Regionalism in the Nineties: What Effect on Trade? #

Transcription

1 Regionalism in the Nineties: What Effect on Trade? # Isidro Soloaga * and L. Alan Winters ** This Draft: November 2000 Abstract We apply a gravity model to annual non-fuel imports data for 58 countries to quantify the effects of recently created or revamped PTAs on trade. We modify the gravity equation to identify separate effects of PTAs on intra-bloc trade, members total imports and their total exports and to test for significant changes in trade patterns following the creation of trade blocs. We find no indication that new regionalism boosted intra-bloc trade significantly and we find trade diversion only for EU and EFTA. The latter also exhibit 'export diversion', which could indicate their imposing welfare costs on other countries. Latin American trade liberalization in the 1990s had a positive impact on bloc members imports and, usually, exports. JEL classification: F10, F13, F15 * Isidro Soloaga is with the World Bank, Development Research Group, Trade Research Team H St. NW, Washington DC, 20433, Room MC Tel.:(202) ; Fax: (202) ; isoloaga@worldbank.org. ** L. Alan Winters is Professor of Economics, School of Social Sciences, University of Sussex, Falmer, BRIGHTON, BN1 9SN, UK. Tel.: +44(0) ; Fax: +44(0) /678466; L.A.Winters@Sussex.ac.uk; Centre for Economic Policy Research, Goswell Road, London, EC1V 7DB, UK; and Centre for Economic Performance, London School of Economics, Houghton Street, London WC2A 2AE, UK. # Much of this work was done while the authors were respectively Consultant and Research Manager in the Development Economics Group of the World Bank. The views expressed are those of the authors and should not be attributed to the World Bank or its member governments. The authors are grateful to Sven Arndt, Andre Azevedo, Vittorio Corbo, Homi Kharas, Carlos Martinez-Mongay and Helene Rey, to colleagues in the World Bank Trade Research Team, and to participants in seminars at the World Bank and Koc University, Istanbul for comments and suggestions. We owe a particular debt to two anonymous referees. We are also grateful to Rosie Bellinger for logistical support.

2 Introduction During the last 10 years, regionalism has re-emerged as a major issue in the policy agenda. In the Americas, the new Common Market of the South (MERCOSUR, 1991) and the North American Free Trade Association (NAFTA, 1994) were created while old Preferential Trade Agreements (PTAs) like the ANDEAN Pact (ANDEAN) and the Central American Common Market (CACM) started a process of renewal in the late 80 s and early 90 s. In Africa new PTAs were formed on the basis of old ones (e.g., in 1994 the Union Economique et Monaitarie de l Africa Occidentale UEMOA was created out of the Communaute Economique de l Afrique Occidentale CEAO; and the Common Market of Eastern and Southern Africa COMESA revived and expanded the Preferential Trade Area for Eastern and Southern African States PTA) and old ones were revamped (e.g., in the early 90 s the Union Duaniere et Economique d Africa Centrale UDEAC). In Asia, countries in the Association of Southeast Asian Nations (ASEAN) formed the ASEAN Free Trade Area (AFTA) in The effect this second wave of regionalism on trade is still an open question. Do they really increase trade among members? Do they contribute to further trade liberalization with nonmembers countries or undermine it? Do they harm non-member countries? This paper aims to provide answers to some of these questions by exploring the effects of recent and revived PTAs on intra- and on extra-bloc trade. We consider nine PTAs- see Annex 1 for details. Five of them were either created (MERCOSUR, NAFTA) or revamped (ASEAN, CACM, ANDEAN) during the 1990 s. One was created in the 1980 s (GCC) and another deepened significantly in that decade (EU). The remaining two, EFTA and LAIA, had existed for some time. By using data up to 1996 we can compare blocs patterns of trade before and after this second wave of regionalism and assess for first time to our knowledge the wave s effect on blocs trade. We use the gravity model to quantify the trade effects, but refine it relative to previous exercises. Existing gravity-model approaches to regional blocs have identified bloc effects on intra-bloc trade and on members extra-bloc trade. We go beyond that by identifying separate effects on intra-bloc trade, members total imports and their total exports, the latter being an important determinant of the blocs effects on the welfare of the Rest of the World (ROW). We also innovate by formally testing the significance of changes in the estimated coefficients before and after blocs formation. 2

3 The paper begins with a statement of the model, continues with a brief description of the data used and recent developments in the PTAs modeled, and then presents the results. The final section summarizes the findings and conclusions. The gravity model In the basic gravity model, trade between two countries depends on their economic and physical size (GDP, population, land area) and on transaction costs (distance, adjacency, cultural similarities). Its empirical robustness has made it the work-horse for investigations of the geographical patterns of trade. Tinbergen (1962), Pöyhönen (1963) and Linneman (1966) provided initial specifications and estimates of the determinants of trade flows and Aitken (1973) applied it to PTAs. More recently, Anderson (1979), Bergstrand (1985), Helpman and Krugman (1985), Deardorff (1997) and Anderson and Mercouiller (1999) have provided partial theoretical foundations for the gravity equation, although none of the models generate exactly the equation generally used in empirical work. In the absence of preferential trade arrangements and ignoring the time dimension and stochastic errors for now, we use the following form of the basic gravity model to explain country i s imports from country j: (1) X ij exp[ β C where 9 = BY ij β 1 i + β 10 N I β 2 i i Y β 3 j + β 11 N I β 4 j j D β 5 i + β 12 D L β 6 β 7 ij i ij ] T Xij is the value of imports of country i from country j (i.e. exports from j to i), T β 8 j Ym is the Gross Domestic Product of country m, Nm is the population of country m, * 3

4 D i is the average distance of country i to exporter partners, weighted by exporters GDP share in world GDP ( remoteness of country i) 1, D ij is the distance between the economic center of gravity of the respective countries, T m is the land area of country m, C ij is a dummy that takes value 1 if countries i and j share a land border and 0 otherwise, I m is a dummy that takes value 1 when country m is an island, and 0 otherwise, and L ij is a dummy for cultural affinities, proxied by the use of the same language in countries i and j (one dummy for each one of the following languages: English, Spanish, Arabic and Portuguese). This equation explains the trade flow from j to i if neither were a member of a PTA. Thus it represents what scholars of international integration have come to call the anti-monde or counterfactual: the volume of trade that would be considered normal between two countries in the absence of the PTAs under investigation. We take an inclusive attitude towards the set of independent variables: our aim is to extend the gravity model to explain regional trade patterns, and to reduce worries about the latter effects proxying for omitted variables, we include virtually all of the explanatory variables that have ever been used by previous researchers. When used to address the effect of a PTA on the direction of trade, the basic gravity model was first extended by including a dummy variable to capture the PTA s effect on intra-bloc trade that is, the sum of Vinerian trade creation and trade diversion for the PTA (e.g. Aitken, 1973; Braga, Safadi and Yeats, 1994). More recently, researchers have added a second set of dummies to capture the PTA effect on the trade of bloc members with non-members (e.g. Bayoumi and Eichengreen, 1997; Frankel, 1997, and Frankel and Wei, 1998). By using two dummies (intra-bloc trade and extra-bloc trade) these authors were able to separate cases where PTAs were only trade-creating (that is, they caused intra-bloc trade to increase above normal levels without changes in extra-bloc trade) from those where a PTA-induced increase in intra-bloc trade came at the expense of lower extra-bloc trade. These authors identified the latter effect as 1 The hypothesis is that, after controlling for distance between i and j, the further is country i from all its partners, the greater will be its imports from country j (Polak, 1996). One might expect to see Australia and New Zealand trading more with each other than another pair of countries separated by the same distance but with lots of other trading partners close to hand (Spain and Poland, for instance). 4

5 trade diversion. Bayoumi and Eichengreen (1997) use trade in both directions as their dependent variable, and hence constrain exports to show the pattern of diversion as imports. This is legitimate, but it does not correspond precisely to most theoretical analyses of trade diversion. Frankel (1997) and Frankel and Wei (1998), on the other hand, include only import effects (à la Viner), and so ignore the possibility that PTAs could affect members export flows at all. We relax these restrictions by introducing separate dummies for members imports from non-members and their exports to non-members. Since the full, three-dummy, specification of the gravity equation is both new and central to our exercise, we consider its derivation in a bit more detail. Our aim is to use the gravity model to explore the effects of regional arrangements on the members internal and external trade flows in terms of Viner s trade creation and trade diversion. As mentioned above, recent researchers have done this by including two dummy variables one to capture excess intra-bloc trade and one to capture excess imports from (or trade with) countries outside the bloc. Trade creation raises intra-bloc trade but has no effect on extra-bloc imports; trade diversion boosts intra-bloc imports and has an equal and opposite effect on extra-bloc imports. Thus, for example, Bayoumi and Eichengreen (1997), Frankel (1997) and Frankel and Wei (1998) write something along the lines of: (2) LnX ij = A ij + b k P kij + m k P ki-j where X ij is imports of i from j, A ij is the anti-monde value of (the logarithm of) that trade as defined by the (log of) equation (1) above, P kij is a dummy taking value 1 if both i and j are members of bloc k and zero otherwise (assume there is only one such bloc for now), b k is the coefficient measuring the extent to which trade is higher than expected if both i and j are members of the bloc (intra-bloc trade), P ki-j is a dummy taking value 1 if i is a member of the bloc but j is not, and m k is the coefficient measuring the extent to which members imports from non-members are higher than expected. 5

6 Pure trade diversion would be identified by a positive value for b k and a negative value for m k amounting to the same volume of imports; trade creation by a positive value for b k indicating greater increases in trade than the negative m k implies decreases. 2 The presumption in this is that the RIA represents the only systematic effect altering trade from its anti-monde values so that m k reflects only trade diversion. We make two innovations: we slightly reparameterize the equation for imports and we also add an export effect to capture what we refer to as export diversion whereby members divert their exports from the rest of the world to other members. We shall argue below that the latter has important implications for the rest of the world s welfare, but first focus on the import story. We shall write: (3) LnX ij = A ij + b k P ki P kj + m k P ki + n k P kj where in addition to the above, P km is a dummy taking value 1 if m is a member of bloc k and zero otherwise, b k is a coefficient measuring the extent to which trade is higher than expected if both i and j are members of the bloc (intra-bloc trade), m k is a coefficient measuring the extent to which members imports are higher than expected from all countries, and n k is a coefficient measuring the extent to which members exports are higher than expected to all countries. Hence relative to the anti-monde, flow ij is raised by m k if i is in the bloc, whether j is a PTAmember or not; by n k if j is in the bloc, whether i is a member or not, and by (m k +n k +b k ) if both are members of the bloc. Comparing the treatment of imports in (2) and (3), P kij = P ki P kj and P ki-j = P ki - P ki P kj, from which we can infer that (2) and [(3) less its last term] would make identical predictions, with 2 In fact, researchers rarely apply these size tests in identifying creation and diversion, relying instead only on the signs of the coefficients. 6

7 b k = b' k - m' k and m k = m' k. Although the numerical implications and statistical properties of the estimates are not affected by the choice between the parameterizations (2) and (3), we feel that ours helps one to conceptualize the regionalism of the nineties. For many of our blocs, regionalism was accompanied by a strong non-discriminatory (mfn) trade liberalization. We view this as being essentially independent of the regionalism itself. Hence we assume that, say, Argentina would have reduced its trade barriers vis-a-vis all its partners after 1990 independent of MERCOSUR, and that MERCOSUR s preference-related effects must be seen as additional to that effect 3. Additionally, our parameterization allows a simple intuitive interpretation of m k and n k as openness. Frankel (1997) uses the term just for extra trade, but most policy-makers view relations with PTA partner countries as part of openness. Of course, while considering the coefficients m k and n k as capturing general openness effects is useful, it is second best. In an ideal world we would have direct measurements on the openness of trade policy (e.g. tariff averages) and hence be much better placed for separating general from PTA trade effects. However, the world is so far from having such data consistently over a period of years and large set of countries that it is not worth spending any longer regretting their absence in this paper. On the basis of this discussion, we think of m k and n k as combining the effects of the general liberalization and trade diversion, while b k captures the increase in intra-bloc trade over and above the general effect. Of course, the numerical values of m k and n k are exactly the same as we would have derived from the traditional approach, and so may be termed trade diversion as in previous studies. In particular, if they are negative i.e. extra-bloc trade is lower despite being confounded by the general liberalization in our sample - we can conclude that the trade blocs have caused trade diversion. Our estimate of the internal trade effects of the blocs, however, will differ from that of the more traditional approach, because we attribute part of any observed increase in internal trade to the underlying general liberalization. It is easy to derive the traditional estimate of the intra-trade effects from ours as (m k +n k +b k ) however, and we report these values in an Annex table. We refer to them as 'gross-intra' trade effects. Consider, now, the second innovation the addition of export effects. Mechanically, the inclusion of the last term in (3) allows us to capture the plausible situation in which j s exports to i 3 The causal relationship between regional and mfn liberalization (if any) is an issue on which neither theory nor empirical work has yet provided a satisfactory answer see, for example, Foroutan (1998). 7

8 increase at the expense of those to non-member countries export diversion. 4 Economically, this is important when it comes to considering the effects of regionalism on excluded countries. Winters (1997) shows that, in assessing the welfare effects of PTAs on non-members, the appropriate indicator is the latter s terms of trade, and that, if one is working only with quantities or values, non-members imports (i.e. bloc exports) are a more appropriate indicator than their exports (bloc imports). Essentially the argument is that welfare is related to what you consume and that, ceteris paribus, this is determined by what you receive from others rather than what you send to them. Of course, what you send helps to determine what you receive, but it is the latter which affects welfare directly. Statements about the effects of trade blocs on non-members welfare that are based on the latter s exports (i.e. the evolution of their shares of the bloc members imports) are quite common. However, despite the emphasis on prices and consumption in the basic theory of integration, ex post empirical studies of the terms of trade effects or of non-members imports (the bloc s exports) are almost entirely lacking. In particular, this is, to our knowledge, the first time the issue has been treated with a gravity equation. 5 A negative coefficient on the dummy for a given PTA s exports to non-members (n k ) indicates that, relative to the anti monde defined by the gravity equation, the PTA is likely to be harmful to non-member countries. 6 For want of a better term, we name this effect export diversion. The gravity variables of the model (GDP, population, area, distance, cultural similarities) control for those factors that are assumed to explain normal trade between countries. Thus, the relationship between trade and these variables for the sample countries defines the anti-monde for PTA members: in the absence of a PTA, members trade would have the same relationship to the gravity variables as the other countries in the sample. In this setting, the bloc-related dummy 4 One can also think of a parallel to trade creation - export creation - whereby exports increase at the expense of domestic consumption or higher production in j, but it does not seem to have any analytical significance and neither would it be separately identifiable from the intra-imports effect. 5 For other non-gravity approaches to this issue, see, Winters and Chang (2000) and Chang and Winters (1999). Winters (1985) observed the phenomenon of export diversion following UK accession the EEC in 1973, but did not discuss its welfare implications. 6 Of course, this effect could be off-set by improvements in the rest of the world s terms of trade, although in general one expects PTAs to worsen these see Winters and Chang (2000). 8

9 variables pick up abnormal levels of trade that could be attributed to a PTA or to unobservable characteristics of country members. Because of the latter, we need to make one further refinement. We define our trade bloc dummies (P km ) by bloc membership in 1996 and measure the effects of trade blocs not by the values of the dummy coefficients per se, but by their movements through time. This recognizes that pairs of countries may have abnormal trade relationships for a variety of reasons, but that provided that these do not change significantly through time these will not affect the evolution of the coefficients through time. In other words, we are interested in whether identifiable instances of regionalism change the intensity with which particular countries trade with each other. The Data It is best to estimate the gravity model on the largest available set of countries and to work over a period of time long enough to describe non-pta years adequately. We used annual non-fuel visible imports data for 58 countries (Annex I shows the list of countries) for 1980 to 1996 from the UN-COMTRADE database. This set of countries represents around 70% of total world imports in the period covered. While such coverage is good overall, certain countries conduct a significant part of their trade with non-sample countries while others are specialized on non-sample commodities (fuels and services). This could complicate the interpretation of our results, for such countries will tend to record abnormally low trade in our sample. However, provided that the causes of these patterns of trade do not change rapidly through time and are not systematically correlated with our explanatory variables, this should not distort our view of regionalism based on changes in the regression coefficients. Long-run trade imbalances may largely be ignored for the same reasons. Shorter-run imbalances, on the other hand, will show up in our annual estimates below, and in the case of MERCOSUR, recognizing them quite clearly influences our view of regionalism. Problems of coverage beset all similar studies to this, and although the trade imbalances problem is less obvious where authors have not separated export from import effects, it is obscured rather than removed. 9

10 The distance variable is the great circle distance between economic centers and was based on distances calculated originally by Linneman. 7 The source for the rest of our variables is the World Bank s Economic and Social Data (BESD). The Econometric Approach Our estimating equation is the logarithmic transform of (1) extended by (3) and with a white noise stochastic error added: (4) LnX ij = α + β1lny i + β 2LnN i + β 3LnY j + β 4LnN j + β 5LnD i + β 6LnD ij + + β 7LnT + m k k P i ki + β8lnt + n k k P kj j + ε + β 9C ij ij + β10i i + β11i j + β12l ij + b k k P ki P kj where, k indicates membership of the kth PTA (k = 1,,9) and e ij is a normally distributed error term. Because trade values are bounded from below by zero, the appropriate estimation procedure is strictly a Tobit model, and we follow this approach in the paper. In truth, however, this refinement does not add much relative to the more normal OLS estimation, because with the logarithmic transform, the truncation occurs at the logarithm of the minimum recorded value of trade, which is $0.001million, i.e. at 6.91, and only about 6% of observations (588 out of 9918 in our pooled regression see below) are recorded at that level. 8 We start our sample in 1980 and explore the existence of both anticipation effects (i.e., the level of trade between country members rising above normal levels before the PTA is formally commenced as indicated, for instance, in Freund and McLaren, 1998), and any non-pta relationships between members that may have been at work since well before the PTAs were 7 These data were generously provided by Lant Pritchett of the World Bank. They refer to the late 1980s. 8 See, for example, Maddala [1992] for the standard discussion of estimation in models with limited dependent variables. We are grateful to Helene Rey for suggesting the relative insensitivity of the results to the estimation method in this case. 10

11 created/revived. While the former can be thought of as a genuine PTA effect, the latter is not; it just reflects the possibility that the PTA is formed between countries that already have long standing economic ties. Table 1 provides a brief description of main developments in the nine PTAs analyzed, and identifies different periods for their (expected) effects on trade. It seems appropriate for our purposes to center our before and after analysis of new-wave regionalism on the years , and also to use the earlier years of our sample for the cases of EU, EFTA and GULFCOOP. We made two different sets of estimates of equation (4). The first is a set of 17 separate regressions one for each year for the annual data , and is presented in Table 2. From these we seek to identify not only the level effect on trade of PTAs but also the variation of this effect through time, in particular around the years marked in the last column of Table 1. This permits us to make an event study around those years, in the belief that seventeen years gives enough time before and after the various PTA events to offer a good chance of determining whether the observed abnormalities in trade are directly associated with preference effects. Second, we averaged the values of all variables for , and , pooled the data and estimated a single regression allowing for all the coefficients to be different in the three periods 9. From this we tested whether the estimates obtained for the period (considered as post-integration/revival years) were different from those obtained for and (the pre-integration/revival years) 10. Results from the pooled data are presented in Table 3. Thus we use annual estimates to visualize the trade patterns, identifying whether or not there are key turning points, and average data to test statistically for the significance of changes. Once we pool data over time, movements in the real exchange rate and competitiveness become important, and so we add a real exchange rate variable to the equation 11. Country i(j) s real exchange rate was defined as the local currency value of 1 US$, multiplied by the US GDP 9 The use of period averages smoothes the effects that transient phenomena (e.g. business cycle, economic shocks, and trade imbalances) may have on any particular year. 10 Additionally, we tested whether parameters for were different from those estimated for This is relevant for the older and well established PTAs in the sample (EU and EFTA) and for GCC. 11 Bayoumi and Eichengreen (1997) also include an exchange rate variable in their inter-temporal model - deviations from PPP but they seem to interpret it in terms of the valuation of GDP rather than as a competitiveness effect. 11

12 deflator and divided by country i(j) s GDP deflator, where i is the importer country and j the exporter 12. Such real exchange rate and price index variables make no sense in a purely crosssectional context, because the data reflect only movements through time (relative to the base year of the index used) with no indication of whether a country s currency is over-valued or undervalued 13. To try to eliminate the spurious cross-section effect, therefore, we specify our real exchange variables such that their means over the three observations ( , and 95-96) are zero. This is equivalent to assuming that countries are in exchange rate equilibrium at the means and identifying the exchange rate effects only by the movements through time relative to those means. We also add time dummies for two of our three periods (the third is, of course rolled into the constant). This makes our model similar to Matyas (1997) fixed-effects model, although he includes time-invariant fixed effects for each individual country where we include dummies for each (bloc x time) combination. Matyas states that in a correctly specified gravity model, bloc dummies are mere linear combinations of the fixed effects (p.365). Even with country-specific dummies this is not correct because the bloc dummies pertain to flows between a set of importers and only a subset of their supplying exporters, and so can not be represented by variables which treat all partners symmetrically. Thus below, contrary to Matyas claim, we can identify, estimate and interpret dummies on trade between bloc members in addition to the fixed effects. Finally, for ease of interpretation we make one superficial change to the dummy variables for LAIA in equation (4). Within our sample, LAIA comprises the countries of the ANDEAN and MERCOSUR blocs plus Mexico. In order to allow the co-efficients of MERCOSUR and ANDEAN to capture the full effects of their respective members integration, we subtract the ANDEAN and MERCOSUR dummies from the respective LAIA dummies. For the total export and import effects (the extra-dummies ), this leaves LAIA reflecting only Mexico s openness. Consequently, the NAFTA extra-effects are determined only by the USA s and Canada s 12 Results (not presented here) did not change when using the IMF s real effective exchange rate measurement, which is a single measure by country that weights all trading partners bilateral exchange rate by their share in imports. 13 Absolute PPP could indeed be useful in a cross-section framework, but such variables have not figured in gravity equations. Bergstrand (1985,1989) uses, incorrectly, we believe, standard exchange rate variables, while Anderson and Mercouiller (1999) argue that prices have a specific role to play, but instrument them using gravity-type variables. 12

13 behaviour, with Mexico s being the sum of the NAFTA and LAIA effects. For intra trade the LAIA dummy captures excess trade between MERCOSUR members, ANDEAN members and Mexico, excluding the intra-bloc trade of the first two. This re-parameterisation affects none of the statistical properties of the estimation but does ease our interpretation. Results Table 2 presents the estimated parameters and the asymptotic significance tests for the set of 17 annual Tobit estimations. As in many other applications the central variables of the gravity model the level of GDP of countries i and j, the area of these countries, and the absolute distance between i and j-- have the expected sign and are all significant at 1%: trade increases slightly more than the proportionately with of GDPs of the importer and exporter countries and decreases with size and distance. The coefficients reflecting population effects (of importer and of exporter) are negative and almost always significant. All of these effects are relatively stable over time. The degree of remoteness of the importer country from its suppliers has the expected positive sign and is always significant, while the estimated parameters for common land borders are not significant in any year of the sample, probably reflecting some colinearity with the parameter for remoteness 14. The coefficient for importer is an island is negative and statistically significant only in the period and in 1995, whereas the coefficient for exporter is an island is generally positive and significant only in and in Regarding the proxies used for cultural similarities (common language), only Spanish and Arabic turned out to be positive and significant all the years of the sample, with English positive and significant only in 1987 and These, less traditional, gravity variables show greater variation through time than the traditional ones, but not alarmingly much given their lack of significance. 14 When the model was estimated without the variable remoteness, border turned out positive and statistically significant. 15 Not all the researchers use a dummy for island. Its inclusion here is based only on a wish to be comprehensive so that our PTA effects are not biased by unintended exclusions. Some authors found Island-effects to be positive and significant for both importer and exporter (Montenegro and Soto, 1996), whereas others found that the sign depends on the direction of trade - positive when imports are modeled as the independent variable, and negative for exports (Havrylyshyn and Pritchett, 1991). 13

14 The model was estimated in logs. Thus the percentage equivalent for any dummy is: [exp(dummy coefficient)-1]*100. For example, the intra-bloc parameter for MERCOSUR in 1996 is 2.77, indicating that MERCOSUR members traded between themselves about fifteen times [=(exp(2.77)-1)*100%] more than expected from the gravity and overall bloc trade variables alone 16. But, because MERCOSUR members overall imports were 66% below what could be expected and their overall exports 30% below expected levels, the net effect of the three dummies the gross intra-trade effect is that in 1996, MERCOSUR members traded 418% [=(exp( )-1)*100] more with each other than would be predicted by the basic gravity model. This is not saying that MERCOSUR increased intra-trade by 418%, however. What matters analytically is less the level of these effects than their changes around the periods of integration. From table 2 it is clear that the results are far from homogenous across PTAs. In the period 1980 to 1996 we see that: In all the cases involving only Latin American countries CACM, LAIA, ANDEAN, and MERCOSUR the intra-bloc trade coefficient was positive and statistically significant for the whole sample. That is, in every dimension we tested, Latin-America countries trade disproportionately heavily with each other. For NAFTA the intra-effect was positive but never significant, and for GULFCOOP it was positive and significant only in 1980 and in The coefficients for the intra-bloc trade effects were negative for EU, EFTA and ASEAN, but consistently significant only in the case of EU. Thus, after controlling for gravity variables and general trade behavior, only the Latin American PTA members trade significantly more with themselves than expected from a snapshot view. The coefficients for overall bloc imports (from members as well as from non-members) were almost always statistically significant (the exception was GULFCOOP). This coefficient was negative for the four Latin American PTAs and positive in the other cases (EU, EFTA, ASEAN and NAFTA). The coefficients for overall bloc exports were negative and almost always statistically significant in five of the nine PTAs (GULFCOOP, NAFTA, CACM, LAIA, and ANDEAN), always positive and significant for ASEAN, and always positive but significant only in As follows from footnote 7 above, it is not necessary to distinguish marginal effects from the coefficient values in this discussion. The adjustment factor is unity to four decimal places. 14

15 86 and 1993 for EFTA. The bloc export coefficients for the EU were positive and significant over and negative after Something similar happened in the case of MERCOSUR, the dummy was positive up to 1991 and negative in , significantly so since Considering the 'gross' intra-trade effects (i.e. b k +m k + n k ), which compare intra-trade with the anti-monde assuming no overall trade effects, we find positive and generally significant effects only for CACM see Annex 2. Those for ANDEAN, MERCOSUR, NAFTA, EFTA and ASEAN are positive, but not always significant, while those for LAIA and GCC negative, and those for the EU falling from positive to negative. These results are similar to some of Frankel s (1997), the piece of literature most closely related to ours, although he estimated several variants of his model and got widely varying results. His estimates for a series of single years suggest significant decreases in extra-bloc openness (e.g. table 4.2), for the EU, NAFTA, MERCOSUR, and ANDEAN, and an increase for ASEAN. Frankel s preferred specification for his policy conclusions, however, assumes constant bloc-effects over the period (!), and suggests little trade diversion (extra-trade effects) and a good deal of additional intra-trade due to the PTAs (e.g. p.226-7). We do not find constant PTA effects over a very plausible approach. To answer the questions posed at the beginning of the paper, we need to go beyond the absolute level of the estimated dummies and consider whether there is a detectable change in their level around the years indicated in Table 1. A useful way of looking at the results is to group the PTAs by levels of development and continent. To ease exposition, the annual dummy coefficients of table 2 are plotted over time in Figure1. In addition, Table 3 reports the results of the pooled Tobit equation on averaged data for , and These allow us to test whether trade patterns varied significantly over a) Europe. The temporal pattern of trade is almost identical for EU and EFTA. Intra-bloc trade is always below normal and has a strong positive trend since 1985 (EU) and 1986 (EFTA). Although the annual coefficients are statistically significant for the EU, the pooled equation shows that for neither bloc were the coefficients for the average of years , , and

16 statistically different from one another 17. For both European PTAs, although overall bloc imports and exports were above expected levels, they showed a strong negative trend since 1986: also for both PTAs the pooled coefficients for imports in were significantly lower than those for the average of , while the propensity to export was lower in and than in The gross-intra-trade effects for the EU and EFTA fall through time, but except for the latter over to the changes are not significant. These results are somewhat similar to those of Sapir (1997), who found that increased integration within the EU impacted negatively on EU imports from European non-members and prompted their application for EU membership. Frankel and Wei (1998) (also reported in Frankel, 1997, Appendix Table B.6.6) also found a decreasing openness to imports for EC countries between 1980 and 1992 coupled with an increase in the within-ec trade bias by the end of their sample (1992). 18 The additional insight that our results contribute is that both European PTAs exhibit export diversion. It is not possible to be categorical without seeing data on the terms of trade, but this pattern is suggestive that European integration has imposed costs on excluded countries. To anticipate our later discussion, over the period European discrimination was not accompanied by strong external liberalization, and so its diversionary effects on trade with excluded countries was not fully offset by a general liberalization. b) South-South PTAs in the Americas: The situation in Latin America is quite different. All four PTAs show intra-bloc trade significantly above expected levels. That is, these countries have traded unusually strongly with each other over the whole period Whether this reflects pre-existing integration arrangements, geographical isolation, regional sentiment or idiosyncratic tastes is impossible to say. The annual estimates suggest, however, that, although these coefficients were always statistically significant, they did not vary much over our sample. This is corroborated by the results from the pooled estimation when comparing coefficients statistically over and That is, the regional arrangements do not seem to have increased the intensity of intra-regional trade beyond the effects of the general trade liberalization. 17 Although the average of was different to that of for the EU at the 90% significance level. 18 By contrast, Bayoumi and Eichengreen (1997) found that during the mid to late 80 s there was a significant growth in intra-ec trade and in EC trade with other non-ec countries in the sample, although by (the last years covered in their study) these effects had vanished. These authors, however, based their model on data only for industrial countries, a specification which frequently suggests positive intra-eu effects. 16

17 (Note that in our model Brazilian-Argentinian trade, say, receives two general impulses: one as Brazil liberalizes and one as Argentina does. The intra effects is additional to these.) Even the gross-intra-trade effects show little significant change: for CACM its value exceeds s at 90% significance, and for LAIA, exceeds both others at 99% and 90% respectively see Annex 3. All four PTAs exhibit a positive trend in members propensity to import since the late 80s, but only for CACM and for MERCOSUR was this coefficient statistically higher in than in The coefficients for bloc exports showed disparate movement across blocs. LAIA, which reflects only Mexico s propensity to export see above shows a significant positive trend over followed by mild decline or stability. CACM and the ANDEAN Pact display slight (and insignificant) positive trends since the late 80 s, possibly reflecting members general liberalization from around Finally, MERCOSUR shows a mild negative trend over the whole period, with the fastest decline corresponding to the growth of preferences and declining competitiveness in the early 1990s. In none of the cases, however, were the estimates for statistically different from those of Thus, when we control for the impact of the gravity variables such as GDP, population, etc, the revamping (CACM and ANDEAN) or launching (MERCOSUR) of PTAs in Latin America does not seem to have been accompanied by a larger than expected increase in intrabloc trade propensities. The positive trend in the estimated coefficients for bloc members imports, significant in the cases of CACM and MERCOSUR, presumably reflects the unilateral trade liberalization that swept Latin America in the late 80 s and early 90 s. The increases in CACM members overall export coefficients in the 1990s also reflect liberalization, while the opposite trend for MERCOSUR suggests that their members trade performance was dominated by competitiveness considerations rather than trade policy. In fact, there is a clear tendency for the total import and total export effects to move in opposite directions in figure 1 for MERCOSUR and this is also evident for the ANDEAN Group. This confirms the importance of competitiveness over the shorter-run also. Such observations square well with simple observations of trade and macro performance, but we can not, of course, completely rule out that they also owe something to our incomplete coverage of commodities and countries. 19 See Foroutan (1998) for an account of tariff reductions in developing countries. 17

18 Two of these four PTAs were also modeled by Frankel and Wei (1998), with qualitively similar results to ours. For ANDEAN countries they find decreasing intragroup trade bias and increasing extrabloc openness by For MERCOSUR, they have intra-bloc trade above the gravity model predictions, although decreasingly so by the end of their sample (1992), whereas bloc members had become more open to trade by the end of the 80 s. For both PTAs they speculate that the greater extra-bloc openness by 1990 was due to unilateral trade liberalization. As noted above, we actually tested whether the observed trends were statistically significant or not (Table 3). Only for the increasing overall openness to imports for MERCOSUR was this answered positively. c) NAFTA. Besides EFTA, NAFTA is the only bloc where the coefficients for intra-bloc trade were never significant. Annual results show an upward trend practically since the beginning of our sample. The coefficient for overall imports (which is determined only by US and Canadian behaviour) showed a negative trend since 1986 and was statistically significant for virtually the whole period. The export coefficients turned from positive in to negative in , without appreciable changes since Although we observe some indication of export-diversion in the annual data (in 1992 and ), none of the three dummies differed significantly in from its value in Thus, it seems that the key developments NAFTA members trade policies (Mexico s unilateral liberalization in mid 80 s, CUSFTA in 1988 and NAFTA itself signed by the end of 1992) were not associated with appreciable changes in intra or extra bloc trade, once we take into account the normal variation in trade levels that follows changes in the gravity variables 21. These results square with Krueger s (1999) recent conclusion that NAFTA has not had much effect on trade patterns so far. However, her gravity modeling is not particularly appropriate because it characterizes the NAFTA effect as comprising a step change and a change in trend dating from 1987, several years before NAFTA was thought of. Frankel (1997) also has results that are comparable to ours (those derived from annual data, table 4.2): they show that NAFTA 20 A different specification (without the remoteness variable) generates a decrease in ANDEAN countries extra-bloc openness by 1992, Frankel (1997), Table The coefficient for exports was statistically lower in and than in , which might be an anticipatory effect of CUSFTA. 18

19 members mutual trade is about what we should expect for countries with similar characteristics in the sample, while bloc-openness is decreasing by the end of the sample (1992). A recent USITC (1997) sector-by-sector study on NAFTA, found that, for 59 out of 68 sectors analyzed, NAFTA had a negligible effect on US trade, due in part to the already low trade-weighted average duties. Imports from Mexico already received preferences under the GSP and duty-free treatment for U.S. inputs; those from Canada were substantially liberalized by the previously agreed (1988) U.S.-Canada FTA. Out of the 9 sectors with a significant effect on trade, the study found evidence of trade diversion in apparel products, where there has been a significant increase in US imports from NAFTA partners and a conmesurate decrease in imports from Asian and Caribbean Basin countries between 1993 and (USITC, 1997, p 5-12) 22. d) ASEAN. The annual estimates show that the intra-bloc trade coefficient was in general negative, with a pronounced negative trend between 1987 and The coefficient for bloc imports was almost always positive, and significant since 1987, while the coefficient for bloc exports was always positive and significant. The estimates on averaged data showed that the bloc s intra-trade propensity was significantly lower and its overall import propensity significantly higher in than in (and than in ) and by roughly equal amounts, since the changes in gross-intra trade propensities are quite insignificant. These results seem exactly what one would expect from such strongly outward oriented countries. The sharp changes in 1996 are too isolated to be taken as anything but anomalies at present, and since subsequent years data will be dominated by the Asian crisis, there seems little likelihood of identifying additional ASEAN effects in the near future. Although our results for extra-bloc openness are qualitatively similar to those of Frankel and Wei (1998), the latter show a positive coefficient for intra-bloc trade, while ours is negative. We believe that this is due to the fact that we are also capturing the overall bloc-openness to exports, which reduces the absolute value of the marginal effect of belonging to a particular bloc Nagarajan (1999) finds that while US exports of clothing to Mexico during increased by 47%, Mexican imports of clothing from all other sources decreased considerably. He suggests that this could be a case of trade diversion under NAFTA. 23 Frankel (1997, in table 4.2 and p. 98) shows that when the ASEAN openness variable is added to the model, the simple ASEAN intra-bloc coefficient becomes smaller. In our case, we are also adding an extra dummy for bloc exports. 19

20 e) GULF COOPERATION COUNCIL. The intra-bloc trade coefficient was always positive (except for 1985), significant in 1980 and in , and trending upwards since The coefficient for bloc imports was only significant in 1996, with a negative trend since 1993, while the coefficient for bloc exports was always negative and statistically significant, showing a sharp positive trend up to In table 1 we marked 1982 as the key year for this PTA. The test run on the pooled data showed that only the export propensity was statistically different (higher) in than in Conclusions We have applied a gravity model to annual non-fuel imports data for 58 countries representing more than 70% of world imports. The effects of PTAs were captured by dummies that reflected intra-bloc trade and members total exports and imports separately. These blocrelated coefficients were plotted through time to identify changes that might be due to PTAs and tested statistically for changes before and after blocs revival/formation. In summary, 1. When we allow for gravity and overall trade effects, we found no indication that the new wave of regionalism boosted intra-bloc trade significantly. When testing intra-bloc trade before and after a bloc revamping/creation we found no statistically significant change in the propensity for intra-bloc trade. 2. Only for EU and EFTA did we find convincing evidence of trade diversion. After controlling for gravity variables, the EU s and EFTA s propensity to import were significantly lower in than in On the other hand, in the four Latin American PTAs we observed a positive trend in the estimated coefficients for bloc members overall imports, although the increment was statistically significant only for CACM and MERCOSUR. 3. We also found evidence of export diversion in EU and EFTA, which would be consistent with their imposing a welfare cost on the rest of the world. In Latin America increasing propensities to export sometimes accompanied increasing propensities to import, suggesting strong effects from general trade liberalization. The main exception was MERCOSUR, for which total import and export propensities seem to display opposite movements. While MERCOSUR members have undoubtedly liberalized since the mid-1980s, these results 20

21 suggest that their trade performance has been influenced more by competitiveness than by trade policy. 21

22 References Aitken, N.D. [1973], The Effect of the EEC and EFTA on European Trade: A Temporal crosssection Analysis. American Economic Review, Vol. 63, pp Anderson, J. [1979] A theoretical foundation of the gravity model. American Economic Review 69(1), Anderson, J.E and Mercouillier, D. [1999], Trade insecurity, and home bias: an empirical investigation. NBER Working Paper, no. 7000, March Bhagwati, J., and Arvind Panagariya [1996] Preferential Trading Areas and Multilateralism- Strangers, Friends, or Foes? Ch. 1 in The Economics of Preferential Trade Arrangements Ed. by Jagdish Bhagwati and Arving Paragariya. The AEI Press. Washington DC. Bergstrand, J. [1989] The Generalized Gravity Equation, Monopolistic Competition, and the Factor-Proportions Theory in International Trade The Review of Economics and Statistics 71, Bergstrand, J. [1985] The gravity equation in international trade: some microeconomic foundations and empirical evidence. The Review of Economics and Statistics 20, Bayoumi T., and Barry Eichengreen [1997]. Is Regionalism Simply a Diversion? Evidence from the Evolution of the EC and EFTA., in Regionalism versus multilateral trade arrangements, Ed. by Takatoshi Ito and Anne O. Krueger. Chicago: U. of Chicago Press. Chang, W. and L. A. Winters [1999]. How regional blocs affect excluded countries: the price effects of MERCOSUR. Policy Research Working Paper, no. 2157, The World Bank. Deardorff, A. [1998] Determinants of Bilateral Trade: Does Gravity Work in a Classical World? In The Regionalization of the World Economy ed. by Jeffrey Frankel. Chicago: University of Chicago Press. 22