The WTO trade effect

https://doi.org/10.1016/j.jinteco.2011.05.011Get rights and content

Abstract

This paper re-examines the GATT/WTO membership effect on bilateral trade flows, using nonparametric methods including pair-matching, permutation tests, and a Rosenbaum (2002) sensitivity analysis. Together, these methods provide an estimation framework that is robust to misspecification bias, allows general forms of heterogeneous membership effects, and addresses potential hidden selection bias. This is in contrast to most conventional parametric studies on this issue. Our results suggest large GATT/WTO trade-promoting effects that are robust to various restricted matching criteria, alternative GATT/WTO indicators, non-random incidence of positive trade flows, inclusion of multilateral resistance terms, and different matching methodologies.

Highlights

► This paper re-examines the GATT/WTO membership effect on bilateral trade flows, using nonparametric methods. ► The nonparametric methods used include pair-matching, permutation tests, and a Rosenbaum (2002) sensitivity analysis. ► This estimation framework is robust to misspecification bias, allows general forms of heterogeneous membership effects, and addresses potential selection bias. ► Our results suggest large GATT/WTO trade-promoting effects. ► The finding is robust to various matching criteria/methodologies, alternative GATT/WTO indicators, zero trade flows, and multilateral trade resistance.

Introduction

Since its creation in 1947, the General Agreement on Tariffs and Trade (GATT) has played an important role in the international trading system. It has sponsored eight rounds of trade-policy negotiations that successfully brought down the average tariff rates on industrial goods and also expanded the set of substantive rules governing international trade (beyond tariffs to nontariff barriers, and beyond trade in merchandise to trade in services). This process culminated in the establishment of the World Trade Organization (WTO) in 1995. Since 1947, the GATT/WTO has also grown in its membership from a small set of 23 (mainly developed) countries to a roster that now includes more than 150 countries. Meanwhile, global trade flows have increased exponentially at a rate above the growth rate of merchandise output. It is against this backdrop that the finding by Rose (2004) came as a surprise.

Based on the gravity model of trade (that hypothesizes that the bilateral trade volume between two countries varies positively with their economic sizes and inversely with their bilateral trade resistance), Rose (2004) conducted parametric estimations and found that the GATT/WTO membership status of a country pair had no statistically significant effect on bilateral trade. This negative finding was partially reversed by Tomz et al. (2007) when they reclassified countries according to their participation status in the GATT/WTO (instead of formal membership), and by Subramanian and Wei (2007) when they differentiated the effects by subsets of the sample (e.g., developed versus developing countries). Although shedding light on possible caveats to the original study by Rose (2004), these studies and other follow-up research in this literature have largely followed the conventional approach of parametric estimation. In this paper, we argue that when the leading gravity theories do not have clear guidance on the parametric (functional) relations of the empirical trade-resistance measures, and when the economic theories of trade agreements (e.g., Bagwell and Staiger, 2010, pp. 245–247) suggest that heterogeneous membership effects on trade are important implications (of uneven levels of trade negotiation participation), these existing parametric studies are at risk of misspecification bias on both accounts. We propose a system of nonparametric methods that is geared toward these concerns to re-evaluate the GATT/WTO trade effect.

In particular, we apply pair-matching methods to obtain point effect estimates. Following the established gravity theories (Anderson, 1979, Bergstrand, 1985, Deardorff, 1998, Anderson and van Wincoop, 2003), empirical researchers have come to adopt a long list of variables as proxies for the theoretical concept of trade resistance between a pair of countries. This list typically includes (foremost) distance, geographic characteristics, language, colonial ties, currency union, free trade agreement, and the GATT/WTO membership status. However, there is no clear theoretical justification for the linear relation (among the various trade-resistance measures) that is often adopted in the empirical studies. In this paper, we conduct matching based on a set of covariates that is exactly the same as the list of regressors used in parametric studies. However, by matching observations that have different treatment status but are otherwise similar in terms of these covariates, we do not have to take a stand on the functional relations among these observed covariates and hence avoid potential parametric misspecifications. In addition, the matching method by design allows for the treatment (i.e., membership) effect to vary with the observed covariates, and thus it can accommodate arbitrary forms of heterogeneous treatment effects. In general, the homogeneous effect estimate in regression approaches does not correspond to the average of subject-wise heterogeneous effects, if the heterogeneity takes on highly nonlinear functional forms.

We also address other potential econometric concerns arising in the current application. First, given a panel of bilateral trade data, which likely have a complicated data structure with serial and spatial dependence, this paper applies permutation tests that circumvent the difficulty in deriving asymptotic tests. Permutation tests are nonparametric and exact inferences (applicable to finite sample sizes). They are also straightforward to implement in the matching framework. We generalize the test to explicitly allow for heterogeneous treatment effects in constructing the confidence intervals. Finally, we complete the estimation procedure with a nonparametric sensitivity analysis à la Rosenbaum (2002) to formally address potential bias due to unobserved self-selection into membership. We put together the above methods in a coherent manner such that they can be easily applied to other treatment effect problems of a similar nature.

Applying the nonparametric methods to the data set of Rose (2004), we reach a conclusion that is in stark contrast with Rose (2004): membership in the GATT/WTO has large and significant trade-promoting effects. We explore robustness of this result to various possible caveats; the general finding continues to hold. First, both parametric gravity and nonparametric matching estimators rely on the assumption of ‘selection on observables’; in other words, non-random selection into membership based on unobservables is assumed away. This assumption may fail if there are important omitted variables. The Rosenbaum (2002) sensitivity analysis partly addresses this problem. Alternatively, we also conduct restricted matching, where we further limit the match to observations from the same ‘dyad’ (where a dyad indicates a pair of trading countries), the same year, or the same relative development stage. This eliminates potential bias arising from unobserved heterogeneity across dyads, years, or development stages.

Second, Tomz et al. (2007) emphasize the importance of de facto participation in the GATT/WTO by colonies, newly independent nations and provisional members, and find strong GATT/WTO effects on trade when this type of nonmember participation is taken into account. We conduct the same nonparametric analysis using the data set of Tomz et al. (2007) and find even stronger results than those based on the Rose (2004) data set.

Third, we verify the robustness of pair-matching by conducting ‘kernel-weighting matching’, which allows multiple matches for a subject while assigning greater weights to closer matches. The kernel-weighting matching effect estimates are very similar to pair-matching estimates.

Fourth, by using the data set of Rose, 2004, Tomz et al., 2007, we have based our analysis on observations with positive trade flows. Studies by Helpman et al., 2008, Felbermayr and Kohler, 2007 suggest that the incidence of positive trade flows may not be random. To address possible bias due to non-random incidence of active trading relationships, we apply our nonparametric procedures to the subset of the data where a dyad has reported bilateral trade flows before either country in the dyad ever joins the GATT/WTO. For these observations, the membership effect on prompting new trading relationships is not relevant, and hence the effect estimates correspond to only the membership effect on trade volumes. We find overall stronger effect estimates based on this refined analysis.

Fifth, relative, rather than absolute, trade resistance is argued by some gravity theories to be more appropriate in explaining bilateral trade flows (Anderson and van Wincoop, 2003); thus, multilateral resistance terms may have to be controlled for. We follow recent studies by Baier and Bergstrand, 2009a, Baier and Bergstrand, 2009b to approximate the endogenous multilateral resistance terms by observable exogenous trade resistance covariates in the matching framework. The strong trade effects of GATT/WTO remain.

Finally, we explore an alternative treatment effect concept, difference-in-difference, which is based on weaker identification assumptions and thus could be more robust to potential bias due to selection on unobservables. This method compares the difference over time in the trade volume of a member dyad to that of a comparable nonmember dyad. The matching estimates indicate that the GATT/WTO trade effects are negligible in early phases of the membership, but become statistically and economically significant five or six years after the GATT/WTO accession. To complete the analysis, we conduct placebo exercises and verify that the time trends of trade flows of matched dyads are the same in advance of membership, dismissing concerns that the difference-in-difference estimates may be picking up systematic differences in time trends between member and nonmember dyads due to unobservables not controlled for.

The discrepancy between the finding of the current nonparametric approach and that of the conventional parametric approach suggests that parametric gravity models may be misspecified. We explore generalizing the parametric gravity model's specifications to reduce the discrepancy. Our limited search suggests that the assumption of homogeneous membership effects could be a major source of misspecification. By allowing the membership dummies to interact with observed covariates (and hence allowing the membership effects to vary with dyad-year characteristics), we find the parametric effect estimates to become significant and positive. However, more research into the nature of heterogeneous membership effects seems desirable and we leave this for future research.

The rest of the paper is organized as follows. Section 2 introduces the nonparametric methodologies. Section 3 explains the data used. 4 Benchmark results, 5 Robustness checks present our benchmark estimation results and robustness checks. Section 6 explores potential misspecifications of the parametric gravity models. Section 7 provides our conclusions.

Section snippets

Mean effects and matching

Recall that a ‘dyad’ indicates a pair of trading countries. In the current application, an observation unit corresponds to a dyad i in a year t, while a matched ‘pair’ indicates two observation units matched on covariates. Let dit denote the observed treatment status of a dyad i in year t, where dit = 1 if the subject it is treated and 0 if untreated. The treatment dummy dit takes on different meanings as the treatment under study changes. For example, a dyad-year is ‘both-in’ treated if both

Data description

We base our analysis on the Rose (2004) data set,5 although we will also use the Tomz et al. (2007) data set in Section 5.2 as one of the robustness checks. Readers are referred to the source for a detailed account of the data. The data set includes 234,597 observations on trade flows among 178 IMF trading entities between 1948 and 1999 (with some “gaps” and missing observations). There are 12,150 distinct dyads and, on average,

Both-in effects

Table 4 reports the estimation results as we apply the nonparametric methodologies described in Section 2 to the data set of Rose (2004). The both-in effects are significantly positive regardless of the caliper choice (which sets the best 100%, 80%, 60%, or 40% of matched pairs to include in the estimation). The estimates suggest that membership in the GATT/WTO by both countries on average raises bilateral trade volume by 74% (= e0.553  1) to 277% (= e1.328  1) for dyads that both chose to be in

Robustness checks

In this section, we conduct an extensive set of robustness checks by considering various restricted matching criteria, alternative GATT/WTO indicators, the non-randomness of zero trade flows, the inclusion of multilateral resistance terms, and different matching methodologies. Overall, the benchmark finding of a significant GATT/WTO effect on trade is strengthened, not weakened, while the GSP effect is qualified.

Potential problems with the parametric gravity estimates

The discrepancy between the current nonparametric matching estimates and the conventional parametric gravity estimates suggests that the empirical gravity models used in the parametric studies may be misspecified. In particular, guided by the pattern of nonparametric effect estimates observed above, we suspect that heterogeneous treatment effects can be important. While our matching estimator allows for heterogeneous effects that vary with the observed covariates, the specifications used in

Conclusion

This paper contributes to the literature on the effects of GATT/WTO membership/participation on actual trade flows. Previous studies of this issue have largely relied on parametric estimation of gravity-based trade models. Concerns about parametric misspecifications, heterogeneous membership effects, and unobserved selection bias are raised by the current paper and addressed by using nonparametric methods. In particular, a pair-matching estimator is used to obtain the point effect estimates,

Acknowledgements

We thank two anonymous referees, Robert Staiger, Jeffrey Bergstrand, and Alan Deardorff for their helpful comments and suggestions. We also thank colleagues, Hian Teck Hoon, Davin Chor, and Tomoki Fujii, at SMU for their helpful feedback at different stages of the project. The research of Myoung-jae Lee was supported by the Korea Research Foundation funded by the Korean Government (KRF-2009-327-B00091).

References (56)

  • J.E. Anderson

    A theoretical foundation for the gravity equation

    American Economic Review

    (1979)
  • J.E. Anderson et al.

    Gravity with gravitas: a solution to the border puzzle

    American Economic Review

    (2003)
  • K. Bagwell et al.

    An economic theory of GATT

    American Economic Review

    (1999)
  • K. Bagwell et al.

    The world trade organization: theory and practice

    Annual Review of Economics

    (2010)
  • J.H. Bergstrand

    The gravity equation in international trade: some microeconomic foundations and empirical evidence

    Review of Economics and Statistics

    (1985)
  • C. Broda et al.

    Optimal tariffs and market power: the evidence

    American Economic Review

    (2008)
  • M. Caliendo et al.

    The employment effects of job creation schemes in Germany: a microeconometric evaluation

  • P.-L. Chang et al.

    The WTO trade effect

  • A.V. Deardorff

    Determinants of bilateral trade: does gravity work in a neoclassical world?

  • M.D. Ernst

    Permutation methods: a basis for exact inference

    Statistical Science

    (2004)
  • G. Felbermayr et al.

    Does WTO membership make a difference at the extensive margin of world trade?

  • R.A. Fisher

    The design of experiments

    (1935)
  • J.A. Frankel

    Regional trading blocs in the world economic system

    (1997)
  • J.L. Gastwirth et al.

    Dual and simultaneous sensitivity analysis for matched pairs

    Biometrika

    (1998)
  • J.J. Heckman

    Sample selection bias as a specification error

    Econometrica

    (1979)
  • J.J. Heckman et al.

    Matching as an econometric evaluation estimator: evidence from evaluating a job training programme

    Review of Economic Studies

    (1997)
  • J.J. Heckman et al.

    Characterizing selection bias using experimental data

    Econometrica

    (1998)
  • E. Helpman et al.

    Estimating trade flows: trading partners and trading volumes

    Quarterly Journal of Economics

    (2008)
  • Cited by (90)

    View all citing articles on Scopus
    View full text