3.1 Institutional background and initial observation of the data

In general, there are two institutional mechanisms for settling a dispute under the WTO DSB. The first is DSB legal proceedings via panel and Appellate Body (AB) procedures. The second is non-litigious bilateral consultations for seeking a mutually agreed solution, official withdrawal, or termination. From a functional perspective, it is insignificant to promote one mechanism as superior to another, as long as member countries adhere to the WTO DSB process instead of taking unilateral actions. From a practical perspective, however, complainants’ main concerns are whether DSB resolutions have a significant effect or not.

The DSB proceedings have four broad stages: consultation between disputing parties, the establishment of a panel, adoption of panel or AB reports, and implementation of the legal rulings. However, when considering the economic analysis of trade flow and economic impact, the more relevant points of a dispute process are (i) before the dispute; and (ii) after dispute initiation upon an official consultation request. The first phase refers to the point when a complainant detects initial evidence of changes to the value or volume of its exports to the respondent. The evidence is usually observed within a year (or shorter in monthly and quarterly data) before a dispute initiation.

The second phase starts from the year of the dispute initiation. This period may also be divided into two periods: (i) the period encompassing the consultation request and settlement before further legal proceedings, and (ii) the period during which a legal dispute officially occurs, and the implementation of the recommendation of the WTO DSB's ruling is carried out. During the consultation period, the WTO DSB provides opportunities for the disputing parties to come together for bilateral negotiation. Prior to a fully fledged legal dispute, the consulting members can adjust their stances on the case. During this period, a case may be closed if the complainant agrees to withdraw. If the parties fail to reach a mutually agreeable solution, the complainant can proceed to establish a panel. At this point, according to WTO DSB Article 4.11, third parties may join the dispute by request via a process set out in Article 10. The precondition for joining the dispute, however, is the possession of a substantial interest in the matter, and written submissions by the third parties to the panel are taken into consideration. Regarding trade flow, any parties participating in a dispute settlement, including third parties, may experience significant changes in trade of disputed products throughout the dispute settlement process. However, we should not overlook trade growth made by non-participants through new market entry (i.e., extensive margin growth) and market expansion (i.e., intensive margins) following the removal of trade restrictive measures.

Table 1 presents preliminary information on the trade flow changes of disputed products. Specifically, it shows the changes in mean values of respondents’ imports of the disputed products per WTO member country for various periods before dispute initiation and after dispute termination. Trade flows are divided into two broad periods: (i) PRIOR_(τ-t) denotes years before dispute initiation (τ) and (ii) POST_(T+t) denotes years following dispute completion (T). For example, the third column of the table (PRIOR_(τ-1~2)) shows the average value of the disputed product imports for respondents prior to the initiation of a dispute; (τ−1~2) indicates the average level of these imports two years before the dispute initiation. The fourth, fifth, and sixth columns – POST_(T+1~2), POST_(T+2~3), and POST_(T+1~3) – represent first- and second-year, second- and third-year, and first- to third-year average imports following a dispute's completion (T). In order to see import changes over time, we calculate the import growth rates (percentage change) of respondent countries in the parentheses for post periods, (T + 1~2), (T + 2~3), and (T + 1~3), which are respectively compared against mean import values 1~2 years before the dispute initiation (τ−1~2). In the POST years, mostly due to new entries to the WTO membership, the number of non-participants changes over time. As such, the number of overall exporting countries to respondents fluctuates accordingly. This table further breaks down these import values according to three groups (complainant, third party, and non-participant) and five categories of final DSB decisions (legal win, mutually agreed, non-legal win, withdrawal, and dropped with no further proceedings).

Table 1.

Trade flow changes pre-and post-DSB dispute and outcomes by group

Unit: Million USD (Percentage change: %Δ)

Note: The real import value is our dependent variables for the regression analysis.

‘N’ is the number of observations for each group over the five year period (τ−2, τ−1, T + 1, T + 2, and T + 3) where τ denotes the point of dispute initiation and T denotes the point of dispute completion. The variations between the three groups (complainant, third-party, and non-participant) and within groups before dispute initiation (PRIOR: τ–1~2) and after dispute completion (POST: T + 1~3) provide the statistical grounds for our subsequent regression analysis.

‘All Groups’ indicates all WTO member countries. ‘Complainants’ are countries that have raised a dispute to the WTO DSB. ‘Third parties’ are countries that join the WTO DSB proceedings due to their substantial interest in the outcome. ‘Non-participants’ are countries not directly participating in the dispute but that may become potential beneficiaries of the outcome.

To see import changes over the dispute period, the import growth rates (percentage change from baseline: %△) of respondent countries are calculated in the parentheses for post periods such as (T + 1~2), (T + 2~3), and (T + 1~3) and are compared against mean import values 1–2 years before the dispute initiation (τ−1~2). In the POST years, mostly due to new entries into the WTO, the number of non-participants varies. As such, the number of overall exporting countries to respondents fluctuates accordingly.

This table illustrates some interesting descriptive characteristics. First, the number of legal participant countries is much smaller than the number of non-participants, but their trade in the disputed products is, on average, much larger. This suggests that the disputed import markets have a somewhat monopolistic structure, meaning that the market has potential growth and welfare in terms of the world economy when the market is more open. Second, as expected, a complainant group has the greatest economic interest in the disputed products while a third party has a substantial interest as well, second only to the complainant group. Third, the average trade quantities are larger for legally contested cases among the DSB proceedings. This is consistent with the fact that WTO legal proceedings are likely to be taken up and pursued to tackle problematic measures that protect high-stakes import markets. Fourth, the table shows a relatively smaller increase in the exports of complainants between the pre-dispute and post-dispute periods following a legal victory than that of other participant groups. Lastly, however, in mutually agreed cases, the exports of all three groups increase substantially in post-dispute period.

3.2 Estimation strategy

We employ two types of variations to detect the economic effect, especially in terms of imports, of the WTO DSB. One is a time variation to check economic impact before and after a dispute. The other is a cross-sectional variation between disputing parties, including third parties and non-participants. The differences in trade value before and after a WTO dispute measured for disputing parties are compared to those measured for non-participants. In addition to combining two dimensions of variation, cases can be divided into those where complainants (1) win, (2) lose, and (3) officially withdraw, and those where disputing parties are (4) settled with a mutually agreed solution. Lastly, cases that are (5) dropped with no further proceedings are used as a reference for the four other dispute outcomes.

In contrast to previous studies that address the trade effects between disputing parties, our analysis shows changes in bilateral import flows of respondent countries with respect to all exporting countries for the period spanning from consultation stages of disputes to implementation of rulings. An empirical estimation of bilateral trade flows for all countries allows a more comprehensive analysis of the economic impact of legal rulings. Since controlling for other factors that may affect respondents’ trade is critical in this comprehensive econometric analysis, this study adopts the following analytical strategies.

First, we include three country-specific effects to fix unobserved characteristics specific to (i) respondents, (ii) complainants, and (iii) WTO members by using corresponding dummies in the regression estimation. The main idea behind including respondent and complainant country-specific fixed effects is due to the differences in time-invariant characteristics that one can observe in their legal roles (one dummy as respondent and another as complainant). ‘Member’ dummy for each WTO member country is included to capture fixed country characteristics, independent of their position in any dispute. The member dummy controls for other unobserved individual country-specific characteristics, which are related to different economic and political conditions (Busch et al., Reference Busch, Reinhardt and Shaffer2009; Sattler and Bernauer, Reference Sattler and Bernauer2011; Shin et al., Reference Shin, Lee and Park2016). Examples of these three fixed effects include the frequency and capacity of how a country makes use of the WTO DSB, as well as income, average trade level, size, and retaliatory power.

Second, we consider bilateral-pair effects to control for the bilateral economic/political relationship between complainants and respondents. These pair effects control any particular tendency formed by the specific relations between complainants and respondents. For example, one pair may have trade disputes more or less often than others due to special political relations (e.g. FTA relations or aid donor–recipient relations) or geographical characteristics (e.g. cultural traits and physical distance) specific to the pair (Ahn and Shin, Reference Ahn and Shin2011; Baier and Bergstrand, Reference Baier and Bergstrand2007; Bown, Reference Bown2009; Shin et al., Reference Shin, Lee and Park2016).

Third, we also take year fixed effects into account to capture any macroeconomic trends in the world trade flow of the disputed products. During the dispute period, there could be time-specific factors such as significant changes in oil prices, financial crises, economic booms, and exchange rate fluctuations that severely influence the dependent variable. Thus, year fixed effects are included to control for these macroeconomic elements.

Fourth, agreement effects are added to control for case-specific factors subject to WTO Agreements. The effects can capture the specificity of different agreements, which do not vary by respondent. It is important to note that the trade levels differ depending on the specific agreements covered. For example, disputes concerning non-tariff barriers such as technical regulations, sanitary and phytosanitary measures, or domestic taxes may relate to smaller and less concentrated trades. On the contrary, disputes dealing with specific tariffs or trade remedy measures such as anti-dumping and countervailing actions tend to relate to greater trade levels as trade reductions become evident.

Fifth, industry fixed effects are included for the industrial characteristics of products at issue. The economic effects of legal rulings are also critically dependent on the industry. Protective measures on industries that are domestically and politically sensitive, such as agriculture, are inclined to show smaller economic effects. In contrast, industries that are less sensitive in this respect can produce larger economic benefits to complainants because it is easier and quicker to remove disputed measures.

Sixth, we carefully address issues concerning the time points. Various time points are used to account for the time-delaying of DSB proceedings and their durational effects. For example, a losing respondent needs a substantial amount of time to implement the recommendations of the WTO DSB; thus, we consider time frames from the year of the DSB decision report to the completion year of implementing the DSB rulings in the post-dispute analysis. On the other hand, we also need to consider the proper duration time of the DSB effect since its inception, which may fade in or out during market adjustment. Factoring into account various time treatment dummies with lags based on the duration times, we investigate the impact of WTO DSB outcomes on trade flow changes. This approach helps to capture the empirical effect of the WTO DSB outcomes, rather than simply detecting the changes in trade flows of disputed products.

Lastly, this paper addresses concerns over the possible endogeneity bias facing estimation for the DSB effect, noting that WTO disputes and DSB outcomes are not random or strictly exogenous. Reverse causality aside, even difference-in-difference estimation with country fixed effects is not a perfect solution for this endogeneity problem (Conconi et al., Reference Conconi, DeRemer, Kirchsteiger, Trimarchi and Zanardi2017). The omitted unobservable time-varying variables are arguably the most significant factors contributing to a potential endogeneity bias (Baier and Bergstrand, Reference Baier and Bergstrand2007). For example, when a DSB outcome is anticipated, trade flows are likely to be adjusted for many other reasons apart from the DSB effect. Particularly, this re-adjustment is driven not only by treatment per se but also by changes of time-varying conditions associated with each respondent and its trade partners.

This estimation technique can be understood as country-and-time fixed effects to account for the changes in the unobservable multilateral resistance terms in the panel gravity modeling of Baier and Bergstrand (Reference Baier and Bergstrand2007). For the treatment of the presence of endogeneity, we make our analysis valid and robust with the full use of fixed effects for time-invariant and time-varying variables using difference-in-difference (differenced) estimations while accommodating the anticipatory/reactionary behaviors of markets using various timings of the dispute to the regression models. Controlling the timings and time-varying situations in respondent markets can crucially help to alleviate the concerns of endogeneity bias.Footnote ²

In summary, we establish an empirical estimation model designed to explore the differential effects of WTO disputes on the dependent variable. The model considers DSB status, case, time, group-specific differences, and timing while controlling for other unobservable factors that affect the variables of interest, employing various fixed effects as explained above. The first equation is used to identify the economic benefit of DSB's legal proceedings, regardless of the different roles of WTO member countries in a disputed case (i.e., treating all countries symmetrically).

(1)$$\eqalign{& \ln \left( {IM_{ij,\; t\left( {{\rm \tau },{\rm \;} T} \right)}} \right) \cr & = \beta _0 + \beta _1 {\rm WTODS}\_{\rm POST}_{it} + \beta _2 {\rm DSB}_{it} + \; \beta _{12} ({\rm WTODS}\_{\rm POST}_{it} \times {\rm DSB}_{it} ) \cr &\quad + {\rm \; \delta} _{r\left( t \right)\;} + {\rm \delta} _{c\;} + {\rm \delta} _{\,j\left( t \right)\;} + {\rm \delta} _{rc\;} + {\rm \; \delta} _{Agreement} + {\rm \delta} _{HS\_industry} + {\rm \delta} _{year} + \epsilon _{rjt}} $$

where the dependent variable, ln(IM _ij,t(τ,T)), is the logarithm of ‘Real Import (RIM)’ of the disputed product and countries in year t (τ, T) ∈ {τ −2, τ −1, τ , T, T + 1, T + 2, T + 3}; i indicates dispute cases where each case corresponds to one respondent r with the dispute-related product(s) p such that i = (r, p); j indicates trading partner countries with respondent r whose exports are product(s) p in year t and j ∈ {complainant, third party, others (non-participant countries)} – all WTO member countries ‘j’ varying by year ‘t’. ‘τ’ indicates the year when a dispute was initiated while ‘T’ refers to the year when a dispute was completed. In our estimation model, all missing observations are replaced with zeros, such as ln(1 + real imports), assuming that all non-reported information for trade values at the HS 6-digit sectoral level is zero (Helpman et al., Reference Helpman, Melitz and Rubinstein2008). Following the heterogeneous firm model (Helpman et al., Reference Helpman, Melitz and Rubinstein2008) to explain zero trade flows across countries with fixed costs of exporting and variable trade costs, we assume that zero trade is associated with high bilateral fixed costs of trade. In the context of our analysis, zero imports of disputed products in respondent countries and asymmetric trade flows between country pairs are products of trade barrier costs stemming from allegedly protectionist measures by respondents.

${\rm WTODS}\_{\rm POST}_{it} \equiv {\rm WTODS}\_{\rm POST}_{\left( {T + t} \right)} $ refers to the time dummy after the WTO DSB's final decision is made; taking a value of 1 when t = 0, 1, 2 and/or 3. We use four period dummies to capture DSB timing effects on imports as opposed to just the base period dummy; taking a dummy which has value 1 in years T0, T + 1 and T + 2 for (T0~2), in T + 1 and T + 2 for (T1~2), in T + 2 and T + 3 for (T2~3), and in T + 1, T + 2, and T + 3 for (T1~3). To test different delayed and timing effects, the regression models use various lagged time variables with different time points. For example, (T + 1~2) refers to a period of one to two years (1~2 years) after a dispute completely ends in WTODS_POST_(T+1~2). To consider a two-year time lag for the effect, (T + 2~3) is used to specifically study a period between two and three years after the DSB's final decision. (T + 1~3) is used to denote a period spanning from the first year to the third year following a dispute's conclusion.Footnote ³

DSB is a dummy variable for measuring the result of WTO DSB proceedings. DSB_i indicates one of the final statuses of a WTO dispute that has concluded for a specific respondent product i. The WTO disputes fall into five non-overlapping possible DSB outcome categories: LW – legal win; MA – mutually agreed or settled; Non-LW – ‘not guilt’; WD – withdrawn; and DROP – dropped. These categories are classified according to official indications by the WTO DSB as ‘Current Status’ of a dispute. There are a small number of subsets under ‘dropped’ such as ‘no further proceeding,’ ‘authority for panel lapsed,’ and so on.Footnote ⁴ For example, LW takes the value of 1 if the measures at issue are found to be inconsistent with the WTO Agreements (i.e., a WTO DSB case where a complainant wins), and equals 0 otherwise.

The interaction term between DSB and POST (β ₁₂) estimates a DSB outcome effect on trade in the post years of DSB, showing whether the WTO DSB's legal ruling actually promotes free trade by increasing overall imports in respondent markets compared to disputes that are initiated but dropped. Under the null that the DSB's legal outcomes (LW, MA, Non-LW, and WD compared to DROP) do not affect imports, β ₁₂= 0, the coefficient β ₁₂ estimates the difference in import changes dependent on DSB's outcomes. Inβ ₁₂(DSB x POST), particularly when the DSB outcome is LW, the coefficient is expected to be positive because a legal win should have a positive effect on overall import liberalization by removing trade barriers and promoting multilateral free trade. However, β ₁₂ for MA is ambiguous and dependent on whether a mutually agreed solution favors freer trade. It is expected to be negative for mutually agreed cases when disputed trade barriers remain in place.

Furthermore, assuming that the variance of variables differs by times and legal groups (complainant, third party, and non-participant), the following regression framework allows us to examine the economic effects of WTO disputes, depending on the status of DSB outcome and participantsFootnote ⁵:

(2)$$\eqalign{& \ln \left( {IM_{ij,\; t\left( {{\rm \tau },{\rm \;} T} \right)}} \right) \cr & = \beta _0 + \beta _1 {\rm COMP}_{ij}^{\rm \;} + \beta _2 {\rm THIRD}_{ij}^{\rm \;} + \beta _3 {\rm WTODS}\_{\rm POST}_{it} + \beta _4 {\rm DSB}_{it} \cr & \quad \quad + \matrix{ \; \beta \cr} _{13} \left( {{\rm COMP}_{{\rm ij}}^{\rm \;} \times {\rm \; WTODS}\_{\rm POST}_{it}} \right) + \beta _{23} \left( {{\rm THIRD}_{{\rm ij}}^{\rm \;} \times {\rm WTODS}\_{\rm POST}_{it}} \right) \cr & \quad \quad + {\rm \;} \beta _{14} \left( {{\rm COMP}_{{\rm ij}}^{\rm \;} \times {\rm DSB}_{{\rm it}}} \right) + \beta _{24} \left( {{\rm THIRD}_{ij}^{\rm \;} \times {\rm DSB}_{{\rm it}}} \right) \cr & \quad \quad + \beta _{34} \left( {{\rm WTODS}\_{\rm POST}_{it} \times {\rm DSB}_{{\rm it}}} \right) + {\rm \;} \beta _{134} \left( {{\rm COMP}_{{\rm ij}}^{\rm \;} \times {\rm DSB}_{{\rm it}} \times {\rm \; WTODS}\_{\rm POST}_{it}} \right) \cr & \quad \quad + \beta _{234} \left( {{\rm THIRD}_{{\rm ij}}^{\rm \;} \times {\rm DSB}_{{\rm it}} \times {\rm WTODS}\_{\rm POST}_{it}} \right) \cr & \quad \quad + {\rm \; \delta} _{{\rm r}\left( {\rm t} \right){\rm \;}} + {\rm \delta} _{{\rm c\;}} + {\rm \delta} _{{\rm j}\left( {\rm t} \right){\rm \;}} + {\rm \delta} _{{\rm rc\;}} + {\rm \; \delta} _{{\rm Agreement}} + {\rm \delta} _{{\rm HS}\_{\rm industry}} + {\rm \delta} _{{\rm year}} + {\rm \epsilon} _{{\rm rjt}}} $$

In this equation, the complainant and third party dummies are expressed as COMP _ij (1 if a country holds the status of complainant(s) in the dispute, 0 otherwise), THIRD _ij (1 if a country holds the status of third party, 0 otherwise) respectively in order to distinguish the role-specific characteristics of affected groups in each dispute case i. Compared to equation (1), equation (2) highlights a distribution of the DSB effect in POST dispute periods (WTODS_POST x DSB) by legal groups and differentiates expected DSB gains according to each country's legal role in the dispute.

The second-level interactions act as a control for changes between different times and within groups. β ₁₃ shows the differences in the complainant's exports (i.e., the respondent's imports) before and after the dispute, whereas β ₂₃ measures the corresponding differences of the third parties. β ₁₄ represents the impact of a dispute outcome on the exports of respondents according to the case outcome and β ₂₄ measures the corresponding impact on third parties when cases reach a final resolution. β ₃₄ controls for the general impact on exports (regardless of group) following final DSB outcomes.

Finally, the third-level interactions, β ₁₃₄ and β ₂₃₄, capture all variation in imports, specific to the affected groups (relative to other countries) in the affected cases (four DSB outcome cases) in the years after a WTO dispute (relative to before the WTO dispute). Country, group, and case specific covariates, especially the time-invariant and time-varying characteristics, are included for the respondent (δ_r(t)), complainant (δ_c), member (δ_j(t)), and respondent-complainant pair (δ_rc) dummies that are fully saturated country-specific fixed effects. Dummies of agreement for 12 agreement categories are denoted as δ_Agreement, and industry dummies for HS 2-digit level are denoted as ${\rm \delta} _{HS\_industry} $, and separate year dummies for 1995 ~ 2010 (δ_year) are included as controls, as discussed above.

5.1 Import changes in post WTO dispute

This section presents the results of estimation based on the discussions above. Tables 4 and 5 contain the results of the estimation equations (1) and (2), respectively. All estimations with robust standard errors clustered by partner–member countries are in parentheses, controlling for country(member), agreement, industry, and year fixed effects. The trend of each group (complainant, third party, and non-participant) before and after the dispute is fixed to take into account the periodical differences among countries under the DD (difference-in-difference) type of application, presented in Table 4 based on equation (1). Moreover, by defining affected cases (i.e. legal win, non-legal win, mutually agreed, and withdrawal cases) and reference cases (i.e. dropped cases) within each group in the estimations (difference-in-difference-in-difference: DDD), the regression results in Table 5 using equation (2) provide statistical evidence that can shed light on our research question of whether the legal effect of WTO DSB has an impact on trade flow.

Table 4.

Q1: Legal winning through WTO DSB on trade flow

Note: Estimated coefficients based on equation (1) are from least squares regressions. * denotes statistical significance at 10%, ** at 5%, and *** at 1% level. Robust standard errors clustered by member countries are in parentheses. InRIM takes natural log (real import value + 1).

WTODS_POST is calculated based on the notification year (T ₀) of the WTO DSB's finalized outcomes. The outcomes are categorized into five cases: (1) LW: legal win (including AB: Appellate Body ruling), (2) MA: mutually agreed, (3) Non-LW: no legal win, (4) WD: withdrawal, and (5) DROP: dropped cases. DROP is used as a reference category.

Columns (5) ~ (8) show the regression results using respondent time-varying (Respondent x Year_T) and membertime-varying (Member x Year_T) instead of country time-invariant fixed effects.

Table 5.

Q2: Effect of legal winning through the WTO DSB on trade flow: who gains more?

Note: Estimated coefficients based on equation (2) are from least squares regressions. * denotes statistical significance at 10%, ** at 5%, and *** at 1% level. Robust standard errors clustered by member countries are in parentheses. InRIM takes natural log (real import value + 1).

WTODS_POST is calculated based on the notification year (T ₀) of the WTO DSB's finalized outcomes. The outcomes are categorized into five cases: (1) LW: legal win (including AB: Appallate Body ruling), (2) MA: mutually agreed, (3) Non-LW: no legal win, (4) WD: withdrawal, and (5) DROP: dropped cases. DROP is used as a reference category.

Participant dummies according to the legal position of WTO DSB are indicated as COMP (complainant), THIRD (third-party); and, non-participants are used as reference group. First- and second-order interaction terms are all included in the estimations, but not reported. Columns (5)–(8) shows the regression result using both respondent time-varying (Respondent x WTODS-POST Year_T) and member time-varying (Member x WTODS-POST Year_T) instead of country time-invariant fixed effects.

Columns (1)–(8) of Table 4 present the estimates from equation (1), in a relative sense, for the DSB outcome dummy in the sample by measuring the DSB effects during the period from year zero (T ₀) up to year three (T ₀+3), the time point (T ₀) where the dispute completely ends. Columns (5)–(8), in particular, show regression results using respondent time-varying (Respondent x Year_T) and member time-varying (Member x Year_T), instead of country time-invariant fixed effects, to account for endogeneity from the omitted variable bias mentioned in Section 3.2. It makes little difference to the results. The WTODS_POST variables are counted based on the notification year (T ₀) of the WTO DSB's outcomes categorized as (1) LW: legal win (including AB: Appellate Body ruling), (2) MA: mutually agreed, (3) Non-LW: no legal win, (4) WD: withdrawal, and (5) DROP: dropped cases. DROP cases are used as a reference category. Additionally, a time dummy, WTODS_INI, is used to control for the trend in a dispute initiated year.

The estimation incorporates the WTO dispute-initiated year dummy (WTODS_INI) to show whether there is a difference in trade behavior in the first year of the WTO dispute initiation. In the fifth row, WTODS_INI estimations show that there are statistically significant positive trends in the year of the dispute initiation except for columns (1) and (5), which include the year of WTO dispute completion (T ₀). Except for the year effects for the dispute initiation year, there are two reasons to include the WTODS-INI variable in the model: to deal with the issues of (i) reverse causality and (ii) any unusual trade flows during the year of dispute. WTODS_POST captures the post-WTO dispute effect by examining the years after completion of a WTO dispute [(T + 0~2), (T + 1~2), (T + 2~3), or (T + 1~3)]. The coefficients on various lagged and durational variables of the WTODS-POST in all columns are positive and significant. According to this result, there is an increase in imports following a dispute's completion.

5.2 Does the WTO DSB's legal win increase trade overall?

To analyze the impact of a legal win through the DSB on trade in the years of WTODS_POST based on equation (1), we need to look at the second-order terms (WTODS_POST x DSB dummies). Rows 1– 4 in Table 4 present estimations of the second-order terms. Dependent on the different DSB outcomes, these second-order terms indicate how much the imports of a respondent (from each source) increase following a dispute's conclusion. When the difference in DSB outcomes is considered, the imports of the LW and Withdrawal cases increase compared to those of the reference case (i.e. Dropped case).

WTODS_POST x LW (the second-order interaction term capturing the impact of legal winning in the post-dispute period) is significant and positive. This implies that LW cases in the post-WTO DSB dispute years are consistently associated with increases in imports as greater trade growth is expected when the case is pursued than when it is dropped. Specifically, in years subsequent to the WTO DSB's ruling implementation, respondents experience higher import values ranging from 8.1% in column (5) to 18.9% in column (2). This result is consistent with the argument that once a dispute is initiated and ended with rulings against a respondent's measures, the respondent abides by the decision and removes trade restrictions, allowing greater access to the market (Bown, Reference Bown2004, Reference Bown2009; Bown and Reynolds, Reference Bown and Reynolds2017). While DSB cases that conclude with a legal win are associated with greater levels of trade liberalization, dropped cases tend to be associated with a smaller degree of potential trade liberalization, not just on the bilateral trade level, but also multilaterally.

Particularly when applying country time-varying fixed effects, the coefficients of WTODS_POST x Non-LW and WTODS_POST x Withdrawal WD are also positive and significant. The findings show that when resolving disputes through the formal DSB, the WD increases overall imports compared to the dropped, non-pursued disputes. For the withdrawal cases, the negatively significant coefficient of WD shows that withdrawal cases have lower trade growth than dropped ones, but the overall growth is still positive because of the strongly positive general WTODS-POST effect. This implies that imports increase in the case of WD once the measures at issue go through the dispute settlement system, assuming that officially withdrawing from a case is a voluntary action of complainants (In our study, unofficially reported withdrawal cases are counted as dropped cases). Since the costly WTO disputes are initiated with an official consultation request by complainants, a notification of withdrawal is likely to be announced either when the market access of respondents becomes sufficient or when complainants, at some point during the DSB proceedings, realize that disputed measures are in fact not in violation of WTO rules. Significantly, positive coefficients for WTODS_POST x Non-LW cases seem to be peculiar, given that Non-LW cases are thought to involve disputed measures at issue and are subject to more stringent legal contest. However, even Non-LW may offer greater legal clarity and so encourage trade by providing a clear interpretation of WTO rules previously subject to dispute. Trade is likely to grow if disputed measures are found to be consistent with WTO agreements, as with Non-LW outcomes. In summary, as opposed to dropped cases, disputes officially resolved as LW, Non-LW, or withdrawal under the WTO DSB lead, on average, to increases in imports from all WTO members in respondent markets and favor multilateral trade liberalization.

Under the DSB system, disputes ending in officially mutually agreed solutions are associated with significant import decreases. Mutual settlements can be made both before and after the WTO ruling, and cases can be ended both with and without official notification. However, in this analysis, only officially announced mutually agreed cases are included. The total number of mutually agreed cases which are used in this estimation is 61, as shown in Tables 2 and 3. In cases where parties seek to end the dispute by settlement in order to reduce legal costs, mutual agreements are usually made right after the initiation of a dispute, within a year of the consultation request. ^{Footnote 8} Parties, for example, may choose to agree to withdraw from the WTO DSB proceedings to enjoy exclusive bilateral benefits. Thus, mutually agreed cases are usually expected to have a positive benefit for the complainant; otherwise, legal proceedings would continue. However, mutual agreements may also limit bilateral trade liberalization and work favorably for respondents. In short, both cases may promote only bilateral trade flows, not multilateral trade liberalization. It is for this reason that our empirical framework estimates a significantly negative MA coefficient.

5.3 Who gains more from the rulings?

The following results focus on our second empirical question of which group is most benefited by the efficacy of DSB outcomes. If we can find a systematic variation in the trade gains of different participant groups following disputes, this could help to understand the overall utility of the system from the perspective of legal contestants. Table 5 shows the results of regressions based on equation (2) when different legal participants are included as the complainant and the third party (with complainant, third party, vs. non-participants as a reference group) with a full model using a three-level interaction of DSB dummies.

Table 5, in the second and third rows under WTODS_POST x LW, displays different legal effects according to the legal status of dispute participants relative to the effect for non-participants.

The first noteworthy points are the third-order interaction terms (β ₁₃₄: WTO_POST x LW x COMP) and (β ₂₃₄: WTO_POST x LW x THIRD). They show what happens to the imports from the complainant and third party groups to the respondents relative to imports from non-participants when the WTO case is legally won. In all columns (1)–(8) of Table 5, the coefficients appear to be strongly significant and negative. There are apparently smaller proportional trade gains from the WTO DSB's rulings for the major legal stakeholders (complainants and/or third-parties) than for non-participants.Footnote ⁹ This interpretation is more precise when viewed with the second-order interaction term: WTODS_POST x LW (which captures the impact of a legal win in the post dispute period) and the three-order interaction term: WTODS_POST x LW x COMP (or THIRD). WTODS_POST x LW is very significant and positive with a large coefficient. This implies that a legal win leads to economic gains (i.e., import liberalization of the respondent) once the DSB ruling (LW) is implemented. However, the gains are more greatly distributed to non-participating WTO members rather than to the complainant and third party groups. The statistically significant and negative coefficients of β ₁₃₄ (WTO_POST x LW x COMP) and β ₂₃₄ (WTO_POST x LW x THIRD) indicate that the distribution of gains between groups is less favorable for legal contestants. For instance, the sum of the two coefficients ($\beta _{34}^{LW} $: WTO_POST x LW and $\beta _{134}^{LW} $: WTO_POST x LW x COMP) represents the effect for COMP, as a complainant group (for the effect for THIRD would be the sum of $\beta _{34}^{LW} $ and $\beta _{134}^{LW} $: WTO_POST x LW x THIRD). Using the estimated coefficients of column (1) for COMP, a precise interpretation is that when a ruling is made against the respondent, the complainant experiences, on average, a 50% (= e^0.116-0.811 −1) smaller proportional increase in trade (exports to respondents) than do non-participants during years (T + 0~2) after a dispute ends.

As discussed in Section 2, the results above may be attributable to the multilateral nature of the WTO system. The MFN principle is a cornerstone for effective trade liberalization in the WTO system. It allows even free riding for non-participant countries to benefit from legal wins and gain new exporting opportunities. Our results reaffirm Ludema and Mayda's (Reference Ludema and Mayda2009) findings on this free-rider problem in relation to WTO MFN externalities. Evidence of the free-rider problem becomes apparent following a legal win for a complainant (LW), when a disputed market becomes more open and competitive (less concentrated) than before.Footnote ¹⁰

The econometric results of Table 5 are in line with the results of Table 4. For the complainants, the proportional additional benefits from the WTO DSB proceedings appear to be relatively small. From each complainant's viewpoint, however, the expected trade gains may not be insignificant considering the absolute level of trade involved and the counterfactual of not contesting a case (actual declines in trade); hence, it may still remain rational to take a complainant to the WTO dispute settlement system. Through WTO disputes, complainants do indeed prevent further trade damage and increase their exports.Footnote ¹¹ We find that, so long as a legal win can result in significant liberalization of the respondent's market and raise total imports for respondents (WTO_POST x LW), complainants and third parties do not necessarily lose their markets. It is simply noted that the proportional trade benefits to other parties are relatively larger than the gains earned by those involved in the legal proceedings.

The results for mutually agreed cases in Table 5 show negatively significant coefficients for both WTODS_POST x MA x COMP and WTODS_POST x MA x THIRD, especially in column (3)/(4) and (7)/(8), indicating that there are no direct benefits to either the complainant or third party. This suggests that mutual settlements may prevent trade liberalization not only between agreed parties but for others as well, as shown in column (7) of Table 5. Rather, mutual settlements seem to hamper the multilateral trade liberalizing effect of the WTO DSB. In contrast to legal winning cases, mutually agreed cases appear to result in neither multilateral liberalization nor increased exports for complainants or third parties.

5.4 Further robustness checks and other issuesFootnote ¹²

In the previous sections, we account for changes in the number of WTO members by assuming that all WTO members are possible beneficiaries from the DSB's rulings. Although not all WTO members have significant economic interests in the products at issue, the inclusion of all exporters that hold WTO membership is important for considering new entry of bilateral trade between the respondent and potential exporters. The following robustness checks are designed to dispel the notion that our results simply reveal the dominance of a large number of small exporters with greater extensive/intensive margin growth relative to a small number of major pre-existing exporters that sell in large volumes.

In this section, we conduct a robustness exercise, first, to extend the DSB effects to non-WTO members and, second, to restrict the impact of the DSB to only ‘major’ exporters. The main findings are that LW cases are associated with trade creation not only from the extensive margin growth for non-WTO members, but also from the intensive margin growth for even non-participants.

To confirm these points, columns (1) to (3) of Table A1 present overall LW DSB effects on trade after the dispute using DD type estimation, while columns (4) to (6) estimate DSB effects according to legal participatory status in the dispute. First, columns (1) and (4) examine the spillover effects of the DSB on ROW (Rest of World). The results of DD and DDD show similar patterns to those in Tables 4 and 5. Second, columns (2)/(3) and (5)/(6), which restrict the data sample to sets of ‘major exporters’ defined respectively as lying within the top 25 or 10 percentiles of all exporters, suggest that our main findings on the overall trade effects of LW captured by second level interaction terms remain robust and are more or less consistent in the third level interaction terms. In the third level interaction terms, however, the coefficients of the third parties in columns (5)/(6) are no longer statistically significant.

In the previous pooled sample of all countries, trade liberalization levels are different following a dispute, especially in cases where the WTO DSB's ruling(s) find the respondent to be in violation. This result turns out to be statistically significant and negative for complainant/third party groups, meaning that in a relative sense, the non-participant group experiences more positive gains. To further check the sensitivity of the overall results, we proceed to another set of regressions by excluding three countries (China, US, and EU) along with AD/CVD dispute cases.

First, the rationale for excluding the sub-samples of the three WTO members is that average and growth of China's exports may be significantly different from other WTO member countries. Also, China has demonstrated rapid export growth and experienced relatively large increases in respondent markets after rulings, whereas the US and EU have been the most frequent users of the WTO DSB as both complainants and respondents.

Second, one of the main concerns is that our results only capture trade diversion effects, rather than trade creation from the rulings. For example, when AD or CVD duties are imposed, other players may replace major exporters in the market – an effect which should not be seen as a trade gain resulting from DSB rulings. Accordingly, we check whether the LW effect is still valid when excluding AD/CVD cases, bearing in mind that AD/CVD cases account for a significant portion of all WTO disputes (approximately 40%). This robustness-check using a sub-group analysis also helps us to verify whether multilateral liberalizing effects also result from removing National Treatment-type non-discriminatory measures after DSB rulings.

Table A2 shows that excluding the AD/CVD sub-group does not change the previous pattern of trade creation. Excluding the AD/CVD sub-group, trade creation from LW effects remains unchanged in the post-dispute years. The rest of the results for the three interaction terms are very similar, though the coefficients in the No-US and No-AD/CVD equations become larger. This means that the market re-adjustment effects, created by the elimination of trade-restrictive features of the NT measures (other than AD/CVD), may generate smaller gains for complainants and larger gains for non-participants. In the sub-group that excludes the US as a respondent, negatively larger coefficients in column (2) and (6) imply that the US has a stronger than average resistance to market liberalization.

Our findings are consistent with the findings of Ludema and Mayda (Reference Ludema and Mayda2009, Reference Ludema and Mayda2013). They argue that GATT/WTO exporting countries have incentives to free ride due to the so-called MFN externality. Connecting WTO MFN tariff-negotiation to trade liberalization, their empirical evidence shows that the higher the import concentration is in a certain product, the greater the market access and import growth that result from the tariff negotiation. Bown and Reynolds (Reference Bown and Reynolds2017) also suggest that the free-rider problem prevails when sectors involve a concentrated number of exporting countries. If WTO dispute settlement is considered as a form of trade liberalizing negotiation, our empirical results on the WTO DSB's LW outcome and import growth are consistent with the findings of Bown and Reynolds (Reference Bown and Reynolds2017) and Ludema and Mayda (Reference Ludema and Mayda2009, Reference Ludema and Mayda2013).

Table A1 presents the size of estimated coefficients from the sub-sample of Major 10th percentile exporters. The coefficient values in column (3) for WTODS_POST x LW are greater than those for less concentrated exporters in the sub-samples of Major 25th percentile and ROW (Rest of World). In addition, given that AD/CVD cases tend to occur in relatively concentrated sectors such as automobile, steel, and electronics sectors, the free-riding problem is expected to be reduced. Our results in Table A2 provide support for this hypothesis. When excluding AD/CVD disputes in the sample, the coefficient estimate of WTODS_POST x LW is the smallest compared to other models, indicating that AD/CVD disputes involve lower competition. In other words, trade gains from DSB rulings are small when excluding these disputes. The estimated results also show that the complainants and third parties make lesser gains than non-participants.

By using only AD/CVD cases, we have conducted another robustness analysis. Table A3 shows that the free-riding problem seems to be weaken in AD/CVD disputes where import markets tend to be more highly concentrated. It is noted that when complainants win a dispute, they reap the biggest trade gain among other exporters: WTODS-POST x LW x COMP is significantly positive. Interestingly, mutually agreed cases provide the most trade gains for complainants, a result which differs from previous regression results that use the full sample. When restricting the analysis sample to AD/CVD cases, DSB disputes ending with a mutually agreed solution (MA) involves a seemingly less free rider problem. As Table A3 shows, the effects of the WTO DSB possibly depend on the behavior of exporters that may vary by sector and agreement. This study, however, has not allowed for such heterogeneity in sector and agreement as it would require precise and sufficient data to do so (Conconi et al., Reference Conconi, DeRemer, Kirchsteiger, Trimarchi and Zanardi2017).