3.1 Baseline Empirical Specification and Variables

Our empirical analysis adopts a gravity model, widely used to study the magnitude and patterns of bilateral trade activities (Anderson and van Wincoop, Reference Anderson and van Wincoop2003; Rose, Reference Rose2004, Reference Rose2005; Behrens et al., Reference Behrens, Ertur and Koch2012). Following Rose (Reference Rose2005) model for trade volatility, we start with a benchmark specification given by:

(1)$$\displaystyle{{\sigma {( {M_{ijt}} ) }_\tau } \over {\mu {( {M_{ijt}} ) }_\tau }} = \beta _0 + \beta _1\mu ( {BothWTO_{ijt}} ) _\tau + {\gamma }^{\prime}\mu ( {Z_t} ) _\tau + \epsilon _\tau , \;$$

where M _ijt represents the log value of imports in country i from country j in year t; $\sigma ({\cdot} ) _\tau $ and $\mu ({\cdot} ) _\tau $ are standard deviation and mean operators over a certain time-interval period τ, respectively. For our baseline regression, we look at trade volatility calculated over five-year periods, where τ ∈ {1950 − 1954, 1955 − 1959, ⋅ ⋅ ⋅ }. Robustness checks are provided to also look at trade volatilities constructed over 10- and 15-year periods.

Proceeding to controls included in equation (1), the dummy variable BothWTO _ijt takes the value of 1 if both importing (i) and exporting (j) countries are WTO members in year t, and 0 otherwise; Z is a vector of other control variables commonly chosen in a gravity model, including log value of real GDP of i and j (lnGDP _it  and lnGDP _jt), log value of real GDP per capita of i and j (lnGDPPC _it and lnGDPPC _jt), log value of geographical distance between i and j (lnDist _ij), whether i and j belong to the same regional trading agreement in year t (RTA _ijt) or the same currency union (CU _ijt), whether i offered generalized system of preferences (GSP) to j in year t (GSP _ijt), whether country i has ever been a colony of j (Colony _ij), whether country i has ever been a colonizer of j (Colonizer _ij), whether i and j have been colonized by the same colonizer (ComColony _ij), whether i was currently a colony of j in year t (CurColony _ijt), whether i was currently a colonizer of j in year t (CurColonizer _ijt), whether i and j share a common language (ComLang _ij), and whether countries i and j share a common religion (ComRelig _ij). In bilateral trade analysis, a measure reflecting trade frictions between a specific country pair and all other trade partners in addition to trade frictions between i and j needs to be included, referred to as the ‘multilateral resistance effect’ in Anderson and van Wincoop (Reference Anderson and van Wincoop2003). Therefore, we follow the literature by including the measure of remoteness (Remote _ijt) to control for the multilateral resistance (Head, Reference Head2003; Hummels, Reference Hummels2007; Bacchetta et al., Reference Bacchetta, Beverelli, Cadot, Fugazza, Grether, Helble, Nicita and Piermartini2012). A country remoteness is defined as the log distance of country i to the rest of the world weighted by all the other countries’ shares of world GDP in year t (Remote _it). The remoteness of a dyad (Remote _ijt) is simply the sum of Remote _it and Remote _jt. We also control for other geographical characteristics for countries i and j in the regressions: whether the pair of countries share a border (Border _ij), log value of the geographical area of i and j (ln(Area _i) and ln(Area _j)), the number of landlocked countries in a pair (Landlock _ij = 0, 1,  or 2), and the number of island nations in a pair (Island _ij = 0, 1,  or 2). Similar to the WTO variable, all other controls are also averaged over time period τ.

The data used in our analysis are from the Direction of Trade Statistics by the IMF, the CEPII, and the Penn World Table 9.1. Our sample covers 189 countries over the period 1950–2017. Appendix Table A (see on-line Supplementary material) lists the countries included in our data set, along with the year of each country's GATT/WTO accession when applicable.

3.2. Modeling Multilateralism of the WTO: The Spatial Dependence

It is important to recognize the potential dependence of trade volatilities across various dyads in an integrated global trading system. To account for spatial correlation of trade volatilities, we modify our baseline regression as equation (2) by including spatial lag trade volatility measures (Li and Vashchilko, Reference Li and Vashchilko2010; Neumayer and Plümper, Reference Neumayer and Plümper2010a; Cho et al., Reference Cho, Dreher and Neumayer2014). A brief description of a general spatial lag model is provided in the Appendix (see on-line supplementary material). For ease of illustration, we represent our trade volatility measure from now on as $Vol( {M_{ijt}} ) _\tau = \sigma ( {M_{ijt}} ) _\tau /\mu ( {M_{ijt}} ) _\tau $:

(2)$$Vol( {M_{ijt}} ) _\tau = \beta _0 + \rho \sum\limits_{km\ne ij} {w_{ij-km}Vol{( {M_{kmt}} ) }_\tau + {\gamma }^{\prime}Z_\tau + \epsilon _\tau , \;} $$

where ij ≠ km, and $Z_\tau = \mu ( {Z_t} ) _\tau $. Equation (2) indicates that, in addition to bilateral characteristics, the volatility of trade between two countries i and j also depends on the volatility of trade between country k and country m (i.e., $Vol( {M_{km}} ) _\tau $), weighted by a matrix capturing the connectedness between dyad i–j and dyads k–m (i.e., w _ij−km). The sign and strength of the spatial dependence are then reflected by the spatial lag coefficient ρ.

Our sample includes 189 countries, conceptually generating about 35,532 trading dyads annually since the volatility of imports in country i from country j is different from the volatility of imports in country j from country i. In this case, the spatial lag measure in equation (2) implies that trade volatility of a dyad i–j, in theory, would be influenced by trade volatilities of other 35,530 trading pairs each year, which can make the calculation of the spatial volatilities over a long sample span extremely computationally costly.Footnote ⁵ As a result, we aim to focus on trading pairs that would most likely influence trade volatility between country i and country j and keep the construction of spatial lags computationally practical. We consider four spatial lag volatility measures to reflect trade volatilities of “other dyads” based on WTO membership. These spatial trade volatilities are represented as follows:Footnote ⁶

(3)$$V_{im\tau }^{{\rm WTO}} = \sum\limits_{m\ne j} {w_m^{{\rm WTO}} \cdot Vol{( {M_{imt}} ) }_\tau } $$

(4)$$V_{im\tau }^{{\rm NWTO}} = \sum\limits_{m\ne j} {w_m^{{\rm NWTO}} \cdot Vol{( {M_{imt}} ) }_\tau } $$

(5)$$V_{kj\tau }^{{\rm WTO}} = \sum\limits_{k\ne i} {w_k^{{\rm WTO}} \cdot Vol{( {M_{kjt}} ) }_\tau } $$

(6)$$V_{kj\tau }^{{\rm NWTO}} = \sum\limits_{k\ne i} {w_k^{{\rm NWTO}} \cdot Vol{( {M_{kjt}} ) }_\tau}. $$

For a trading dyad i–j, our first two spatial measures, $V_{im\tau }^{{\rm WTO}} $ and $V_{im\tau }^{{\rm NWTO}} $ in equations (3) and (4), are the trade volatilities between the same importing country i and other exporting countries m, where m ≠ j. The connectedness variable $w_m^{{\rm WTO}} $ in equation (3) is a dichotomous measure, equal to 1 if exporting country m is a WTO member and 0 otherwise. The connectedness variable $w_m^{{\rm NWTO}} $ in equation (4) takes the value of 1 if exporting country m is not a WTO member, 0 otherwise.

For example, the volatility of imports in the US (country i) from Thailand (country j) could be correlated with the volatility of imports in the US from India as well as with the volatility of imports in the US from Venus (a hypothetical non-WTO member). In this example, the term $V_{im\tau }^{{\rm WTO}} $ in equation (3) is the volatility of imports in the US from India (a WTO member), and the term $V_{im\tau }^{{\rm NWTO}} $ in equation (4) is the volatility of imports in the US from Venus. The actual $V_{im\tau }^{{\rm WTO}} $ and $V_{im\tau }^{{\rm NWTO}} $ terms we include in our model are calculated based on the import volatilities of all dyads i–m in our sample.

Moving on to equations (5) and (6), for a dyad i–j, these spatial measures represent trade volatilities between other importing countries k and the same exporting country j, where k ≠ i. The dichotomous connectedness measure $w_k^{{\rm WTO}} $ in equation (5) is set to 1 when importing country k is a WTO member and 0 otherwise. The connectedness measure $w_k^{{\rm NWTO}} $ in equation (6) is 1 when importing country k is not a WTO member and 0 otherwise. These measures suggest, for instance, the volatility of imports in the US from Thailand is influenced by the volatility of imports in Germany from Thailand and also by the volatility of imports in Mars (a hypothetical non-WTO member) from Thailand. In this example, the term $V_{kj\tau }^{{\rm WTO}} $ captures the volatility of imports in Germany (a WTO member) from Thailand, and the term $V_{kj\tau }^{{\rm NWTO}} $ is the volatility of imports in Mars from Thailand.Footnote ⁷ Again, the actual $V_{kj\tau }^{{\rm WTO}} $ and $V_{kj\tau }^{{\rm NWTO}} $ terms in our model are calculated based on the import volatilities of all dyads k–j in our sample. Figure 2 illustrates graphically how we define these four spatial trade volatilities for a dyad i–j and how those spatial measures can affect the trade volatility between countries i and j. Based on the above discussion, we rewrite the panel data model (2) as follows:

(7)$$Vol( {M_{ijt}} ) _\tau = \beta _0 + \rho _1V_{im\tau }^{{\rm WTO}} + \rho _2V_{im\tau }^{{\rm NWTO}} + \rho _3V_{kj\tau }^{{\rm WTO}} + \rho _4V_{kj\tau }^{{\rm NWTO}} + {\gamma }^{\prime}Z_\tau + \epsilon _{\tau}.$$

We present the correlation matrix of variables in our baseline regressions in Table B and the correlation matrix of the bilateral trade volatility and four measures of spatial dyadic trade volatilities in Appendix Table C (see on-line supplementary material).Footnote ⁸ Summary statistics are provided in Table 1.

Figure 2. An example of spatial measures for dyad i–j, where the US represents importing country i and Thailand exporting country j.

Table 1. Descriptive statistics

4.1. Baseline Regressions without Spatial Dependence

Our baseline regression results regarding the effect of WTO membership on trade volatility without spatial measures are reported in Table 2. The three columns in Table 2 give results from OLS regression, regression with importer and exporter fixed effects (Ctry FE), and regression with dyadic fixed effects (Dyad FE), respectively. We prefer the specification with dyadic fixed effects, which controls for the unobserved heterogeneity in bilateral volatility for each dyad. Time fixed effects are included in all regressions.

Table 2. Results of WTO membership and trade volatility

Notes: All regressions are controlled for geographical factors: ln(Area _i), ln(Area _j), Border _ij, Landlock _ij and Island _ij. Clustered standard errors by countries i and j are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

The coefficient on BothWTO in Table 2 is consistently negative across all regressions and significant at the 1% level. These results suggest that if both trade partners are WTO members, they have more stable trade. Regression 2.3 indicates that the volatility of trade between two WTO members is, on average, 0.0058 units smaller than other dyads, a reduction equivalent to 8% of the average trade volatility reported in Table 1. The trade-stabilizing effect of the GATT/WTO, as discussed previously, can result from the GATT/WTO's role in creating and promoting a more transparent and predictable global trade system. The negative and significant coefficients on RTA, GSP, and CU suggest dyads participating in regional trading agreements, Generalized System of Preferences program, or currency unions also experience trade that is more stable. These findings on the effects of RTA, GSP, and CU on bilateral trade volatility are consistent with those in Rose (Reference Rose2005). The stabilizing effect of WTO membership appears small relative to the trade-stabilizing effect of RTA or GSP in regressions without dyad fixed effects. When dyad fixed effects are included, only GSP program participation seems to have a stronger negative effect on trade volatility than WTO membership. For instance, regression 2.3 shows that if an importing country offers GSP to its trade partner, trade volatility of this dyad would be on average 0.0073 units lower, approximately 10% of the sample average trade volatility, than a dyad where an importer does not provide GSP treatment to an exporter. These results suggest that, to a large extent, dyadic heterogeneities explain the effect of RTAs and CU on bilateral trade volatilities. In addition, the stronger effect of RTAs or GSP on trade volatilities in some cases might be caused by weak disciplines involved in WTO agreements in certain sectors, as suggested in Cadot et al. (Reference Cadot, Olarreaga and Tschopp2008). Further, many WTO members belong to at least one RTA. RTAs (or GSP) can offer more extensive product coverage and/or further reduction of trade barriers to member countries than to regular MFN countries, which might also explain the different effects of WTO and RTAs (or GSP) on trade volatilities (Crawford and Laird, Reference Crawford and Laird2001).

We also adopt a standard event-study approach to estimate the effect of WTO membership on the trade volatilities.Footnote ⁹ In particular, we closely follow the event-study methodology suggested by MacKinlay (Reference MacKinlay1997) to compare trade volatilities of a dyad before and after both countries become WTO members. We estimate the following model (see also Head and Mayer, Reference Head, Mayer, Gopinath, Helpman and Rogoff2014; Rose, Reference Rose2019):

(9)$$M_{ijt} = {\beta }^{\prime}Z + \mu _i + \chi _j + \delta _t + \epsilon _{ijt}, \;\;{\rm over}\;t = \{ {T-( {s + 10} ) , \;\cdots , \;T-10} \} $$

where μ _i, χ _j, and δ _t are importer, exporter, and time dummy variables, respectively, and ε _ijt represents the stochastic error term. The window of estimation (in years) is represented by s, and T is the year of the event when both countries i and j are members of WTO.

After we obtain coefficient estimates from equation (9), they are used to form trade volatilities around the time of the event (Moser and Rose, Reference Moser and Rose2014). In this case, we calculate trade volatility as the standard deviation of the residuals from equation (9) for any given pair of countries over the nth five-year period since the event in year T, referred to as σ(ε _ijt)_n. For example, σ(ε _ijt)₀ represents the initial trade volatilities over the period of T + 1 and T + 5, or over the first five-year period after both countries i and j join the WTO; σ(ε _ijt)₁ is the trade volatility in the period of T + 6 and T + 10, and so on. Similarly, σ(ε _ijt)₋₁ is the trade volatility over the period T − 4 and T; and σ(ε _ijt)₋₂ denotes the trade volatility over the period of T − 9 and T − 5. We then test if the trade volatility in any period n is significantly different from that in the period right before both countries were WTO members (i.e., n = −1). If WTO membership does reduce trade volatilities, the difference in trade volatilities, $\tilde{\sigma }_n = \sigma ( {\epsilon_{ijt}} ) _n-\sigma ( {\epsilon_{ijt}} ) _{{-}1}$, should be insignificant when n < 0, and become negative and significant when both countries join the WTO (i.e., n ≥ 0).

To ensure that the trade-stabilizing effect of the WTO does not depend on the length of our estimation windows, we estimate the model with 15-year (i.e., s = 15) and 25-year (i.e., s = 25) estimation windows. The results with 15-year and 25-year estimation windows are illustrated in Figures 3a and 3b, respectively. Both figures show that the difference in dyad trade volatilities is not statistically significant before both trade partners are WTO members (i.e., n < 0), and it becomes negative and significant once both countries become WTO members (i.e., n ≥ 0). These results again provide evidence supporting that WTO members tend to experience lower trade volatilities than non-WTO members.

Figure 3. Event study on trade volatility.

We also compare the changes in bilateral trade volatilities after one country joining the WTO and after both countries joining the WTO in an event study setting. The left panel of Figure 4 (4a) shows the change in trade volatility after one country joins the WTO, and the right panel (4b) displays the change in bilateral trade volatility after both countries become WTO members. Figure 4a shows that, in general, the bilateral trade volatility decreases after one country becomes a WTO member. However, the effect is not significant at the 5% level immediately after one partner joins WTO (i.e., n = 0 and 1). The difference in trade volatility becomes significant in n = 2 and 3. On the other hand, Figure 4b shows that, after both countries join the WTO, the bilateral trade volatility drops significantly compared to the case where only one country is a WTO member.

Figure 4. Trade volatilities after one country vs. two countries joining the WTO.

4.2. Regressions with Spatial Dependence

Next, we include in our model the spatial volatility measures defined in equations (3)–(6). To better interpret coefficients on these spatial measures and answer whether the comovement of trade volatility differs based on partners’ WTO membership, we restrict our sample of estimation to dyads within the WTO. Note that we still construct spatial volatilities with all dyads in the sample but run regressions only including trading pairs with both partners being WTO members. In this way, the estimated coefficients on $V_{im\tau }^{{\rm WTO}} $ and $V_{kj\tau }^{{\rm WTO}} $ will capture the strength of trade volatility dependence or comovement among WTO trading pairs and the estimated coefficients on $V_{im\tau }^{{\rm NWTO}} $ and $V_{kj\tau }^{{\rm NWTO}} $ will show the strength of trade volatility dependence between a dyad within the WTO (i.e., dyad i–j) and dyads with one non-WTO member.Footnote ¹⁰

As both countries i and j are WTO members in our new regressions, the variable BothWTO is irrelevant and excluded from all regressions. We also construct and include two combined terms $V_\tau ^{{\rm WTO}} $ and $V_\tau ^{{\rm NWTO}} $ for our analysis, where $V_\tau ^{{\rm WTO}} = V_{im\tau }^{{\rm WTO}} + V_{kj\tau }^{{\rm WTO}} $ and $V_\tau ^{{\rm NWTO}} = V_{im\tau }^{{\rm NWTO}} + V_{kj\tau }^{{\rm NWTO}} $.Footnote ¹¹ To save space, we only report coefficients on our spatial measures. Estimated coefficients on other controls are available upon request.

The results with spatial measures for WTO pairs are reported in columns 3.1 and 3.2 in Table 3. The estimated coefficients on spatial volatility measures are all positive and significant at the 10% level or better, well supporting the inclusion of spatial terms to control for the inter-dyad trade volatility dependence and an empirical specification allowing for multilateral interactions.

Table 3. Spatial dyadic trade volatility for WTO countries and non-WTO countries

Notes: All regressions are controlled for other control variables presented in regression 2.3 in Table 2. We define: $V_\tau ^{{\rm WTO}} = V_{im\tau }^{{\rm WTO}} + V_{kj\tau }^{{\rm WTO}} $ and $V_\tau ^{{\rm NWTO}} = V_{im\tau }^{{\rm NWTO}} + V_{kj\tau }^{{\rm NWTO}} $. Clustered standard errors by countries i and j are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

More importantly, we find a strong comovement of trade volatilities among trading pairs in general. The positive coefficients on spatial terms indicate that trade between two WTO members i and j is more stable if trade between the same country i and other WTO or non-WTO exporters m is more stable. Similarly, trade between countries i and j is more stable if trade between other WTO or non-WTO importers and the same exporting country j becomes more stable.

We then estimate the trade volatility comovements by restricting our sample to dyads outside the WTO (i.e., both countries i and j are non-WTO members) and report the results in columns 3.3 and 3.4 in Table 3. The estimated coefficients on $V_{im\tau }^{{\rm NWTO}} $ and $V_{kj\tau }^{{\rm NWTO}} $ in this regression show the trade volatility comovement among pairs outside the WTO, and those on $V_{im\tau }^{{\rm WTO}} $ and $V_{kj\tau }^{{\rm WTO}} $ capture volatility comovement between a non-WTO pair (i and j) and pairs with one WTO member.

For ease of discussion, we focus on coefficients on $V_\tau ^{{\rm WTO}} $ and $V_\tau ^{{\rm NWTO}} $ in regressions 3.2 and 3.4. The estimated coefficient on $V_\tau ^{{\rm WTO}} $ drops from 0.485 in regression 3.2 to 0.328 in regression 3.4. These numbers show that the trade volatility of a WTO pair goes up by 0.485 units if the trade volatility of other WTO pairs goes up by one unit (regression 3.2). The trade volatility of a non-WTO pair goes up by 0.328 units if the trade volatility of other pairs with a WTO member goes up by one unit (regression 3.4). The estimated coefficient on $V_\tau ^{{\rm NWTO}} $ rises from 0.018 in regression 3.2 to 0.111 in regression 3.4. These numbers indicate that WTO dyad trade volatility only increases by 0.018 units if the trade volatility of other dyads with a non-WTO member goes up by one unit (regression 3.2). In comparison, the trade volatility of a non-WTO pair increases by 0.111 units if the trade volatility of other non-WTO pairs goes up by one unit (regression 3.4). The results suggest that the comovement of trade volatility appears strongest among WTO members. We also observe that the trade volatility comovement among non-WTO pairs is stronger than that between WTO pairs and pairs including a non-WTO member – we indeed observe the weakest trade volatility interdependence between WTO dyads and dyads with a non-WTO member based on the empirical results on volatility comovement across various groups in Table 3.

Over our sample period, the GATT/WTO has gone through several big formative stages, following rounds of negotiations on tariff- and non-tariff trade barriers reduction. Because lower trade barriers should apply to all GATT/WTO members according to the MFN principle, an effective round of trade negotiations should have made trade among GATT/WTO members more integrated than trade between members and non-members or trade among non-members. Lower trade barriers and more integrated trade among members, in turn, may lead to the comovement of their trade volatilities. This result suggests that the comovement should be stronger in periods with more progress toward reducing trade barriers over our sample time span.

Following Felbermayr and Kohler (Reference Felbermayr and Kohler2010), we look at four subsample periods: pre-Kennedy Round (1948–1967), Kennedy to Tokyo Round (1968–1978), Tokyo to Uruguay Round (1979–1994), and WTO (1995–2017) and report subsample regression results in Table 4. The subsample regressions allow us to tease out whether the spatial correlation of trade volatilities across trading pairs varies over time. In light of our previous findings that the coefficients on $V_{im\tau }^{{\rm WTO}} $ and $V_{kj\tau }^{{\rm WTO}} $ are qualitatively similar and the coefficients on $V_{im\tau }^{{\rm NWTO}} $ and $V_{kj\tau }^{{\rm NWTO}} $ are qualitatively similar, we include two combined terms $V_\tau ^{{\rm WTO}} $ and $V_\tau ^{{\rm NWTO}} $ in Table 4 for brevity. Similarly to regressions in Table 3, we report results with both countries i and j being WTO members and with both countries i and j being non-WTO members.

Table 4. Subsample regression results, with dyad fixed effects

Notes: All regressions are controlled for other control variables, dyad and time fixed effects similar to regression 2.3 in Table 2. We group our sample period (1948–2003) into four periods based on the GATT rounds and the WTO, following Felbermayr and Kohler (Reference Felbermayr and Kohler2010).

We define: $V_\tau ^{{\rm WTO}} = V_{im\tau }^{{\rm WTO}} + V_{kj\tau }^{{\rm WTO}} $ and $V_\tau ^{{\rm NWTO}} = V_{im\tau }^{{\rm NWTO}} + V_{kj\tau }^{{\rm NWTO}} $. Clustered standard errors by countries i and j are in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

By and large, the results in Table 4 are qualitatively similar to those in Table 3. In the regressions with both countries i and j being WTO members (regressions 4.1, 4.3, 4.5, and 4.7), six of the eight coefficients on spatial terms are positive and significant at the 5% level or better. In the regressions with both countries i and j being non-WTO members (regressions 4.2, 4.4, 4.6, and 4.8), seven of the eight spatial coefficients are positive, and three are statistically significant. We also observe that the strength of trade volatility comovement changes over time. For example, in the regressions for WTO members, the magnitude of the positive coefficient on $V_\tau ^{{\rm WTO}} $ is similar in regressions 4.1 and 4.3 over the pre-Kennedy Round (0.329) and the Kennedy to Tokyo Round (0.329). It declines to 0.197 over the Tokyo to Uruguay Round in regression 4.5 and then goes up to 0.487 during the WTO period in regression 4.7. The change in the strength of trade volatility dependence across GATT/WTO member pairs over time is, in general, consistent with our hypothesis related to the relationship between progress in GATT/WTO trade liberalization and trade volatility comovement across trading dyads. For instance, tariff reductions by major industrial countries in the first few rounds of negotiations under the GATT spurred fast growth in international trade in the 1950s and 1960s. However, economic recessions in the 1970s drove governments to devise other forms of protection (non-tariff barriers or NTBs) for sectors facing increased foreign competition, which could undermine the GATT's credibility and effectiveness and reduce the predictability of trade.Footnote ¹² These might lead to a decline of trade volatility comovement among WTO trading pairs shown in the subsample results. Over the latter subsample periods, new developments facilitating international trade and a more formal and effective trade dispute settlement mechanism under the WTO could result in more integrated and more stable trade and hence a stronger interdependence of trade volatilities shown in the results. For non-WTO members, the coefficients on both spatial terms, $V_\tau ^{{\rm WTO}} $ and $V_\tau ^{{\rm NWTO}} $, are insignificant in regression 4.2, suggesting that trade volatility comovement of dyads outside the GATT/WTO with other dyads is rather weak before the Kennedy round. The coefficient on $V_\tau ^{{\rm WTO}} $ in regression 4.4 is positive and significant, showing a strong comovement of trade volatility between non-WTO dyads and dyads with a WTO member between 1968 and 1978. Interestingly, the coefficient on $V_\tau ^{{\rm NWTO}} \;$is positive and significant at the 5% level, while the coefficient on $V_\tau ^{{\rm WTO}} $ is not statistically different from 0 in regressions 4.6 and 4.8. These results indicate that since the Tokyo round, trade volatility of dyads outside the GATT/WTO co-moves strongly with that of other dyads outside the GATT/WTO. When there is deeper trade integration among GATT/WTO members, trade patterns of countries outside the system may appear to diverge from member countries. As a result, we could see the volatility of trade of non-members co-move with that of other non-members. These results, however, need to be interpreted with caution given that the sample size of regressions 4.6 and 4.8 is quite small relative to that of regressions 4.5 and 4.7.

In summary, our results in Table 2 and those from the event study suggest that the GATT/WTO reduces members’ trade volatility. Regressions with the spatial correlation of trade volatilities in Table 3 indicate that the comovement of trade volatilities exists in general. There is a stronger interdependence of trade volatilities among WTO trading pairs than among other dyads. Our results imply that the WTO may stabilize bilateral trade even further through the strong comovement of trade volatilities among members. For example, joining WTO directly helps to reduce a dyad's trade volatility. A more stable trade between this dyad likely leads to more stable trade among other dyads due to the positive dependence of trade volatilities among members. In turn, spatial dependence indicates that stable trade among other dyads will further stabilize trade of our dyad of interest. In other words, the trade-stabilizing effect of the WTO is magnified due to spillovers when spatial dependence exists.Footnote ¹³ Focusing on WTO members, they can benefit from growing international trade with each other and a more transparent and integrated system.