A tale of parallel integration processes. A gravity analysis of EU trade with Mediterranean and Central and Eastern European Countries more

Review of Middle East Economics and Finance Volume 5, Issue 2 2009 Article 2 A Tale of Parallel Integration Processes: A Gravity Analysis of EU Trade with Mediterranean and Central and Eastern European Countries Francesco Pastore∗ Anna Maria Ferragina† Giorgia Giovannetti‡ ∗ Seconda Universit` di Napoli and IZA, fpastore@unina.it a Universit` di Salerno, aferragina@unisa.it a ‡ Universit` di Firenze, giovanne@dicea.unifi.it a † Copyright c 2009 The Berkeley Electronic Press. All rights reserved. A Tale of Parallel Integration Processes: A Gravity Analysis of EU Trade with Mediterranean and Central and Eastern European Countries∗ Francesco Pastore, Anna Maria Ferragina, and Giorgia Giovannetti Abstract Despite the EU emphasis on the 1995 Barcelona process, trade integration with the Mediterranean (MED) countries is still underdeveloped. To contrast the success of EU integration with MED countries and that with the new EU members, we compute the trade potential of these EU partners from 1995 to 2002 using an “out-of-sample” methodology. The coefficients are taken from different panel estimators of the gravity equation relative to intra-EU trade. Our analysis suggests the existence of sizeable, unexploited trade potential with both groups of partners, although the ratio of potential to actual trade with the MED countries is much larger (from 1.7 to 2.5 times), more dispersed and stable compared to that with the CEECs. KEYWORDS: MED agreements, Europe agreements, Eastward EU enlargement, gravity equation, trade potential Previous versions of this paper have been presented at the European Trade Study Group, University of Nottingham, 9-11 September 2004; at the CNR Study Group on “International Economics and Development” Universit` Commerciale Luigi Bocconi, Milan, 5-6 November 2004; at the a Romanian Academy of Science, 10 November 2004; and at the XII Economic Research Forum Conference, Cairo, 18-22 December 2005. We thank Sergio Alessandrini, Edward Christie, Paolo Epifani, Ahmed Gonheim, Rodolfo Helg, Lelio Iapadre, Paolo Malanima, Saad Nassar, Mariana Nicolae, Lucia Tajoli and Elisa Valeriani for useful comments. However, the usual disclaimer applies. ∗ Pastore et al.: A Tale of Parallel Integration Processes 1. Introduction In the last decade, gravity models have been extensively used to forecast potential bilateral trade relations and integration effects between EU (or OECD) countries and the former CMEA1 members and less frequently in the case of trade relations between the EU and the South Mediterranean countries (MED from now). The persistence of high barriers to trade as well as remarkable social and economic differences between the two shores of the Mediterranean made the perspective of enhanced financial and commercial integration less likely than that with Eastern Europe. The average rate of protection in the MED countries is indeed high, at 17.5% for all products, except agricultural products and services. Moreover, the protection of Morocco has increased, from 21% in 1997 to 31% in 2001, while that of Tunisia remained constant at 20% and that of Egypt at a rate above 20%. These figures compare with a rate of 5.2% in the case of CEECs (Femise, 2002). According to some observers, the emphasis and commitment on the Eastward enlargement of the EU in the second half of the nineties resulted in a sort of “crowding out” of the Euro-Mediterranean partnership launched in 1995 at the European Council held in Barcelona. However, EU signed Association Agreements with Malta (1971), Cyprus, (1972), Tunisia and Israel (1995), Morocco and Turkey (1996), Jordan (1997), Palestine (1997), Egypt (2001), Lebanon and Algeria (2002), Syria (2003). All these agreements entered into force, except that relative to Syria and their application has had a new acceleration with the European Council of Thessalonica in June 2003. Since 2004 Malta and Cyprus have become member states of the EU and in 2005 Turkey started accession negotiations. Libya and Mauritania have obtained the status of observer. Association Agreements entail commitments to political, economic and human rights reforms and have given birth to a process of trade liberalisation consisting of a gradual levying of tariffs and trade barriers for manufacturing products and of a gradual opening up to international trade for agricultural products and services. It should be noted that the EU has opened its markets to MED long before the Barcelona agreements were signed, gradually reducing tariffs for a large number of industrialised products. As a result, liberalisation has essentially regarded tariffs imposed by the Mediterranean countries on goods imported from Europe (see, for further details, Melad, 2008; Romagnoli and Mengoni, 2009). The EU initiated the Europe Agreements (EAs) with each CEEC separately. The first agreements with Poland, Hungary and Czechoslovakia were signed in December 1991 and came into force in 1994. In February 1993, similar agreements were signed with Bulgaria, Romania as well as the newly established countries of the Czech Republic and Slovakia. They came into force in 1995. In 1 Council for Mutual Economic Assistance. Published by The Berkeley Electronic Press, 2009 1 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 1998, EAs were implemented with the three Baltic States, followed by Slovenia in February 1999. The EAs were aimed at fostering trade integration, but also the political dialogue and cultural and economic cooperation between the partners, while providing a basic outline for the gradual EU accession of CEECs. Over the period before the agreements came into force, Interim Agreements provided for an anticipated and temporary application of trade provisions. Their aim was to establish a free trade area for industrial goods for ten years on a reciprocal but asymmetric basis: the EU had to remove its trade barriers more quickly than the CEECs. This led to the total removal of all tariff barriers on industrial products from the EU in January 2002. However, a special discipline was created for some “sensitive” industrial sectors, in particular for textiles, iron and steel, car industry (only for Poland) and a much more gradual liberalization was applied to agricultural goods and fisheries (EC, 1997)2. The aim of this paper is to provide an assessment of the relative success of the EAs versus the Mediterranean Association Agreements (MAAs from now), using the same modelling strategy and type of data. The structure of the MAAs is analogous to the EAs under many respects. Moreover, they were built almost over the same span of time. This analysis aims at answering three main questions: 1) What amount of trade could be achieved with these two country groups if the trade elasticity with respect to economic and geographic variables (relative mass, physical distance, common language, common land border, colony links) were like those achieved in intra-EU trade (trade potential)? 2) Do we observe a reduction of the gap between potential and actual trade over the two liberalisation processes (trade creation)? 3) What would be the timing for convergence to the potential for the two areas if integration would be pushed further (convergence)? Gravity analysis is used to measure the different speed of trade integration of the EU with the two areas now mentioned. This is the best alternative when intertemporal extrapolation of trade patterns cannot be based on historical trade performance, as it is the case of both CEECs and MED due to their past economic isolation, distorted pricing structures, state intervention, scant international openness and recent transition from a planned to a market economy. From a methodological point of view, this study differs from previous ones on CEECs and MED, because it uses an “out-of-sample” computation to estimate trade potential3. More precisely, we use the parameters extracted from a gravity equation of intra-EU bilateral trade flows to predict trade potential between five EU members (Italy, Germany, France, UK, Spain) and, respectively, MED and CEE countries. For an overview of the recent debate, see the special issue of the Journal of International Trade and Economic Development published in 2009 on: “The Economics of the Fifth Enlargement”. 3 2 To our knowledge, only Gros and Gonciarz (1996) use an out-of-sample method, while Baldwin (1994), Matyas (1997) and Egger (2000; 2002) use an in-sample method. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 2 Pastore et al.: A Tale of Parallel Integration Processes The results, which are remarkably stable across different specifications and estimation methods, suggest the existence of substantial unexploited trade between the EU and both the areas considered. Not surprisingly, the gap is especially sizeable in the case of EU imports. The ratio of potential to actual trade between EU and MED countries is between 3.5 and 5, a factor of almost 2 with respect to that with CEECs, which is between 2.0 and 2.6. Moreover, while the gap was recently narrowing in EU exports and imports with CEECs, approaching 2 in some cases (the Czech Republic, Hungary and Romania), the gap with MED economies is highly dispersed and rising. While this finding could be expected for the MEDs, it contrasts with previous results relative to trade between EU and CEECs, suggesting that trade potential was almost fully exploited already in 1992. The differences might depend on the fact that: a) this study focuses on intra-EU trade, rather than on trade with a larger, but less homogeneous group of countries, as a reference to estimate the elasticity of trade determinants with MED and CEECs; b) it applies an out-of-sample method to compute potential trade and, therefore, does not constrain the estimates of the ratio of potential to actual trade to average one; c) it concentrates on a later period, 1995-‘02, when GDP in most CEECs was rapidly increasing. The paper is organised as follows. Section 2 surveys recent relevant literature. Section 3 presents the econometric modelling. Section 4 describes statistically the recent evolution of trade patterns and provides the estimates of the ratio between potential and actual trade for MED and CEECs and of the time necessary for convergence. The concluding remarks summarise the main findings and discuss some policy implications. 2. The state of the art 2.1. The recent evolution of trade patterns Over the first ten years of the Barcelona partnership, the participation of the south Mediterranean countries to international trade has certainly increased. The annual rate of growth has increased from 4.9 in 1995 to 6.1% in 2003, but it is still lower than that of other less developed countries (6.4%) (IMF, 2004). According to the estimates of the World Trade Organisation relative to 2003, the share of the MED on world merchandise exports is still negligible, at around 2.3%, increasing by only fifteen percent, from the 2.0% in 1990. Two thirds of it comes from Turkey, Israel and Algeria. Import shares increased by the same amount reaching around 2.8% in 2003 from 2.5% in 1990. The progress is slow if compared to that achieved by the new EU member states of CEE, whose Published by The Berkeley Electronic Press, 2009 3 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 share in world trade has increased by more than one third, from 1.7 in 1995 to 2.8% in 2003. Although MED economies are a natural outlet for European trade and capital flows, the current trade with the EU is quite static and asymmetric. They keep today the same share of trade as in the early nineties, about 8% of total extra-EU trade (around 9% for exports and 7% for imports). Conversely, the economies of Eastern Europe have increased the corresponding shares from 7.5 to around 12% over the same period. The EU accounts for about 45% of total MED foreign trade (for the Maghreb area the corresponding figure is 70% for exports and 60% for imports) (Eurostat, 2004; Melad, 2008; Romagnoli and Mengoni, 2009). Another feature of MED trade relationships is the high level of concentration on both sides. The countries with higher trade intensity are Malta, Turkey and the Maghreb area4. Among EU exporters to MED the most important trade partners, namely France, Germany, Italy, Spain and the UK, together represent more than 70% of total EU-MED exports and imports thanks to historical and geographical links, such as, for some of them, the access to the Mediterranean Sea. Also the growth of trade volumes is distributed asymmetrically across countries and shows large differences: during the period 1996-’02, for instance, EU exports scored a high increase in Turkey (81%), and Algeria (71%), but only a small one in Lebanon (20%). The increase in EU imports was in general much lower and took place mostly with Turkey (+138%). EU imports from Cyprus even decreased (Eurostat, 2003). Among EU partners, Spain is the one which has benefited the most from trade intensification, with an increase in exports equal to 92% (against 60% for France and the UK, 46% for Italy and 42% for Germany). France, Germany and Italy have increased their imports from the MED economies by 12% and Spain by 7%. Overall, from 1995 to 2002, MED exports to the EU have increased by a factor of around 1.4, a much slower rate than other regions, most notably the CEECs, whose exports to the EU grew by a factor of 5 and more during the same period. 2.2. A survey of the literature Gravity models have been frequently used over the past decade to evaluate integration of CEECs into the EU (in particular for the 10 new EU members), the impact this might have in terms of trade potential, and the ratio with the actual level of trade. The results of these studies are mixed, depending on the period 4 This study considers the Maghreb countries of Morocco, Algeria and Tunisia (but not of Libya and Mauritania) and the Mashrek countries of Lebanon, Syria, Jordan and Egypt. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 4 Pastore et al.: A Tale of Parallel Integration Processes considered, the econometric specification, the estimator and the computation method used to calculate trade potential. The literature of the early 1990s predicted a sizeable increase of trade exchanges between the EU and the CEECs (Baldwin, 1994). Wang and Winters (1992) found that East-West trade was only a fraction of what it would have been in an integrated Europe. Hamilton and Winters (1992) predicted an increase in trade with Western Europe by up to five times. Later studies suggested that the actual integration between Eastern and Western countries was above the potential already in the early 1990s (Gros and Gonciarz, 1996; Brenton and Di Mauro, 1999; Nilsson, 2000). For instance, Gros and Gonciarz (1996) corrected Baldwin’s (1994) estimates on the grounds that the latter used a GDP which was overvalued because it was calculated on pretransition data. Combining the parameters from Baldwin (1994) with the 1992 data, Gros and Gonciarz ended up with a downward revision of Baldwin’s projections. Using the same procedure as Baldwin on data relative to the years 1995-’96, when GDP had overcome the negative pick of the J-curve of transition, Nillson (2000) found some unused trade potential even if the gap between potential and actual trade appeared greatly reduced (just 1.1 on average) as compared to Baldwin’s, and several EU countries (except Austria, Portugal, France, Ireland, Spain and the UK) exported more to the candidate countries than the gravity model would predict. Also the average ratio relative to exports to the EU was 0.9 (exceeded export potential) and only for Cyprus, the Czech and the Slovak Republics it was above one. Turning to the literature on MED, the few studies available invariably suggest that both MED trade with most non-MED countries and intra-MED trade are low in relation to what would be predicted on the basis of a gravity model. AlAtrash and Yousef (2000)5 find that, in recent years, despite EU, Gulf Cooperation Council and Arab Maghreb Union trading arrangements, trade has not increased, but rather decreased. Moreover, Mashreq countries exhibit higher levels of integration both within the area and with the rest of the world as compared to Maghreb and Gulf countries. 5 They apply cross-section maximum likelihood Tobit estimates of bilateral trade between 18 Middle East and North African countries (MENA) and 43 other countries (over 90% of total MENA trade) using 1995-1997 average data for trade (IMF-DOTS), total GDP and per capita GDP (World Economic Outlook), free trade agreements (ASEAN, EU, the GCC, the AMU), common border, common language, openness and a set of dummies for groups of MENA countries. Published by The Berkeley Electronic Press, 2009 5 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 IMF (2002, p. 118) contrasts the trade potential of MED economies, Asia, Sub-Saharan Africa, South America and Central America6 and finds that the MED region exhibits the highest degree of under-trading, after South East Asian countries. However, South East Asian countries have a sizeable volume of trade in services, which is not included in the gravity estimates. A more recent study by Miniesy et al. (2004) finds that the MED region is an “underachiever”, especially considering trade with the EU and with Eastern Europe. MED trade with the EU was much more developed back in the ‘70s and in the ‘80s up to 1984. Nonetheless, MED economies have cumulated a wide and increasing gap with respect to their potential in the following period. Intra-regional integration is also very poor: the Arab common market, the Gulf Cooperation Council and the three dummy variables for sub-regional trading arrangements within MENA have virtually no effect on the parameters and have a negative and significant coefficient. There are strong "underachievers" in intra-MENA trade like Algeria, Egypt, Kuwait, Qatar, Sudan, Syria, all oil exporting countries, while Jordan, Morocco, Oman, the UAE, Turkey are overachievers. Similar studies confirm these results using more recent data (see, among others, Fazio, 2006; Taleb and Younes, 2008; Miniesy and Nugent, 2008; Boughanmi, 2008, and the references therein). 3. The gravity equation framework 3.1. Specification In what follows, we estimate the parameters of a gravity model of intra-EU trade7 and then plug them into equations of trade between eleven MED8 and ten CEECs, respectively, and their main European trade partners (Italy, Germany, France, UK, and Spain) to calculate potential trade relative to the period 1995-’02. This potential or “normal” trade is then compared to actual trade volumes to assess the 6 This study calculates the potential of MENA, Asia, Sub-Saharan Africa, South America and Central America using the parameter estimates from the pooled gravity model of world trade averaged for 1995-99. 7 The analysis includes 13 EU members, because of missing data for Belgium and Luxembourg for some years. 8 In the Eurostat definition, MED includes 12 countries of the Euro-Mediterranean Partnerships: Algeria, Cyprus, Egypt, Jordan, Israel, Lebanon, Malta, Morocco, Syria, Tunisia, Turkey, Palestina and Gaza which is represented by the Palestinian Authority. Libya has the status of observer within the Euro-Mediterranean process. In the empirical analysis, due to the lack of data, we do not include Palestina and Gaza. As far as ex-COMECON countries are concerned, this study considers Poland, Hungary, Czech Republic, Slovak Republic, Slovenia, Estonia, Lithuania, Latvia, Bulgaria and Romania. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 6 Pastore et al.: A Tale of Parallel Integration Processes dimension of trade potential not exploited in the short run. The period considered includes the years of the output recovery after the transitional shock, before EU accession. The gravity model has been frequently adopted to analyse trade flows and has proved successful in a variety of contexts. The general idea behind it comes from the gravity theory in physics, from which it derives its name. In the context of international trade, the physical bodies are the exporting and importing countries, and the sizes of their economies are their masses. The larger the economies of the countries involved, the larger the volume of their trade. However, distance causes a resistance to trade, because of transportation and delivery costs, as well as of time lags between order and delivery, among other reasons. Additional trade resistance factors are import tariffs, border controls, quantitative restrictions, which represent indeed artificial trade costs (see, for surveys of the literature, Greenaway and Milner, 2002; Anderson and van Wincoop, 2003; Carrère, 2006). The gravity model is particularly suited to analyze the “trade gap” as it provides a sort of “natural benchmark”, taking into account both trade enhancing elements, such as market size (population) and factor endowment (per capita GDP), and trade resistance factors, such as distance. The use of panel data allows us to exploit the additional information that the data may convey introducing time invariant parameters and dummy coefficients able to capture country specific features, including historical and cultural linkages (gravity model with fixed effects). The following specification was preferred as it produced the lowest value of the variance inflationary test (VIF) for multicollinearity among a number of alternative cross-section specifications: X ijt = α ij + β 1 POP it + β 2 GDPPC + β 6 CLB ij + β 7 CL ij + ε ijt it + β 3 POP jt + β 4 GDPPC jt + β 5 D ij (1) where: i are the countries of origin, j are the destination countries and t = 1995-2002 is the period under examination; Xij are exports of country i to country j in real terms; αιj is the bilateral constant; POPit and POPjt denote the population at time t of country i and j respectively; GDPPCit and GDPPCjt are per capita GDP of country i and j at time t in real terms; Dij is the (natural log of) geographical distance in Km between the capital city of country i and of country j; CLB is a dummy equal to one, if the two countries share a common land border, and zero otherwise. In the computation of trade potential, CLB takes a value of one also if the countries involved have ex-colony links, and zero otherwise.; CL is a dummy taking a value of one if the trade partners speak the same language, and Published by The Berkeley Electronic Press, 2009 7 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 zero otherwise; finally, ε ijt ∼ IID(0, σ ε2 ). Except for the dummies, all the variables are in natural logarithms and, therefore, the estimated parameters can be interpreted as elasticities. Data sources are given in Table A1 in the Appendix. For robustness checks, total GDP was used instead of population as a measure of the mass of the partners. Total income and population were however not used together because of possible multicollinearity. By applying the specification in [1], we expect the coefficients of the population variables to be positive, as they catch the effects usually caught by absolute GDP. The estimates of trade potential were computed using several different specifications, but the main results did not change. The factors which are expected to affect positively bilateral exports are: a) the importer demand and exporter supply, proxied respectively by their population (POP) and per capita GDP (GDPPC). A higher per capita income – a proxy of the economic development – means a higher import demand and export supply. Trade grows more than proportionally as the economy gets richer: in fact, demand for variety grows with income and leads to a higher share of intraindustry trade in similar goods, since scale economies favour specialisation in differentiated goods (Helpmann and Krugman, 1985). The effect of population is more ambiguous: in fact, on the one hand, a larger population means a large domestic market, a higher degree of self-sufficiency and less need to trade. At the same time, however, a large population entails also a deeper division of labour and scale economies in production, which are generally associated with a larger need for trading. Therefore, the effects of this variable are ambiguous. b) dummies such as sharing a land or a sea border, ex-colony links, common language capture the geographical closeness, the better information, the lower cultural differences, the lower “home bias” and search and communication costs associated with proximity (familiarity with custom regime, institutions, legal systems). Conversely, distance is expected to affect negatively bilateral exports. Distance can be measured as geographical remoteness, but also as the surface of host markets, the level of trade costs, the presence of a “home bias” effect and of time and search costs. Like in several previous studies, this research relies on (natural logarithms of) straight-line distances between capital cities. Such measure does not take into account the level of development of infrastructures, but has the advantage of being transparent and comparable with previous studies. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 8 Pastore et al.: A Tale of Parallel Integration Processes 3.2. Panel estimators Table 1 reports the results of estimates of different panel models, namely the between effect (BE), the fixed effect (FE) and the random effect (RE) models9. Column (1), reports the coefficients of the BE model, which can be thought of as an OLS estimate of cross-section equations of average data relative to the years 1995-‘02. The Huber/White/sandwich estimator of variance is used to correct for heteroskedasticity. Column (2) reports the results of a FE, which is equivalent to a least square dummy variable (LSDV) model. In both cases, it is assumed that the error term, υ ijt , has a complex structure, since it includes a component, ε ijt , which is white noise and a component vijt , which is dependent on unobserved bilateral country-level effects: υ ijt = vi + ε ijt , where only ε ijt ∼ IID 0, σ ε2 . Equation (1) takes the following form: X ijt = Z ij β + υ ijt (2) ( ) where β represents the vector of coefficients and Z the matrix of independent variables as defined in (1). The aim of the FE and of the LSDV is to cancel out the time-invariant effects. While the LSDV reaches this result by including different intercepts for each yearly bilateral trade relation, the FE, instead, takes the distance of every variable from its specific mean for every bilateral trade relation: (X ijt − X ij ) = (Z ijt − Z ij )β + (ε ijt − ε ij ) (3) This model is sometimes called a within group estimator since it uses only the variation within a set of observations representing specific bilateral trade relations. The unwanted consequence of the FE model is that all the timeinvariant terms, including the dummy for common land border, common language and distance, are dropped and caught by the bilateral constant term10. 9 One of the problems with the FE and the RE models is that they do not allow controlling for the possible autocorrelation of residuals. To check for this, we also estimated the FE and the RE with an AR(1) disturbance. The results, not reported here for reasons of space, are not significantly different from the one shown in the table and are available on request. 10 The estimator removes the fixed effects by subtracting the time average for each variable for each country and then estimating by ordinary least square the coefficients of these transformed variables. Published by The Berkeley Electronic Press, 2009 9 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 Table 1: Gravity model estimates of intra-EU trade (1995-2002) BEM (1) Const Lgdppci Lgdppcj Lpopi Lpopj Ldist Clg Clb Sd(ui+ei) Corr(ui,XB) σu σe ρ (fraction of σ2 due to ui) F (all ui=0) RE of ui θ R2-within R2-between R2-overall F (all coeffs) Hausman test Breusch and Pagan Lagrange Multiplier test for RE Nobs N. of groups Obs per group Notes: A FEM (2) -6.75 0.68*** 1.59*** 2.01*** -0.02 *** REM (3) -1.36 1.15*** 1.10*** 0.84*** 0.91*** -0.64*** 0.26 0.17 0 0.56 0.13 0.95 Gaussian 0.91 0.63 0.87 0.86 2635*** 76.9*** 2846*** FEM (4) - 0.63 0.37*** 1.27*** 2.03*** 0.00 REM (5) 3.88** 0.53*** 1.06*** 1.01*** 1.04*** -0.61*** 0.08 0.47*** 0 0.36 0.13 0.89 Gaussian 0.87 0.67 0.95 0.95 4752.9*** 4.85 2359.2*** 6.00 1.27*** 0.41*** 0.82*** 0.88*** -0.78*** 0.24 0.15 0.56 *** -0.68 1.88 0.13 1.00 124.0*** -0.69 1.88 0.13 1 50.1*** 0.59 0.89 0.88 166.6*** 0.64 0.31 0.31 414.3*** 0.67 0.30 0.30 187.7*** 1085 155 7 1085 155 7 1085 155 7 1085 155 7 1085 155 7 Column (1) reports the coefficients of the between-effect model. The Huber/White/sandwich estimator of variance is used to correct for heteroskedasticity. The columns (2) and (4) report the results of fixed-effect models, of which the latter includes dummies for years. The columns (3) and (5) report estimates of random-effects models, of which the latter includes dummies for years, exporting and importing countries. B *** denotes a 1% significance level; ** denotes a 5% significance level and * denotes a 10% significance level. No stars means not significant. C The R2 have a panel meaning and cannot be interpreted exactly like the OLS R2. Two properties of this last are missing in the panel R2. This last does not represent the squared correlation between the fitted and actual values and is not, therefore, equal to the fraction of the variation in the dependent variable explained by the variation of the fitted value. In other words, they do not denote the fraction of variance of the dependent variable explained by the variance of the estimated variable. The R2-within, the R2-between and the R2-overall represent the within, between and overall interpretation of the estimated model. D Notice that in the REM the Corr(ui,XB) is assumed to be zero. E The Breusch and Pagan test for random effects is a Lagrange Multiplier test on the H0: σ 2 = 0 . Rejecting this ui hypothesis suggests that the REM model is appropriate. F H0 for the Hausman test is that all the common coefficients between the FEM and the REM are not statistically different. G θ represents the estimated value in the following REM: yit − ϑyi = 1 − ϑ α + xit − ϑxi β + 1 − ϑ vi + ε it − ϑε i , where xit is the matrix of independent ( ) ( ) ( ) {( ) ( )} variables and vi are the time invariant fixed effects. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 10 Pastore et al.: A Tale of Parallel Integration Processes Finally, column (3) reports estimates of a RE model of this type: (X ijt − ϑX ij ) = (1 − ϑ )α + (Z ijt − ϑZ ij )β + {(1 − ϑ )vij + (ε ijt − ϑε ij )} (4) This model differs from the FE model in as much as it assumes that the error term is correlated with random, not with fixed effects. In other words, the bias would be due to time-varying factors, which are omitted in a normal crosssection estimate. 3.3. Results The between estimator does not take into account time-varying factors (for instance, by way of year dummy variables), but provides a term of comparison for significance levels. This leaves the choice between the FE and the RE models. As shown in Table 1, a Hausman test rejects the hypothesis that the coefficients of the FE model in columns (2) and the RE model in column (3) are equal, which suggests relying on the more consistent FE estimates. In fact, the RE estimates are more efficient than the FE estimates, but they are inconsistent. On the other hand, though, the FE model is a within estimator and, therefore, does not control for time varying factors. Moreover, it does not explicitly provide estimates of either country effects or of other time-invariant factors, such as distance, common land border or common language. These are fixed effects and are therefore removed in estimates at the difference. To overcome these shortcomings of the FE model, and taking into account Màtyàs (1997) intuition that both time varying and time invariant effects should be taken into account in the gravity model, the FE and the RE models have been estimated also adding to the general specification a set of dummy variables for the years and one for the exporting and for the importing country. The columns (4) and (5) report the results. All the test statistics support the specification in model (5), namely a RE model with exporting and importing countries’ and years’ dummies. Again, the Breusch and Pagan Lagrange Multiplier test rejects the hypothesis of absence of random effects. However, this time, the Hausman test cannot reject the null hypothesis that the coefficients in column (5) are not statistically different from those in column (4)11. Inspection of the R2 also suggests this conclusion, with the R2-within of the RE model tending to that of the FE model; in addition, the 11 This result is partly due to the similarity of the coefficients for the year dummies, but also mirrors the tendency of the coefficients of the REM to become closer to those of the FEM, when including dummies for exporting and importing countries. These dummies seem to be able to clean the estimates of most of the fixed effects. Published by The Berkeley Electronic Press, 2009 11 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 overall R2 increases. Therefore, the tests concurrently support the view that the RE can be safely chosen instead of the FE model. An advantage of model (5) is that it may control for the presence of country (fixed) effects, like a FEM, through the inclusion of country dummies, while maintaining also the advantages of a REM, namely the higher degree of efficiency and estimated coefficients of the variables CL, CLB and distance. In other words, the model in column (5) has consistent coefficients like that in column (4), while being more efficient. The biggest difference when comparing the models (4) and (5) comes from the coefficient of the population variables, which, in the FE estimate, is not statistically significant. Most likely, this can be attributed to the role of the time invariant factors. Removing such factors might make the population variable statistically not significant. In the preferred specification, all the variables have the expected sign and are highly significant. The explanatory power of the model is also high. The statistical significance of distance in explaining the trade intensity of EU countries suggests that transport costs still have an important impact on the export performance. The estimated parameter for distance is stable across different specifications and is in line with that typically found in earlier studies (– 0.7). Also the numerical values of the coefficients are reasonable. In line with previous studies, the coefficient for the per capita GDP of the exporting country is lower than that of the importing country. This is to be expected, considering that exports are more related to the income level of the importing, rather than of the exporting country. The coefficient of the per capita GDP of the importing country is slightly greater than 1, which is what one would expect if the propensity to import from EU countries was constant over time and the share of EU over total imports was increasing. 4. Estimates of trade potential 4.1. Methodological issues The next step consists of applying the estimated coefficients of the gravity equation in column (5),12 relative to intra-UE trade, to the same specification for EU-MED (CEECs) trade. These parameters are used as a “benchmark” to 12 The same calculations have been implemented using the coefficients of the FE model in column (2) and in column (4). The results, which are not reported for sake of shortness, but are available on request, are very similar to those reported in the paper: they slightly differ only in terms of level, but not in terms of trends and ranking of the countries level of integration. More specifically, the estimates of the trade potential based on model (2) tend to generate values that are between 10 and 20% higher, perhaps because of the fact that the model presented here allows for more accurate measures of such variables as, among others, distance. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 12 Pastore et al.: A Tale of Parallel Integration Processes estimate the potential integration that MED countries might obtain if the elasticities of trade determinants were the same as those observed in the case of intra-EU trade. These trade volumes are considered “normal” flows to which actual flows are compared. The same procedure has been applied also to EU-CEE trade. This out-of-sample projection approach, namely the projection of trade relations for groups of countries different with respect to those for which the parameters have been estimated, provides an interesting and novel testing ground to compare the different speeds of integration for individual countries in each area and the relative success of the EAs versus the MAAs. As Papazoglou, Pentecost and Marques (2006, p. 1088) note in a similar exercise, the out-of-sample projection approach proposed here might imply higher values of trade potential than those based on estimates on a sample including also other third countries due to the high degree of economic integration and similarity existing among EU members. However, such a choice has an important motivation as it is policy relevant. In fact, first of all, the declared objective of the EAs and the MAAs was to lead the third countries involved to a close integration with the EU. Second, in addition to generating absolute measures of trade potential, this study provides also an estimate of the relative size of unexploited trade, by comparing the case of two groups of countries with similar level of development.13 An in-sample projection approach, namely the projection of trade relations for some of the countries included in the sample for which the parameters have been estimated, is simply unfeasible as consistent and efficient estimators should exhibit white-noise residuals and, therefore, on average the ratio between observed and fitted values of the regression should equal one. In other words, in-sample methods do not allow disentangling whether the results depend on untapped trade or on a wrong specification of the model (Egger, 2002). 14 A third alternative would be to use a larger sample of countries and compare the coefficients of dummy variables. Trade potentials can be calculated with the use of dummy variables to proxy for trade creation and trade diversion 13 It is useful to remark that a shared view in the literature is that, no matter which methodology is adopted, while gravity models have strong power in explaining trade patterns between countries, they are less reliable as an instrument to obtain punctual estimates of the trade level, because of the sensitivity of the estimated coefficients to valuation errors. Most of all, the estimation accuracy of the constant term can have a strong influence on the predicted level of trade flows. Therefore, the prediction accuracy for trade flows in absolute terms is rather low and we should take the following calculation as an indication of the presence of negative or positive gaps of certain intensity but not as a precise measure of them. 14 In-sample methods can still be used when the model is well specified with the use of appropriate econometric tools. For a more in-depth analysis of this issue, see, among others, De Benedictis and Vicarelli (2005). Published by The Berkeley Electronic Press, 2009 13 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 effects (see, as examples of this last methodology, De Benedictis and Vicarelli, 2005; De Benedictis, De Santis and Vicarelli, 2005; Carrère, 2006; Miniesy and Nugent, 2008; Boughanmi, 2008 and the references therein). However, also this choice would be exposed to the same criticisms as the out-of-sample methods. Pooling data for industrial and developing countries might also create heterogeneity problems, sub-sample instability and heteroskedasticity. In fact, first, developed countries are more affected than less developed countries by sudden shifts and distortions in trade policy; second, transaction costs are different in countries with more articulated markets as compared to countries with less articulated ones (Egger and Pfaffermayr, 2003; Màrquez-Ramos and Martinez-Zarzoso, 2005). More importantly from the point of view of this paper is the fact that the findings of such a third type of exercise would measure the degree of integration of the economies considered in world trade, not in EU markets. However, this is not the declared objective of the agreements signed by the EU with the countries considered. 4.2. Results Figures 1 through 10 in the Appendix depict the trends in the trade ratio of potential and actual trade between each of five EU main trade partners (France, Germany, Italy, Spain and UK) and each country in the two areas (MEDs and CEECs) over the period 1995-02. A ratio of one suggests that potential trade equals actual trade. The higher is the ratio, the higher is the gap that has to be filled and therefore the possibility to create new trade. A decreasing (increasing) trend of this ratio over time suggests that trade is increasing (decreasing) and tends to approach its potential level. The ratio between potential and actual trade with the MED countries generally shows a constant or, sometimes, an increasing trend. The main exceptions are Tunisia and Turkey (where it is weakly decreasing). This suggests that actual exports between each EU partner and the MED economies has further increased the distance with respect to the “normal” level that they might have reached given the economic, cultural and geographic conditions. The trade potential between EU and MED area is far from being exploited in 2002: the ratio is much higher than 1, ranging between a minimum of around 3.5 and a maximum of 5 for both exports and imports. In other words, the actual volume of trade is only from 20 to less than 30 percent of the potential level. Conversely, the CEECs show a trend marked by a large decline of the ratio. Not only they start from a ratio much lower than the MED economies (around 2), they also further close the gap. This is especially the case of countries starting from the worst positions, such as the Baltic Republics and Bulgaria, http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 14 Pastore et al.: A Tale of Parallel Integration Processes which show the most dynamic trend. The trade potential between EU and CEECs is close to be exploited in 2002, with a ratio ranging between 2 and 2.6 The projected/actual ratio of imports systematically exceeds that of exports. This result is consistent with the aforementioned evidence of EU trade surpluses with the CEE and MED areas, which are partly due to the low growth rate of EU members. It indicates that there is wide scope for an increase in imports more than in exports if the EU economies were to start a new era of economic growth. Moreover, it might also suggest that MED firms still face many barriers to entry in EU markets, or, in other words, that the CEECs have not benefited of a total and preferential opening for their exports to the EU within the EAs. Consider now the trade relations of individual EU partners with MED countries, first, and then with CEECs. France is less distant from its potential with Algeria, Malta, Morocco and Tunisia than with Mashrek countries. Germany has developed trade relations especially with Turkey. Spanish trade is less far from its potential with Tunisia, Turkey and Morocco. The UK exploits its trade potential more intensely with Israel, Malta, Cyprus and Turkey. Finally, compared to the other four partners, Italy has a trade structure spread more evenly across all MED economies. Generally speaking, all the MED countries are in a similar position, although Malta, Israel, Tunisia, and Algeria show a greater degree of integration in both imports and exports (ratio around 4). The degree of underachievement is higher with Cyprus, Jordan and Lebanon, especially as far as imports are concerned. Trade relations with CEECs are by far more developed. French trade with Slovenia, Poland and Romania is closer to its potential than that with other countries, especially the Baltic Republics. Nonetheless, a clear process of convergence seems to be in place. Italy’s case is similar to that of France with the addition of Bulgaria among the favourite partners. Germany is slightly different, as it shows a greater degree of integration with Central European countries, especially the Czech Republic, Hungary and Poland, but in addition to them also with Slovenia. With some of these countries, the ratio approaches 2 in recent years. For Spain, trade seems to be closer to exhaustion especially with Hungary, the Czech Republic and Slovenia (for exports). The ratio is generally higher for the UK, especially with Bulgaria and Estonia. As expected, overall the estimates provide high values of trade potential with both groups of EU trade partners. Such values are realistic considering that the EU trade partners have weak economic structures and therefore the most important component of intra-EU trade, namely intra-industry trade, is still very low. Published by The Berkeley Electronic Press, 2009 15 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 Concluding remarks The EU is placing remarkable political emphasis on the integration process with the Mediterranean countries still involved in the 1995 Barcelona process (MED9, after Malta, Cyprus accession and Turkey accession process started). However, trade integration with these countries seems to be still underdeveloped. Conversely, many observers assume that a significant level of trade integration with the new EU members has been already achieved. To contrast the success of EU integration with these two areas, this paper computes the trade potential of a sample of old EU members (Italy, Germany, France, UK, Spain) from 1995 to 2002 using an “out-of-sample” methodology. The coefficients are taken from different panel estimators of the gravity equation relative to intra-EU trade. The results, which are remarkably stable across different specifications and estimation methods, suggest the existence of important unexploited trade potential with both groups of partners, although the ratio of potential to actual trade with the MED countries is much larger and dispersed compared to that with the CEECs and, unlike these countries which are rapidly converging to their potential, this gap is quite stable. While our results converge with previous studies which also found a large “under trading” between EU and MEDs (Rose, 2002; IMF, 2002) they contrast the existing literature in the case of the CEECs (but are in line with Baldwin, 1994 and 1997). The differences from previous gravity studies depend on the period of analysis, on the estimation method and the procedure used to compute potential trade. Previous research almost universally deployed “in sample” methods on early 1990s data, when, as Baldwin (1997) notes, GDP in the CEECs was depressed by the transition phase. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 16 Pastore et al.: A Tale of Parallel Integration Processes APPENDIX Figures 1 France-CEECs: Ratio of potential to actual trade (1995-'02) Bulgaria 2.2 2.4 2.6 2.8 Estonia Hungary Latvia Potential trade / Actual trade Lithuania 2.2 2.4 2.6 2.8 Poland Romania Slovenia 1996 1998 2000 2002 1996 1998 2000 2002 The Czech Rep. 2.2 2.4 2.6 2.8 The Slovak Rep. 1996 1998 2000 2002 1996 1998 2000 2002 France Export Graphs by cee1 Import Figures 2 France-MED: Ratio of potential to actual trade (1995-'02) Algeria 3.5 4 4.5 5 Cisjordan Cyprus Egypt Potential trade / Actual trade Israel 3.5 4 4.5 5 Lebanon Malta Morocco 1996 1998 2000 2002 Syria 3.5 4 4.5 5 Tunisia Turkey 1996 1998 2000 2002 1996 1998 2000 2002 1996 1998 2000 2002 France Export Graphs by med1 Import Published by The Berkeley Electronic Press, 2009 17 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 Figures 3 Germany -CEECs: Ratio of potential to actual trade (1995-'02) Bulgaria 2.1 2.2 2.3 2.4 2.5 Estonia Hungary Latvia Potential trade / Actual trade Lithuania 2.1 2.2 2.3 2.4 2.5 Poland Romania Slovenia 1996 1998 2000 2002 1996 1998 2000 2002 The Czech Rep. 2.1 2.2 2.3 2.4 2.5 The Slovak Rep. 1996 1998 2000 2002 1996 1998 2000 2002 Germany Export Graphs by cee1 Import Figures 4 Germany -MED: Ratio of pot ent ial to actual trade (1995-'02) Algeria 3.5 4 4.5 5 Cisjordan Cyprus Egypt Potential trade / Actual trade Israel 3.5 4 4.5 5 Lebanon Malta M orocco 1996 1998 2000 2002 Syria 3.5 4 4.5 5 Tunisi a Turkey 1996 1998 2000 2002 1996 1998 2000 2002 1996 1998 2000 2002 Germany Export Graphs by med1 Import http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 18 Pastore et al.: A Tale of Parallel Integration Processes Figures 5 Italy -CEECs : Ratio of potential to actual trade (1995-'02) 2.2 2.4 2.6 2.8 Bulgaria Estoni a Hungary Latvia Potential trade / Actual trade 2.2 2.4 2.6 2.8 Lithuania Poland Romania Slovenia 1996 2.2 2.4 2.6 2.8 1998 2000 2002 1996 1998 2000 2002 The Czech Rep. The Slovak Rep. 1996 1998 2000 2002 1996 1998 2000 2002 Italy Export Graphs by cee1 Import Figures 6 Italy -MED: Ratio of potential to actual trade (1995-'02) Algeria 3.5 4 4.5 5 Cisjordan Cyprus Egypt Potential trade / Actual trade Israel 3.5 4 4.5 5 Lebanon Malta Morocco 1996 1998 2000 2002 Syria 3.5 4 4.5 5 Tunisia Turkey 1996 1998 2000 2002 1996 1998 2000 2002 1996 1998 2000 2002 Italy Export Graphs by m ed1 Import Published by The Berkeley Electronic Press, 2009 19 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 Figures 7 Spain-CEECs: Ratio of potential to ac tual trade (1995-'02) Bul garia 2.2 2.4 2.6 2.8 Estonia Hungary Latvia Potential trade / Actual trade Lithuania 2.2 2.4 2.6 2.8 Pol and Romania Slovenia 1996 1998 2000 2002 1996 1998 2000 2002 The Czech Rep. 2.2 2.4 2.6 2.8 The Slovak Rep. 1996 1998 2000 2002 1996 1998 2000 2002 Spain Export Graphs by cee1 Import Figures 8 Spain-MED: Ratio of pot ential to actual trade (1995-'02) Al geria 4 4.5 5 Cisjordan Cyprus Egypt P te tia tra e/ A a tra e o n l d ctu l d 3.5 Israel 4.5 5 Lebanon M alta M orocco 3.5 4 1996 1998 2000 2002 Syria 4 4.5 5 Tunisia Turkey 3.5 1996 1998 2000 2002 1996 1998 2000 2002 1996 1998 2000 2002 Spain E xport Graphs by med1 Import http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 20 Pastore et al.: A Tale of Parallel Integration Processes Figures 9 UK-CEECs: Ratio of potential to actual t rade (1995-'02) 2.2 2.4 2.6 2.8 Bul garia Estonia Hungary Latvia Potential trade / Actual trade 2.2 2.4 2.6 2.8 Lithuania Pol and Romania Slovenia 1996 1998 2000 2002 1996 1998 2000 2002 The Czech Rep. 2.2 2.4 2.6 2.8 The Slovak Rep. 1996 1998 2000 2002 1996 1998 2000 2002 UK Export Graphs by cee1 Import Figures 10 UK-MED: Ratio of potential to actual trade (1995-'02) Algeria 3.5 4 4.5 5 Cisjordan Cyprus Egypt Potential trade / Actual trade Israel 3.5 4 4.5 5 Lebanon M alta M orocco 1996 1998 2000 2002 Syria 3.5 4 4.5 5 Tunisi a Turkey 1996 1998 2000 2002 1996 1998 2000 2002 1996 1998 2000 2002 UK Export Graphs by med1 Import Published by The Berkeley Electronic Press, 2009 21 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 Table A1. Data source for the empirical analysis Variables Source Bilateral export flows (current Direction of Trade Statistics, price, US$ millions) International Monetary Fund, 2003 World Bank, World Development Per capita GDP (US$ at 1995 Indicators 2003 constant prices and exchange rates ) Population (in millions) World Bank, World Development Indicators 2003 Distance (straight line distance www.wcrl.usda.gov/cec/Java in KM between capital cities) export (and import) unit values International Financial Statistics, International Monetary Fund, 2003 Sample period 1995-2002 1995-2002 1995-2002 1995-2002 1995-2002 References Al-Atrash, Hassan M., Yousef, Tarik, 2000. Intra-Arab trade: is it Too Little? International Monetary Fund Working Papers, 00/10. Anderson, James and Eric van Wincoop, 2003. Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review 93(1): 170-192. Baldwin, Richard E., 1994. Towards an Integrated Europe. London: Centre for Economic Policy Research. Baldwin, Richard E., 1997. Comment on Gros and Gonciarz. European Journal of Political Economy 13(1): 187-188. Brenton, Paul and Di Mauro, Francesca, 1999. The Potential Magnitude and Impact of FDI Flows to CEECs. Journal of Economic Integration 14(1): 5974. Boughanmi, Houcine, 2008. The Trade Potential of the Arab Gulf Cooperation Countries (GCC): A Gravity Model. Journal of Economic Integration 23(1): 42-56. Carrère, Cèline, 2006. Revisiting the Effects of Regional Trade Agreements on Trade Flows with Proper Specification of the Gravity Model. European Economic Review 50(2): 223-247. De Benedictis, Luca, Roberta De Santis and Claudio Vicarelli, 2005. Hub-andSpoke or else? Free Trade Agreements in the “Enlargemed” European Union. European Journal of Comparative Economics 2(2): 245-260. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 22 Pastore et al.: A Tale of Parallel Integration Processes De Benedictis, Luca and Claudio Vicarelli, 2005. Trade Potentials in Gravity Panel Data Models. Topics in Economic Analysis and Policy 5(1): 1-31. Egger, Peter, 2000. A Note on the Proper Econometric Specification of the Gravity Equation. Economics Letters 66(1): 25-31. Egger, Peter, 2002. An Econometric View on Estimation of Gravity Models and the Calculation of Trade Potential. World Economy 25(2): 297-312. Egger, Peter and Michael Pfaffermayr 2003. The Proper Panel Econometric Specification of the Gravity Equation: A Three-Way Model with Bilateral Interaction Effects. Empirical Economics, 28: 571-580. Eurostat, various years. Statistics in focus, Brussels. European Commission, 1997. “Agenda 2000”, Brussels. Fazio Giorgio, 2006. Euro-Mediterranean Economic Integration: An Empirical Investigation of Trade Flows, Ersa, Conference Paper n. 610. Femise, various years. Report on the Euro-Mediterranean Partnership. Marseille: Femise. Greenaway, David and Milner, Chris, 2002. Regionalism and Gravity. Scottish Journal of Political Economy 49(5): 574-585. Gros, Daniel and Gonciarz, Andrzej, 1996. A Note on the Trade Potential of Central and Eastern Europe. European Journal of Political Economy 12(4): 709-721. Hamilton, Carl B. and Winters, Alan L., 1992. Opening Up International Trade with Eastern Europe. Economic Policy 14: 77-116. Helpman, Elhanan and Krugman, Paul, 1985. Market Structure and Foreign Trade. Cambridge Mass: MIT Press. International Monetary Fund, various years. World Economic Outlook. Washington: IMF. Màrquez-Ramos, Laura and Immaculada Martinez-Zarzoso, 2005. Does Heterogeneity Matter in the Context of the Gravity Model. Economics Bulletin 6(17): 1-7. Matyas, Laszlo, 1997. Proper Econometric Specification of the Gravity Model. World Economy 20(3): 363-368. Melad, Khaled, 2008. The Euro-Mediterranean Trade Relations. MPRA discussion paper, n. 7085. Published by The Berkeley Electronic Press, 2009 23 Review of Middle East Economics and Finance, Vol. 5, Iss. 2 [2009], Art. 2 Miniesy, Rania S., Nugent, Jeffrey B. and Yousef, Tarik M., 2004. Intra-regional trade integration in the Middle East. Past performance and future potential. In Hassan, Hakimian, Nugent Jeffrey B. (eds.) Trade Policy and Economic Integration in the Middle East and North Africa. Economic boundaries in flux. London: Routledge. Miniesy, Rania and Jeffrey Nugent, 2008. Are there Shortfalls in MENA Trade? If so, to what can they be attributed? in Ferragina, A. (ed.), Bridging the Gap: The Role of Trade and FDI in the Mediterranean, CUEN, Napoli, 197-238. Nilsson, Lars, 2000. Trade integration and the EU economic membership criteria. European Journal of Political Economy 16(4): 807-27. Papazoglou Christos, Eric J. Pentecost and Helena Marques, 2006. A Gravity Model Forecast of the Potential Trade Effect of EU Enlargement: Lessons from 2004 and Path-Dependency in Integration. World Economy 29(8): 1077-1089. Romagnoli, Alessandro and Luisa Mengoni, 2009. The Challenge of Economic Integration in the MENA Region: from GAFTA and EU-MFTA to small Scale Arab Unions. Economic Change and Restructuring 42(1-2): 69-83. Rose, Andrew K., 2002. Estimating Protectionism from the Gravity Model. Mimeo. Washington: IMF. Taleb, W. and H. Younes 2008. Augmented gravity model: An empirical Application to Arab Mediterranean Countries European Union trade flows, in Ferragina, A. (ed.), Bridging the Gap: The Role of Trade and FDI in the Mediterranean, CUEN, Napoli, 239-275. Wang, Z. K. and Winters, L.A. 1992. The Trading Potential of Eastern Europe, Journal of Economic Integration, 7 (2): 113-136. http://www.bepress.com/rmeef/vol5/iss2/art2 DOI: 10.2202/1475-3693.1228 24