Do Customs Union Members Indulge In More Bilateral Trade Than Free Trade Agreement Members?

Transcription

1 Do Customs Union Members Indulge In More Bilateral Trade Than Free Trade Agreement Members? Jayjit Roy * Abstract Fiorentino et al. (2007) question the popularity of customs unions (CUs) relative to that of Free Trade Agreements (FTAs) and consider the former to be out of tune with today s trading climate. In such a scenario, a comparison of FTAs and CUs seems especially relevant. This paper provides the first empirical analysis to directly compare the effect of the two preferential regimes, on members bilateral trade, while addressing the biases arising from log-linearization of the gravity model and omission of time-invariant unobservables. Using the gravity model and data from Baier and Bergstrand (2007), striking results are obtained. Baier and Bergstrand (2007) find that on average, an FTA approximately doubles two members bilateral trade after 10 years. However, findings of this paper show that it is actually a CU, and not an FTA, which doubles the members bilateral trade after 10 years. JEL: F1 Keywords: Free Trade Agreements, Customs Unions, Gravity Model * Department of Economics, Southern Methodist University, Dallas, TX jroy@smu.edu. The author wishes to thank Scott Baier and Jeffrey Bergstrand for generously providing the data, Daniel Millimet for helpful comments and seminar participants at SMU.

2 1 Introduction In today s trading climate, the relevance of analyzing Preferential Trade Agreements (PTAs) cannot be emphasized enough. The e ective number of such international agreements exceed two hundred with Mongolia being the only World Trade Organization (WTO) member not party to one (Saggi and Yildiz 2007). The noti cation of more than fty of these to the WTO between January 2005 and December 2006, coupled with the ongoing negotiations of numerous agreements indicate their recent proliferation and unabated rise in years to come. If all the PTAs currently under negotiation and proposal are implemented, then one would be looking at over four hundred PTAs by 2010 (Fiorentino et al. 2007). Hence, policy issues associated with trade agreements seem relevant for some time to come. Any PTA is essentially an arrangement among countries whereby members engage in trade at reduced tari rates. In most cases, quantitative restrictions are also dismantled (Krueger 1999). Such bene ts are typically not extended to non-members. The arrangements may be partial or total with respect to the extent of duty reduction or commodity coverage (Krueger 1997). However, partial scope agreements, which typically involve a reduction or elimination of duties on certain goods only, are not explicitly considered in this paper. 1 The remaining set of agreements can be classi ed into two categories. If the members eliminate tari s internally while maintaining their individual external tari s, a Free Trade Area or Free Trade Agreement (FTA) is formed. In case they also unify their external tari s, the arrangement is termed a Customs Union (CU). It is this common external tari (CET) which essentially separates a CU from an FTA. Another important distinction between the two, which follows from the CET, is the extent of the role of rules of origin (ROO). The ROO are essentially restrictions on the preferential treatment of goods not produced or su ciently transformed by a member country. 2 In case of an FTA, due to the absence of a CET, they perform the additional function of preventing a good from being imported by an FTA member, with the minimum duty for it, and subsequently reexported to the other members preferentially. Prominent FTAs include the North American Free Trade Agreement (NAFTA) whereas the MERCOSUR comprises an example of a CU. 3 The literature on preferential agreements, dating back to at least as early as Viner (1950), has addressed a host of associated policy issues. These issues include but are not limited to the building and stumbling bloc e ects of PTAs with 1 Fiorentino et al. (2007) consider partial scope agreements to be characterized by poor implementation record and scarce visibility. In fact, they also consider such agreements to be often accompanied by a commitment to future negotiations on further integration. 2 Bhagwati et al. (1999, p.543) provide an account of the alternative criteria used in de ning ROO. 3 One may also go further up the scale of integration and consider a common market, which allows considerably free movement of factors of production as well. However, a common market is essentially an arrangement where countries form a CU and then permit increased mobility of the factors (Krueger 1997). Accordingly, such regimes of further integration (e.g. The European Union) are considered as CUs in this paper. 1

3 respect to multilateral liberalization (see, e.g., Bhagwati 1991; Limao 2006), and their trade creation and trade diversion e ects (see, e.g., Clausing 2001; Eicher et al. 2007). The contributions, both theoretical and empirical, have usually relied on regional or preferential trading regimes in general (see, e.g., Magee 2003) or, a particular type of PTA (see, e.g., Saggi and Yildiz 2007) in order to draw relevant conclusions. Some studies have also analyzed speci c trade agreements (see, e.g., Romalis 2007; Chang and Winters 2002) or trade agreements with respect to one particular country (see, e.g., Limao 2006). However, analyses pertaining to a comparison of the generic types of PTAs have received relatively less attention. Perhaps Krueger (1997) best expresses this, stating: Surprisingly... there has been little analysis of di erent types of preferential arrangements, and in particular, of free trade agreements in contrast to customs unions. Clausing (2000), the only contribution after Krueger(1997) in terms of directly comparing FTAs and CUs, also alludes to this lack of attention. However, both Krueger (1997) and Clausing (2000) are theoretical contributions. While Krueger (1997) nds a CU to be Pareto-superior to an FTA, Clausing (2000) generates conditions that determine when customs unions are preferred to free trade areas. Accordingly, the empirical literature seems to be even more lacking in this respect. 4 This paper lls the gap by analyzing whether countries belonging to a CU engage in more bilateral trade (in goods) than countries belonging to an FTA, on average. Issues pertaining to trade creation, trade diversion and welfare comparisons are beyond the scope of this paper. A comparison of FTAs and CUs seems to be of greater relevance today. Although the emergence of FTAs is a more recent phenomenon compared to that of CUs, the former account for 84% of all the PTAs noti ed and in force (Fiorentino et al. 2007). 5 The proportion of FTAs relative to CUs is even higher if one considers the PTAs currently under negotiation. Accordingly, Fiorentino et al. (2007) question the popularity of CUs and consider them to be out of tune with today s trading climate. The more restrictive nature of CUs in terms of the members trade relations with countries outside the union, requirement of a greater degree of harmonization among members, and longer implementation periods are o ered as possible explanations. In this scenario the ndings of this paper are especially signi cant. Using the gravity model and the data from Baier and Bergstrand (2007), the paper compares the e ect of FTAs and CUs on the members volume of bilateral trade. Once the biases arising from log-linearization of the gravity model or, the omission of time-invariant unobservables are addressed, the results are striking. Baier and Bergstrand (2007) nd that on average, an FTA approximately doubles two members bilateral trade after 10 years. However, using the same data, the results of this paper indicate that it is a CU, instead of an FTA, which is responsible for this doubling e ect. The nding is extremely relevant 4 Ghosh and Yamarik (2004) and Magee (2007) are empirical contributions, which allow for di erential e ects of the degree of integration. Both analyses primarily allude to the trade creation and trade diversion issue. However, Magee (2007) nds CUs to in uence trade over a longer period of time than FTAs. 5 CUs and partial scope agreements constitute 8% each. 2

4 for policy makers and interesting in the wake of diminishing popularity of CUs. Hence, analyses which do not allow the e ects of FTAs and CUs to di er, fail to capture this crucial aspect of trade policy decisions. The remainder of the paper is organized as follows. Section 2 describes the empirical methodology. Section 3 discusses the data. Section 4 presents the results, while Section 5 concludes. 2 Empirical Methodology 2.1 Cross-section Analysis Gravity models are estimated - in levels and logs - to compare the e ects of FTAs and CUs. The level speci cation is given by T ij = 0 D 1 ij exp( 2lang ij + 3 adj ij + 4 F T A ij + 5 CU ij (1) + i ctry i + j ctry j ) ij Here T ij is the nominal value of exports from country i to country j; D ij is the distance between i and j; lang ij is a dummy variable taking the value one if i and j share a common language (zero otherwise); adj ij is a binary variable assuming the value unity if i and j share a land border (zero otherwise); F T A ij (CU ij ) is a dummy variable taking the value one if i and j are part of an FTA (CU) and zero otherwise; and ctry i and ctry j are country-speci c dummies. 6 Santos Silva and Tenreyro (2006) show that (1) may be estimated using an estimator that is numerically equivalent to the Poisson pseudo-maximum likelihood (PPML) estimator, provided E ij jd ij ; lang ij ; adj ij ; F T A ij ; CU ij ; ctry i ; ctry j = 1: (2) The log speci cation is instead given by ln (T ij ) = ln ln D ij + 2 lang ij + 3 adj ij (3) Consistent estimation of (3) requires + 4 F T A ij + 5 CU ij + i ctry i + j ctry j + ln ij E ln ij jdij ; lang ij ; adj ij ; F T A ij ; CU ij ; ctry i ; ctry j = 0: (4) However, as noted by Santos Silva and Tenreyro (2006), (2) does not imply (4) (invoking Jensen s inequality); in fact the elasticity estimates from the log-linearized model may be biased if the level speci cation su ers from heteroskedasticity. Henderson and Millimet (fothcoming) also nd this concern well-founded and recommend estimating the gravity model in levels. Also, estimating the model in levels avoids the omission of observations with zero trade ows or the use of other ad hoc measures to address it. 6 The country-speci c dummies are usually used to control for country-speci c unobservables that do not vary across trading partners. In this case they also require the GDP variables to be dropped. An alternative, which is not considered here, is to impose unit income elasticities and consider the dependent variable as the volume of bilateral trade relative to the product of the GDPs. 3

5 2.2 Panel Analysis The cross-section estimates are likely to su er from a bias due to the endogenous trade agreement dummies. An excellent account of the endogeneity issue of the trade agreement dummies and the failure of previous studies to address it can be found in Baier and Bergstrand (2007), who further state that the omitted variable (selection) bias is the major source of endogeneity facing estimation of FTA e ects in gravity equations using cross-section data. Although Magee (2003) attempts to address the issue by relying on the instrumental variables (IV) method, the quality of the instruments used is clearly suspect. It is unlikely that the instruments like GDP similarities between two countries or di erences in their relative factor endowments would be uncorrelated with the unobservables a ecting their volume of trade with each other. The relative di culty of coming up with an instrument, which is correlated with the likelihood of two countries forming a trade agreement and also uncorrelated with the unobservables a ecting their volume of bilateral trade compels Baier and Bergstrand (2007) to conclude that IV estimation is not a reliable method for addressing the endogeneity bias of the trade agreement dummies. The Heckman control function approach also su ers from the lack of a suitable exclusion restriction. 7 However, the xed e ects panel approach, with pairs of countries as the basic units of observation, addresses the endogeneity issue to a certain extent. It allows one to control for time invariant unobservables, which a ect the volume of trade between a pair of countries and are also correlated with their decision to form a trade agreement. For example, as discussed in Baier and Bergstrand (2007), the volume of trade between a pair of countries depends on their domestic policies. A strict domestic policy in the form of internal shipping regulations may reduce the volume of goods that they trade with each other. However, the countries are likely to form a trade agreement if they expect welfare gains from potential trade creation, provided the agreement deepens liberalization beyond tari barriers and into domestic regulations. Such domestic regulations constitute an example of bilateral or pairwise unobservables, which are correlated with the volume of trade a pair of countries engage in along with their decision to form a trade agreement. Estimates from the panel xed e ects method do not su er from a bias due to the presence of such time-invariant unobservables and hence are a de nite improvement over the cross-section estimates. 8 However, the bias arising from the logs versus levels speci cation is a separate issue. Panel estimates from the log-linearized model may still be biased in the presence of heteroskedasticity in the levels model. The fact that the location-speci c dummies also fail to address this bias can probably be best expressed in the words of Santos Sliva and Tenreyro (2006), who state that even controlling for xed e ects, the presence of heteroskedasticity can gen- 7 Although it is possible to apply the control function approach without using an exclusion restriction, it would entail a non-linear term in the gravity equation. However, Henderson and Millimet (forthcoming) fail to reject the linear functional form of the gravity equation. 8 The method also addresses the issue of measurement error in the time-invariant regressors, such as distance. 4

6 erate strikingly di erent estimates when the gravity equation is log-linearized, rather than estimated in levels. Following Santos Silva and Tenreyro (2006), Henderson and Millimet (forthcoming) and the cross-section results (to be discussed in a later section), the levels version of the gravity model is used for the panel xed e ects method. Accordingly, the panel speci cation is given by T ijt = 0 D 1 ij exp( 2lang ij + 3 adj ij + 4 F T A ijt + 5 CU ijt (5) + it ctry it + jt ctry jt ) ijt ij The dependent variable is the real value of exports from country i to country j, at time t. The other variables have the same notation except the subscript t on variables which are not time-invariant. Accordingly, F T A ijt (CU ijt ) takes the value one if i and j are part of an FTA (CU) at time t and zero otherwise; and ctry it and ctry jt are the country-by-time dummies. The unobservable term is decomposed into time-varying and time-invariant components such that ijt = ijt ij : The panel xed e ects method provides consistent estimates even in the presence of any correlation between the bilateral time-invariant unobservables ij and the trade agreement dummies. Since trade agreements usually have a phase-in period, some of the panel speci cations, discussed in the results section, include lag and lead terms of the trade agreement dummies to capture any lagged or anticipatory e ects of the agreements. Baier and Bergstrand (2007) recommend the inclusion of lag terms also due to the fact that trade agreements involve changes in countries terms of trade, which tend to have lagged e ects on trade volumes. In other words, some panel speci cations include variables like F T A ij;t k and CU ij;t k or, F T A ij;t+k and CU ij;t+k : The variable F T A ij;t k (CU ij;t k ) is simply the k th lag of F T A ijt (CU ijt ) and captures the lagged e ects of the FTA(CU). Similarly, the variable F T A ij;t+k (CU ij;t+k ) is the k th lead of F T A ijt (CU ijt ) and captures any anticipatory e ects of the FTA(CU). 3 Data The data come from Baier and Bergstrand (2007); thus, only limited details are provided. The bilateral trade ows (nominal) are from the International Monetary Fund s Direction of Trade Statistics for the years 1960 to 2000, at ve year intervals. For the panel analysis, exporter GDP de ators are used to generate the real trade ows. The bilateral distances and the language and adjacency dummies are calculated from the CIA Factbook. However, the trade agreement dummies used here, are di erent from Baier and Bergstrand (2007). While they considered a single FTA dummy variable, which included... full (no partial) FTAs and customs unions pooled together, here the e ects of the two types of PTAs are allowed to di er. This is unambiguously relevant for trade policy decisions, when the popularity of CUs is being questioned, and relatively more FTAs are being negotiated.. It is also more interesting than 5

7 comparing CUs and other higher forms of integration. This is because the essential di erence between an FTA and a CU, in terms of the CET and the role of ROO, has a direct in uence on countries volume of trade in goods. 9 The trade agreement dummy in the original dataset involved errors, which have been corrected for using the sources listed in Baier and Bergstrand (2007) Results Tables 1, 2 and 3 report the cross-section results for the years 1960, 1970,..., Table 1 reports the results from the log-linearized model excluding the observations with zero bilateral trade. The results in Table 2 use the same speci cation with the zero values of the dependent variable replaced by ones. The issue of dealing with the zero trade values does not arise in the level speci cation. Accordingly, Table 3 reports the results from the level speci cation, which includes the observations with zero trade ows. For all the cross-section results, columns (a) and (b) do not consider separate dummies for FTAs and CUs. While (a) uses the original trade agreement dummy from Baier and Bergstrand (2007), (b) reports the results after incorporating the necessary corrections in it. 11 The coe cient estimates and their statistical signi cance hardly di er across (a) and (b) in any of the cross-section tables. Thus, the results of the paper are not sensitive to the corrections made in the original trade agreement dummy. Column (c) considers the di erential e ects of FTAs and CUs. Results from the log speci cations (Tables 1 and 2) indicate mostly negative coe cients on the FTA and CU dummies. This is similar to the cross-section ndings in Baier and Bergstrand (2007), who consider the trade agreement coe cients to be underestimated solely due to an omitted variables bias. However, results from the level speci cation are strikingly di erent. The signi cant coe cients on FTA and CU are all positive. Santos Silva and Tenreyro (2006) also nd striking di erences in their Monte Carlo simulations using the gravity equation. They further nd that log-linearization yields signi cantly larger e ects for geographical distance. The results in Tables 1, 2 and 3 are in consonance with this nding as well. Thus, regardless of the endogeneity due to omitted variables, the potential endogeneity arising from log-linearization of the gravity model seems to be of signi cant relevance by itself. The cross-section ndings further recommend estimation in levels. Accordingly, the level speci cation is adopted for the panel xed e ects method For example, Krueger (1997) considers that ROO can act as additional trade barriers under an FTA in ways that they cannot do under a customs union. But the characteristics which separate a CU from a higher form of integration seem to have a less direct in uence on trade in goods (see footnote 3). 10 The author would like to thank Scott Baier and Je rey Bergstrand for their cooperation in this. 11 Baier and Bergstrand (2007) use the log-linearized model but drop the observations with zero trade. Accordingly, the estimates in column (a) of Table 1 are very similar to the crosssection estimates in Baier and Bergstrand (2007). However, very slight di erences arise due to the imposition of unit income elasticities in Baier and Bergstrand (2007). 12 The panel xed e ects estimates of the log-linearized model are available upon request. 6

8 Unlike the cross-section estimates, the results using the panel method, reported in Table 4, allow for an unambiguous ranking of FTAs and CUs with respect to their e ect on member countries volume of bilateral trade. Column (a) does not consider any lagged or anticipatory e ects of the trade agreements; (b) allows for a single lag of FTA and CU; (c) considers two lags of FTA and CU; and (d) allows for a single lag and lead of FTA and CU. 13 Across all speci- cations, the CU coe cients are signi cantly greater than the FTA coe cients. Individual and joint tests reject the equality of the coe cients on FTA and CU, and on their lag and lead terms. Column (a) indicates that an FTA increases members bilateral trade by less than 17%, 14 on average, relative to countries not belonging to a CU or an FTA. However, a CU increases the same by about 77%. This di erence in the extent of the volume of bilateral trade seems to be even more stark when the lagged and anticipatory e ects are considered in (b), (c) and (d). In each of these speci cations, the cumulative e ect of an FTA or a CU is obtained by adding the coe cients on the (signi cant) lag and lead terms. The coe cients in (b) imply that while an FTA increases members trade by about 25%, a CU brings about an increase of more than 90%. 15 Similarly, the results in (c) nd FTA members to indulge in 26% more bilateral trade, while CU members are found to indulge in more than 110% of it. This is reminiscent of the principal result in Baier and Bergstrand (2007), who nd that... on average, an FTA approximately doubles two members bilateral trade after 10 years. However, the results here show that it is CUs and not FTAs, which double members bilateral trade after 10 years. Unlike Baier and Bergstrand (2007), the ndings are robust to any potential bias arising from log-linearizing the gravity model. Thus, pooling all FTAs and CUs into a single trade agreement dummy masks this crucial information, which is extremely relevant for policy decisions pertaining to PTAs in today s trading climate. The coe cients in (d) depict a very similar nding to (c) when the cumulative e ects of FTAs and CUs are considered. Unlike the ndings in Baier and Bergstrand (2007), but similar to those in Magee (2007), trade agreements are found to have signi cant anticipatory e ects. Table 5, reexamines the ndings in Table 4 by using the same lag and lead speci cations but splitting the sample. While columns (a), (b), (c) and (d) are with respect to the years 1960 to 1985, (e), (f), (g) and (h) correspond to the years 1990 to The results further strengthen the ndings in Table Baier and Bergstrand (2007) argue that trade agreements typically have a phase-in period of ten years and since the observations are at ve year intervals, it is reasonable to include one or two lagged levels of the trade agreement dummy. Accordingly, the lags and leads used in this paper are in similar to the ones in Baier and Bergstrand (2007). 14 exp(0.154)= exp( )=1.25; exp( )= This split is interesting and worth analyzing since Bhagwati et al. (1999) consider the US-Israel FTA, of 1985 (year of entry into force) as the start of the Second Regionalism and the main driving force for regionalism today. Hitherto, the United States abstained from forming FTAs, and was considered by Bhagwati et al. (1999) as the key defender of multilateralism. 7

9 In both samples the CU coe cients are unambiguously greater than the FTA coe cients. For the years 1960 to 1985, CUs more than double the volume of members bilateral trade across all the speci cations except (a), where the increase is by about 85%. The di erences in the FTA and CU coe cients are smaller when only the years 1990 to 2000 are considered. However, the e ect of CUs is nearly twice as that of FTAs across (e), (f), (g) and (h). Thus, even after splitting the sample, the ranking of the two PTA regimes, in terms of members bilateral trade, remains unaltered. Individual and joint tests continue to reject the equality of the coe cients on the two trade agreement dummies, and on their lag and lead terms. 5 Conclusion According to Clausing (2000) policy makers have become increasingly concerned with how best to design preferential agreements. Clausing (2000) further considers the choice between a customs union and a free trade area as an essential part of this consideration. The policy issue seems to be of even greater relevance today, when Fiorentino et al. (2007) consider CUs to be out of tune with today s trading climate. This paper is the rst empirical contribution to directly analyze this policy issue while addressing any bias due to the omission of time-invariant unobservables or log-linearization of the gravity model. While Baier and Bergstrand (2007) address the former, the latter is found to be of signi cant relevance as well. Once both are addressed, the results are striking. Baier and Bergstrand (2007) nd that on average, an FTA approximately doubles two members bilateral trade after 10 years. However, this paper uses the same data in concluding that it is actually a CU instead of an FTA, which doubles members bilateral trade. In general, members of a CU are found to indulge in more bilateral trade than FTA members. The nding is extremely relevant for trade policy decisions and especially interesting in the wake of declining popularity of CUs relative to FTAs. Moreover, separate considerations of the First Regionalism or Second Regionalism, as termed in Bhagwati et al. (1999), do not alter this nding. 8

10 References [1] Anderson, J. E. and E. van Wincoop (2003), Gravity with Gravitas: A Solution to the Border Puzzle, American Economic Review, 93, [2] Baier, S.L. and J.H. Bergstrand (2007), Do Free Trade Agreements Actually Increase Members International Trade, Journal of International Economics, 71, [3] Bhagwati, J. (1991), The World Trading System at Risk, Princeton, Princeton University Press. [4] Bhagwati, J., A. Panagariya and P. Krishna (1999), Trading Blocs, Cambridge, MIT Press. [5] Chang, W. and L. A. Winters (2002), How Regional Blocs A ect Excluded Countries: The Price E ects of MERCOSUR, American Economic Review, 92, [6] Clausing, K.A. (2000), Customs Unions and Free Trade Areas, Journal of Economic Integration, 15, [7] Clausing, K. A. (2001), Trade Creation and Trade Diversion in the Canada-U.S. Free Trade Agreement, Canadian Journal of Economics, 34, [8] Eicher, T., C. Henn and C. Papageorgiou (2007), Trade Creation and Diversion Revisited: Accounting for Model Uncertainty and Natural Trading Partner E ects, unpublished manuscript, University of Washington. [9] Fiorentino, R. V., L. Verdeja and C. Toqueboeuf (2007), The Changing Landscape of Regional Trade Agreements: 2006 Update, WTO Discussion Paper No. 12. [10] Ghosh, S. and S. Yamarik (2004), Are regional trading arrangements trade creating? An application of extreme bounds analysis, Journal of International Economics, 63, [11] Henderson, D. J. and D. L. Millimet (forthcoming), Is Gravity Linear?, Journal of Applied Econometrics. [12] Krueger, A. O. (1997), Free Trade Agreements versus Customs Unions, Journal of Development Economics, 54, [13] Krueger, A. O. (1999), Are Preferential Trading Arrangements Trade- Liberalizing or Protectionist? Journal of Economic Perspectives, 13, [14] Limao, N. (2006), Preferential Trade Agreements as Stumbling Blocks for Multilateral Trade Liberalization: Evidence for the United States, American Economic Review, 96,

11 [15] Magee, C. S. (2003), Endogenous Preferential Trade Agreements: An Empirical Analysis, Contributions to Economic Analysis & Policy, 2. [16] Magee, C. S. (2007), New Measures of Trade Creation and Trade Diversion, unpublished manuscript, Bucknell University. [17] Romalis, J. (2007), NAFTA S and CUSFTA S Impact On International Trade, Review of Economics and Statistics, [18] Saggi, K. and H. Yildiz (2007), Bilateral Trade Agreements and the Feasibility of Multilateral Free Trade, unpublished manuscript, Southern Methodist University. [19] Santos Silva, J.M.C. and S. Tenreyro (2006), The Log of Gravity, Review of Economics and Statistics, 88, [20] Viner, J. (1950), The Customs Union Issue, Carnegie Endowment for International Peace, New York. 10

12 Table 1. Cross-section estimates of the log specification excluding observations with zero trade 1960 (a) 1960 (b) 1960 (c) 1970 (a) 1970 (b) 1970 (c) 1980 (a) 1980 (b) 1980 (c) ln(distance) * * * * * * * * * (0.039) (0.039) (0.039) (0.038) (0.038) (0.038) (0.04) (0.04) (0.04) Language * * 0.38 * * * * * * * (0.104) (0.104) (0.104) (0.107) (0.107) (0.107) (0.104) (0.104) (0.105) Adjacency * * * * * * 0.44 * * (0.123) (0.123) (0.124) (0.151) (0.151) (0.153) (0.148) (0.148) (0.148) FTA * * * * * (0.118) (0.117) (0.136) (0.183) (0.185) (0.184) (0.137) (0.137) (0.14) CU * * * (0.182) (0.345) (0.213) Test FTA = CU [p = 0.007] [p = 0.175] [p < 0.001] N (a) 1990 (b) 1990 (c) 2000 (a) 2000 (b) 2000 (c) ln(distance) * * * * * * (0.042) (0.042) (0.042) (0.039) (0.04) (0.04) Language * * * * * * (0.099) (0.099) (0.099) (0.093) (0.093) (0.093) Adjacency 0.57 * * * * * * (0.146) (0.146) (0.145) (0.168) (0.168) (0.168) FTA * * * * (0.117) (0.118) (0.131) (0.09) (0.085) (0.096) CU * * (0.158) (0.12) Test FTA = CU [p < 0.001] [p = 0.002] N Notes: For all years (a) uses the Baier and Bergstrand (2007) FTA dummy; (b) uses the corrected trade agreement dummy with FTAs and CUs pooled together; (c) uses separate dummies for FTAs and CUs based on the correction in (b). Standard errors in parentheses are robust. The p-values are reported for the test of equality between the coefficients on FTA and CU. Each regression also includes country dummies. * denotes statistical significance at the 5% level.

13 Table 2. Cross-section estimates of the log specification including observations with zero trade 1960 (a) 1960 (b) 1960 (c) 1970 (a) 1970 (b) 1970 (c) 1980 (a) 1980 (b) 1980 (c) ln(distance) * * * * * * * * * (0.056) (0.056) (0.056) (0.052) (0.052) (0.052) (0.056) (0.056) (0.056) Language * * * * * * * * 1.46 * (0.15) (0.15) (0.15) (0.135) (0.135) (0.135) (0.137) (0.137) (0.137) Adjacency * * * (0.244) (0.244) (0.245) (0.237) (0.237) (0.239) (0.26) (0.26) (0.259) FTA * * * * * * * * * (0.318) (0.304) (0.358) (0.315) (0.322) (0.329) (0.229) (0.225) (0.243) CU * * (0.514) (0.595) (0.348) Test FTA = CU [p = 0.234] [p = 0.788] [p < 0.001] N (a) 1990 (b) 1990 (c) 2000 (a) 2000 (b) 2000 (c) ln(distance) * * * * * * (0.051) (0.051) (0.051) (0.051) (0.052) (0.052) Language * * * 1.68 * * * (0.127) (0.127) (0.128) (0.119) (0.119) (0.12) Adjacency * * * 0.66 * * * (0.215) (0.215) (0.215) (0.219) (0.219) (0.22) FTA * * * * * (0.16) (0.165) (0.195) (0.126) (0.116) (0.134) CU * * (0.22) (0.163) Test FTA = CU [p = 0.003] [p < 0.001] N Notes: For all years (a) uses the Baier and Bergstrand (2007) FTA dummy; (b) uses the corrected trade agreement dummy with FTAs and CUs pooled together; (c) uses separate dummies for FTAs and CUs based on the correction in (b). Standard errors in parentheses are robust. The p-values are reported for the test of equality between the coefficients on FTA and CU. Each regression also includes country dummies. * denotes statistical significance at the 5% level.

14 Table 3. Cross-section estimates of the level specification (including observations with zero trade) 1960 (a) 1960 (b) 1960 (c) 1970 (a) 1970 (b) 1970 (c) 1980 (a) 1980 (b) 1980 (c) ln(distance) * * * * * * * * * (0.075) (0.076) (0.076) (0.047) (0.047) (0.047) (0.042) (0.043) (0.043) Language * * * * * * * * 0.35 * (0.126) (0.126) (0.122) (0.088) (0.088) (0.085) (0.072) (0.072) (0.074) Adjacency * * 0.51 * * * * * * * (0.141) (0.141) (0.141) (0.084) (0.084) (0.088) (0.077) (0.077) (0.079) FTA * * * (0.112) (0.112) (0.175) (0.068) (0.068) (0.106) (0.099) (0.106) (0.12) CU * (0.161) (0.109) (0.117) Test FTA = CU [p = 0.039] [p = 0.017] [p = 0.051] N (a) 1990 (b) 1990 (c) 2000 (a) 2000 (b) 2000 (c) ln(distance) * * * * * * (0.044) (0.043) (0.045) (0.045) (0.045) (0.046) Language * * * 0.33 * * * (0.084) (0.084) (0.086) (0.075) (0.075) (0.076) Adjacency * * * 0.32 * * * (0.079) (0.078) (0.078) (0.081) (0.081) (0.082) FTA * * * * * * (0.089) (0.085) (0.085) (0.083) (0.083) (0.082) CU * * (0.11) (0.108) Test FTA = CU [p = 0.828] [p = 0.568] N Notes: For all years (a) uses the Baier and Bergstrand (2007) FTA dummy; (b) uses the corrected trade agreement dummy with FTAs and CUs pooled together; (c) uses separate dummies for FTAs and CUs based on the correction in (b). Standard errors in parentheses are robust. The p-values are reported for the test of equality between the coefficients on FTA and CU. Each regression also includes country dummies. * denotes statistical significance at the 5% level.

15 Table 4. Panel estimates of the level specification (including observations with zero trade) (a) (b) (c) (d) FTA * * 0.08 * * (0.0001) (0.0001) (0.0001) (0.0001) Lag FTA * * * (0.0001) (0.0001) (0.0001) Lag2 FTA * (0.0001) Lead FTA * (0.0001) CU * 0.32 * * 0.35 * (0.0001) (0.0001) (0.0001) (0.0001) Lag CU * * 0.37 * (0.0001) (0.0001) (0.0001) Lag2 CU * (0.0001) Lead CU * (0.0001) Test FTA = CU [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] Test Lag FTA = Lag CU [p < 0.001] [p < 0.001] [p < 0.001] Test Lag2 FTA = Lag2 CU [p < 0.001] Test Lead FTA = Lead FTA [p < 0.001] Joint test [p < 0.001] [p < 0.001] [p < 0.001] N Notes: Due to the panel fixed effects approach, coefficients on the time-invariant regressors are not reported. (a) does not include any lags or leads of FTA or CU; (b) includes one lag of FTA and CU; (c) includes two lags of FTA and CU; (d) includes one lag and one lead of FTA and CU. The standard errors are reported in parentheses. The p-values are reported for the test of equality between the coefficients on - FTA and CU, their corresponding lag and lead terms. The p-values for the joint test of equality between these coefficients are also reported. Each regression also includes country-by-time dummies. * denotes statistical significance at the 5% level.

16 Table 5. Panel estimates of the level specification (including observations with zero trade), after splitting the sample (a) (b) (c) (d) (e) (f) (g) (h) FTA * * * * * * * * (0.0001) (0.0001) (0.0002) (0.0002) (0.0001) (0.0001) (0.0001) (0.0002) Lag FTA * * * * * * (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0002) Lag2 FTA * * (0.0001) (0.0001) Lead FTA * * (0.0001) (0.0002) CU * * * * * * * * (0.0001) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0003) Lag CU * * * * * * (0.0001) (0.0002) (0.0002) (0.0001) (0.0001) (0.0002) Lag2 CU * * (0.0002) (0.0002) Lead CU * * (0.0002) (0.001) Test FTA = CU [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] Test Lag FTA = Lag CU [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] Test Lag2 FTA = Lag2 CU [p < 0.001] [p < 0.001] Test Lead FTA = Lead FTA [p < 0.001] [p < 0.001] Joint test [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] [p < 0.001] N Notes: The results pertain to analyses similar to that in Table 4, where the full sample from 1960 to 2000 was considered. Here (a), (b), (c) and (d) correspond to the years 1960 to 1985; (e), (f), (g) and (h) correspond to the years 1990 to For the years 1960 to 1985, there are 520 observations with FTA=1 and 322 with CU=1. For the years 1990 to 2000 there are 592 observations with FTA=1 and 624 with CU=1.