WPS4220 Evaluating the Trade Effect of Developing Regional Trade Agreements: A Semi-parametric Approach Souleymane Coulibaly World Bank* 1818 H Street NW, 20433 Washington DC, USA Tel: +1 202 473 9845 Email: scoulibaly2@worldbank.org Abstract Many recent papers have pointed to ambiguous trade effects of developing regional trade agreements (RTAs), calling for a reassessment of their economic merits. We focus on seven such agreements currently in force in Sub-Saharan Africa (ECOWAS and SADC), Asia (AFTA and SAPTA) and Latin America (CACM, CAN and MERCOSUR), estimating their impacts on their members' trade flows. Instead of the usual dummy variables for RTAs, we propose a variable taking into account the number of years of membership. We then combine a gravity model with kernel estimation techniques so as to capture the non-monotonic trade effects while imposing minimal structure on the model. The results indicate that except for SAPTA, all these RTAs have had a positive impact on their members' intra-trade over the estimation period (1960-1999). AFTA seems to be the most successful among them with an estimated positive impact on its members' imports from the rest of the world (ROW), but its impact on their exports to the ROW is rather limited. During its first ten years of existence, ECOWAS appears to have had a positive impact on its members' imports from the ROW, but this positive impact vanished over time. SAPTA's negative impact on its members' intra-trade is probably an implicit effect of the India-Pakistan tensions over the estimation period. J.E.L Classification: F11, F15, O50 Keywords: regional trade agreement, kernel regression, trade impact World Bank Policy Research Working Paper 4220, May 2007 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. Acknowledgments: I thank Jeffrey Bergstrand, Paul Brenton, Marius Brülhart, Celine Carrère, Renato Flores, Lionel Fontagné, Guillaume Gaulier, Thierry Mayer, Daniel Mirza, Eric Toulemonde and an anonymous referee for helpful comments and suggestions. 1 Introduction According to official rhetoric, countries involved in a regional trade agreement (RTA) expect a welfare gain. This expectation is so strong that most engage in many different agreements leading to what Bhagwati called the "spaghetti bowl" phenomenon, that is the crisscrossing of many regional agreements differing in their schedules of phasing out tariffs, rules of origin and excluded products. Recent studies of trade effects of developing RTAs come to different conclusions, sometimes for the same RTAs, as depicted in Table 1. Table 1: Trade impact of some developing RTAs Net trade creation Net trade diversion AFTA/ASEAN Carrère (2004) Dee & Gali (2003) Elliott & Ikemoto (2004) Soloaga & Winters (2000) Gosh & Yamarik (2004) Cernat (2001) LAFTA/LAIA Dee & Gali (2003) Carrère (2004) Gosh & Yamarik (2004) Soloaga & Winters (2000) Soloaga & Winters (2000) MERCOSUR Gosh & Yamarik (2004) Carrère (2004) Cernat (2001) Dee & Gali (2003) Soloaga & Winters (2000) Krueger (1999) For instance, AFTA, LAIA and MERCOSUR appear to have been net trade creating in some studies and net trade diverting in others. These studies use different estimation methods, different databases and different dynamic specifications to measure trade effects, and they focus on the number of years these RTAs have existed to estimate their trade impact. Freund and McLaren (1999) introduced an alternative way of looking at RTAs trade effect by focusing on the dynamic of trade orientation when a country joins a regional trade agreement and over the number of years of membership. This paper follows this idea of evaluating the participation effect of each RTA's member. To carry out such analysis, we propose an RTA variable taking into account the number of years of participation of each member, and we use a two-step estimation approach combining a gravity model estimation and a kernel regression of the estimated trade residuals. We focus on seven developing RTAs covering Sub-Saharan Africa (ECOWAS and SADC), Asia (AFTA and SAPTA) and Latin America (CACM, CAN and MERCOSUR) over the period 1960-1999.1 The results indicate that except for SAPTA, all these RTAs have had a positive impact on their members' intra-trade over the estimation period (1960-1999). AFTA 1Appendix 1 describes these RTAs. seems to be the most successful among them with an estimated positive impact on its members imports from the ROW (hence no trade diversion), but its impact on their exports to the ROW is rather limited. During its first ten years of existence, ECOWAS appears to have had a positive impact on its members imports from the ROW (hence no trade diversion), but this positive impact vanished over time. SAPTA's negative impact on its members' intra-trade is probably an implicit effect of the India-Pakistan tensions over the estimation period. The remainder of the paper contains a theoretical and an empirical part. In the theoretical part (section 2), we first describe the RTA variable, then we present the two- step estimation approach. In the empirical part (section 3), we estimate and discuss the trade effect of the selected developing RTAs. Section 4 concludes the paper. 2 Theoretical investigation XRTA-RTA RTA XRTA-ROW XROW-RTA ROW XROW-ROW export flows Figure 1: Geography of World Trade Flows To properly measure the RTAs' trade effect, we focus on export flows of the trading partners in a general equilibrium framework as described in Figure 1. The subset RTA comprises the member countries of one of the seven RTAs under consideration and the subset ROW represents all the remaining countries in the world. 2.1 The RTA variable The usual RTA's dummy variable assesses the impact of the RTA year after year. In this paper, we propose a variable designed to assess the impact of the RTA after a given period of membership. The variable we propose is based on the count of the number of years each member has participated. We thus combine the expansion dimension of the RTA (the evolution of the membership over time) and the cumulative cooperation experience of the members over time. 2 For instance, let us consider the membership of the Central American Common Market (CACM): El Salvador, Guatemala, Honduras and Nicaragua created this RTA in 1960, and Costa Rica joined in 1962. Let us call YP(i,t) the number of years of participation of member country i in the RTA at date t. Table 2 illustrates CACM member participation in 1988, 1990 and 1992. Table 2: Number of years CACM members have participated Years of participation: YP(i,t) 1988 1990 1992 Year: t El Salvador 29 31 33 Guatemala 29 31 33 Honduras 29 31 33 Nicaragua 29 31 33 Costa Rica 27 29 31 Member: i To compute the RTA variable, we distinguish between the exporter (country i) and the importer (country j). Each RTA is thus characterized by three variables representing respectively export flows from a member to a non-member (VRTA-ROW), export flows from a non-member to a member (VROW-RTA), and export flows between members (VRTA-RTA). These variables depend on i, j and t: VRTA (i, j,t)= YP(i,t) if i belongs to RTA and j does not, 0 otherwise -ROW (1) VROW (i, j,t)= YP( j,t) if j belongs to RTA and i does not, 0 otherwise -RTA (2) VRTA (i, j,t)= Min{YP(i,t),YP(j,t)}if i and j belong to RTA, 0 otherwise -ROW (3) To take account of anticipation effects from the beginning of the negotiation of the RTA to the end of the first year of existence, we can start the analysis a certain number of years ahead of the date of entry into force. We arbitrarily choose ten years. This is sufficient to capture any anticipation effect following Freund and McLaren (1999) who estimate this period to be approximately 12 years. Under this hypothesis, the RTA variables become: V~RTA (i, j,t)= YP(i,t)+10 if i belongs to RTA and j does not, 0 otherwise -ROW (4) V~ROW (i, j,t)= YP( j,t)+10 if j belongs to RTA and i does not, 0 otherwise (5) -RTA V~RTA (i, j,t)= Min{YP(i,t),YP(j,t)}+10 if i and j belong to RTA, 0 otherwise (6) -ROW These measures help to take into account the variation in membership and the cumulative 3 cooperation effect over time of the RTA. 2.2 The two-step estimation approach The gravity equation is the most used tool to analyze the trade impact of RTAs. However, regardless of any theoretical base, most of the empirical papers addressing RTAs' trade impact impose a linear relationship between RTAs and trade flows through the inclusion of dummy variables. A non-parametric approach would let the data impose the relevant structure to the RTA-Trade flows relation and this paper proposes an estimation approach in this vein. We proceed in two steps. 2.2.1 First step: the gravity equation estimation First, we have to estimate a simple gravity model not including any RTA measures. In the empirical trade literature, many recent papers have revisited the formulation of the gravity equation by proposing different set of dummy variables to be included to control for the price and the remoteness term. Among these papers, we can mention Baier and Bergstrand (2002), Anderson and van Wincoop (2003), Martinez-Zarzoso and Nowak- Lehmann (2003), and Cheng and Wall (2005). The recent paper by Baldwin and Taglioni (2006) summarizes this debate and describes the common errors made in this empirical literature. Basically, three current errors are made: the inadequate deflation of trade flows by CPI, a misleading bilateral trade average (taking the log of average bilateral trade instead of the average of the log of bilateral trade), and the omission or the incorrect inclusion of the multi-lateral resistance term. All these errors lead to biased estimates of the trade impact of any trade policy. Baldwin and Taglioni propose some improvements of the empirical estimation of gravity equations: use unidirectional trade flows and include country-pair and time dummy variables, or country-time dummy variables. Both options correct for the inadequate deflation of the trade flows, but correct for omission of the multi-lateral resistance only partially. Baldwin and Taglioni's preferred specification is to include both country-pair and country-time dummy. However, they acknowledge that since most of the trade policies examined by trade economists are country-pair specific, this approach alters the estimation of the trade impact of these policies. Against this backdrop, we propose two specifications incorporating most of the suggestions of Baldwin and Taglioni. The first specification includes country-pair and time dummies, the second includes country-time dummies and some bilateral geographical variables to partially control for the omitted country-pair dummies: LnXijt =1LnGDPit +2LnGDPjt + 1LnPOPit + 2LnPOPjt +LnRERijt +t+0 + FEij +FEt +ijt (7) 4 LnXijt =0LnDistij +1LnGDPit +2LnGDPjt + 1LnPOPit + 2LnPOPjt +LnRERijt +t+0 + FEit + Geoij +ijt k (8) k where Xijt is country i's export to country j at period t, Distij is the distance between country i and j, GDPit is the GDP of country i in year t, POPit is the population of country i in year t, RERijt is a measure of the real exchange rate between country i and j in year t, t is the time trend so that measures the long term effect of time on trade flows, 0 is an intercept common to all years and country-pairs, FEij (with FEij FEji) is the country- pair fixed effects, FEit is the exporter-year fixed effects, FEt is year fixed effects, and Geoij is a set of k bilateral geographical variables.2 Following Rose (2003), we consider k the following bilateral geographical variables: Border (sharing a common border), Colony (colonizer-colony relationship), Comcol (sharing a common colonizer), Comlang (sharing a common language), and Curcol (currently in a colony-colonizer relationship). ijt is the error term. 2.2.2 Second step: the non-parametric estimation The estimated residuals of these two equations are extracted and used in the second step for the non-parametric part of the estimation. Imagine a scatter plot depicting the estimated trade residuals (^ijt ) against one of the three RTA variables described in the previous section (V~RTA (i, j,t)). The point is to evaluate the non- -ROW (i, j,t),V~ROW -RTA(i, j,t), or V~RTA -RTA parametric function f () underlining the variation of ^ijt in accordance with . V~RTA (i, j,t)) by using a kernel estimator: -ROW (i, j,t),V~ROW -RTA(i, j,t), or V~RTA -RTA E ijt VRTA ( (i, j,t)) -ROW (i, j,t))= f^(VRTA -ROW (9) E ijt VROW ( (i, j,t)) -RTA (i, j,t))= f^(VROW -RTA (10) E ijt VRTA ( (i, j,t)) -RTA (i, j,t))= f^(VRTA -RTA (11) where: 2Our real exchange rate variable is inspired by Soloaga and Winters (2000): RERijt = e× US / i . e× US / ( )( ) where ,t ,t ,t j,t e is the value of 1 US $ evaluated in the currency of country i and is the GDP deflator. 5 f^(x) = n i=1K(xi ).^ijt / n (12) n i=1K(xi )/ n where n is the number of observations, n is an a-priori chosen sequence of positive numbers called the window width parameter and K() . is an a-priori chosen real function called the kernel, and satisfying K(x)dx < and K(x)dx = 1. Bieriens (1994) analyses the asymptotic property of this estimator, and shows that it is asymptotically normal, that is: n f (x)- f (x) N(0,V (x)) [^ ] (13) where V (x) depends on the characteristics of the kernel function K(x). Bierens shows that the specific choice of the kernel function is not crucial: any Gaussian kernel is relevant. More important is the choice of the bandwidth that controls the trade-off between bias and variance of the estimated trade effects. Since the RTA variables are discrete variables (number of years of participation), we choose a bandwidth n =1so as to smooth trade effects over a one-year period. Bierens (1994) describes in detail how to use equation (13) to directly build the Confidence Interval of the estimated trade effects. 3 Empirical Analysis In this section, we present and discuss the data used to evaluate the trade effect of the seven developing RTAs under consideration. 3.1 Data and estimation issues Our database comes from Rose (2003) completed with data on export price index from IFS. We divided the export values by the export price indices to obtain export quantities. The final database is an unbalanced panel containing 56 exporter and 90 importer countries over the period 1960-1999 (see the list in Appendix 2). It contains no zero trade flow and only 8% of export values are missing. We thus use a simple regression model to estimate the gravity models (7) and (8). Since we are using fixed effects, the estimators are not biased because of the unbalancedness of the database; however, we use the Huber/White estimator of the variance to correct for the potential heteroscedasticity problem. 6 The estimation results of the gravity equations are reported in Appendix 3.3 Specification 1 corresponds to equation (7) including country-pair and year fixed effects and Specification 2 corresponds to equation (8) including exporter-year fixed effects. In Appendix 3, a parameter with an upper index a is significant at the 1% level, that with an upper index b is significant at the 5% level and that with an upper index c is significant at the 10% level. The traditional gravity variables (distance, GDP and Population) depict the expected sign and magnitude in the two specifications. The estimated coefficients of the real exchange rate variable are negative, indicating a slight decreasing competitiveness among trading partners over the period 1960-1999 after controlling for the traditional gravity variables. The time trend is not statistically significant in Specification 1. The bilateral geographical variables' coefficients are statistically significant with the expected signs and magnitudes. In the second step, we extracted the estimated trade residuals from equations (7) and (8), and run a kernel regression as described in Section 2.2. We then used equation (13) to build the confidence interval as follows: for each grid point, we consider the standard deviation () from equation (13) and use it to compute the 95% confidence interval of the trade effects defined as ±1.96× , where = 1/(12n) , n being the total number of years of existence of a given RTA and 1/12 being the variance of the uniform distribution.4 The results are presented graphically in Appendix 4. Following Baldwin and Taglioni (2006), we choose the specification including the exporter-year fixed effects as our preferred estimation and comment the results in the next section. 3.2 The trade effect of some developing RTAs The ASEAN Free Trade Agreement (AFTA) was created in 1992 by six members of the Association of South East Asian Nations (Brunei Darussalam, Indonesia, Malaysia, Philippines, Singapore and Thailand), four other members joined subsequently (Vietnam in 1995, Laos and Myanmar in 1997, Cambodia in 1999). The AFTA members included in the estimation as exporter and importer are Indonesia, Malaysia, Philippines Singapore and Thailand, the remaining members being included only as importers. Figure 2 of 3We do not report exporter-year, country-pair and year fixed effects to save space. Note also that since equation (8) introduces exporter-years fixed effects, the variables with the index it are absorbed in this specification. 4See Bierens (1994) for a full explanation of this process. 7 Appendix 4 plots the estimated trade residuals against the AFTA membership evolution over time: the top panel focuses on intra-AFTA exports ( X AFTA-AFTA ), the middle panel focuses on AFTA's imports flows from the ROW ( X ROW -AFTA ) and the bottom panel focuses on AFTA's exports to the ROW ( X AFTA -ROW ). The dashed lines represent the estimated 95% confidence interval. These graphs clearly show an anticipation effect of AFTA members which started increasing their intra-trade five years before the official year of joining this RTA. In addition, the trade effect of AFTA seems to be globally positive over the estimation period since its effect on intra-AFTA exports and imports from the ROW are estimated to be positive and increasing. However, its impact on export flows remained neutral. The Central American Common Market (CACM), was created in 1960 by El Salvador, Guatemala, Honduras, Nicaragua. Costa Rica joined in 1962. It is notified at the WTO as a Customs Union. Except for El Salvador included as importer only, all the CACM members are both exporter and importer in the database. Figure 3 of Appendix 4 plots the estimated trade residuals against the number of years of each CACM member's participation. The RTA's impact on intra-CACM exports are estimated to be negative during the first years of its existence, and then it became positive and increasing over time. The tensions between El Salvador and Honduras in the late sixties may explain this initial negative impact.5 The RTA's impact on its members' exports to and imports from the ROW are estimated to be negative and sometimes decreasing, a result suggesting an overall ambiguous trade effect of the CACM. The Andean Community (CAN) is a preferential agreement signed in 1988 by Bolivia, Colombia, Ecuador, Peru and Venezuela. Except for Venezuela included as importer only, all the other CAN members are both exporter and importer in the database used for the estimations. Figure 4 of Appendix 4 plots the estimated trade residuals against the number of years of CAN members' participation. Intra-CAN exports seem to have started increasing three years before the official date of entry into force of this RTA. It remained positive and increasing over the estimation period. However, the RTA's effects on imports from and exports to the ROW are estimated to be negative or neutral. The Economic Community of West African States (ECOWAS) is a political association created in 1975 by fifteen members (Mauritania withdrew in 1999): Benin, Burkina Faso, Cape Verde, Côte d'Ivoire, Gambia, Ghana, Guinea, Guinea-Bissau, Liberia, Mali, Niger, Nigeria, Senegal, Sierra Leone, Togo. Except for Burkina Faso, Côte d'Ivoire, Liberia, Nigeria, Senegal and Togo that are included as exporter and 5In fact, the CACM collapsed in 1969 after a five-day war that had been known as the "soccer war" between El Salvador and Honduras. After this episode, the partners tried to slowly re-establish their collaboration. This may explain the abnormal trade effects observed. We may also notice that in Figure 4 of appendix 5, the CACM trade flows are limited to two years before the official date of entry into force (1962 for Costa Rica) because the database used is limited on the period 1960-1999. 8 importer, the remaining members are included in the database as importers only. Figure 5 in Appendix 4 plots the estimated trade residuals against the number of years these countries have participated in the ECOWAS. These graphs indicate a slight anticipation effect of ECOWAS members five years before the official date of its creation. The RTA's impact on intra-ECOWAS trade flows is estimated to be positive and increasing over the estimation period, while its impact on its members exports to the ROW is negative and decreasing over time. During the first ten years of the existence of the RTA, its impact on its members imports from the ROW was estimated to be positive, but this result was reversed after. The overall trade impact of the ECOWAS is thus ambiguous. The Southern Common Market (MERCOSUR) was established in 1991 between Argentina, Brazil, Paraguay and Uruguay. Except for Uruguay included as importer only, the other MECOSUR members are included as exporter and importer in the database. Figure 6 of Appendix 4 plots the estimated export volume residuals against the number of years of member participation. These graphs indicate that MERCOSUR members were very involved in intra-trade five years before the official date of implementation of this RTA. The RTA's impact on its members intra-trade is estimated to be positive and increasing over time, while its impact on their imports from the ROW is negative. The RTA appears to have had no impact on its members exports to the ROW. The South African Development Community (SADC) is a political association created in 1992 by fourteen members: Angola, Botswana, Democratic Republic of Congo, Lesotho, Malawi, Mauritius, Mozambique, Namibia, Seychelles, South Africa, Swaziland, Tanzania, Zambia, Zimbabwe. Except for Malawi, Mauritius, Seychelles, South Africa, Zambia and Zimbabwe included as importer and exporter in the database, the other SADC members are included as importers only. Figure 7 in Appendix 4 plots the estimated trade residuals against the number of years of SADC member participation. Figure 7 reveals an anticipation effect of SADC members depicted by a continuous increase in the intra-SADC trade flows five years before the official implementation date. The RTA's impact on its members' intra-trade is estimated to be positive and increasing. However, its impact on their exports to or imports from the ROW are estimated to be slightly negative. The last RTA analyzed is the South Asian Preferential Trade Agreement (SAPTA) comprising Bangladesh, Bhutan, India, Maldives, Nepal, Pakistan, and Sri- Lanka. The treaty creating the SAPTA was signed on April 1993, and it enters into force in December 1995. Except for India, Pakistan and Sri-Lanka included as importers and exporters, the other members are included only as importers in the database. Figure 8 of Appendix 4 plots the estimated trade residuals against the number of years of SAPTA members' participation. The RTA's impacts on its members intra-trade and imports from the ROW are estimated to be negative, while its impact on the members exports to the ROW is estimated to be neutral. The recurrent tensions between India and Pakistan over the estimation period may explain the negative impact on intra-SAPTA trade flows. 4 Conclusion 9 This paper proposes two contributions to the evaluation of RTAs' trade impacts. First, we use an RTA variable that takes into account the number of years each member has participated instead of the usual RTA dummy variable. Second, we combine traditional gravity regressions with non-parametric estimation techniques so as to capture the non- monotonic trade effects while imposing minimal structure on the model. We focus on a panel of seven developing RTAs covering Africa, Asia and Latin America. Except for SAPTA, all these RTAs appear to have had a positive impact on their members' intra-trade over the estimation period (1960-1999). AFTA seems to be the most successful of these RTAs with an estimated positive impact on its members imports from the ROW (hence no trade diversion), but its impact on their exports to the ROW is rather limited. During its first ten years of existence, ECOWAS has had a positive impact on its members imports from the ROW (hence no trade diversion), but this positive impact vanished over time. SAPTA's negative impact on its members' intra-trade is probably an implicit effect of the India-Pakistan tensions over the estimation period. This work is based on the up-to-date formulation of the gravity model and the proposed semi-parametric estimation approach can be easily implemented to rigorously assess the trade impact of developing RTAs. It could be improved and used as a key diagnostic tool to evaluate the trade impact of the various RTAs signed between many World Bank clients. 10 References Anderson, J.E., van Wincoop, E., 2003. Gravity with Gravitas: a Solution to the Border Puzzle. American Economic Review, 93:170-192. Baier, S.L., Bergstrand, J.H, 2002. On the Endogeneity of International Trade Flows and Free Trade Agreements. American Economic Association Annual Meeting. Bierens, H.J., 1994. Topics in Advanced Econometrics. Cambridge University Press. Baldwin, R., Taglioni, D., 2006. Gravity for Dummies and Dummies for Gravity Equations. NBER working papers No. 12516. Carrere, C., 2006. Revisiting the Effects of Regional Trade agreements on Trade Flows with Proper Specification of the Gravity Model. European economic Review, 50:223- 247. Cernat, L., 2001. Assessing Regional trade Arrangements: Are South-South RTAs More Trade Diverting? Global economy Quarterly, 2(3):235-59. Cheng, I.H., Wall, H.J., 2005. Controlling for Heterogeneity in Gravity Model of Trade and Integration. The Federal Reserve Bank of St. Louis Review, 87(1):49-63. Dee, ph., Gali, J., 2003. The Trade and Investment Effect of Preferential Trade Arrangements. NBER Working Paper No. 10160. Elliot, R., Ikemoto, K., 2004. AFTA and the Asian Crisis: Help or Hindrance to the ASEAN Intraregional Trade? Asian Economic Journal, 18(1):1-10. Freund, C.L., McLaren, J., 1999. On the dynamics of Trade Diversion: Evidence from Four Trade Blocs. International Finance Discussion Paper no. 637. Gosh, s., Yamarik, S., 2004. Are Regional Trading Arrangements Trade Creating? An application of Extreme Extreme Bounds Annalysis. Journal of International Economics, 63(2):369-395. Krueger, A., 1999. Trade Creation and Trade Diversion under NAFTA. NBER Working Paper No. 7429. Martinez-Zarsoso, I., Nowak-Lehman, F., 2003. Augmented Gravity Model: An Empirical Application to MERCOSUR-EU Trade Flows. Journal of applied economics, 6(2):291-316. Rose, A., 2003. Does the WTO Make Trade More Stable? NBER Working Paper No. 10207. Soloaga, I., Winters, A.L., 2000. Regionalism in the Nineties: What Effect on Trade? North American Journal of Economics and Finance, 12(1):1-29. 11 Appendix Appendix 1: A Panel of Developing RTAs Agreement Full name Membership evolution Type ECOWAS Economic 1975: Benin Political Community 1975: Burkina Faso Association Of West Africa 1975: Cape Verde 1975: Côte d'Ivoire 1975: Gambia 1975: Ghana 1975: Guinea 1975: Guinea Bissau 1975: Liberia 1975: Mali 1975: Niger 1975: Nigeria 1975: Senegal 1975: Sierra Leone 1975: Togo Agreement Full name Membership evolution Type SADC South African 1992: Angola Political Development 1992: Botswana Association Community 1992: DR Congo 1992: Lesotho 1992: Malawi 1992: Mauritius 1992: Mozambique 1992: Namibia 1992: Seychelles 1992: South-Africa 1992: Swaziland 1992: Tanzania 1992: Zambia 1992: Zimbabwe Agreement Full name Membership evolution Type CAN Andean 1988: Bolivia Preferential Community 1988: Columbia Arrangement 1988: Ecuador 1988: Peru 1988: Venezuela 12 Agreement Full name Membership evolution Type CACM Central 1960: El Salvador Customs American 1960: Guatemala Union Common 1960: Honduras Market 1960: Nicaragua 1962: Costa Rica Agreement Full name Membership evolution Type MERCOSUR Southern 1991: Argentina Customs Common 1991: Brazil Union market 1991: Paraguay 1991: Uruguay Agreement Full name Membership evolution Type AFTA ASEAN 1992: Brunei Darussalam Political Free Trade 1992: Indonesia Association Agreement 1992: Malaysia 1992: Philippines 1992: Singapore 1992: Thailand 1995: Vietnam 1997: Laos 1997: Myanmar 1997: Cambodia Agreement Full name Membership evolution Type SAPTA South Asia 1995: Bangladesh Preferential Preferential 1995: Bhutan Agreement Trade 1995: India Agreement 1995: Maldives 1995: Nepal 1995: Pakistan 1995: Sri Lanka 13 Appendix 2: Exporter and importer countries Code Country Exporter Importer Code Country Exporter Importer 111 United States yes yes 522 Cambodia no yes 112 United Kingdom yes yes 524 Sri Lanka yes yes 122 Austria no yes 534 India yes yes 124 Belgium yes yes 536 Indonesia yes yes 128 Denmark yes yes 542 Republic of Korea yes yes 132 France yes yes 544 Lao People's Dem Rp no yes 134 Germany yes yes 548 Malaysia yes yes 136 Italy yes yes 556 Maldives no yes 137 Luxembourg no yes 558 Nepal no yes 138 Netherlands yes yes 564 Pakistan yes yes 142 Norway yes yes 566 Philippines yes yes 144 Sweden yes yes 576 Singapore yes yes 146 Switzerland yes yes 578 Thailand yes yes 156 Canada yes yes 582 Viet Nam no yes 158 Japan yes yes 614 Angola no yes 172 Finland yes yes 616 Botswana no yes 174 Greece yes yes 624 Cape Verde no yes 176 Iceland yes yes 636 Congo, Dem Rep of no yes 178 Ireland yes yes 638 Benin no yes 181 Malta yes yes 648 Gambia no yes 182 Portugal yes yes 652 Ghana no yes 184 Spain yes yes 654 Guinea-Bissau no yes 186 Turkey yes yes 656 Guinea no yes 193 Australia yes yes 662 Côte D'Ivoire yes yes 196 New Zealand yes yes 664 Kenya yes yes 199 South Africa yes yes 666 Lesotho no yes 213 Argentina yes yes 668 Liberia yes yes 218 Bolivia yes yes 676 Malawi yes yes 223 Brazil yes yes 678 Mali no yes 233 Colombia yes yes 682 Mauritania no yes 238 Costa Rica yes yes 684 Mauritius yes yes 248 Ecuador yes yes 688 Mozambique no yes 253 El Salvador no yes 692 Niger no yes 258 Guatemala yes yes 694 Nigeria yes yes 268 Honduras yes yes 698 Zimbabwe yes yes 273 Mexico no yes 718 Seychelles yes yes 278 Nicaragua yes yes 722 Senegal yes yes 288 Paraguay yes yes 724 Sierra Leone no yes 293 Peru yes yes 728 Namibia no yes 298 Uruguay no yes 734 Swaziland no yes 299 Venezuela no yes 738 Tanzania no yes 513 Bangladesh yes yes 742 Togo yes yes 514 Bhutan no yes 746 Uganda no yes Brunei 516 Darussalam no yes 748 Burkina Faso yes yes 518 Myanmar no yes 754 Zambia no yes 14 Appendix 3: Gravity equation estimations Dependent variable: LnXijt 1 2 LnDistij -1.23a LnGDPit 1.34a LnGDPjt 0.90a 1.39a LnPOPit -0.06c LnPOPjt -0.53a -0.57a LnRERijt -0.002a -0.004a t -0.008 Border 0.16a Colony 1.29a Comcol 0.86a Comlang 0.37a Curcol 1.50a Constant -23.44a 1.96a N 123,205 123,205 R2 0.45 0.29 P-value 0.00 0.00 15 Appendix 4: Figures Exporter-time fixed effects Country-pair and year fixed effects 1 1 rt t poxe expor ATFA 0 a- ntrI ATFAartnI 0 -1 -1 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Years Years Exporter-time fixed effects Country-pair and year fixed effects 1 1 WOR WOR omrf 0 omrft 0 port mI pormI -1 -1 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Years -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Years Exporter-time fixed effects Country-pair and year fixed effects 1 1 WOR WOR totropxE 0 totrop 0 Ex -1 -1 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Years Years Figure 2: AFTA trade effects 16 Exporter-time fixed effects Country-pair and year fixed effects 3 3 2 2 tropxe 1 tropxe 1 ATFA 0 a-tr In -1 MCACatrIn 0 -1 -2 -2 -3 -3 -9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36 39 -9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36 39 Years Years Exporter-time fixed effects Country-pair and year fixed effects 3 3 2 2 WOR 1 1 WOR omrftr 0 omrft 0 po Im -1 pormI -1 -2 -2 -3 -3 -9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36 39 -9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36 39 Years Years Exporter-time fixed effects Country-pair and year fixed effects 3 3 2 2 1 1 WOR WOR tot 0 porxE totropxE 0 -1 -1 -2 -2 -3 -3 -9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36 39 -9 -6 -3 0 3 6 9 12 15 18 21 24 27 30 33 36 39 Years Years Figure 3: CACM trade effects 17 Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 tropxe tropxe NAC-atr 0 In NACatrIn 0 -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 Years Years Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 WOR WOR mofrtrop 0 morftrop 0 mI Im -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 Years Years Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 WOR tot 0 WORott 0 porxE porxE -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 Years Years Figure 4: CAN trade effects 18 Exporter-time fixed effects Country-pair and year fixed effects 2 1 tro 1 xpe tropxe SA SA WOCE-art 0 WOCEatrIn 0 In -1 -1 -9 -6 -3 0 3 6 9 12 15 18 21 24 -9 -6 -3 0 3 6 9 12 15 18 21 24 Years Years Exporter-time fixed effects Country-pair and year fixed effects 1 1 WOR WOR romftr 0 omrft 0 po mI pormI -1 -1 -9 -6 -3 0 3 6 9 12 15 18 21 24 -9 -6 -3 0 3 6 9 12 15 18 21 24 Years Years Exporter-time fixed effects Country-pair and year fixed effects 1 1 WOR WOR totropxE 0 totr 0 poxE -1 -1 -9 -6 -3 0 3 6 9 12 15 18 21 24 -9 -6 -3 0 3 6 9 12 15 18 21 24 Years Years Figure 5: ECOWAS trade effects 19 Exporter-time fixed effects Country-pair and year fixed effects 2 2 tropxe 1 tropxe 1 RUS RUS OCRE 0 OCRE 0 M-a M-atr Intr -1 In -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 Years Years Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 OWR WOR omrftrop 0 omrft 0 por Im mI -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 Years Years Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 WOR WOR totropxE 0 ott 0 porxE -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 Years Years Figure 6: MERCOSUR tradeeffects 20 Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 tr t poxe porxe CDAS-atr 0 In CDASarntI 0 -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Years Years Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 WOR WOR omrftr 0 omrft 0 po mI pormI -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Years Years Exporter-time fixed effects Country-pair and year fixed effects 2 2 1 1 WOR ottrop 0 WORott 0 Ex porxE -1 -1 -2 -2 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Years Years Figure 7: SADC trade effects 21 Exporter-time fixed effects Country-pair and year fixed effects 1 1 tr tr poxe poxe ATPAS 0 ATPAS 0 a-rtnI a-rtnI -1 -1 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 Years Years Exporter-time fixed effects Country-pair and year fixed effects 1 1 WOR mofrtrop WOR 0 ottrop 0 mI Im -1 -1 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 Years Years Exporter-time fixed effects Country-pair and year fixed effects 1 1 WOR WOR totropxE 0 totropxE 0 -1 -1 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 Years Years Figure 8: SAPTA trade effects 22