Why Does MENA Trade So Little? *

Transcription

1 Why Does MENA Trade So Little? * Jeffrey B. Nugent, Department of Economics University of Southern California Los Angeles, CA Nugent: nugent@usc.edu August 30, 2002 (Work in progress - Please do not quote) Abstract: The purpose of this paper is four-fold: (1) to examine the extent to which trade of MENA countries, both intra-regional trade and extra-regional, has lived up to its potential based on the gravity model and other standard determinants of trade patterns, (2) to determine the extent to which that has varied across MENA s subregions, (3) to examine alternative explanations for such discrepancies, and (3) to offer quantitative simulations of the extent to which the existing patterns could be changed by alternative trade policy or regime scenarios such as strengthened intra-mena trade, strengthened MENA-EU trade or reduced tariffs, controls on current account and capital account transactions and foreign exchange retention. * Background Paper presented to the Middle East Region Group at the World Bank ** The author expresses his appreciation for the collaboration of Tarik Yousef and Rania Miniesy in previous as well as on-going work, to Grace Lim, Sawanang Wanvimol and especially Paul Hughes for their research assistance, to T.G. Srinivasan for the telecommunications data used in the study, and for their useful comments and suggestions to Dipak Dasgupta, Mustapha Nabli, Aristomene Varoudakis, T.G. Srinivasan, Constantine Glezakos, Michael Kevane, Alfred Tovias and other participants in seminars at the World Bank, the Third Mediterranean Social and Political Research Meeting at the European University in Florence and the Western Economic Association International in Seattle, Washington. The findings, interpretations and conclusions expressed in this paper are entirely those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent.

2 TABLE OF CONTENTS 0.0 Executive Summary... 4 I. INTRODUCTION: THE TRADE INITIATIVE OPTIONS AND LITERATURE REVIEW ON THE USE OF THE GRAVITY MODEL FOR THE MENA REGION Different Analysts, Different Results, Lack of Comparability Use of the Gravity Model and Standard International Data II. THE GRAVITY MODEL, DATA SOURCES AND APPLICATION OF THE MODEL TO BILATERAL TRADE OF MENA AND OTHER REGIONS Rationale for the Gravity Model Gravity Model Specification and Database of Andrew Rose and Its Extension to 1992 and Initial Results Using Rose s Data Extended Only To Extension to Include Additional Observations and Explanatory Variables and to Disaggregate into Energy and Non-Energy Trade Estimation of Bilateral Trade Intensity via Equation (2) Extending the Data Base to III. SIMULATIONS OF THE DIFFERENCES BETWEEN ACTUAL AND PREDICTED BILATERAL TRADE FLOWS WITH DIFFERENT MODELING ASSUMPTIONS AND POLICY SCENARIOS Estimation of Gaps between Actual and Gravity-Predicted Trade Flows of MENA Simulated Effects of Alternative Trade Liberalization and Integration Scenarios IV. CAVEATS, CONCLUSIONS AND POLICY IMPLICATIONS...33 V. References...38 Tables Table 1: Regression Coefficients Obtained for the Basic Gravity Model for Total Trade from Estimating Equation (1) for Different Samples Table 2: Full Sample Panel Regressions of Basic Gravity Model with Added Variables Distance Squared (Ln D 2 ), Landlocked (LL), Sum of the Areas (SAREA) Table 3 Smaller Sample Panel Regressions of Extended Gravity Model with Tariff Data Table 4: Larger Sample Panel Regressions of Base Gravity Model with Additional Variables for Both Transaction Costs and Oil Exports Included Table 5 Larger Sample Panel Regressions of Extended Gravity Model Allowing for Trade Diversion and Including Oil Export Variables Table 6 Panel Regressions for Equation (2) Version of Extended Gravity Model Allowing for Trade Diversion and Oil Export Dummy Variables Table 7 Panel Regressions of Trade Residuals from Extended Gravity Model on Tariffs and Other Barriers Table 8 Panel Regressions of Extended Gravity Model with and Without Outtraffic Based on Pooled Data Table 9 Regression of Extended Base Model with Outtraffic by Year Table 10 Regression Results from with and without New Trade Deflation and Extensions of the Extended Base Model Table 11 Actual Bilateral Trade Values Relative to Their Predicted Values in 1992 (Based on Data) Table 12 Ratios of Actual Trade to Predicted Trade for MENA Subregions 1992 and 1997 for Various Modeling Assumptions Table 13 Simulated Effects (in Percentage Changes) on Trade Resulting from Alternative FTA Arrangements Table 14 Simulation Experiments with Trade Liberalization

3 3

4 0.0 Executive Summary This study provides an overview of recent studies attempting to ascertain how the actual trade flows of the countries of the MENA region compare with those that would be expected on the basis of the gravity model. Although the results turn out to be quite sensitive to the specification of the gravity model, according to most versions of the model, most of the MENA trade flows are well below those that would be expected on the basis of the model. Both variations in the specification and simulations are used to identify factors that would seem to be responsible for these trade shortfalls and policy changes that could be used to overcome them. The principal findings are: (1) With both the widely used Rose data set and our own extension of it in various directions, and with all the modifications to the gravity model undertaken, there is evidence that the MENA region trades less than would be predicted on the basis of the gravity model. (2) The biggest and most consistent shortfall of actual relative to the predicted level of trade is with respect to MENA s trade with the EU. Trade with NAFTA and the intraregional (intra-mena1) trade are also well below what they would be predicted to be according to most specifications of the gravity model. On the other hand, according to all specifications of the gravity model, MENA trades more than would be predicted by the gravity model with the other OECD countries (OOECD) and LDCs. (3) Thanks in part to some trade liberalization by MENA countries during the 1990s, there has been some decline in the size of these trade shortfalls between 1992 and The largest decline in these shortfalls has been in the intraregional trade flows. The gaps remain (as of 1997 at least) very large, however, in MENA trade with the EU and NAFTA. (4) Several factors rather than any single factor seem to contribute to the explanation of the MENA trade shortfalls. One of these is the region s substantial natural resource endowments. This is partly dealt with by the inclusion of dummy variable for one or both partners being oil exporters. These variables have negative and statistically significant effects in virtually all specifications of the model. A second factor contributing to the MENA trade shortfalls is the relatively low scores of MENA countries on the telecommunications indicator OUTTRAFFIC. Still another is the relatively high tariff rates and controls that several MENA countries practice over foreign exchange and both current and capital account transactions. Fourth, is the 4

5 price of oil. When the price of oil is low in real terms (as it has been throughout most of the 1990s), it adversely affects the real value of several important trade flows. The higher real price of oil from the mid-1970s into the mid-1980s helps account for the fact that for 1975, 1980 and 1985, actual trade flows of MENA countries were actually greater than predicted by the gravity model. (5) Another and perhaps surprising factor that contributes to the trade shortfalls of MENA, especially in the case of the EU and NAFTA, would seem to be the trade diversion effects of FTAs like the EU, NAFTA and MERCOSUR. When these effects are accounted for, there is evidence that their influence falls disproportionately on the MENA region. This effect may not be so much one of the FTA itself as of the Common Agricultural Policy of the EU. Without these trade diversion effects, gravity considerations would predict much larger trade between MENA and the EU than they do once these effects are included. (6) A full-fledged FTA with the EU could have a very significant influence, in part since this would eliminate these harmful trade diversion effects and add trade creation effects. This change could exert such an influence only if the EU s Common Agricultural Policy were substantially reformed. At the same time, however, these arrangements would further exacerbate the shortfall in inter-mena trade. (7) Only by reducing the many subtle sources of transaction costs as well as formal barriers to trade among MENA countries can the intra-mena trade shortfall be overcome. (8) There is strong evidence that the existing subregional trade arrangements within the MENA region, such as the ACM and AMU but probably not the GCC, have failed to increase trade.. 5

6 regions. 1 The second and perhaps strongest of these initiatives has been the Euro-Med trade initiatives I. INTRODUCTION: THE TRADE INITIATIVE OPTIONS AND LITERATURE REVIEW ON THE USE OF THE GRAVITY MODEL FOR THE MENA REGION The purpose of this paper is four-fold: (1) to examine the extent to which trade of MENA countries, both intra-regional trade and extra-regional, has lived up to its potential based on a standard gravity model of the determinants of trade patterns, (2) to examine the extent to which this varies across sub-regions of MENA, (3) to examine alternative explanations for such discrepancies actual and predicted trade, (4) to offer quantitative simulations of the extent to which the existing patterns would be changed by alternative trade policy and other scenarios, in particular, strengthened intra-mena trade, strengthened MENA-EU trade, and/or other liberalization measures such as tariff reduction and removing controls on the capital or current account. The basic finding is that MENA trades too little so the focus is on why this is the case. The first two purposes are primarily of interest because they are needed to be able to fulfill the last two objectives. Hence, we start with the rationale for them. If MENA trades too little, what can and should be done about it? To what extent, might better infrastructure help? To what extent, would reduced tariffs help? To what extent would establishing wellenforced FTAs with the EU or among each other help? On the trade policy front, in recent years MENA countries have been considering three quite different kinds of initiatives. In some cases, these initiatives have been at least partially implemented. One has been trade liberalization as part of an overall policy of deregulation and liberalization. This has involved some lowering of tariffs and tariffication of some quotas either on a multilateral basis within the context of the Uruguay round or on a unilateral basis under the pressure of donor organizations or the WTO, (in the latter case as a condition for membership). Yet, as many analysts have recognized, trade and other forms of liberalization in the MENA region have lagged behind those of virtually all other leading to the development of FTAs between the EU as a whole and some individual MENA countries. Among the reasons for assigning priority to EU-MENA links are: (1) to take advantage of the large size of the EU market, (2) to offset the trade diversion effects that have arisen from the EU and its tariffs and quotas on products exported by MENA countries, (3) to avoid waiting until after the anticipated major expansion of the EU membership to the East (for afterwards, the EU may have much less interest in further agreements) and (4) to allow MENA countries to take advantage of the dynamic benefits 1 See Dasgupta, Keller and Srinivasan (2002) 6

7 associated with FDI and technology transfer. For the EU, of course, an additional reason is to reduce the pressure for immigration both legal and illegal from the MENA region. The third initiative, though more on-again, off-again rather than steady, has been the rebirth of interest in inter-arab or inter-mena trade. At least four reasons have been given for pursuing inter-arab or inter-mena trade: (1) to serve as a possible counterbalance to already initiated EU-MENA trade arrangements, (2) to help MENA countries prepare for those trade arrangements, greater competition within the region being seen as a useful training ground for the more difficult competition with European producers and in how to interest European consumers, (3) to provide a more politically and socially acceptable means of maintaining some degree of openness and liberalization in the face of accumulated resentment arising from the presently very dim prospects for the peace process and cooperation between Israel, the Palestinians and the other nation states in the region, and (4) to complement the EU-MENA initiatives which, otherwise, would not be likely to attract much FDI because of the hub and spokes character of the EU-MED links. 2 As far as the inter-mena trade arrangements are concerned, these may not necessarily be at the level of MENA as a whole or ever the Greater Arab Free Trade Association among only the Arab countries of the MENA region. Indeed, it could be at the subregional level (the Arab Maghreb Union, the Arab Common Market, the Gulf Cooperation Council (GCC)) or some other grouping of countries. Given the three alternative trade initiatives, in what follows we shall investigate the possible effects of all three on MENA trade. 1.1 Different Analysts, Different Results, Lack of Comparability Despite a large number of papers and books on such matters (see below), there still exists considerable variation in views about the relative net benefits for MENA countries of pursuing any of the above three alternative trade strategies. The relative desirability of the three alternatives quite naturally hinges on the extent to which actual trade of MENA countries exceeds or falls short of its full potential and if it does, why. If MENA trade were less than its potential, presumably one could identify ways in which that potential might be more fully realized. On the other hand, if MENA s actual trade were equal to or even greater than what might be regarded as its full potential, it would cast doubt on the usefulness of such efforts to increase such trade. It is also of obvious relevance as to whether any shortfalls in actual trade relative to potential trade have been in intra-mena trade, MENA-EU trade or MENA trade with other countries. Even if attempts were made to boost MENA trade through regional trading arrangements 2 By hub and spokes is meant that the investors would have much greater incentive to locate their investment in the EU so as to serve both the EU and all MENA markets than to locate in a MENA country to serve only that MENA country and the EU. 7

8 either within MENA or between MENA and the EU, since such arrangements necessarily involve trade diversion as well as trade creation, would such regional arrangements be welfare enhancing? On all these points, there are substantial differences in opinions and conclusions in the literature. One reason for differences in opinion is the fact that in the past there has been such a wide gap between the way trading arrangements have been designed and the way they have worked in practice. Many countries of the MENA region have had a variety of bilateral trading arrangements (with each other, with European countries and with other developing and developed countries) that in practice have been very poorly implemented. Often they have been virtually disregarded soon after they were signed. Similarly, or even especially, there has been a large gap between what the multilateral agreements within the MENA region say on paper and what they do in practice. Indeed, there has been a long history of failure in multilateral trading arrangements in the region (the Arab Common Market, other initiatives of the Arab League and the Council of Economic Unity, various subregional arrangements such as among Maghreb countries and Turkey, Iran and Pakistan. Not only have these special arrangements been far from fully realized, but also even normal trade among countries of the region has been periodically and unexpectedly interrupted by trade boycotts, ad hoc border closings and so on. This implies that businesses set up to take advantage of regional trade have been adversely impacted by such failures. 3 Without a radical change from past experience, such businesses may justifiably believe it better not to think about the possibilities of such trade in making their investment and production decisions. Doubts about the wisdom of even well-implemented free trade agreements (FTAs) and customs unions (CUs) for promoting trade and welfare also derive from the economic integration theory literature. This literature points to the possibilities for welfare-reducing trade diversion of free trade agreements, especially in conditions not unlike those within the MENA region where differences in factor endowments are relatively small, tariff rates vary considerably from one country to another, trade taxes constitute a relatively large part of total government revenue, technological efficiency is relatively low and differences in wage rates are relatively small. In addition, the past experience with such arrangements, in developing countries at least, suggests that there could well be conflict-generating polarization effects, given the relatively large differences in initial levels of income, market size and policy environments of relevance to trade and competitiveness among countries in the MENA region. But the polarization effects could also apply a fortiori to FTAs or CUs with developed countries like the EU, especially given the hub and spokes character of EU-Med arrangements that arises from their being concluded separately with individual MENA countries rather than with groups of MENA countries. Finally, since factor movements (of both labor and capital) are rather high among countries of the region, and factor movements can substitute for trade, some feel that such flows undermine the basis for trade 3 See especially United Nations Economic Commission for Western Asia

9 within the region. Some have even argued that the promotion of factor movements is a better way of achieving regional integration than trade. Even at best, many see regional trading arrangements as coming at the expense of the more advantageous broader multilateral trading arrangements (such as those within the WTO). Since economic integration theory specifies that the net benefits of any FTA or CU to any member country depend on the relative magnitudes of trade creation benefits and trade diversion losses both of which are hard to estimate, it is easy to understand how there can exist disagreements in the resulting evaluations of different analysts. This is because the outcomes of such studies depend on assumptions concerning functional forms, the values of hard-to-estimate elasticities of supply and demand, and so on about which there are likely to be varying estimates and judgments. Even if there should be consensus on the estimates of partial equilibrium analysis, the fact that the more satisfactory general equilibrium analysis would involve thousands of assumed parameter values of unknown validity further increases the odds that different analysts will come to very different conclusions of the net benefits to an individual country or set of countries about any such arrangement, intraregional or extraregional. For example, a commonly used modeling approach to such issues has been computable general equilibrium (CGE) models. 4 Since these models are highly dependent on many modeling assumptions and parameter estimates of unknown realism, not surprisingly different modelers have derived different conclusions about the effects of trade policy from these models. In many cases, moreover, the results derived from these models are more attributable to modeling features than to trade issues. Often the trade scenarios used in simulating such models are quite naïve and based on ad hoc assumptions about future trade patterns and alternative trade policies and strategies. 5 Even if there were complete agreement on how to model trade options of a particular country, very different evaluations of these alternatives might arise depending on whether or not the model is left as a national model or it is inserted into a regional or world system of national models. Since the magnitude of the research effort involved in any such study is very large, each is likely to be done by a large research team and different studies by different teams. Each such team may use a different database and set of modeling assumptions, and employ different research strategies, thereby making it difficult to pin down how and why they arrive at their different evaluations of the options. Hence, in comparing different results for the same question obtained by different research teams, seldom is it possible to distinguish the effects of different modeling assumptions 4 Hertel (1997) provides a rather encyclopedic description of the construction and application of CGE models to international trade and Robinson and Thierfelder (1999) a survey of the empirical literature on the application of CGE models to regional economic integration in particular. 5 This is certainly not always the case. For example, Bayar, Ben-Ahmed, de Boer, Diao and Yeldan, (forthcoming) examine the effect of trade liberalization in manufactures and other sectors under two alternative scenarios about MENA. In one of these the Euro-Med agreements are carried out with no complementary integration among MENA countries and then again with trade liberalization among MENA countries. 9

10 from those of different data sets. Moreover, many of the parameter values used in CGE models are merely guesses. Even if they are estimates based on micro data sets or time series analysis, seldom is concern given to econometric issues such as possible sources of specification bias, errors in variables, identification problems, multicollinearity, etc. Moreover, as indicated above, many of the parameters that are estimated (as opposed to only assumed) are often based on single observations, e.g., an average inputoutput coefficient from a social accounting matrix (SAM table) for a particular (usually badly outdated) year. The estimates of trade flows obtained from world econometric models (like Project Link or other simpler versions of econometric and simulation models) suffer from similar problems, again limiting the comparability of their results and raising serious methodological questions as to the internal consistency of the models and the validity of any conclusions drawn Use of the Gravity Model and Standard International Data In order to help put the analysis of such effects of different types of trade strategies on a firmer basis, the present study makes use of a far simpler and more transparent model of trade, namely, the gravity model. Section II (to follow) provides a complete description of the gravity model and the rationale for its popularity. In particular, in this study the gravity model is used to explain the determinants of trade patterns within the world as a whole arising from the bilateral trade flows between each pair of countries in the world over the period The database for this study is an updated version of a large trade matrix that has now been used in a number of the studies referred to below. The gravity model has been applied to a wide variety of countries, regions, and levels of commodity aggregation. Gravity models focus on the effects on trade patterns of the geographic distance between, and respective market sizes, of each pair of trading partners. Important advantages of the gravity model are its relatively high explanatory power and its modest data requirements. One study using a gravity model applied to some countries of the MENA region is that of Ekholm et al. (1996). This study argued that the potential for trade growth within the region, even within a more peaceful one, is quite small. This study computed four sets of predicted to actual export ratios (across all trade partners in the sample) for twelve MENA countries (Algeria, Egypt, Iran, Israel, Jordan, Mauritania, Morocco, Saudi Arabia, Sudan, Syria, Tunisia and 6 For example, in large models of this sort it is obviously impossible to estimate all variables simultaneously. Usually a bloc procedure is used with little information about the reliability of the assumptions made in the use of such simplifying procedures. 10

11 Turkey). The first set of predicted to actual trade ratios was intra-regional on the basis of no integration. A second was again intra-regional but with full intra-mena integration. The third set was for trade of each individual MENA country with EU countries as a whole without any special trade liberalization measures with the EU. Finally, the fourth set was for trade with EU countries but in this case assuming full integration with the EU. For no MENA country was full intra-regional integration found to increase the predicted to actual export ratio by more than 15 percent. In most countries, moreover, the predicted to actual export ratio was below unity, indicating that intra-mena trade was already larger than might be expected on the basis of country income, distance and various other transaction cost variables. 7 The results for the trade levels of individual MENA countries with other countries were very different than those for intra-regional trade even though the estimated effects of integration were quite similar. 8 These calculations were made on the basis of parameter estimates obtained from a different set of countries and then applied to data for the explanatory variables taken from MENA and other countries. To estimate the parameters, the authors used a gravity model equation along the lines of that used in the next section but with a smaller set of explanatory variables 9. They estimated the parameters of the model on the basis of cross section data for the year 1989 (that year being deemed to be a normal trade year for MENA and other countries) for a relatively small set of countries, in particular, 13 developed countries 10 and 11 developing countries 11. The latter sample includes only relatively open countries whose trade values would not have been suppressed or distorted. Being quite restrictive in coverage, the total sample amounts to only 469 observations. While most of the parameter estimates were of the expected sign and magnitude and were statistically significant (at the five percent level), certain key ones were not, namely, the coefficients of common language and European Community. Both of these 7 Fischer (1993) was also pessimistic about the prospects of increased trade through integration within the MENA region. 8 In particular, the predicted to actual trade levels are typically much larger than unity. For a brief comparison of the Ekholm et al (1996) estimates with some others for MENA and other regions, see Havrylyshyn (1997). 9 Specifically, the explanatory variables included are the GDP s and GDP s per capita of each trading partner, distance, difference in mean years of schooling, the shares of primary products in total exports of both countries, common language, common border and a dummy variable for membership in the European Community. 10 The developed countries in the sample are Austria, Belgium-Luxemburg, Canada, Finland, France, Germany, Italy, Japan, Netherlands, Norway, Sweden, United Kingdom and United States. 11 These are: Brazil, Chile, Hong Kong, Israel, Korea (South), Malaysia, Singapore, Thailand, Tunisia, Turkey and Uruguay. 11

12 variables have been found to have had positive and significant effects on bilateral trade flows in other studies. Yet, in the Ekholm et al. (1996) study, neither of these important variables was found to have had a statistically significant effect. The small size and lack of significance of the European Community dummy variable was obviously of considerable importance since it was the coefficient used to simulate the effect of complete intra-regional integration. The small size of the sample, the fact that most countries in the sample were rich (either developed or high income developing countries), that EC countries make up a significant portion of the sample, could contribute to the explanation for why the integration variable is so small and insignificant compared to estimates obtained in other studies and why the intra-mena trade values were so large relative to the values predicted from the gravity model. Moreover, other studies using more traditional Vinerian methods have been much more optimistic about the potential for increased trade in the MENA region. 12 One of the few other studies applying the gravity model to Arab countries, though not the broader MENA region is Al-Atrash and Yousef (2000). Once again, this study makes use of data for only a portion of world. Yet in this case, it includes 18 Arab countries and 43 other countries that account for over 90 percent of trade of Arab countries. Also since it deals with the years , it is much more up to date. The dependent variable in this case is the average value of the bilateral trade over the years and explanatory variables are many of the same ones used above inspired by the gravity model. By restricting the country coverage and averaging the bilateral trade flows over three years, they are better able to distinguish between zeros and missing values in the bilateral trade flows. Whereas the data set used above and based on Rose (2000) had no zero observations for bilateral trade, Al-Atrash and Yousef (2000) report that almost 15 percent of the entries in their bilateral trade matrix were indeed zero. For this reason, they employ a Tobit procedure instead of OLS used in the other studies in the estimation stage. They included in their analysis dummy variables for subregions within MENA and subregional FTAs. The maximum likelihood estimates obtained by Al-Atrash and Yousef (2000, Tables 4 and 5) do not suffer from the rather unfortunate choice of sample used by Ekholm et al (1996). Perhaps as a result, their results seem more plausible than those obtained by Ekholm et al (1996) and generally quite comparable to those reported here below, even though based on a far smaller 12 For example, Zarrouk (1992) estimated that the elimination of tariff barriers in the MENA region would induce intra-regional trade to grow somewhere between 4 and 8 percent per annum faster over the first ten years faster than it would without such an action. See also Abed (2000) and El-Naggar (1992). 12

13 sample and somewhat different specification of variables. They found the effects of membership in the AMU and GCC (Table 4) on bilateral trade of members countries to be negative and significant and as such very different from estimates for other FTAs. So too do the dummy variables for oil exporter (Arab2) and for both countries being Arab countries (Arab3) in Table 5. We specify a gravity model that is more general, includes additional measures than in the Ekholm et al (1996) and Al-Atrash and Yousef studies. Instead of including only a small set of countries for a single year, neither of which may be very representative, we use the experience of virtually all countries in the world and for a number of different years taken at the decades of the 70s, 80s and 90s. We also attempt to greatly mitigate or even altogether eliminate the problems of endogeneity and uncertain directions of causality that often arise in the use of models when trade variables depend on variables that may be endogenous, such as GDP, relative prices, demand patterns or even policy variables like tariffs and quotas which are also endogenous within a political economy framework. The gravity model is used as the basis for simulating the levels of trade that would be predicted among each pair of countries, including the MENA countries, under alternative assumptions about trade regimes and other conditions. This allows us to determine the extent to which actual trade flows among partners exceeds or falls below those that would be expected. The general finding is that MENA countries tend to trade less than would be expected on the basis of the gravity model, especially so in the case of intra-regional trade. Different estimating equations are used and alternative regimes assumed in making these predictions. In particular, we examine the effects of two scenarios, the Euro-Med trade agreements and an FTA among Arab or MENA countries, independently as well as jointly. The remainder of the paper proceeds as follows: Section II outlines the methodology and its application to the questions at hand, including the data sources. Section III presents some results obtained from the gravity model, and summarizes the answers to the first two objectives of the paper. Section IV contains the simulations and derives conclusions concerning the third objective. Section V concludes. II. THE GRAVITY MODEL, DATA SOURCES AND APPLICATION OF THE MODEL TO BILATERAL TRADE OF MENA AND OTHER REGIONS 13

14 As already mentioned, to help evaluate the different trade initiatives identified above and estimate the extent to which actual MENA trade patterns differ from those that would be expected on the basis of determinants thought to be important in the applied trade literature, we use the well-known gravity model developed in the 1960s and which has been extremely widely applied ever since Rationale for the Gravity Model The use of the gravity model is especially attractive in this context because of its consistency with underlying theoretical micro-foundations, 14 its demonstrated applicability to many different kinds of countries and regions, its robustness over time, and to various different specifications. According to this model, the trade flows between any pair of countries should be affected by their mass (the product of their respective GDPs) as well as by the distance between them 15. The latter is because transport and transaction costs can be assumed to rise with distance. Factors such as exchange rate variability, common language, common colonial or other historical experience, common currency, free trade agreement, having a common border, the physical size of the country, and whether or not the country is landlocked, can also be included since these, too, can affect these transaction or transportation costs, and thereby also the trade flows between each pair of bilateral trade partners. While hypotheses concerning the importance of trade for growth have been popular for many years, they remain very controversial since the construction of tests of the impact of trade on income growth have been subject to obvious problems of simultaneity and causality. The gravity model has received considerable recent attention in this context since it permits a measure of openness to trade that is not dependent on potentially endogenous variables. Endogeneity arises because actual exports and/or imports would depend heavily on income, tariff rates and quotas which are certainly also variables that could depend on the level of trade flows. This is obvious in the case of income or GDP, but even tariffs and quotas are endogenous in a 13 See for examp le, Anderson (1975), Bergstrand (1985, 1989), Deardorff (1998), McCallum (1995), Rose (2000), Anderson and van Wincoop (2001) and numerous other references contained therein. To our knowledge, however, the only previous studies that have used the gravity model in the context of the Middle East have been Ekholm et al (1996), Al-Atrash and Yousef (2000), Limam and Abdalla (1999) and Sayan (1998). Their results are of limited usefulness either because they either focused on aggregate trade flows for one period or utilized small samples of countries and few explanatory variables. 14 See for example, Anderson (1975, 1979, 1998) and Bergstrand (1985, 1989). 15 As noted by Rose (2000), Leamer and Levinsohn (1995, p. 1384) attribute to the gravity model some of the clearest and most robust empirical findings in economics. 14

15 political economy sense. 16 To get around this long-standing problem, Frankel and Romer (1999) used a gravity model containing only exogenous variables as instruments in order to predict trade in relation to GDP, thereby moving GDP to the left hand side of the equation. They employed data for the year 1985 from 150 countries to estimate the parameters of the gravity model and then used the estimated parameters to predict the trade flows for that year. The results were then used as the new and improved measure of openness to be inserted into a standard cross-country model for the level of per capita GDP. Other variables included in the equation for per capita income were investment rates, population, area and in some cases other variables suggested in standard cross-country growth models. The relationship between the new and improved measure of openness and per capita income (or growth) was shown to be positive and significant. 2.2 Gravity Model Specification and Database of Andrew Rose and Its Extension to 1992 and 1997 Recent papers by Rose (2000) and Frankel and Rose (2000) have applied just such a model to bilateral aggregate trade flows for a large, incomplete panel data set of 186 countries taken at five year intervals between 1970 and These authors compiled the relevant data on all the relevant variables and have made that data set available to other researchers. For this reason we make use of this data but extend it in three important directions. First, we extend it in time to 1992, based on the international trade matrix for 1992 put together by the National Bureau of Economic Research (Feenstra, Lipsey and Bowen 1997), and subsequently to 1997 based on Feenstra (2000). Second, we fill in several thousand observations that were missing in the original data set with information from more recent standard international sources, especially for explanatory variables like country size and distance and for MENA countries whose coverage was rather spotty in the original data set. Third, we extend it by disaggregating the trade matrices of each year from the aggregate flows constructed by Rose and his colleagues into energy (specifically oil and gas flows) and non-energy subtotals. The 1992 and 1997 values for GDP, population and other variables are taken from standard sources such as the World Bank s World Development Indicators. Detailed sources for the earlier years are given in Rose (2000). The exchange rate volatility index was computed as the standard deviation of the first difference of the monthly natural log of the bilateral exchange 16 In particular, if imports are very high into a particular country, that country may impose tariffs in order to avoid serious balance of payments problems. 15

16 rate (based on line ae from the International Monetary Fund s International Financial Statistics in the preceding five years). As in Rose, the EEC/EC, NAFTA, EFTA, Australia/New Zealand, ASEAN, CACM, CARICOM, PATCRA, SPARTECA, Cartagena Agreement (Andean Group) and Israel/US Free Trade Agreements are all treated as examples of FTAs and hence signified by values of 1 in the FTA dummy variable. Significantly, Rose did not include any of the aforementioned trade arrangements among MENA countries as examples of FTAs. Presumably, this was because their effectiveness was deemed to be too slight to be treated on an equal footing with those other arrangements. Rose (2000) found the results of the basic gravity model to be strong and to be remarkably robust to alternative specifications such as the inclusion of other variables or alternative measures of some of the included variables. Among the additional variables were regional dummy variables, the sum or product of the two countries tariff rates, a dummy variable for whether one of the countries is landlocked, the sum or products of the two countries land areas, and dummy variables for one or both countries having controls on the current account and foreign exchange surrender and for one (or both) being an island economy. Yet, since few of these variables were reported to have been very significant, he did not include these additional variables in his basic analysis. To start the present analysis as close to his well-trodden way, neither did we. Specifically, we first specify the model as follows: Ln( X ijt ) = β 0 + β Ln( YY ) + β 6 1 FTA ijt i j t + β 7 + β Ln( YY / Pop Pop ) + β LnD 2 ComNat ij i j + β ComCol 8 i j ij ij + β Cont + β Colony + γcu 9 3 ij 4 ijt ij + β 5 Lang + δv( e ) ij t ij + ε ijt (1) where i and j denote countries, t denotes time, and the variables are defined as follows: X ij denotes the value of bilateral trade between i and j, Y is real GDP, Pop is population, D ij or DIST is the distance between i and j, Cont ij is a binary variable which is 1 if i and j share a border, Lang ij is a binary variable which is 1 if i and j share an official language, FTA ij is a binary variable which is 1 if i and j belong to a FTA ComNat ij is a binary variable which is 1 if i and j are part of the same nation, ComCol ij is a binary variable which is 1 if i and j share the same colonizer after 1945, 16

17 Colony ij is a binary variable which is 1 if i colonized j or vice versa, Cu ijt is a binary variable if i and j use the same currency at time t, and V(e ij ) t is the volatility of the bilateral of the exchange rate between i and j in period t. 2.3 Initial Results Using Rose s Data Extended Only To 1992 In the course of the work reported here, despite seeming similarities in results obtained after making various changes in data manipulation, model specification and data coverage, we have found that the results are in fact quite sensitive to such changes. For this reason, the presentation which follows starts with a data set, data manipulation procedures and model specification that are as close as possible to those used in the studies by Rose. Then, we add data and variables to the specification, one change at a time, allowing us to find at virtually each stage of the analysis how that change affects the results. This specification from Rose (2000) was used in Miniesy, Nugent and Yousef (2001) and is the starting point for the present analysis. The results with this specification are reproduced in Table 1 for each different period, i.e., 1970, 1975, 1980, 1985, 1990 and 1992 as well as for the pooled sample. Note that for each period there were at least 4000 observations and the pooled sample contains over 27,000 observations. Note also that, although the magnitude of the parameter estimates varies somewhat from one sample to another, with perhaps a single exception the signs of the coefficients do not vary across samples. The numbers in parentheses below the corresponding parameter values are White s heteroskedasticity-consistent standard errors. In virtually every case, the estimated coefficient has the sign expected according to the gravity model, i.e., negative for exchange rate volatility and distance and positive in each of the other cases. In most cases, moreover, as indicated by the asterisks next to the coefficients, the coefficients are all significant at the 5 percent level or below. The value of adjusted R 2 varies from 0.57 to 0.73, indicating explanatory power comparable to that obtained by Rose (2000) and Frankel and Rose (2001). As indicated by the form of equation (1), the model is loglinear in the continuous variables so that the estimated coefficients represent elasticities, i.e., showing the percentage changes in the bilateral trade values with respect to given percentage changes in the continuous explanatory variables. The parameters of the dummy variables indicate the percentage change in the dependent variable attributable to a unit increase in the dummy variable (from 0 to 1 ). An idiosyncrasy of this data set is that only positive trade flows are reported. In other words, for 17

18 those cells in the matrix for which no bilateral trade is reported, no information on the explanatory variables is provided. Presumably, this is because the authors suspected that in fact there were some trade flows but that the information was missing. As a result, the parameters can be estimated by ordinary least squares (OLS) just as they were by Frankel and Romer (1999), Frankel and Rose (2000) and Rose (2000). 2.4 Extension to Include Additional Observations and Explanatory Variables and to Disaggregate into Energy and Non-Energy Trade In this paper, however, we make use of some additional gravity-type variables that are available in the Rose (2000) data set. These include: Distance squared (D 2 ij or DISTSQ), Landlocked (LL), sum of the areas (SAREA), sum of the tariffs (TARS), an index of telecommunications infrastructure (OUTTRAFFIC) and dummy variables if one of the partners is an oil exporter (ONEOIL), or if both partners are oil exporters (BOTHOIL). Finally, subsequently, to capture possible trade-diverting effects of both FTAs and currency unions, we include dummy variables if only one partner is a member (ONEFTA), and ONECU). Irrespective of the specific variables included on the right hand side, this shall be called equation (1). However, as mentioned above, these changes are made individually or in groups, not altogether. Alternatively, the dependent variable could be the bilateral trade flows relative to the product of the GDPs ln (Xij/(YiYj), in which case the ln(yiyj) term is removed from the right hand side. This has the effect of removing the most important source of endogeneity in the equation. By comparing the results of both specifications in Frankel and Romer (1999), however, it can be seen that there is a price to be paid for this in terms of a lower R 2. This specification shall be referred to as equation (2). After filling in observations that were missing in Rose s data but available from the sources indicated above, especially for MENA countries, as shown in Table 2 the panel data set is increased from 27,291 to 35,621 for the same set of 186 countries. 17 This table includes only three additional variables to those included in the basic gravity model of Table 1. These variables are distance squared (D 2 ), the sum of the geographic areas of the two trade partners (SAREA) 17 A recent application of the gravity model to the evaluation of the effects of intraregional trade arrangements on extraregional trade arrangements by Al-Mashat 2000 for NAFTA showed these effects to be much more positive than suggested in the standard trade literature. 18

19 and landlocked (LL) if at least one of the partners is landlocked. Table 2 also provides separate estimates for trade flows broken down into energy trade and non-energy trade. Although subject to the possible criticism that in a political economy sense, it could be another source of endogeneity, in Table 3 we include one more additional variable, namely, the sum of the tariff rates between each pair of trading partners i and j (TARS). Since the data on tariff rates, especially for the earlier sample years, is quite incomplete, this has the unfortunate effect of reducing the number of available observations to 15,992 for total trade (from 35,621 in Table 2), to 6586 for oil trade, and to 15,734 for non-oil trade. Due to a high correlation between the tariff rate variable (TARS) and Cont for this data set, the Cont variable had to be omitted from the model. By happenstance also, TARS is also fairly highly correlated with D. For this smaller data set, the values of the adjusted R 2 are slightly higher for each set of estimates in this table than for the comparable models in Table 2. In other respects, however, the results are quite similar. As expected, TARS has a negative effect on bilateral trade flows in all three types of trade, significantly negative for both total trade and non-oil trade. Another difference is that the coefficient of the linear distance term (LnD) in each case is much weaker and even positive for all three types of trade. This is compensated, however, by the fact that the coefficient of the quadratic distance term (LnD 2 ) is now more strongly negative in each case. To retain the larger sample and to avoid potential endogeneity in the use of TARS as an explanatory variable, in Table 4 we exclude TARS and return to the larger sample of 35,621 (for total trade). Given, the much poorer application of the simple gravity model to energy trade and the fact that quite a few MENA countries are in fact energy exporters, the specification used in generating the results of this table includes two additional dummy variables for being an energy exporter. ONEOIL takes the value of 1 if only one of the trade partners is an oil exporter (and the value of 0 in all other cases). BOTHOIL takes a value of unity if both trade partners are oil exporters (and 0 in all other cases). We would expect the coefficient of ONEOIL to be positive in the case of oil trade and ambiguous in other trade types and that of BOTHOIL to be negative in all cases. Since oil exports depend on being well-endowed in this form of natural resources, both of these variables can safely be assumed to be exogenous to the level of trade flows. For total and non-energy trade, the inclusion of these variables in Table 4 has the effect of increasing the explanatory power modestly and sharpening some of the other effects. More importantly, both variables (but especially BOTHOIL) have significant negative effects on 19