Joining the World Trade Organization: It s All About the Exports By: Christopher Balding University of California, Irvine August 31, 2005 The author would like to thank the insightful comments and assistance of Bernie Grofman and Priya Ranjan. Thanks to Steven Carter for all his patience, technical assistance, and suggestions. D.A. Elliott and Gloria Nash provided tremendous help with their research assistance and editorial comments. Finally, thanks to my wife, Daisy for putting up with me through the long hours crunching data and running regressions.
Abstract Recent research (Rose 2003) has called into question the impact of the World Trade Organization on trade. According to Rose, the WTO has no statistically significant impact on trade. The Rose research however, has been called into question on both modeling grounds for regressing against total trade rather than imports and failing to utilize fixed country effects. Subramanian and Wei (2003) find that when these factors are accounted for imports rise by economically and statistically significant amounts. This paper seeks to reconcile these works. Regressing against imports utilizing the panel OLS with fixed year method utilized by Rose and the fixed country effects of Subramanian and Wei, this research finds that the WTO has a larger impact on exports than imports. The results also indicate that the WTO frequently causes imports and exports to move in opposite directions negating any increase in overall trade. The panel data with fixed year effects regressions demonstrate surprising consistency with the fixed country effect regressions. The results presented here imply that countries may not be as interested in liberalizing trade as selling to the world. The results indicate that trade rises moderately between members but falls slightly when only one country of a trading pair is a member. 2
Introduction Everyone seems to have a love hate relationship with the World Trade Organization (WTO). Economists, politicians, and anti-globalization protesters all seem to find something to dislike about the WTO. To its supporters, the WTO has freed global trade by lowering tariffs and reducing the use of non-tariff barriers, ushering in unprecedented prosperity and growth. Only recently, however has research been conducted on the impact of the WTO on trade flows between countries (Rose 2003). Rose concludes, we currently do not have strong empirical evidence that the GATT/WTO has systematically played a strong role in encouraging trade. This seems at odds with popular wisdom. How could the institution responsible for liberalizing international trade arrive with more of a thud than a bang? Subramanian and Wei (SW) put forth an argument that Rose is incorrect on both modeling and methodological grounds. SW argues that the standard gravity model should be regressed against imports rather than the average value of real trade and should include fixed importer and exporter effects. SW concludes that, once these important changes are made, imports rise significantly in developed countries while increasing slightly in developing countries. Though they correctly diagnose part of the overall trade puzzle as unevenly felt across countries, they fail to account for the WTO s impact on exports. Tomz, Goldstein, and Rivers focus on WTO classifications used by Rose, specifically concerning developing countries, but this fails to solve the problem of how trade flows and how the WTO impacts imports and exports differently. Rose, however, in his rebuttals demonstrates not only is it not possible for both SW and TGR to be right, but 3
that these classification issues have little impact on overall trade levels. 1 Do we really know then, that the WTO increases overall trade? In this paper, I argue that Rose arrives at an insignificant finding for overall trade because the WTO impacts imports and exports in different ways for different types of members. Running panel data ordinary least squares (OLS) regressions used by Rose, it is clear that the WTO impacts imports and exports frequently in opposite ways. When utilizing importer and exporter effects, as specified by SW, the difference is less pronounced, but again imports and exports react to WTO membership differently. Highincome countries are the only income group to demonstrate a clear and unambiguous rise in both imports and exports across time, methodological specification, or changes to the data. The other income groups, supporting the SW findings, have either stagnant or declining levels of trade. The major finding of this paper and explanation to reconcile these conflicting findings is that the WTO impacts imports and exports in different ways causing the non-finding when regressing against overall trade and the significant finding when utilizing real imports. The results indicate that trade rises between members but falls when only one country of a trading pair is a member. The WTO and its Discontents The WTO is a controversial institution. The defender and promoter of free trade to its supporters, the WTO touts its own achievements in opening up markets and facilitating the unparalleled growth of economic interdependence. 2 A defender of the rich and privileged to its critics, the WTO tramples on local needs and fails to assist in 1 Please go to haas.berkley.edu/arose to download his rebuttals with supporting output and data. 2 Please go to www.wto.org and the 10 Benefits of the WTO Trading System. 4
economic development adopting the Washington consensus. Until recently however, little systematic research had been conducted to study the impact of the WTO. The research that did take place did not attempt to systematically determine the impact of the WTO on all its members and their trading patterns(bagwell and Staiger 1999, Rivera- Batiz and Xie 1992). Rose (2003) filled in this omission by producing a bilateral gravity trading model with dummy variables for whether one or both of the countries were members. Rose regressed total bilateral trade against the gravity model with variables for distance between trading partners, land area, income, and numerous dummy variables representing factors such as language and colonial history. Rose notes that his empirical strategy is to control for as many natural causes of trade as possible. After conducting a wide variety of sensitivity, robustness, and classification tests on the initial and derivative models, Rose concludes, there is little reason to believe that the GATT/WTO has had a dramatic effect on trade. This conclusion matters all the more because research indicates that trade is important to economic growth(lee, Ricci, Rigobon 2004). This conclusion then seems at odds with the popular wisdom of economists, politicians, and protesters. Subramanian and Wei (2003) produced a paper that on the face of their results seems to dispute the Rose conclusions. SW using a bilateral gravity trade model and the much of the Rose data, but make one important model change and one econometric change. First, SW argues that Rose fails to include country and year effects, which fundamentally change the results. Due to the nature of the model and data considerations, there does seem to be significant theoretical reasons to include importer 5
and exporter effects. 3 The important point, however, when comparing these two methods, is that clear trends reveal themselves in both methods. The results that seem most prone to significant changes between methods are those that are most fragile to alternative model specifications and most affected by data exclusions. High income countries display strong import and export growth across, time, model, and method. Second, SW argue that a gravity model is better understood and grounded in theory if regressed against the natural log of real imports rather than real average bilateral trade as with Rose. There are important theoretical reasons to agree that a gravity model is better understood when regressed against imports. The Rose method of regressing against total trade demonstrates that overall the WTO may not have the impact its supporters or critics contend. His method however, may obscure some important points about the flows of trade and types of trade because, as will be demonstrated, the WTO impact is different across income classification and types of trade. Rose even notes that others may wish to explore the impact of the WTO on imports and exports. The results demonstrate that the WTO does impact a country s exports. Rose and SW make important decisions regarding data construction. Rose creates a model with one trading relationship between two countries where country n has n-1 total trading relationships; country n-1 has n-2 trading relationships and so on. This method makes no distinction between the types of trade between the two countries and prevents a better understanding of trade flow between members. SW regress against imports of a specific country for all import relationships at time t. This differs from the Rose model in that all countries n, have an n-1 trading relationships, except where data issues preclude an observation. SW regress against imports, rather than total trade, 3 I will use the language country effects though this means controlling for importer and exporter effects. 6
allowing the isolation of importer specific variables. This is not an irrelevant difference in modeling. The data construction method of Rose raises two issues. First, it obscures differentiation in trade patterns. As will be seen, significant differences exist in member trade. Second, this prevents the isolation of importer and exporter specific variables. As will be demonstrated, importing members imported less from non-members but exported more to non-members. Though there is no theoretical reason why joining the WTO should increase exports, as SW note, there are many practical reasons why it may impact exports levels. Without presenting a complete analysis of this claim, there are many reasons to expect that joining the WTO will impact exports. First, countries join the WTO and negotiate accession based upon the knowledge of their comparative advantages and disadvantages. Countries pick winners or industries that they hope to protect but also try to gain access under most favored nation status for their competitive industries. If countries only relaxed import controls and did not gain greater market access, few would have an incentive to join. Countries that join actively tout the access to new markets they gain when joining the WTO. Second, countries work hard and file costly litigation to protect their rights to others markets. One of the major advantages of the GATT/WTO system is its dispute resolution mechanism, which permit aggrieved countries to file costly and time consuming litigation based upon substantial evidence that their products are not being accorded their rights under WTO law. Even after joining, countries seriously pursue their ability to export and conversely attempt to reduce imports. Third, joining the WTO frequently involves a costly restructuring of domestic economies. Whether this comes 7
through reduction in tariffs, ending of subsidies, or legal reform, joining the WTO frequently involves large and significant economic reform. This may mean a surge in imports and it may mean that competitive industries are both freed to pursue growth opportunities and obtain access to new markets. Fourth, high income members that are capital intensive currently block large amounts of products which lesser developed countries make or could make, agriculture the most obvious example. Conversely, lesser developed countries import capital intensive products under high tariff levels. The lesser developed countries face high exporting hurdles to developed countries while needing capital intensive products. This pattern comes through clearly as lesser developed country members have high positive import member coefficients and negative or insignificant export member coefficients. Though SW are correct in asserting that there are no theoretical reasons that WTO membership may impact exports, there are practical reasons to believe it may influence exports. OLS vs. Country One of the major differences between the Rose and SW research, is the importance of including country effects in the regression estimation. Research has indicated that the gravity model may over estimate some variables as a result when failing to control for importer and exporter effects. McCallum (1995), omitting fixed importer and exporter effects, found a 2,200% increase in intra-canadian trade due to the border with the United States. As others have demonstrated, and as these results will support, including country effects can change the outcome of results and provide a better estimation of the data (Anderson and van Wincoop 2003, Feenstra 2002, Egger 2000, 8
Egger 2002, Matyas 1997, Matyas 1998). More important, however, is the question about which method provides a better estimation of the data in question. In this specific instance, there are significant reasons to believe that utilizing country effects may better estimate the impact of the WTO on trade. First, as Rose notes, while he attempts to control for as many natural variables such as language, distance, and land area, he does not control for unnatural country specific variables such as policy and openness to trade. It is exceedingly difficult to measure openness and policy variables that impact trade, but there are undoubtedly a wide range of policy, time, and country specific environments that impact trade(rodriguez and Rodrik 1999, Anderson 1998, Pritchett 1996). These unnatural influences impact trade through variety of methods including trade costs and political instability that are specific to each country (Anderson and Marcouiller 1999, Anderson and van Wincoop 2004). Including importer and exporter effects will help control for these unnatural variables. Rose does control for year effects in his regressions, but fails to control for importer and exporter effects. Time controls are included for a 49 year study and it would therefore seem only reasonable to include importer and exporter controls for 177 countries. 4 Second, as Rose (2002) notes WTO members do not as a whole have a more liberal trade policy than non-members. This does not mean that no WTO member has a liberal trade policy, but rather taken as a whole, they do not have a more liberal trade policy than non-members. It would then make sense to include controls that hold constant changes in the specific importer and exporter environment. Importer and 4 It is worth noting that in his original paper and in subsequent output provided on his website Rose demonstrates that including fixed country effects does not change his results. My results are broadly supportive of his assertion that including fixed country effects makes minimal changes, as the results demonstrate. More important however, is the data construction method used. 9
exporter controls will increase the overall specificity of the results. Third, as numerous others have noted (Feenstra 2002, Anderson and Van Wincoop 2003), simply utilizing a gravity model may not correctly estimate key variables. Many possibilities have been proposed to correct for friction, remoteness, policy, country, and time. An international border unquestionably brings about additional variables that impact the flow of trade as demonstrated most notably by McCallum (1995). These other affects require controls if an accurate estimation is to be arrived at. Fourth, due to the nature of the research, a study with 177 countries across 49 years requires time and country controls. Neither the year nor countries are constant, either in natural or unnatural variables, and require additional controls to mitigate their impact. This does not mean the method used by Rose is without value, but rather implies that different countries and income groupings will react differently to WTO membership. The Three Questions To arrive at a better understanding of the impact of the WTO building upon the work of Rose and SW, I will focus on three areas. 1. How important is the modeling difference of regressing against imports rather than average real trade? 2. How important is the inclusion of fixed country effects to the results? 3. How does the WTO impact trade if at all? Based upon their divergent results, the differences between Rose and SW must stem from either the model, the method, or some other unaccounted issues or variables. The modeling differences does allow for some difference due to its inability to isolate certain 10
variables and its tendency to smooth others out. Many key results presented here, even when regressing against real imports, are strikingly close to results obtained by Rose. The methodological differences between panel OLS regressions with fixed year effects and panel OLS regressions with fixed year and country effects are smaller than might be expected. Most of the patterns are unchanged and changes between methods that do take place typically happen with insignificant or fragile results. The results here also support the theory that fixed country effects have a moderating tendency on gravity model estimations. Finally, this paper demonstrates that only studying imports in relation to the WTO, overlooks its significant impact on exports. The WTO has a greater impact on exports than imports. Data and Methodology The data comes from Rose (2003) downloaded via his website. 5 To study the impact of the WTO on trade, Rose uses a bilateral gravity model controlling for the natural determinants of trade. The gravity model has been used by a wide variety of authors to study a wide variety of trade issues (Feenstra, Markusen, and Rose 2000, Rose and Spiegel 2003, Anderson and van Wincoop 2003, 2004, Feenstra 2002, Glick and Rose 2001, Rose 2003, Rose 2004b, Frankel and Romer 1999). The STATA dataset covers 177 countries with controls for such natural variables as distance, GDP, and land area. 6 It also includes a comprehensive set of dummy variables that control for such variables as common language between the trading pair, colonial history, and geographic factors such land locked countries. This research will use the SW model of regressing 5 To download the data, paper drafts, and supporting output for STATA please go to haas.berkley.edu/arose. 11
against imports rather than average bilateral trade as Rose does. International Monetary Fund Direction of Trade data was extracted from the online database Webstract for the years 1950 to 1999. 7 The natural log of real imports for the importing country was arrived at by averaging the exports of country two with the imports of country one, deflating by the 1982-1984 Urban Consumer CPI, and taking the natural log. Variables were added using Rose data for import and export country WTO membership with further classification based upon income and regional dummy variables. Following the SW method, country effects will be used as a means of comparison to demonstrate that the fundamental conclusions will hold regardless of method. Other than the natural log of real imports and dummy variables added for importing and exporting country membership with income level classification, all variables in the data used come from or are based on the Rose dataset. The major methodological change from Rose comes in two areas. First, whereas Rose regresses against total trade, this study will regress against imports. This provides the basis to reconcile the Rose and SW findings. As will be demonstrated, the WTO s impact on imports and exports support Rose s conclusions that it has no impact on overall trade when only one country of a trading pair is a WTO member, but that trade rises between WTO members. Second, as mentioned previously, country one import data was averaged with country two export data and vice versa. This produced two numbers: average country one imports and exports or averaged country two exports and imports, 6 For a complete explanation of the Rose dataset please see Rose (2003). 7 Please note that in my data set due to direction of trade data, not all countries from the Rose data set have been included. For instance Bhutan, Namibia, and Swaziland were not included as there were not four trade numbers from which to arrive at an average of two import data statistics. 12
depending on the point of view. 8 The country two imports, or country one exports, were then inserted as the dependent variable and all necessary variables inverted, similar to the data set of SW. This change did not affect most of the bilateral variables such as distance, language, and border. If the countries shared a common language and border before they will still share a common language and border. It does however, necessitate an inversion of the country specific dummy variables such as WTO membership, income level, and regional classifier. This change has two major effects. First, it significantly enlarges the dataset. Rose has 234,597 observations of overall trade, this change creates a data set with 419,910 observations. Second, this permits an examination of the impact of the WTO on exporting country membership. Whereas with Rose, the United States and the United Kingdom had one relationship of overall trade and did not differentiate between exporter or importer, now there is an import and export relationship. This is not a minor point. Many trading relationships, especially ones involving lesser developed countries, have goods moving in one direction but not both. The Rose method of turning all trades into one trading relationship tends to smooth these numbers out and obscure offsetting trades. As will be seen, it is not uncommon for the WTO to impact member trade in offsetting ways. Also, it is worth noting that this data set does not exclude, except where noted, small observations of trade. Where real imports equaled zero, the natural log of one was used as the import value. In other words, many observations of real trade are zero or lower. 9 Though this may be a point of contention for some, this more accurately reflects actual 8 Though the imports of country 1 should equal the exports of country 2, this is not always the case. As I am regressing against the natural log of real imports, averaging the imports of country 1 with the exports of country 2, and vice versa, creates a smoother number for analysis. 13
trade observations, without excluding the lack of trade as a non-observation. As the data was based upon the Rose observations, this includes many observations where imports were zero resulting in negative natural log of real trade. Finally, this study will utilize panel OLS regression with year effects, as used by Rose, and the country effects method preferred by SW as a point of comparison. 10 The regressions will be run with both so that a common point of comparison can be established between the two methods. Though the results will diverge at some points, especially with less robust findings, there is a striking degree of commonality between their results. The fixed country effects issue raised by SW does not seem to cause the divergence between their results. The Model This paper will argue that more than imports, the WTO impacts member exports. To differentiate the importance of the WTO on exports, it is necessary to control for exporting country membership. The basic model will be specified as follows: Ln(M ijt ) = lnd ij + ln(area i Area j ) + ln(y i Y j ) + ln(y i Y j /Pop i Pop j ) + Lang ij + Border ij + Landl ij + Island ij + ComCol ij + CurCol ij + Colony ij + Comctry ij + Custrict ijt + FTA ijt + T t + MWTO i + XWTO j where i and j denote trading partners, t denotes times, and the variables are 11 : M ijt is the real imports of i from j at time t D is the distance between i and j 9 The natural log of small numbers is negative therefore many observations of real imports are negative observations. 12.8% of all observations of the natural log of real imports were zero or below. 10 Rose includes fixed year effects but omits country effects except in specified regressions. SW include both year and country effects in all their regressions. 14
Y is real GDP Pop is population Lang is a dummy variable which is unity if i and j have a common language Border is a dummy variable which is unity if i and j share a land border Landl is the number of land locked countries in the country pair (0,1,2) Island is the number of island nations in the pair (0,1,2) Area is the area of the country (in square kilometers) Comcol is a dummy variable which is unity if i and j were ever colonies after 1945 with the same colonizer Curcol is a dummy variable which is unity if i is a colony of j at time t or vice versa Colony is a dummy variable which is unity if i ever colonized j or vice versa Comctry is a dummy variable which is unity if i and j remained a part of the same country during the sample Custrict is a dummy variable which is unity if i and j use the same currency at time t FTA is a dummy variable which is unity if i and j belong to the same regional trading agreement T is a comprehensive set of time fixed effects MWTO is a dummy variable which is unity if the importing country is a member of the WTO at time t XWTO is a dummy variable which is unity if the exporting country is a member of the WTO at time t 11 The models, variables, dataset, and descriptions are almost completely from Rose (2003) except as noted previously. 15
This model allows the freedom to not just study whether trade increased or decreased when one or both countries joined the WTO, but more importantly how did those countries trade? SW explicitly assumes that the WTO impacts only imports, but the data of this study does not support that assumption. The Good News and the Bad News The WTO does impact trade but not in the way assumed. The Baseline Results are presented in Table 1 and the gravity variables all yield expected results. Distance is significant and negative while real GDP, currency union, and colonial variables are positive and significant. The gravity coefficients are similar to the results obtained by Rose and SW for the comparable variables and to the gravity literature. A few variables change sign or significance as a result of method. For instance, land area under the Rose method yields a moderately negative and statistically significant coefficient. The fixed country effects method returns an economically and statistically significant coefficient of.37 to.53 depending on exact specification. The fixed country method appears to have the most impact on variables with coefficients near zero or less robust results. The general model however, returns pretty much the expected. The WTO impacts exports significantly across methodological specification, though differently across income specification and member trading pairs. In the all the regressions, member exports rose by an economically and statistically significant amount. Only Country post-1970 had a smaller increase for exports than imports. Even then however, there is virtually no difference between their increase. The difference between regression methods for importing and exporting country membership without 16
income classification was stark. The regressions without country effects yielded an economically and statistically significant decrease in imports, while fixed importer and exported effects yielded a moderate, but statistically significant rise in imports. 12 Both the panel OLS and fixed country effects yield economically and statistically significant rise in exports. In every baseline regression, membership was highly significant. When classified by income the findings vary more but still yield interesting and consistent findings. High income member imports and exports increased by economically and statistically significant amounts regardless of time or method specification. Least developed members saw decreases in exports across method and time. Many of the middle and low income member results are insignificant or fragile to data or modeling changes. For instance, middle income exporting members using Rose OLS regressions experienced an insignificant drop, while under fixed country effects registered a positive and statistically significant coefficient of.35, a 42% rise in exports. Least developed countries post 1970 using an OLS regression increased imports under fixed year effects, while adding country effects makes imports negative. The inconsistent findings across methods are susceptible to modeling or data changes, so it is not surprising that a method change would cause a change in the variable. The highly significant variables that are robust to changes demonstrate a high degree of consistency between the Rose fixed year specification and the SW country effects method. A major argument of this paper is that Rose obtained an insignificant finding when regressing against total trade, because imports and exports act differently under the 12 It is worth noting that when creating a variable for importing country membership in the Rose data and running a regression without country effects, the coefficient returned is.19 with a robust standard error of. High income country members yields a coefficient of.67 with a robust standard error of. This is strikingly close to the results obtained here. Though not an exactly fair comparison, it is not unreasonable 17
WTO. Least developed members under the Rose method saw imports rise significantly and exports drop by a similar amount. Opposite signs on import and export country membership may explain why Rose found no WTO impact on total trade. The middle and low income results are frequently insignificant or somewhat fragile, for both the Rose and the fixed country effects method, implying middle and low income countries do not trade with each other due to WTO membership. This supports the finding made by SW that membership matters to industrialized countries, but it does not speak well of the WTO s ability to stimulate trade for its lesser developed members. How Does Trade Then Flow? When grouping countries by income and WTO membership to determine how the WTO impacts trading flows, the results generally reflect trade theory and previous findings in this paper. High income countries results on Table 2 indicate positive and mostly significant coefficients for imports and exports trading with members and nonmembers alike. High income countries have increased imports and exports significantly across all income categories, econometric method, and changes to data. Out of the 42 trade flow coefficients involving high income members, only 10 coefficients were negative and most of those are insignificant. High income countries have higher import and export levels, with members and non-members alike. This is not a surprising finding and supports the findings of SW that the WTO has an asymmetric impact on countries. The middle and low income countries do not seem to have benefited from the WTO the way the high income countries did. This occurred in two ways. First, middle either. Rose divides total trade (x1 + x2 + m1 +m2) by 4, creating almost an import variable. In this paper, I use average real imports (m1 + x2) divided by 2. 18
and low income members export primarily to high income countries and almost all other export categories were negative or insignificant. Middle and low income countries did not, for the most part, increase trade with other middle and low income countries. In other words, the WTO did not create trade for middle and low income members, independent of high income members. Least developed members have divergent numbers, but support the previous results that imports rise while exports decline. Second, middle and low income members saw exports to high income members rise more than imports from high income members. This does not entirely support the SW idea that the WTO is a rich country club. This indicates that trade diversion may occur away from natural trading partners, according to gravity theory, and shift to high income members. High income exporting members do not demonstrate higher levels of exports to middle and low income members. The WTO has allowed lesser developed members to liberalize trade at a slower rate than developed countries and the data seems to bear that out. Interestingly enough, high income members do not appear to demonstrate significantly different import patterns between middle and low income members or non-members. 13 The small difference and statistically borderline results, do not indicate that middle and low income members have economically and statistically higher export levels to high income members than middle and low income non-members. This is all the more important as research has shown that trade composition matters to economic growth(arora and Vamvakidis 2004). The trade patterns between income classes reflect many of the difficulties encountered by the WTO in making substantial changes to trade policy. 19
The clearest trends that the data reveals about membership, is its impact on trade between members and trade with non-members. Trade between members increases significantly, but has some areas of insignificance or negative coefficients. Interestingly, the member to member trading results are fragile to method change. Nine of the sixteen income and method pairs have opposite signs between coefficients. For instance, high income member trade with other high income members has a small negative and statistically insignificant result without country effects and a moderate, though not as large as SW, but statistically significant increase with country effects. Overall, member to member trade demonstrates a significant increase that is robust to method and changes to the data. The more interesting regressions focused on trade between members and nonmembers. A clear and unchanging result of this data is that members do not import significantly from non-members though they export heavily to non-members. The import result is even greater if only considering low and middle income countries. With the exception of high income members, virtually every sign for members importing from non-members are negative and the coefficients that were not negative were mostly insignificant. Conversely, members increased their exports to non-members by economically and statistically significant amount, though this result was less robust for least developed and low income countries. It is somewhat surprising that WTO members increased exports to non-members more than to members, this implies that variables other than the WTO are factoring in to export levels for members. Rose obtains an insignificant finding when regressing against 13 The results of regressions not presented here confirm this conclusion. Lesser developed member versus lesser developed non-members exports to developed members was economically small and of borderline 20
total trade because membership does not expand average real trade, rather it changes the composition of trade. It is worth noting that in the regressions where only one country was a WTO member, trade shrank by a small amount, which was insignificant with fixed country effects. This closely matches Rose results for similar regressions. It is also worth nothing that many of the trading pair statistics found in SW are similar to the results produced in Table 2 Trade Flow Results. Around the World in Under a Page Breaking down the regression by region and WTO membership largely confirms the previous findings. WTO membership is widely divergent in its impact, both positive and negative, and in its impact on exports and imports. Of the five regional membership dummy variables in Table 4, four of the import coefficients are negative in both the panel OLS and fixed country effects regressions. Export coefficients fared only slightly better. Only East Asia members have robust, economically and statistically significant higher exports. The rest of the regional member variables used here are either negative or insignificantly positive. The regional dummy variables are not incredibly robust and are susceptible to changes in method, data, or country exclusions. The coefficients for Sub- Saharan Africa, for example, go from economically large and statistically significant increased member imports and decreased exports under the Rose OLS method, to negative imports and positive exports with borderline significance. Three of the export and two of the import coefficients have different signs between methods. These specific results are fragile and prone to change based upon country exclusions and data changes. significance in both methods utilized. 21
The country effects method seems to confirm, though moderate, highly robust findings while potentially making large changes to more fragile or insignificant results. SW have argued that not only is a fixed country effects method more appropriate, but also that WTO impact is extremely different across income classes. The results presented here, support that argument, but also support the claim that the WTO impacts imports and exports differently. Five of the ten import and export pairs have opposite signs, under OLS and fixed country year effects regressions. The impact of the WTO may vary widely, but its impact on the type of trade also varies dramatically. It does not uniformly raise trade across regional classification or econometric method. Exports for East Asian members rose dramatically regardless of method, while imports fell or rose depending on your preferred method. Interestingly, under fixed country regressions, only East Asian variables were positive, highly significant, and robust. Can the Results Take the Heat? The overall member results are robust to the removal of high income countries and other interesting data exclusions. As SW note, the WTO was designed and developed under the careful watch of high income members, so they are therefore essential to any understanding of its management and its impact. Table 4 provides a list of changes made to the basic model testing model sensitivity and method specification issues on baseline results. Excluding high income members from the general member results, either in the OLS or with country and year effects, had little effect on the fundamental results. Exports were significantly and positively impacted by WTO membership. The only case in Table 4 where this does not hold is the Rose OLS method 22
excluding real imports of less than $100,000 and $500,000. It is also worthy of note that the developing country and high income coefficients excluding imports under $100,000 and $500,000 are rather close to the SW results. The main argument of this paper however seems to be largely confirmed both across method or income class. Countries join the WTO for what they can sell the world, not what they can buy from it. Though middle and low income members may only have a slight advantage over non-members exporting to high income members, they have higher export levels to many exclusions of data and changes of method. For instance, even for trade entirely within Africa, members have higher export levels in line with the general results. This finding is robust across data exclusions and method. The results in Table 4 detail a wide variety of income and sorting regressions to test whether the models are susceptible to changes in the data. Without detailing every regression, one over arching theme emerges: exporting country member coefficients are consistently positive, significant, and higher than importing country membership. This finding is robust to many changes in the regression and more importantly method. The Rose panel data OLS method and the fixed country effects used by SW yield strikingly consistent results in support of the theory that the WTO has a greater positive impact on exports than imports. The WTO matters more to exports than to imports. The Stress Tests The results were subject to a battery of tests detailed in Table 5 including random dummy variables for WTO membership, weighted least squares regressions with a variety of different weightings with and without country effects, and random effects 23
regressions. In all instances, the results held up well. A few brief details on some of the stress testing. First, just to make sure it wouldn t be possible to obtain even a small amount of either economic or statistical significance random dummy variables for importing and exporting country members were created. Regressions under with and without fixed country effects were run with the random country member dummy variables. Their coefficients came back close to zero and statistically insignificant. Second, the weighted least squares regressions with both real GDP and real GDP per capita for high income countries reflected the results of SW. Third, the weighted least squares, fixed effects, and random effects regressions reflect the conclusions that the WTO positively and significantly impacts exports. Fourth, WTO membership has a greater impact on exports, both economically and statistically, than imports under these methods. In every case concerning overall membership, imports drop or rise insignificantly, while exports rise in both economically and statistically significant ways. Fifth, the other major argument of this paper is that WTO membership will frequently impact imports and exports in different ways. In fact, in every case except one, weighted least squares with country frequency weighting, overall import and export member coefficients are the opposite sign. This helps explain the Rose finding of insignificance. A Few Cautionary Notes To temper any jubilation or despair that these results might invoke, there are a number of qualifiers that need to accompany this research. First, the impact of the WTO seems to be highly correlated with income. Though industrialized countries would be the most obvious example, as noted by SW and supported here, East Asian members also 24
have significantly higher trade levels. Income may not be the defining variable in WTO success, as it also depends significantly on the domestic institutions that deal with trade policy (Rodrik 2000). Interestingly enough, high income members trading with each other put forth a mixed bag of results depending on method use. It is also worth noting, that high income members trading with high income non-members uniformly have higher coefficients than high income member trade. Least developed members saw lower exports and mixed results for imports. These results imply that the WTO may depend more on income or geography for its impact, than trade depends on the WTO. Second, the WTO has not brought the expected gains to lesser developed members. Though SW find significant levels of high income member imports from developing members when excluding absolute trade under $500,000, the results here find similar levels of imports by high income members from developing non-members. In fact, developing members have only slightly higher export levels to high income members than developing nonmembers. 14 As research has shown, economic liberalization is not significantly correlated with increased trade and growth(hausmann, Pritchett, and Rodrik 2004). If an objective of the WTO is to integrate developing countries in the global economy, it would seem that it could do a better job. Third, countries appear to join the WTO more for the most favored nation status than the domestic tariff reductions. High income members are the only income group to demonstrate economically and statistically higher import levels across time, trading partners, and method. Middle and low income members have mixed import records. In 14 Creating a developing member exporter dummy variable for high income country imports, yields a coefficient of.14 under the Rose method with a robust standard error of.06. The fixed year method returns a coefficient of.23 and a robust standard error of.07. A moderate economically significant return, though 25
some instances, the results seem to imply that the WTO acts as a defense to protection more than a method to liberalize. Even when considering inter-african trade for instance, imports dropped and exports increased. This indicates that countries liberalize only to the degree necessary to gain most favored nation status and stay out of trouble. Fourth, though typically used as robustness checks, a weighted least squares model might prove more accurate than the panel OLS by taking into account trade levels, GDP, or country data quality. It is heartening to note however, that the weighted least squares regressions returned similar results to the baseline data. Conclusion The WTO impacts trade but not in the way assumed. The WTO has a greater impact on exports than imports. Though there is no theoretical reason as SW rightly note, there are practical reasons to believe it does and the data supports this argument. The asymmetric distinctions between developed and lesser developed countries hold, but more importantly, specifically when trying to reconcile the Rose and SW result, the WTO impacts types of trade. The generalized WTO variables that positively and significantly impact overall trade, are when both importing and exporting countries are WTO members. In addition to concluding that the WTO significantly impacts exports, this paper also reconciles many of the seemingly contradictory findings of Rose and SW. It is also important to note the difference between fixed country effects and the panel OLS with fixed year method preferred by Rose. First, strong robust findings held up across method. A surprising amount of the results held closely across specification. borderline statistically, middle and low income countries do not seem to receive the unambiguous boost to trade promised by the WTO. 26
Second, the results that changed from method to method, were somewhat fragile. Robust finding results that could withstand data or modeling alterations in the original method remained. Fragile results manipulated by data exclusion, underwent significant changes with a different econometric method. Third, the results here support previous work comparing the importance of fixed country effects that failing to include them in a gravity model may overstate the impact of key variables. Fourth, this area requires additional theoretical work to arrive at a preferred methodological specification for the gravity model. The OLS and fixed year and country effects were presented here because they remained strikingly consistent, but also as a means of comparison. Fifth, an important difference between Rose and SW seems to be their choice of variables to measure WTO impact and data construction methods. Though econometric method undoubtedly raises some discussion points and estimation questions, the major trade patterns when using similar variables hold across method. The results indicate that the WTO has a small but positive impact on trade when both members of a trading pair are members of the WTO. The results also indicate the joining the WTO is all about the exports. 27
Table 1-Baseline Results Membership Income Post 1970 Post 1970 OLS Country OLS Country OLS Country OLS Country Regional 1.319 (.196).602 (.227) 1.29 (.216).635 (.225) 1.26 (.195).372 (.232) 1.25 (.213).416 (.228) Currency Union 1.619 (.165) 1.461 (.287) 1.80 (.171) 1.456 (.282) 1.788 (.189) 1.68 (.383) 1.92 (.199) 1.67 (.378) Distance -1.437 (.033) -1.66 (.038) -1.376 (.033) -1.668 (.038) -1.500 (.032) -1.80 (.036) -1.44 (.032) -1.803 (036) Real GDP 1.156 (.013).648 (.078) 1.154 (.012).699 (.077) 1.167 (.013).993 (.086) 1.167 (.012).982 (.087) Real Per Capita GDP.534 (.02).119* (.072).138 (.022).02 (.072).513 (.019) -.483 (.081).087 (.024) -.489 (.081) Common Language.472 (7).65 (.063).596 (6).648 (.063).515 (9).66 (.063).630 (8).659 (.063) Border.698 (.167) -.108 (.235).788 (.175) -.121 (.234).754 (.161) -.146 (.216).837 (.169) -.154 (.216) Landlock -.018 (7) -.188 (7) -7 (6) -.245 (6) Island.029 () -.189 (.291) -.105 (9) 4 (.271) 4 () -.681 (333.2) -.088 ().718 (.268) Land Area -.12 (.01).534 (5) -.117 (.01).488 (5) -.12 (.01).372 (8) -.115 (.01).475 (7) Common Colonizer.621 (.087).693 (.09).822 (.088).694 (.09).536 (.087).746 (.09).752 (.09).746 (.089) Current Colony 1.241 (.326) 1.214 (.419) 1.33 (.308) 1.21 (.394) 1.58 (.537).954 (.812) 1.56 (.469) 1.126 (.711) Colony post-1945 1.776 (.167) 1.55 (.176) 1.77 (.179) 1.561 (.174) 1.98 (.159) 1.65 (.167) 1.99 (.168) 1.66 (.163) Common Country -.472 (1.17).366 (1.192) -.492 (.941).67 (1.02) -1.284 (1.02).261 (1.18) -1.12 (.739).443 (.976) Import Country Membership -.25.18 -.31.22 High Income Import Members.75.64.09.76.42.12 Middle Income Import Members -.55.01 -.57.22 Low Income Import Members -.06 -.12.10.06 -.15.09 Least Developed Import Members.53.06 -.21.11.49.07 -.31.10 Export Country Membership.89.38.86.21 High Income Country Members 1.36.55.08 1.72.10 Middle Income Country Members -.35.08 -.30 Low Income Country Members -.42 -.11.11 -.49.06 -.21.15 Least Income Country Members -.64.07 -.24.15 -.69.07 -.21.18 R-squared.525.634.540.634.546.666.564.667 Observations 419,910 381,625 419,910 381,625 331,651 298,166 331,651 298,166 Robust coefficients with standard error in parentheses. 28
Table 2-Trade Flow Between Members OLS Member Country Exports to Other Members.83 Member Country Exports to Non-Members 1.11.07 Member Country Imports from Non-Members -.02.07 Member Country Imports from Other Members.28 One Country of Trading Pair WTO Member -.15 Both Countries WTO Members.32 High Income Middle Income Low Income Importing Member Importing Member Importing Member Exporting OLS Country OLS Country OLS Country NonMem High.64 1.01 -.69 -.13 -.53.65 Income.23.41.26.34.28.38 Middle 1.27.56 -.25.28 -.88 -.39 Income.13.20.12.13.15.35 Low 1.28 1.13 -.95.71-1.10 1.61 Income.20.33.21.26.23.74 Least 1.73.57 -.71.89 -.96 2.00 Developed.24.31.23.33.25.71 Country.31.47.08.40.09.13 -.03.20 Least Developed Importing Member OLS Country.74.91.28.53 -.29 -.15.19.51.24.38.29.96 -.27 1.93.28.69 Exporting Members-Importing Non-Members High Income Importing Non Middle Income Importing Non Low Income Importing Non Least Developed Importing Non Exporting Memb. OLS Country OLS Country OLS Country OLS Country High Income 1.52.26.64.41 2.63.11.70.16 2.26.19.45.27 1.63.22.51.33 Middle Income.78.25.17.28.86.11.50.12.48.18 -.07.22.02.21 -.19.29 Low Income.34.32.80.52 -.94.14 -.02.38 -.52.20 1.08 1.29 -.34.24-2.03 1.78 Least Developed -.19.46 -.54.72-1.67.19 -.83.62 -.82.28 1.59 2.67 -.29.29-2.02 1.78 High Income Importing Member Middle Income Importing Member Low Income Importing Member Least Developed Importing Member Exporting Memb OLS Country OLS Country OLS Country OLS Country High Income -.12.08.39.09 -.39.06 -.13 -.39.08.17.09 -.09.09 -.10 Middle Income.99.10.21.20 -.22.09.11.09.99.11 -.08.22 -.23.13 -.22.28 Low Income 1.83.14.71.24 -.39.13.41.16 -.13.14.50.30.76.16.50.38 Least Developed 2.46.20.94.41 -.31.19.52.27.37.21-1.12.49.60.22-1.28.52 Robust coefficients with standard error in parentheses. Coefficients only presented for ease of presentation 29