothesis in Pacific NIC Trade 1965-1990. Applied Economics. 31: 175-82. Dabrowski, M., Gorski, U., and Jarocinsky, M. (2000). The inflationary consequences of the devaluation crisis in Russia and Ukraine, Problems of Economic Transition, 43, 32-57. Deardorff, A. V. (1984). Testing Trade Theories and Predicting Trade Flows. Handbook of International Economics, (Eds.) R.W. Jones and P.B. Kenen, Elsevier Science Publishers, Amsterdam, I. Deardorff, A. V. (1995). Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World? NBER Working Paper 5377, December. Dell?Ariccia, G. (1999). Exchange Rate Fluctuations and Trade Flows: Evidence from the European Union. IMF Staff Papers, (September/December) 46(3): 315-33. Dickey, D., and Fuller, W. (1979). Distribution of the Estimates for Autoregressive Time Series with a Unit Root. Journal of the American Statistical Association, 74, 427-31. Egger, P. (2002). An Econometric View on the Estimation of Gravity Models and the Calculation of Trade Potential. The Word Economy. (February) 25: 297-312. Eichengreen, B. (1993). ?European Monetary Unification.? Journal of Economic Literature 31: 1321-57. Eichengreen, B. (1993). European Monetary Unification. Journal of Economic Literature, 31: 1321-57. 167 Eichengreen, B. and Irwin, P. (1995). The Role of History in Bilateral Trade Flows, in The Regionalization of the World Economy, J. Frankel, editor, National Bureau of Economic Research: Chicago, 1998. European Central Bank (ECB) website. Statistics. Eurostat website. National Accounts, External Trade. Evenett, S. J. and Keller, W. (2002). On Theories Explaining the Success of the Gravity Equation.? Journal of Political Economy. 110: 281-316. Feenstra, R. C. (1999). Facts and Fallacies about Foreign Direct Investment. In Martin Feldstein (ed.), International Capital Flows. Chicago: University of Chicago Press and NBER. Feenstra, R. C., Markusen, J. R., and Rose, A. K. (2001). Using the Gravity Equation to Differentiate among Alternative Theories of Trade. Canadian Journal of Economics. 34 (May): 430-447. Felmingham, B. S. (1988). Where is the Australian J-curve? Bulletin of Economic Research, 40:1, 43-56. Frankel, J. (1997). Regional Trading Blocs in the World Economic System. Washington, D.C.: Institute for International Economics. Frankel, J. (1998). The Regionalization of the World Economy. National Bureau of Economic Research: Chicago. Frankel, J., Stein, E., and Wei, S. (1993). Continental Trading Blocs: Are They Natural, or Super-Natural? (December). NBER Working Paper No. W4588. Frankel, J., Stein, E., and Wei, S. (1995). Trading Blocs and the Americas: The Natural, the Unnatural, and the Super-Natural. Journal of Development Economics. 47: 61-95. Frankel, J., Stein, E., and Wei, S. (1997). Continental Trading Blocs: Are They Natural or Supernatural? in The Regionalization of the World Economy. J. Frankel, editor, National Bureau of Economic Research: Chicago, 1998. Frankel, J. and Wei, S. (1997). Regionalization of World Trade and Currencies: Economics and Politics, in The Regionalization of the World Economy, J. Frankel, editor, National Bureau of Economic Research: Chicago, 1998. Geraci, V. J. and Prewo, W. (1977). ?Bilateral Trade Flows and Transport Cost.? The Review of Economics and Statistics. 59: 67-74. Geraci, V. J. and Prewo, W. (1982). ?An Economic and Supply Model of Multilateral Trade,? The Review of Economics and Statistics. 64: 432-441. Giambiagi, F. and Averburg, B. (2000). The Brazilian Crisis of 1998/1999: Origins and Consequences. Rio de Janeiro: BNDES. 168 Glick, R. and Rose, A. K. (2001). ?Does a Currency Union Affect Trade? The Time Series Evidence.? NBER Working Paper No. 8396, National Bureau of Economic Research, July. Greene, W. H. (2003). Econometric Analysis. 5 th Edition, Pearson Education. Greytak, D. and McHugh, R. (1977). Linder?s Trade Thesis: An Empirical Investigation. Southern Economic Journal, January, 1386-89. Greytak, D., and Tuchinda, U. (1990). The Composition of Consumption and Trade Intensities: An Alternative Test of the Linder Hypothesis, Weltwirtschaftliches -Archiv, 126, 50-57. Grubel, H. G. (1970). The Theory of Intra-industry Trade, in I. A. McDougall and R. H. Snape (ed.), Studies in International Economics. Amsterdam: North Holland. Gupta-Kapoor, A. and Ramakrishnan, U. (1999). Is There a J-curve? A New Estimation for Japan. International Economic Journal, 13, 71-9. Hacker, R. S. and Abdulnasser, H.-J. (2003). Is the J-curve Effect Observable for Small North European Economies? Open Economies Review, 14, 119?34. Hanink, D. M. (1988). An Extended Linder Model of International Trade. Economic Geography, October, 322-34. Hanke, S. H. (2002). The Great Argentine Train Robbery. The International Economy, 18-21. Harrigan, J. (1994). Scale Economies and the Volume of Trade. The Review of Economics and Statistics. 76(2): May, 321-28. Harris, M. and M?ty?s, L. (1998). The Econometrics of Gravity Models. Melbourne Institute Working Paper No.5/98. Helpman, E. (1999). The Structure of Foreign Trade. The Journal of Economic Perspectives, 13:2, Spring, 121-44. Helpman, E. and Krugman, P. (1985). Market Structure and Foreign Trade: Increasing Returns, Imperfect Competition, and the International Economy. Cambridge, MA: MIT Press. Himarios, D. (1985). The effects of Devaluation on the Trade Balance: A Critical View and Reexamination of Miles?s (New Results). Journal of International Money and Finance, 4, 553?63. Himarios, D. (1989). Do Devaluations Improve the Trade Balance? The Evidence Revisited, Economic Inquiry, 27, 143-68 Hirsch, Z. and Lev, B. (1973). Trade and Per Capita Income Differentials: A Test of the Burenstam-Linder Hypothesis. World Development, 1, 11-17. 169 Hoftyzer, J. (1975). Empirical Verification of Linder?s Trade Thesis: Comment. Southern Economic Journal, April, 694-98. Instituto Brasileiro de Geografia e Estad?stica (IBGE) website. Indicadores: Contas Nacionais Trimestrais. Instituto Nacional de Estadisticas y Censos (INDEC) website. Cuentas Nacionales. Sector Externo. Inter-American Development Bank (IADB) (2002). Beyond Borders: The New Regionalism in Latin America. John Hopkins University Press. Washington, D.C. Junz, H. B. and Rhomberg, R. R. (1973). Price-Competitiveness in Export Trade among Industrial Countries. American Economic Review, 63(2), 412?18. Karadeloglou, P. (1990). On the Existence of an Inverse J-curve. Greek Economic Review, 12(2), 285-305. Kennedy, T. E. and McHugh, R. (1983). Taste Similarity and Trade Intensity: A Test of the Linder Hypothesis for U.S. Exports, Weltwirtschaftliches-Archiv 119, 84-96. Kravis, I. (1971). The Current Case for Imports Limitations, in Commission of International Trade and Investment Policy. United States Economic Policy in an Interdependent World. Washington: Government Printing Office. Krueger, A. D. (1983). Exchange Rate Determination, Cambridge University Press, Cambridge. Krugman, P. (1979). Increasing Returns, Monopolistic Competition, and International Trade. Journal of International Economics, 9: 469-79. Krugman, P. (1980). Scale Economies, Product Differentiation, and the Pattern of Trade. American Economic Review, 70: December, 950-9. Krugman, P. (1981). Intra-industry Specialization and the Gains from Trade. Journal of Political Economy, 89: October, 959-73. Krugman, P. (1989). The J-curve, the Fire Sale, ad the Hard Landing. American Economic Review, 79, 31-5. Krugman, P. (1990). Rethinking International Trade. Cambridge, MIT Press. Kulkarni, K. G. (1996). The J-curve hypothesis and currency devaluation: Cases of Egypt and Ghana. Journal of Applied Business Research, 12, 1-9. Lal, A. K. and Lowinger, T. C. (2002). The J-curve: Evidence from East Asia, Journal of Economic Integration, 17, 397-415. Learmer, E. (1974). ?The Commodity Composition of International Trade in Manufactures: An Empirical Analysis.? Oxford Economic Papers 26: 350-374. 170 Leontief, W. (1953). Domestic Production and Foreign Trade: The American Capital Position Re-Examined. Proceeding of the American Philosophical Society, 97: 332-349. Levi Yeyati, E., Schmukler, S. L, and Van Horen, N. (2004). The Price of Inconvertible Deposits: The Stock Market Boom during the Argentine Crisis. Economics Letters, 83, 7- 13. Linder, S. B. (1961). An Essay on Trade and Transformation, John Wiley & Sons, New York. Linneman, H. (1966). An Econometric Study of International Trade Flows. Amsterdam: North Holland. MacDonald, R., and Taylor, M. P. (1994). Reexamining the Monetary Approach to the Exchange Rate: The Dollar-Franc, 1976-90, Applied Financial Economics, 4, 423-429. Magee, S. P. (1973). Currency Contracts, Pass Through and Devaluation, Brooking Papers on Economic Activity, 1, 303-25. Martinez-Zarzoso, I. And Nowak-Lehmann, F. (2002). Explaining Mercosur Sectoral Exports to the EU: The Role of Economic and Geographical Dustance. Ibero-American Institute for Economic Research. Discussion Paper No 85. Martinez-Zarzoso, I. And Nowak-Lehmann, F. (2003). Augmented Gravity Model: An Empirical Application to Mercosur-European Union Trade Flows. Journal of Applied Economics, 6(2): 291-316. Marwah, K. and Klein, L. R. (1996). Estimation of J-curves: United States and Canada. The Canadian Journal of Economics, 29:3, 523-39. M?ty?s, L. (1997). Proper Econometric Specification of the Gravity Model. The World Economy. 20: 363-369. M?ty?s, L., K?nya, L., and Harris, M. (1997). Modelling Export Activity in a Multicountry Economic Area: The APEC Case. Working Paper, Monash University, Department of Econometrics and Business Statistics, No.1/97. M?ty?s, L., K?nya, L., and Harris, M. (2001). Modelling Export Activity of Eleven APEC Countries. Melbourne Institute Working Paper No.5/00. McCallum, J. (1995). National Borders Matter: Canada-U.S. Regional Trade Patterns.? The American Economic Review. 85 (June): 615-623. McPherson, M. A., Redfearn, M. R., Tieslau, M. A. (2000) A Re-examination of the Linder Hypothesis: A Random-Effects Tobit Approach. International Economic Journal, 14, 3, 123-136. McPherson, M. A., Redfearn, M. R., Tieslau, M. A. (2001) International trade and developing countries: an empirical investigation of the Linder hypothesis. Applied Economics, 33, 649-657. 171 Ministerio de Econom?a (MECON) website. Macroeconomic Data. Millimet, D. L. and Osang, T. (2004). Do State Borders Matter for U.S. Intranational Trade? The Role of History and Internal Migration.? Working Paper, Southern Methodist University, May. Nelson, C. and Plosser, C. (1982). Trends and Random Walks in Macroeconomics Time Series: Some Evidence and Implications, Journal of Monetary Economics, 10, 130-62. Nicholson, W. (1998). Microeconomic Theory: Basic Principles and Extensions. 7 th Edition, Fort Worth: Dryden Press. Pakko, M. R. and Wall, H. J. (2001). Reconsidering the Trade-Creating Effects of a Currency Union? FRB St. Louis Review 83:5, 37-45. Pardo, M. (2002). Tipo de cambio real cruzado de Colombia con los principales pa?ses con los que existen acuerdos de integraci?n. Grupo Consultivo de la Comunidad Andina. Mimeo. Perron, P. (1989). The Great Crash, The Oil Price Shock, and the Unit Root Hypothesis, Econometrica, 57, 1361-1401. Pesaran, M. H. and Shin, Y. (1995). An Autoregressive Distributed Lag Modeling to Cointegration Analysis, in Centennial Volume of Rangar Frisch (Eds) S. Strom, A. Holly, and P. Diamond, Cambridge University Press, Cambridge. Pesaran, M. H., Shin, Y., and Smith, R. J. (2001). Bound Testing Approaches to the Analysis of Level Relationships, Journal of Applied Econometrics, 16, 289-326. Pindyck, Robert S. and Rubinfeld, D. L. (1998). Econometric Models and Economic Forecasts. 4 th edition. Irwin McGraw-Hill. Pollak, R. A. (1970). Habit Formation and Dynamic Demand Functions, Journal of Political Economy, 78, 745-763. Quispe-Agnoli, M. and Kay, S. (2002). ?Argentina: The End of Convertibility.? EconSouth, Q1: 14-19. Qureshi, U. A., French, G. L., and Sailors, J. W. (1980). Linder?s Trade Thesis: A Further Examination. Southern Economic Journal, January, 933-36. Rivera-Batiz, F. L. and L. A. (2002). International Finance and Open Economy Macroeconomics. 2 nd ed. Macmillan Publishing Company, New York. Rose, A. K. (2000). One Money, One Market: The Effect of Common Currencies on Trade. Economic Policy: A European Forum, April, 30: 7-45. Rose, A. K. and Yellen, J. L. (1989). Is There a J-curve? Journal of Monetary Economics, 24 53-68. 172 Soloaga, I. and Winters, A. (2001). Regionalism in the Nineties: What Effect on Trade? North American Journal of Economics and Finance. 12: 1-29. Sukar, A. and Zoubi, T. A. (1996). Real Exchange Rates and US Bilateral Trade. Journal of Applied Business Research, 12:4, 138-45. Thompson, H. (2001). International Economics: Global Markets and International Competition, World Scientific Publishing Co., New Jersey. Thoumi, F. (1989). Trade Flows and Economic Integration among the LDCs of the Carribean Basin. Social and Economic Studies 38(2), 215-33. Thursby, J. G. and Thursby, M. C. (1987). Bilateral Trade Flows, the Linder Hypothesis, and Exchange Risk. The Review of Economics and Statistics. 69 (August): 488-495. Tinbergen, J. (1962). Shaping the World Economy: Suggestions for an International Economic Policy. New York: The Twentieth Century Fund. Tongzon, J. L. and Felmingham, B. S. (1998). Bilateral Trade in the Asia-Pacific: A Case Study for Australia, USA, Japan, and Singapore. Asian Economic Journal. 12:4, 379-94. United States Census Bureau. International Database. Upadhyaya, K. P., and Dharmendra, D. (1997). Devaluation and the Trade Balance: Estimating the Long-Run Effect, Applied Economics Letters, 4, 343-345. Wall, H. J. (1999). ?Using the Gravity Model to Estimate the Costs of Protection.? Federal Reserve Bank of St. Louis Review. (January/February) 81: 33-40. Wall, H. J. (2000). ?Gravity Model Specification and the Effect of the Canada-U.S. Border.? Working Paper No. 2000-024A , Federal Reserve Bank of St. Louis, September. Wall, H. J. (2002). ?Has Japan Been Left Out in the Cold by Regional Integration?? Bank of Japan, Monetary and Economic Studies, April, 20(2): 117-34. Wall, H. J. (2003). NAFTA and the Geography of North American Trade. Federal Reserve Bank of St. Louis Review, March/April, 85(2): 13-26. Wilson, P. (2001) Exchange Rate and the Trade Balance for Dynamic Asian Economies: Does the J-curve exist for Singapore, Malaysia and Korea? Open Economies Review, 12, 389- 413. Wilson, P. and Tat, K. C. (2001). Exchange Rates and the Trade Balance: The Case of Singapore 1970 to 1996, Journal of Asian Economics, 12, 47-63. Yeats, A. J. (1998). Does Mercosur?s Trade Performance Raise Concerns about the Effects of Regional Trade Agreements? World Bank Economic Review, 12(1): 1-28. 173 Zahniser, S. S., Pick, D., Pompelli, G., and Gehlhar, M. J. (2002). ?Regionalism in the Western Hemisphere and Its Impact on U.S. Agricultural Exports: A Gravity-Model Analysis.? American Journal of Agricultural Economics. 83, August: 791-797.