Procedure for generating candidate multi-cue associates

The advantage of including three-way fixed effects is that it allows the estimation of a theo-retically consistent structural gravity model that controls for any observable and unobservable heterogeneity across countries. For instance, differences in institutions and in political systems that could affect trade flows are captured and are not subject to data availability (Bigler and Raess 2019). The drawback is clear, exporter-time and importer-time fixed effects absorb all the coefficients of observable country-specific characteristics such as GDP, national policies and exchange rates, which cannot be estimated. For our purposes, this means that is not possible to include a simple dummy indicating whether the country has a LABPTA with the US or the EU (LABP T A_{U S,i,t} or LABP T A_EU,i,t) and examine the coefficient to evaluate its effects on trade

22They develop the ppml_panel_sg Stata^® command that implements multi-way clustering, although the estimates reported in this study are calculated using the ppmlhdfe command of Correia, Guimarães, and Zylkin (2020) which allows the same type of clustering.

flows.²³ LABP T Ai,t would be perfectly collinear with the exporter-time fixed effects α_i,t; hence it would be unidentified in the context of this structural gravity framework. However, Heid, Larch, and Yotov (2020) recently demonstrated that scholars can exploit the presence of data on intra-national trade flows in gravity equations to ‘identify the country-specific variables even in the presence of exporter and importer fixed effects’ (at p. 3). Provided that T_ij also include intra-national trade flows, it is possible to interact LABP T A_i,t with an indicator variable IN T L_ij that is equal to one for international trade and set to zero for intra-national trade.²⁴ Including this interaction LABP T A_IN T L_ij,t allows us to identify the impact of LABPTAs on international exports relative to domestic trade. A positive coefficient on this variable would support the pro-ductivity argument, indicating that, on average, countries with LABPTAs have more international exports than domestic trade. In contrast, a negative coefficient would support the comparative advantage argument, indicating that countries with LABPTAs become increasingly unable to target foreign markets and have to rely on domestic commerce.²⁵

A similar strategy allows estimation of the effects of LABPTAs on exports towards high-income economies relative to other (emerging) economies. I interact LABP T A_i,t with a dummy variable LM IC_j identifying whether the importer is a lower- or a medium-income country, defined according to the World Bank classification.²⁶ The expectation is that LABP T A_LM IC_ij,t = LABP T A_i,t × LM IC_j will be negative, indicating that, on average, countries that have a LABPTA trade less with LMICs than with high-income countries. Note that this identification strategy is only able to examine whether LABPTAs have a differential impact on exports towards high-income countries relative to LMICs; it is not able to reveal whether this differential impact is caused by an increase in exports towards higher-income economies or by a decline in exports towards LMICs. Either of these causal mechanisms could explain a negative coefficient of LABP T A_LM IC_ij,t. To adjudicate between these arguments, one can interpret this

23Indeed, if LABPTAs have the cost, productivity or demand effect implied by the theory, the LABPTA can be considered as a country-level variable affecting exports towards all destinations.

24Note that the trade literature is becoming increasingly aware of the importance of including intra-national trade in structural gravity estimation (Yotov et al. 2016; Heid, Larch, and Yotov2020; Baier, Yotov, and Zylkin2019;

Bergstrand, Larch, and Yotov2015; Dai, Yotov, and Zylkin2014). Scholars argue that examining only international trade flows may bias PTA estimates downwards (Yotov et al. 2016).

25Cf. Heid, Larch, and Yotov (2020) and Beverelli et al. (2018) for proof of the fact that this variable is identified.

Note that this coefficient refers to the effect of LABPTAs on international trade relative to domestic trade as shown by Beverelli et al. (2018).

26In the baseline model, I use the World Bank classification of 1995, which is the median year of the sample. I would like to thank Prof. Marcelo Olarreaga for suggesting this identification strategy in one of the numerous conversations we had.

3.3. EMPIRICAL APPROACH CHAPTER 3

coefficient simultaneously with LABP T A_IN T L_ij,t. Indeed, if a LABPTA has a positive effect on international exports, it is likely that it is attracting demand from high-income economies. In contrast, if a LABPTA harms international exports, it is likely that it is hurting exports to LMICs.²⁷

The paper also aims to examine whether LABPTAs promote international trade and affect export destinations of countries competing with signatory countries. With this aim, I create a new variable that measures competitors’ engagement in LABPTAs (CELABPTAs), defined thus:

CELABP T A_i,t =^X

z6=i

W_iz,t× LABP T A_z,t (3.2)

CELABP T Ai,t is the sum of all LABPTAs signed by the competitors z of the country i, weighted by the level of competition W_iz,t between the two countries at time t.²⁸ The basic idea is that the more two countries are close competitors, the more they will affect each other’s exports. If two countries have very similar export profiles, changing the productivity/comparative advantage of one of them will probably affect the exports of the other. At the same time, if they trade completely different goods, this effect will be smaller or non-existent. To build the competition weight, I follow Guler et al. (2002) and much of the subsequent literature and measure competition by looking at countries’ sectoral-level export profiles (i.e. by looking at product similarities in their export portfolios), with no discrimination on export destination (Chatagnier and Kavaklı 2017;

Wang 2017; Baccini and Koenig-Archibugi 2014; Cao 2010; Simmons and Elkins 2004; Elkins, Guzman, and Simmons 2006; Polillo and Guillén 2005; Guler et al. 2002).²⁹ The weighted sum of competitors’ engagement varies at the country-year level i, t and, hence, like LABP T As_i,t, its effects on export volumes and destinations can be identified interacting with an indicator variable individuating international trade (IN T L_j) and with an indicator variable individuating LMICs (LM IC_j). Note that the inclusion of the variable CELABP T A allows estimation of the general

27Note that this identification approach is, to my knowledge, the only one that can be estimated in a structural gravity-consistent framework. To evaluate the effects of policy variables on export destination, scholars often simply split the sample and look at the effects of LABPTAs on exports towards high-income countries and low-income countries separately (Carrère, Olarreaga, and Raess2017). However, this approach does not overcome the issue that LABP T Ai,tis a country-level variable and hence not identified in the presence of exporter-year fixed effects.

28Note that in creating this weighted sum, I exclude all the US and all the EU countries from the sample of competitors. In other words, when the competitor z is the EU or the US, Wiz,t is equal to zero.

29Product-level data is drawn from the United Nations (UN) Standard International Trade Classification (SITC).

The similarity in export profiles is measured at the three-digit level because most granular data for exported products suffer from severe problems of missing observations for developing countries. I use different specifications of competition as robustness checks. Note that, for endogeneity reasons, it is not possible to measure competition by looking at the similarities in export destinations.

equilibrium effects of LABPTAs in the context of a partial equilibrium model.

The baseline econometric specification we use to test the hypothesis of this paper is:

Tij,t=exp



αi,t+ α_j,t+ α_ij + κP T A_ij,t+

β1LABP T A_IN T Lij,t+ β₂LABP T A_LM ICij,t+ β₃CELABP T A_IN T L_ij,t+ β₄CELABP T A_LM IC_ij,t

 + ε_ij,t

(3.3)

where T_ij,tare nominal trade flows, which include international and intra-national trade at non-consecutive year t (Yotov et al. 2016); α_i,t, α_j,t and α_ij are the three-way fixed effects discussed above; and P T A_ij,t is a dummy variable that controls for the existence of a trade agreement. It is important to note that this measure includes all bilateral trade agreements, taking the value of one for LABPTAs with the EU and the US. Hence, P T A_ij,t controls for the increase in bilat-eral exports with the US and the EU that these agreements generate.³⁰ LABP T A_IN T Lij,t and LABP T A_LM IC_ij,t capture the effects of LABPTAs on the international exports and destina-tions of the signatory countries. CELABP T A_IN T L_ij,t and CELABP T A_LM IC_ij,t capture the effects of competitors’ engagement with LABPTAs. The comparative advantage argument pre-dicts that β₁ will be negative because LABPTAs lead to increased labour costs, whereas β₃ will be positive because these rising costs promote competitors’ exports. Conversely, the productivity argument predicts that β₁ will be positive because LABPTAs will lead to increased productivity, while β₃will be negative because competitors will suffer from a loss in their comparative advantage.

Finally, the demand-side argument predicts that β₂ will be negative because LABPTAs attract the demand from high-income economies while reducing demand from lower-income ones, whereas, for specular reasons, β₄ should be positive. Competitors with lower labour costs, unable to meet the social standards required by the demand of high-income economies, will export more to emerging markets. In sum, using an innovative identification strategy, Equation (4.5) allows the unbiased estimation of the effects of LABPTAs and CELABPTAs on export flows and destinations in the context of a structural gravity framework.³¹

30Hence, the coefficients on LABPTA variables will not be inflated by the bilateral trade-enhancing effects of these agreements between signatory countries.

31The Annex discusses the issue of endogeneity in further detail.