Policy Research Working Paper 10277 Deep Trade Agreements and Heterogeneous Firms Exports Matteo Neri--Lainé Gianluca Orefice Michele Ruta Macroeconomics, Trade and Investment Global Practice January 2023 Policy Research Working Paper 10277 Abstract This paper studies the effect of regional trade agreements on trade impact of deep trade agreements depends on firm’s firms’ exports. Using detailed information on the content characteristics. The impact is stronger for large firms and of trade agreements and firm-level exports for 31 develop- firms involved in global value chains and is negative for ing countries between 2000 and 2020, the analysis shows small firms. Robustness tests and an Instrumental Vari- that the depth of trade agreements matters for the export able strategy confirm the causal interpretation of the results. performance of firms. Moving from shallow to deep trade These heterogeneous impacts on firms’ exports imply a selec- agreements boosts firms’ exports, on average, by 3.6 percent. tion effect of deep trade agreements with significant welfare In line with models of trade with heterogeneous firms, the consequences. This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at mruta@imf.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Deep Trade Agreements and Heterogeneous Firms Exports∗ Matteo Neri--Lainé† Gianluca Orefice‡ Michele Ruta§ Keywords: Deep Trade Agreements, Export Performance, Firm Heterogeneity, Developing Countries JEL-Classification: F13, F14, F15 ∗ We would like to thank seminar participants at the World Bank, Indiana University, University of Mainz and the European Trade Study Group for helpful comments and suggestions and to Ana Fernandes for help with the data. The views expressed in this paper are the authors’ and do not necessarily reflect those of the institutions they belong to. † University Paris-Dauphine PSL, Place du Maréchal de Lattre de Tassigny 75016 Paris, France. ‡ University Paris-Dauphine PSL and Cesifo, Place du Maréchal de Lattre de Tassigny 75016 Paris, France. § International Monetary Fund, 700 19th Street N.W., Washington, D.C. 20431, USA. 1 Introduction During the last two decades, many developing countries have signed Regional Trade Agreements (RTAs) in the attempt to better integrate their economies into regional and global markets and improve the export performance of their firms – see Freund and Ornelas (2010); Limão (2016). In particular, out of the 190 RTAs signed during the period 2000-2020, 175 involved at least one developing country as a member. Over the same period, the content of RTAs has widely changed (Hofmann et al., 2017; Mattoo et al., 2020). The average RTA signed by a developing country at the end of the period covered roughly 30 percent more policy areas than at the beginning of the period, with issues like technical and sanitary standards, investment, and intellectual property rights protection being added to the more traditional areas covered by RTAs (such as tariff liberalization and the reduction of other border barriers). We refer to these more complex trade agreements as “deep RTAs” or simply “Deep Trade Agreements” (DTAs hereafter). The effects of regional trade agreements on trade have been largely studied, but there remain some important gaps in the literature. In part motivated by the changing depth of RTAs, a number of studies have empirically investigated the trade effects of different types of trade agreements. A key finding of this literature is that the heterogeneity of trade agreements matters (Baier et al., 2018, 2019), and that the depth of RTAs can help explain different trade outcomes (Orefice and Rocha, 2014; Mattoo et al., 2017). At the same time, theoretical models have emphasized a different type of heterogeneity, namely that the reduction of trade costs affects different types of firms differently. Larger and more productive firms can more easily take advantage of the change in trade costs than smaller and less productive firms (Melitz, 2003; Melitz and Redding, 2015). As deep trade agreements can reduce trade costs among members by eliminating tariffs and by reducing other frictions due to regulatory differences and/or policy uncertainty, a natural extension of this literature is that DTAs should affect firms differently based on their characteristics, promoting the export performance of larger firms and leading to a selection effect. This paper empirically investigates the effect of RTAs’ depth on the export performance of heterogeneous firms. The analysis combines firm-level export data for 31 developing countries from the World Bank Exporter Dynamics Database (Fernandes et al., 2016) and information on the detailed content of more than 300 RTAs from the World Bank Deep Trade Agreements database (Hofmann et al., 2017). The richness of the data allows to precisely take into account the heterogeneous consequences of deep RTAs on the exports of different firms. We use a decomposition à la Berman et al. (2012) to study the extent to which the effects of deep RTAs 1 on trade come from the intensive and the extensive margins of trade. Importantly, we carefully investigate the endogeneity of trade and deep RTAs through different methods. We use an Instrumental Variable approach based on the domino effect of RTAs (Baldwin and Jaimovich, 2012), a plausible exogeneity test à la Conley et al. (2012), and we employ an event study approach to check the parallel trend assumption and confirm the validity of the key results. To assess the impact of deep RTAs on the export performances of firms, we adapt the standard gravity model for trade (Head and Mayer, 2014) to firm-level analysis and include a variable capturing the depth of RTAs. Using information from the Deep Trade Agreements database, we construct different measures of RTA depth based on the policy areas covered by the agreements and their legal enforceability. Our specification allows to identify the impact on firms’ exports of a change in the depth of RTAs (i.e. newly signed RTAs or amendment of pre- existing ones) between two countries, controlling for any firm-year and country-specific factor that may affect the export performance of firms. The baseline results show that one additional legally enforceable policy area in RTAs boosts the exports of firms by 0.3%; corresponding to a 3.6% increase in firms’ exports when moving from shallow to deep RTAs.1 This effect is larger for policy areas in RTAs that are not regulated by the agreements of the World Trade Organization (WTO-extra), such as investment or competition. One additional legally enforceable WTO- extra provision increases the firm’s exports by 0.7%. The instrumental variable approach and the plausible exogeneity test both suggest the causal interpretation of our results. This average effect of deep RTAs on the export performance of firms may hide substantial heterogeneity across firms with different characteristics. In New New Trade Theory models à la Melitz (2003) firms are heterogeneous and a reduction in fixed and/or variable export cost is expected to favor exports of large and highly productive firms at the expenses of low-productivity firms, leading to a selection effect. This selection effect is magnified in presence of heterogeneous demand elasticity: if large and more productive firms face smaller demand elasticity and higher mark-ups, the pro-competitive effect induced by deep RTAs exacerbates the reduction in foreign sales of less productive firms (in line with Atkeson and Burstein 2008 and Crowley et al. 2022). Our second set of estimations tests the effect of deep RTAs on heterogeneous exporters.2 In line with the findings of the New New Trade Theory, we uncover strong heterogeneous effects of deep RTAs on firms with different characteristics. While large and high-productivity firms benefit from deep RTAs, small and less productive firms suffer the increased competition 1 We consider shallow those RTAs including only the two tariff related provisions that are always included in RTAs (i.e. tariff cut on agriculture and industrial sectors), and deep those RTAs including 14 provisions (i.e. the 75 percentile in the depth of RTAs). 2 For our large sample of developing countries, we do not have data on the balance sheet of firms to compute direct measures of productivity, so we resort to several proxies discussed in detail in section 5.2. 2 at destination and export less – a selection effect of deep trade agreements. On average, the inclusion of an additional legally enforceable provision in RTAs stimulates the exports of large firms by 0.4%-0.6%, while it reduces the exports of small firms by 0.5%-0.6%.3 Interestingly, high-productivity firms participating in Global Value Chains (GVC), i.e. firms that export and import to/from the same country, benefit the most from the signature of deep RTAs. For these firms, an additional legally enforceable provision in RTAs stimulates exports by 0.8%. The selection effect of deep trade agreements shows that using firm-level data is key to understand the welfare implications of RTAs. Specifically, by favoring exports of large and high-productivity firms and reducing exports of small and less productive firms, deep RTAs promote a reallocation of resources from the latter to the first which entails adjustment costs but also leads to an overall increase in the average productivity of firms. We next investigate how deep trade agreements affect the margin of trade at the firm level. Deep RTAs could reduce both the variable and the fixed trade costs between member countries. This in turn impacts on the margin of trade that is affected. Previous work by Baier et al. (2014) uses aggregate trade data to study how different types of RTAs affect the intensive and the extensive margin of trade. Here, we use a decomposition approach à la Berman et al. (2012), to disentangle the average export sales effect of deep RTAs into intensive versus extensive (firm-based) margin component. Namely, we use the firm-specific export behavior to impute the extensive versus intensive margin effect of deep RTAs. We show that the extensive margin channel has only a slightly larger contribution (60%) than the intensive margin channel (40%). This evidence, in light of the theoretical predictions in Chaney (2008), suggests that indeed deep RTAs represent a reduction in both the fixed and the variable export cost components. This is the second contribution of this paper which provides novel firm-level evidence on the channel through which deep RTAs boost aggregate exports. Finally, we study the dynamic effects of deep trade agreements on firms’ exports. The positive effect of bilateral deep RTAs may vanish over time because of the worldwide increase in RTAs and their depth and the inclusion in RTAs of non-discriminatory provisions that de facto can reduce trade costs for exporters in third countries (Lee et al., 2019). By adopting an event study approach in the vein of Fajgelbaum et al. (2020), we show that a change in the depth of RTAs makes firms export more during the two years after the shock. The effect of the change in depth of RTAs vanishes afterwards. The event study approach also shows the validity of the 3 This finding is the average effect of deep trade agreements on firms’ exports. It is still possible that individual provisions in DTAs have effects that are more favorable to small firms. For instance, Fernades et al. (2021) find that for a sample of Latin American countries firms’ exports increase significantly in destination markets with RTAs that promote regulatory cooperation and that the effect is stronger for smaller firms. They find that this effect is driven by entry into new product markets and increases in export quality of smaller firms. 3 parallel trend assumption and hence reinforces the causal interpretation of our baseline results. The rest of the paper is organized as follows. Section 2 discusses the possible theoretical channels underlying the firm-specific effect of deep trade agreements. Section 3 presents the data used in the analysis and proposes some descriptive evidence. Section 4 discusses the empirical strategy. Section 5 shows the baseline results and the heterogeneity tests. In section 6 we address the potential endogneneity problem. Section 7 tests the dynamic effect of deep RTAs. Concluding remarks follow. 2 Theoretical framework Deep trade agreements are wider in scope and more complex than shallow trade agreements as they go beyond a standard improvement in market access via reductions in bilateral import tariffs. Specifically, by filling regulatory gaps among member countries, DTAs may among other things reduce fixed export costs and uncertainty in bilateral trade relationships. These features make any attempt of setting a general theoretical framework extremely complicated (and likely unsatisfactory). However, some general predictions can be drawn. The reduction in fixed or variable export costs and the reduction in uncertainty associated to DTAs unambiguously increase market access at destination, thus make it easier for firms of a given origin to export to destination. In a world with homogeneous firms, this simply implies larger export values for firms at destination. But firms are heterogeneous (Bernard et al. 2007) and the expected trade effect of DTAs is not as simple. In a standard model of trade with heterogeneous firms à la Melitz (2003), where all firms face the same demand elasticity, an improved market access at destination makes large and more productive firms export relatively more than less productive firms. But firms differ also along other relevant dimensions. First, firms that participate into global value chains (i.e. GVC firms) can be more affected by DTAs either because they tend to be high-productivity firms (Goldberg et al., 2010; Feng et al., 2016) and/or because they import and export from the destination market, so that the reduction of trade costs and uncertainty associated to DTAs affects them both on the import and export side. A second way firms differ is in the demand elasticity they face. Large and high-productivity firms face lower demand elasticity (Spearot, 2013) and have larger mark-ups than small and less productive firms (Atkeson and Burstein, 2008). This implies that, after the entry into force of a DTA, large and high-productivity firms at origin may reduce their export price at destination to a larger extent to gain market share, exerting a tougher “within origin” competition on small and less productive firms. Such a strong heterogeneous impact of DTAs on firms with different characteristics is even starker in a framework in which 4 the degree of substitutability across varieties of a given origin country is stronger than that between origin countries—see Crowley et al. (2022). In these settings, DTAs may have very different consequences on the export performance of large and small firms at destination, with large firms unambiguously benefiting the most from the trade agreement. The impact of DTAs on small and less productive exporters also depends on whether the new agreement has an impact on the market structure of the origin country. If the number of exporters is unaffected by the entry into force of a deep trade agreement, small incumbent exporters may benefit through an increase in exports even if these gains are milder relative to large exporters. Conversely, if the DTA has an impact on the market structure in the origin country by increasing the number of exporters, small exporters may also experience a reduction in their exports (and market share) at destination, and eventually may be induced to exit from the export market due to enhanced competition. All in all, while DTAs are expected to have an unambiguous positive effect on the export value of large and high-productivity firms, the effect on small and less productive firms is an empirical question. The within-origin competition effect of DTAs can be so strong to offset the reduction of fixed and variable costs and induce the exit of small firms from the destination market. This effect has relevant welfare implications. If small and less productive firms exit from a specific destination market and large and high-productivity firms survive and thrive, resources are reallocated from low- to high-productivity firms. This selection effect implies adjustment costs, but also an increase in the average productivity of firms. The heterogeneous trade effects of DTAs on low- versus high-productivity firms, and the consequent selection effect, is at the core of the empirical investigation in the rest of this paper. 3 Data and descriptive evidence Our empirical analysis is based on two main World Bank data sources: (i) the Exporter Dynamics Database (Fernandes et al., 2016) providing firm-level exports for 55 developing countries in the period 1996-2020, and (ii) the Content of Deep Trade Agreements (Hofmann et al., 2017) on the content of RTAs in force and notified to the WTO for the period 2000-2020. Moreover, we complete our dataset by including gravity-related variables (such as distance, common border, language, etc.) from the CEPII gravity database, and data on the effectively applied tariffs faced by each exporter at the destination from the MacMap (CEPII) database. We build our final dataset in several steps. First, we reorganize the original DTA database into a country-pair-year-specific dataset. Indeed, since a given pair of countries (origin and desti- nation) may have two (or more) RTAs contemporaneously in force, we collapse trade agreement 5 provision dummies indicating the presence of each policy area in RTAs by country-pair-year, and take the maximum value of each provision dummy across multiple RTAs (if any) within each country-pair-year combination. As a second step, we construct several measures of RTAs’ depth based on the provisions included in the trade agreement that each country pair shares in a given year. Specifically, we consider different count variables, differentiating by type of provision: (i) number of provisions (independently of their legal enforceability), (ii) number of legally enforceable provisions (i.e. whose implementation is supported by strong legal language and by the availability of a dispute settlement mechanism), (iii) number of WTO-plus provisions (WTO+) – i.e. provisions covered by the current mandate of the WTO, (iv) the number of WTO-extra provisions (WTO-X) – i.e. provisions not covered by the current mandate of the WTO, and (iv) number of core provisions – i.e. provisions directly related to trade enhancing factors.4 While in our empirical exercise, we use all these proxies for the depth of RTAs, our baseline measure of RTA depth is the count of legally enforceable provisions in each RTA.5 The simple count of provisions in RTAs gives equal weight to each clause included in the agreement, and one may want to give more weight to “rare” provisions (i.e. provisions included less frequently in RTAs and likely to signal stronger market integration). To address this concern, as a robustness check, we also use alternative measures of RTA depth based on the weighted sum of provisions (using one minus the frequency to which each provision appears in the 300 RTAs covered by the DTA database).6 One drawback of the DTA database is that it does not cover RTAs that are no longer in force.7 So, when controlling for the presence of an RTA between two countries (see section 4), we complement the DTA database with the CEPII data covering active and inactive RTAs. The RTA adjusted dummy is equal to one if a RTA is observed (from CEPII and/or DTA dataset) and zero otherwise. As for the depth of RTAs not mapped in the DTA database, we construct two proxies of depth. First, we (arbitrarily) set the depth measures for these RTAs as equal to 4 See Horn et al. (2010) for definition of WTO+ and WTO-X provisions. Core provisions are defined in Hofmann et al. (2017) and include all WTO+ provisions plus clauses that regulate competition policy, bilateral investment, movement of capital and intellectual property rights. 5 It should be noted that in this paper we aim to capture the impact of the overall depth of RTAs on firms with different characteristics. Different provisions in trade agreements, however, have different effects on aggregate trade outcomes (Fernandes et al., 2021) and are likely to have heterogeneous effects on firms’ exports. For example, provisions that lower fixed entry costs on the destination market, such as provisions aimed at improving trade facilitation or reducing regulatory divergence, can make it easier for small firms to export. Fernades et al. (2021) finds some preliminary evidence of this focusing on provisions on technical and sanitary standards in trade agreements involving a sample of Latin American countries. Other provisions may have just the opposite effect. For instance, requirements in DTAs to meet higher environmental or labor standards could make it easier for larger exporters relative to smaller firms to export in the destination market, reinforcing the competition effect that we stress in this paper. 6 In Figure A1 we show how frequently each provision is included in Regional Trade Agreements. 7 The World Bank provides information on the content of RTAs active in the year of the creation of the DTA database. RTAs that were active in the past, but inactive at the time of the creation of the database are not covered. 6 one. Second, we set the depth of non-mapped RTAs as the value of the closest-in-time (positive) RTA depth observed for the considered country-pair. The Exporter Dynamics Database (EDD) takes the form of multiple databases: one per country.8 So, as a last step in the construction of the final data set, each country-specific EDD is merged with the CEPII gravity database. Finally, we merge the tariffs data from MacMap (CEPII), and pool all the 55 country-specific databases to get a single proper and complete firm level base. This dataset initially contains 29,009,865 observations (firm-destination-product- year specific) spanning from the late 1990s to 2020. However, the DTA database covers only the period 2000-2020; so we keep only firm export data over that period. It must be noted that the coverage of the EDD varies by country: for some countries we have data for the entire 2000-2020 period, for some other countries we have a more limited time period (cf. table A1 in appendices). Although original firm-level exports data are (also) product (HS6) specific, our main variable of interest is not, so we aggregate (sum) export data by firm-destination-year across HS6 products. As described in Table 1, the majority of exporters included in the EDD have at least one active RTA in force (i.e. 48 out of 55). However, only a sub-sample of countries (and country- pairs) have changed the depth of their RTAs in the period 2000-2020 (i.e. a newly signed RTA or amendment of pre-existing RTAs). Namely, 31 exporting countries signed new RTAs or amended a pre-existing one, giving within variation to our measures of RTAs depth. This translates into 701 country-pairs having time variation in the depth of RTAs in the period 2000-2020. This time variation comes mainly from the entry into force of new trade agreements.9 Since in the empirical strategy we rely on the within country-pair variation in RTAs’ depth, the final estimation sample considers only such 31 exporting countries and contains 4,659,362 observations with non-missing information on export value and content of RTAs. The estimation sample shrinks to 2,924,126 observations because of missing values in the tariff data. – Table 1 about here – Among the 31 exporting countries having within variation in the RTA depth values, 15 countries account for 87% of the overall variation in the RTA depth. These are: Croatia, Georgia, Colom- bia, Chile, South Africa, Serbia, Slovenia, Peru, Mauritius, Guatemala, Madagascar, Nicaragua, Tanzania, Ecuador, Malawi. In Figure 1 we show the empirical distribution of country-pairs exports (panel a and b) and firm-destination specific exports (panel c and d) for the sub-sample 8 In case of breaks in the firms’ identifiers over the time period covered, the Exporter Dynamics Database contains two separate data sets per country (before versus after the break in the firms’ identifier). In case of breaks in the firms’ identifiers, we keep the most recent period (i.e. after-break period). 9 In our final sample, 72 RTAs involve more than two members, while 2 are bilateral for a total of 74 agreements contributing to the within variation of our RTA depth variable. 7 of the 15 countries that provide much of the variation in RTA depth.10 In panel (a) of Figure 1 we show the empirical distribution of bilateral exports for country-pairs with and without RTA in force (respectively dash and continuous line in the figure), while in panel (b) of Figure 1 we show the empirical distribution of bilateral exports for pairs having deep versus shallow RTA in force. It clearly emerges that the presence of a RTA and its depth matters for the aggregate country-pair exports. In panel (c) of Figure 1 we show the empirical distribution of firm-destination export sales in 2020 depending on whether the exporting country has (or not) an active RTA with the destination market. In panel (d) of Figure 1 we consider only RTA partners (i.e. pairs with RTA in force), and distinguish between destinations at the top- and bottom-quartile of the RTA depth. There is only small evidence that RTA depth matters for the export of firms. The positive effect of RTAs and their depth is less stark at firm level than at aggregate country-pair level, suggesting a strong heterogeneous effect of DTAs on firm level export. While some firms largely benefit from deep RTAs (boosting the aggregate exports of the country), other firms are hindered by the presence of deep RTAs. The empirical distribution of firms’ exports is thus only marginally affected by the presence of deep trade agreements. In what follows we carefully test such a heterogeneous effect of deep RTAs. – Figure 1 about here – Finally, in Table 2 we report the in-sample descriptive statistics for the variables included in our econometric exercise (overall RTA depth, WTO-plus, WTO-extra and core provisions). It clearly emerges the large variation in the depth of RTAs across country-pairs. Regardless of the type of provisions considered to define the RTAs’ depth (i.e. count of provisions), the standard deviation is almost equal to the average value. We exploit the variability in the depth of RTAs in the econometric exercise reported in the next section. – Table 2 about here – 4 Identification strategy This section discusses the empirical strategy adopted to test the effect of deep RTAs on the exports of firms in developing countries. Our baseline specification is as follows: Xf ijt = exp [θf t + θjt + θij + β1 DT Aijt + β2 ln (1 + τf ijt )] × εf ijt (1) where the subscripts f , i, j , and t stand respectively for firm, origin country (i.e. the country 10 In Figure A2 we provide the same evidence for the full set of 31 countries with time variation in RTA depth. 8 where the exporting firm is located), destination country and year. The explanatory variable of interest – DT Aijt – is the depth of the RTA (if any) that country i has with destination j at time t. As a first, coarse measure of RTA depth we use the count of any type of provisions included in the RTA. However, our preferred measure of RTA depth is the count of legally enforceable provisions. This measure is then refined and we use the count of WTO+, WTO-X and “core” legally enforceable provisions in RTAs. The firm-year fixed effects (θf t ) control for any unobserved time-variant firm specific characteristics, such as productivity shocks, size and workforce composition (i.e. quality of the management, etc.). Since each firm is unambiguously located in a country i, the firm-year fixed effects subsume origin-year fixed effects and capture any origin country-year specific shock affecting the export performances of all firms in country i (i.e. multilateral resistance term on the exporter side). Any country specific technological or productivity shock, as well as the distribution of firms productivity in the exporting country, are implicitly captured by firm-year fixed effects. In equation (1) we also control for the multilateral resistance term in the importer side by including destination-year fixed effects θjt capturing also any demand shock that affects the import demand at destination. Finally, any country-pair (time-invariant) factor, such as geographical distance and any other gravity-type covariate is captured by country-pair fixed effects θij . Given the set of fixed effects included in equation (1) our variable of interest DT Aijt is identified on the change in the depth of RTAs (i.e. newly signed RTAs or amendment of pre- existing ones) between country i and j , controlling for any firm- and country-specific factor that may affect the export performances of firms. Specifically, we compare a given firm’s exports towards destinations with versus without changes in RTA’s depth (conditional on any firm- and destination-specific shock). The omitted variable concern is therefore very reduced here. Moreover, the concern that a specific firm in country i may affect the signature and the content of a trade agreement between country i and j is in general remote, and the reverse causality argument is unlikely to bias our baseline estimations. In Table A2 we perform a pre-trend test correlating the (average) firms’ export growth before a change in RTA depth and extent of the change in the depth of RTAs. Table A2 shows the absence of correlation between firms’ export growth and the change in RTAs’ depth supporting qualitatively the absence of reverse causality issue. Nevertheless, in section 6 we propose an Instrumental Variable (IV) approach aimed to reduce further any residual endogeneity concern. This could come, for instance, from large exporters lobbying for deep trade agreements. A plausible exogeneity test à la Conley et al. (2012), discussed in section 6.1, supports the robustness of our 2SLS results to deviations from the perfect validity of the exclusion restriction hypotheses and reinforces the causal interpretation 9 of our results. The parallel trend assumption in the pre-treatment period is tested in section 7. The main empirical challenge here is the high-collinearity between the mere presence of an active RTA (abstracting from its content, RT Aijt ) and the depth of the agreement (DT Aijt ). This problem is exacerbated when country-pair fixed effects (θij ) are included in the estimation and absorb any cross country-pair variability in the presence of RTAs and depth. For this reason in our baseline estimations we do not control for the presence of an RTA; this is de facto subsumed by the DT Aijt variable when larger than zero. However, since the main and ever- present objective of any RTA is to reduce bilateral tariffs, we control for the presence of an RTA by including in all the estimations the weighted average applied tariffs faced by a given firm f into a given destination j across exported products, ln (1 + τf ijt ). We use the product share of firm’s exports in the initial year as a weight in averaging the applied tariffs faced by each firm at destination across exported products. As an alternative test, we disregard the collinearity problem and propose a robustness check explicitly controlling for the RT Aijt dummy (namely the RTA adjusted dummy discussed in section 3). In order to test the heterogeneous effect of deep RTAs on firms with different characteristics, we extend eq. (1) by interacting the DT Aijt variable with a firm characteristic indicator I (kf > ¯) as follows: k Xf ijt = exp θf t + θjt + θij + β1 DT Aijt + β2 ln (1 + τf ijt ) (2) ¯) + β3 DT Aijt × I (kf > k × εf ijt . ¯. ¯) is equal to one if a given firm’s characteristics kf is above a threshold k The indicator I (kf > k ¯). First, as a proxy for Four firm specific characteristics are used to define the indicator I (kf > k the firm’s size, we use the total exports of the firm (across destinations and years),11 and define dummy variables equal to one if the size of the firm is in turn above the 75th and 90th percentile of the distribution. This is an intuitive but coarse proxy of firm size. It can be endogenously affected by the presence of deep trade agreements among firm’s destinations. To address this problem, we use the total exports of the firm in the initial year t0 (i.e. the first year in which the firm is observed in the data). Firm’s total export at time t0 is used to define our second ¯). Percentiles 75th and 90th of the initial firms’ exports distribution indicator variable I (kf > k ¯. Third, we capture the GVC status of the firm by a dummy equal to one are used as threshold k if the exporting firm is also an importer, indicating that the firm is likely to use imported inputs 11 High-productive firms export more. So the total exports of the firm (across all products and destinations) is a plausible proxy for its productivity. See Fontagné et al. (2015). 10 in production for exports. Finally, we refine the GVC nature of the firm by using a dummy equal to one if the firm exports and imports to/from the same country j – GVC bilateral. This last firm characteristics is meat to capture the importance of deep RTAs for firms in developing countries having bilateral (import-export) relations with destination j . We adopt a PPML estimator to address the heteroskedasticity problem in structural grav- ity model for trade (Santos-Silva and Tenreyro, 2006), and cluster standard errors by origin- destination-year (i.e. the source variation of our main variable of interest). As a benchmark, in Appendix Table A3 we show OLS estimations.12 5 Results This section discusses our main results on the effect of deep RTAs. We start by showing the results obtained by estimating our baseline specification, equation (1), on the full sample of firms in developing countries facing changes in their RTAs’ depth - section 5.1. In the same section we propose a robustness check using the weighted count of provisions as an alternative measures of RTA depth. In section 5.2, we estimate equation (2) and show the heterogeneous effect of deep RTAs on firms with different characteristics. Finally, in section 5.3, we disentangle the effect of deep RTAs into extensive versus intensive margins of exports. 5.1 Baseline results Table 3 shows our baseline results. The depth of RTAs has a positive and significant effect on the export performance of firms in developing countries no matter the type of provisions considered to approximate the depth of the RTA (all, legally enforceable, WTO+, WTO-X or core). In particular, one additional legally enforceable provision in the agreement boosts the exports of firms by 0.3%. This means that, by moving from a shallow agreement (here defined as an RTA including only legally enforceable provisions related to tariff cuts in manufacturing and agriculture sectors) to a deep RTA containing legally enforceable provisions at the 75th percentile of the distribution of the RTAs depth implies a 3.6% increase in firm exports.13 If we consider the count of WTO+ provisions, moving from shallow to deep RTAs implies a 2.7% increase of firm exports (i.e. one additional legally enforceable WTO+ provision boosts the exports of firms by 0.3%).14 The effect per-provision is larger for legally enforceable WTO-X provisions: one additional WTO-X provision boosts the export of firms by 0.7% (moving from shallow to 12 In the log-linear OLS estimations we set to zero the log of zero firms’ exports. 13 RTAs at the 75th percentile of legally enforceable provisions contain 14 provisions. 14 WTO+ provisions contain standard tariff cut provisions on agriculture and manufacturing sectors, so we con- sider RTAs with two WTO+ provisions as shallow. The 75th percentile in the count of WTO+ provisions is equal to 11. 11 deep RTAs in WTO-X provision implies a 2.1% increase in firm exports).15 As expected, the applied tariffs at destination have a negative and significant effect on the exports of firms. The point estimates on applied tariffs are lower than commonly obtained in the previous literature. This is due to the aggregation bias. Indeed, we disregard the product dimension of both export and tariffs, and the consequent aggregation bias produces tariff elasticity that are smaller in magnitude (Redding and Weinstein, 2019).16 – Table 3 about here – Using the simple count of provisions to approximate the RTAs’ depth implicitly gives the same importance to any type of provision. One may want to assign relatively higher value of depth to those RTAs including rare provisions. So, as a robustness check, in Table 4 we show results by using weighted count to approximate the depth of RTAs (the weight is equal to one minus each provision’s frequency in the matrix of the RTAs mapped by the World Bank Deep Trade Agreements database). The resulting index weights relatively more RTAs containing rare provi- sions. Results, reported in Table 4, support the robustness of our baseline results. Interestingly, the estimations coefficients in Table 4 point to a stronger impact of deep RTAs on exports when approximated by a weighted sum. This suggests that the inclusion of rare provisions in RTAs is a good signal of the extent of trade costs reductions associated to deep RTAs between member countries. – Table 4 about here – The effect of RTA depth is robust to the inclusion of dummies for the presence of a RTA between country i and j at time t. See results reported in Table A6. To take into account the presence of non-mapped active RTAs (i.e. those RTAs not included in the World Bank database), in Table A6 we take the list of RTAs from CEPII and assign a value of depth respectively equal to one (see columns 1-2), or equal to the closest (in time) country-pair’s RTA depth to non-mapped active RTAs (see columns 3-4). As expected, the presence of an empty (active) RTA – i.e. a RTA with zero depth – has null effect on the export of firms (see coefficient on RTA adjusted in table A6 columns 1-3). The presence of a RTA has positive effects on export only if the agreement has some depth (i.e. positive number of provisions) - see columns (1),(2) and (3). While reassuring, 15 WTO-X provisions do not contain standard tariff cut provision, so we consider RTAs with zero WTO-X provision as shallow. The 75th percentile in the count of WTO-X provision is equal to 3. 16 The negative and significant coefficients on applied tariffs reassure us on the accuracy of our estimations. This accuracy is also supported by point estimates on standard gravity controls reported in appendix Tables A4 and A5. In these appendix tables we remove country-pair fixed effects and include standard gravity model controls (such as distance, colony, common language and border) to have a benchmark with previous literature on gravity controls’ coefficients (Head and Mayer, 2014). These variables have the expected sign and magnitude in line with previous studies. 12 results in Table A6 must be taken cum grano salis because of the high collinearity between the RTA dummy and the measure of depth in regional trade agreements. As a last robustness check, we estimate equation (1) on the sub-sample of country-pairs that change RTA status during the period 2000-2020. Results, reported in Table A7 support the robustness of our baseline results. The baseline and the robustness check regressions discussed so far suggest that deep RTAs have a statistically significant positive effect on the exports of firms in developing countries. However such a positive effect may hide strong heterogeneity across firms of different types, and dynamic effect after the change in depth of a RTA. We dig more into the heterogeneous effects of deep RTAs in the next section. 5.2 Firm heterogeneity In line with the theoretical discussion in section 2, deep RTAs may have a different effect on low- versus high-productivity firms and/or on firms with different involvement in GVCs. This section explores the heterogeneous effects of deep RTA across different types of firms based on size and GVC participation. Namely, we interact the DT Aijt variable with four firm specific indicators: (i) large firm dummy variables equal to one if the total exports of the firm (across destination and years) is in turn above the 75th and 90th percentile of the distribution, (ii) large firm dummy if the exports of the firm at t0 is in turn above the 75th and 90th percentile of the distribution, (iii) a dummy equal to one if the exporting firm imports some products (i.e. GVC firm indicator), and (iv) a dummy variable equal to one if the firm exports and imports to/from the same destination j (GVC bilateral ). Results reported in Table 5 show an interesting and robust pattern: large and GVC firms benefit from deep RTA.17 For small firms, and firms that do not participate in GVCs, the depth of RTAs has a negative effect on exports. Specifically, for large high-productivity firms, one additional legally enforceable provision implies a 0.3% - 0.6% increase in exports (see columns 1 - 4). This means that, moving from shallow to deep RTAs implies a 3.6% - 7.2% increase in exports. For GVC firms exporting/importing to/from the same country j , moving from shallow to deep RTAs implies a 9.6% increase in exports (see columns 6). These results uncover an interesting selection effect à la Melitz (2003). Deep RTAs reduce variable and fixed trade costs between members, and more productive firms are more likely to take advantage of this policy change. In addition to this size/productivity channel, GVC firms benefit as they see a reduction in trade costs on imports of intermediate products used in production for exports. 17 The same conclusion holds by using a non-parametric binned model where the DT A variable is interacted by three firm size bins based on whether the total exports of the firm (across destination and years) are below the 25th percentile (small firms), above the 75th percentile (big firms), or in between (medium size firms). See appendix table A8. 13 The heterogeneous effect of deep RTAs is confirmed also by using the weighted sum of provision as a proxy for the depth of RTAs (see Table A9). – Table 5 about here – The negative effect on small firms is in line with trade models with heterogeneous elasticity of demand (Atkeson and Burstein, 2008; Crowley et al., 2022) that predict a strong within-country competition effect hurting less productive firms, and implying an increase in the concentration of foreign market shares among high-productivity firms. So, deeper RTAs favor the export sales of firms with certain characteristics (large, more productive and GVC firms) at the expense of small and less productive firms. Such a selection effect of deeper RTAs is shown in Table 6 where we estimate the effect of deep RTAs on: (i) the number of firms in i that keep exporting at destination j at time t (i.e. surviving exporters), and (ii) the average export per firm.18 Deep RTAs reduce the number of firms surviving in the export market (columns 1-2) but increase the average export (value) per firm (columns 3-4). This further confirms the selection effect of DTAs in developing countries, and points to welfare effects associated to adjustment costs and improved productivity deriving from the re-allocation of resources from less- to more-productive firms. – Table 6 about here – 5.3 The extensive and intensive margin channels The export sales effect of deep RTAs may come from the intensive and/or the extensive margin of trade. Indeed, the reduction in trade costs associated to deep trade agreements may lead incumbent firms to boost their exports in destinations with deep RTAs (i.e. intensive margin channel), or may allow new firms to start exporting in such destination (i.e. extensive margin channel). In order to disentangle the overall export sales effect into the extensive and intensive margin channel, we calculate the change in aggregate country-pair specific exports from: (i) incumbent firms – i.e. firms always exporting toward a given destination (intensive margin channel); and (ii) entry-exit firms to/from a specific market (extensive margin channel). Then, we adopt the decomposition approach as in Berman et al. (2012). Namely, we regress the aggregate exports by origin-destination-year for respectively incumbent and entry-exit firms on RTAs’ depth, and country-year fixed effects. The coefficients on RTA depth are respectively ˆentry . We then calculate the share of total exports from respectively incumbent ˆincumbent and β β 18 PPML estimator is used in Table 6. 14 and entry-exit firms (Vincumbent /Vtot and Ventry /Vtot ), and multiply such shares for the respective ˆentry . ˆincumbent and β elasticity β The results of this calculation are reported in Table 7. As expected, the elasticity to deep RTAs for both incumbent and entry firms are positive and significant, with a larger effect for incumbent firms. But since the share of total exports is larger for entry-exit firms than for incumbent firms, we obtain a lower contribution from the intensive margin channel to the aggregate effect. Namely, the intensive (extensive) channel accounts for 40% (60%) of the total impact of deep RTAs. The not too different contribution of the intensive and extensive margin channel suggests that deep RTAs represent at the same time a reduction in both the fixed and the variable export cost component. This is consistent with the view that several deep provisions in RTAs, such as the ones on technical standards, sanitary measures or services, allow to reduce fixed entry costs associated to divergent regulations and policy uncertainty. – Table 7 about here – In a trade model with Pareto distributed firm productivity as in Chaney (2008), the aggregate exports of country i to country j , and hence the extensive and the intensive margin of trade, depend on the firm productivity dispersion parameter. When low- and medium-productivity firms represent a large fraction of firms in the country (i.e. skewed productivity distribution), a change in the productivity threshold induced by higher market access is expected to have large aggregate export consequences because many firms enter the export market (large extensive margin effect). Conversely, when the distribution of firm productivity is more dispersed, and large and high-productive firms represent a larger fraction of firms in the country, a change in the productivity threshold has only a marginal effect on the extensive margin, and the aggregate exports of the country increases only through the intensive margin channel. Thus, no matter the specific type of export costs impacted by deep RTAs, the aggregate effect of deep RTAs depends on productivity distribution of firms in the exporting country. In Table 8, we interact the RTA depth variable with two proxies of country-specific Pareto distribution parameter (falling with increase in the share of high-productive firms). One obtained by following the QQ approach in Head et al. (2014),19 the other by following Gabaix and Ibragimov (2011).20 To facilitate the interpretation of the results, we use a dummy variable for countries having above-the-median Pareto shape parameter. In line with the intuition, 19 As in Head et al. (2014), we retrieve the Pareto-shape parameter of firm size distribution by using the QQ estimator. We regress the empirical quantiles of the sorted log exports on the theoretical quantiles (i.e. −ln(1 − ((k − 0.3)/(n + 0.4)), where k is the firm’s ascending order of exports and n the rank of the firm having the highest export value). The coefficient of such regression, 1/θ˜, gives us the inverse of the Pareto shape parameter, that we ˜ recover as θ = (σ − 1)θ. We use elasticity of substitution σ = 5. 20 To reduce any endogeneity concern, we use beginning of the sample data to calculate Pareto share parameters. 15 deep RTAs have a strong positive effect on the extensive margin of countries having very skewed productivity distribution (i.e. high Pareto shape parameter) – see columns (1)-(2). Interestingly, in countries with a larger share of highly productive firms (i.e. Pareto shape parameter below the median) the extensive margin of export is negatively affected. This is likely due to the selection effect of deep RTAs discussed above. When the exporting country is populated by a large share of high-productivity firms (low Pareto shape parameter), the within-origin pro- competitive effect of DTAs makes less-productive firms exiting the market. Results for the intensive margin channel are showed in columns (3)-(4). Deep RTAs have a strong positive effect on the intensive margin of countries having large share of highly productive firms (i.e. Pareto parameter below the median) and tiny negative effect on countries having very skewed firm productivity distribution (i.e. large Pareto shape parameter).21 Results on total aggregate exports are reported in columns (5)-(6) and reflect results on the extensive and intensive margin channels discussed above. – Table 8 about here – 6 Endogeneity The inclusion of the set of fixed effects discussed above strongly reduces any omitted variable concern. Namely, any firm specific productivity shock, as well as any import demand shock and country-pair specific transaction cost are captured by fixed effect. In this setting, the only endo- geneity concern may come from the presence of unobserved factors affecting contemporaneously a change in the RTA depth and in the exports of firm f in country i. To further reduce any endogeneity concern, we propose an Instrumental Variable (IV) approach based on the idea that each country (i or j ) sets the depth of new RTAs based on the depth of existing ones (in the vein of the domino effect of RTA formation by Baldwin and Jaimovich 2012). Our instrumental variable is therefore the following: 1 1 IVijt = DT Aikt × DT Azjt (3) K − 1 k̸=j Z −1 z ̸=i where the two terms in brackets represent: (i) the average depth of RTAs signed by country i with trade partners k ̸= j within the j ’s macro region, and (ii) the average depth of RTAs signed by country j with trade partners z ̸= i within the i’s macro region.22 We use the leave-one-out means to construct the average depth in brackets to address the finite sample bias coming from 21 In the regressions reported in table 8 we include origin-year, destination-year and origin-destination fixed effects. 22 Macro-regions are: East Asia and Pacific, Europe and Central Asia, Latin America and Caribbean, Middle East and North Africa, North America, South Asia, Sub-Saharan Africa. 16 using own-observation information. The exclusion restriction is based on: (i) the absence of a direct effect of firms’ exports toward destination j on the average depth of RTAs signed by country i and j with third-countries, and (ii) on the absence of a direct effect of average RTA depth signed with third countries on firms’ exports toward j . While condition (i) is likely to hold, condition (ii) deserves careful discussion. Indeed, the presence of a trade diversion effect of deep RTAs threatens the validity of our IV: if country i signs a deep RTA with third country k (k ̸= j ) and this diverts firm’s exports from j to k , the validity of the IV is challenged. Notice, however, that the average reduction in trade costs of country i and country j with third countries is captured here by respectively firm-year and destination-year fixed effects. We are therefore confident of the exclusion restriction validity in our empirical setting. Still, in the next section we present a plausible exogeneity test aimed at supporting the validity of our results even in presence of weak deviation for the perfect validity of exclusion restriction assumption. Our baseline results are confirmed by the 2SLS estimations addressing any residual endo- geneity concerns - see Table 9 column 1. The (instrumented) measure of RTA depth has a positive and significant effect on the exports of the average firm. Also our results on the hetero- geneous impact of DTAs on firms with different characteristics is confirmed by 2SLS estimations - see Table 9 columns 2-7.23 The interaction terms are instrumented by simply interacting the ¯) with the IV discussed above. The bottom part of Table 9 shows the firm indicators I (kf > k first stage results of the 2SLS approach. Our IVs (one for the RTA depth and the other for its interaction with the firm type dummy) are good predictors of the endogenous variables (i.e. significant first stage coefficients). Also, the joint F-stat statistics well above 10 support the absence of a weak instrument problem. – Table 9 about here – 6.1 IV validity As discussed above, the validity of our instrumental variable is based on the absence of a direct effect of the depth of RTAs signed with third-country on f ij -specific exports. In this section we test the robustness of our baseline results to a deviation from the perfect validity of the exclusion restriction (Conley et al., 2012). A degree of deviation from the exclusion restriction can be obtained by regressing the exports of firms Xf ijt on our main DT Aijt and IV variable. The coefficient associated to the IV represents an approximation of the direct effect of the IV on the outcome variable (i.e. degree of deviation from the exclusion restriction) – van Kippersluis 23 Point estimates from linear 2SLS estimations in Table 9 cannot directly compared with our baseline non-linear PPML estimations. 17 and Rietveld (2018). We obtain a small and not statistically significant direct effect of IV on firm exports - see parameter ν in Table 10. We therefore plug such a degree of deviation from exclusion restriction in the plausible exogeneity test à la Conley et al. (2012), and obtain a lower- and upper-bound coefficients that do not cross the zero. Thus, we can safely argue that the depth of RTAs has an unambiguous positive causal effect on the export of firms in developing countries.24 – Table 10 about here – 7 The dynamic effect of deep RTAs: An event study approach We relied so far on a fixed effects approach delivering the average effect of deep RTAs on the exports of firms (i.e. pre- versus post- change in RTA depth). However, deep RTAs may stimulate firms’ exports dynamically and for a limited amount of time. In this section we adopt an event-study approach to visualize the dynamic effect of deep RTAs. Namely, we follow Fajgelbaum et al. (2019), and compare the targeted varieties (i.e. firm-destination combinations that face a change in RTA depth) to non-targeted varieties, using the following specification: 3 Xf ijt = exp θf + θjt + θij + β0z I (eventijt = z ) (4) z =−2 3 + β1z I (eventijt = z ) × targetf jt × εf ijt z =−2 . The event-study specification includes firm (θf ), destination-year (θjt ) and origin-destination (θij ) fixed effects. The indicator variable I (eventijt = z ) captures the event time coefficient before (z = −2, −1) and after (z = 1, 2, 3) the change in the RTA depth (z = 0).25 The target variable, targetf jt , is a dummy for varieties (i.e. firm-destination) that experience a change in the RTA depth during the period. The presence of firm fixed effects implies that β1z coefficients are identified using the variation between targeted and non-targeted destinations in each point in time z . For targeted varieties, the event date is the year of the first change in the RTA depth 24 For computational reasons (i.e. maximum number of covariates allowed by plausexog STATA command) we had to reduce the number dummies (fixed effects) included in the estimations. First, we replaced destination-year fixed effects by destination-period fixed effects, with periods containing 6-year each. Second, we restricted the number of destinations by: (i) using top-50 destinations for the all the 31 exporting countries of our sample (panel a in Table 10), (ii) keep only destinations representing at least the 0.5% of total exports of all the 31 exporting countries of our sample. 25 We adopt asymmetric time period before versus after change in RTA depth because in our data the pre- treatment period is limited. 18 between country i and j . For non-targeted varieties we assign the event date to be the earliest year in which the destination market j experiences a change in RTA’s depth with at least one of its trade partners. As in the baseline specification we adopt a PPML estimator and cluster standard errors by origin-destination-year. Figure 2 reports the impact of RTA depth on targeted varieties. On impact, we find a positive but imprecisely estimated effect of RTA depth on the export value of firms. The effect becomes statistically significant one and two years after the change in RTAs’ content. The positive effect vanishes after two years from the change in RTA depth. Interestingly, the event study approach also addresses concerns on the anticipation of changes in RTA depth (and/or RTA signature). During the pre-treatment period targeted and non-targeted varieties show a parallel trend. The absence of pre-trend reassures on the causal interpretation of our baseline results. 7.1 Two-way (robust) fixed effects estimations As recently argued by De Chaisemartin and d’Haultfoeuille (2020), two-way fixed effects esti- mations with heterogeneous treatment across groups and over time may be biased by negative weights. Indeed, comparing the outcomes of treated country-pairs with those of non-treated pairs that may (or may not) be treated afterwards may cause negative weights in the difference-in- difference estimator. Moreover, RTAs’ depth is a continuous treatment that can change over time and may have a dynamic effect (i.e. the initial variation in RTA depth may affect fu- ture variations in depth and the outcome variable), implying a second possible source of bias.26 Therefore, we follow De Chaisemartin and d’Haultfoeuille (2020) and perform a robust two-way fixed effects estimation of the trade effect of the number of legally enforceable provisions in RTAs. Since we are interested in the dynamic effect, we specifically adopt the estimator pro- posed by De Chaisemartin and D’Haultfoeuille (2020). The intuition behind this estimation is that, to avoid negative weights that may bias standard two-way fixed effects estimators, one should compare the outcome change from t − 1 to t + l for only the first time switchers’ (i.e. firm-destination pairs at first change in RTA depth occurred at t), to the outcome change of firm-destination pairs whose treatment has remained stable until t. 27 We present the results of the estimation with l ∈ [0, 3] in Table 11.28 In line with the event study approach reported in Figure 2, the depth of RTAs has a weak positive effect on the export of firms in the year of the first variation in RTA depth (i.e. t). The effect of RTA depth becomes strongly significant the year after t + 1, increases in magnitude at time t + 2 and vanishes at 26 Notice that this second source of bias is already taken into account in the event study exercise where we focus exclusively on the first variation in RTAs depth. 27 See De Chaisemartin and D’Haultfoeuille (2020) for more details on the robust two-way fixed effects estimator. 28 In Table A10 we show the same estimations including firm’s tariffs as a controls and results hold. 19 t + 3. While the pattern is the same as in the standard event study approach reported in the previous section, the point estimates in Table 11 differ with respect to those reported in Figure 2. However, such a difference is small if standard errors are considered; suggesting the presence of a (very) small negative bias in the standard two-way fixed effects event study approach. Conclusion This paper studies the effect of deep trade agreements on the export performance of firms in 31 developing countries. We show a moderate but statistically significant effect of RTA depth on the exports of the average firm. Namely, one additional legally enforceable provision boosts the export of firms by 0.3%. This implies that moving from shallow to deep RTAs leads to a 3.6% increase in firms’ exports. This average effect is however strongly heterogeneous across firms with different characteristics. Large and GVC firms are more positively affected by deep RTAs. For firms belonging to the top-quartile of size distribution, moving from shallow to deep RTAs implies 4.8% increase in their exports. For GVC firms importing and exporting from/to the same country, moving from shallow to deep RTAs implies a 9.6% increase in their exports. Conversely, small and less productive firms are negatively affected by deep RTAs as they suffer the higher degree of competition induced by deep RTA at destination. These results are robust to a number of extensions and robustness checks that confirm the causal impact of deep RTAs on firms’ exports. These findings have relevant welfare and policy implications for developing countries. While the selection effect, through the reallocation of resources toward more productive firms, is expected to improve welfare in countries joining deep trade agreements, the negative export performance of small firms signals that the adjustment process in developing countries can be significant. 20 Bibliography Atkeson, A. and A. Burstein (2008). Pricing-to-market, trade costs, and international relative prices. American Economic Review, 98 (5), 1998–2031. Baier, S. L., J. H. Bergstrand, and M. W. Clance (2018). Heterogeneous effects of economic integration agreements. Journal of Development Economics 135, 587–608. Baier, S. L., J. H. Bergstrand, and M. Feng (2014). Economic integration agreements and the margins of international trade. Journal of International Economics 93 (2), 339–350. Baier, S. L., Y. V. Yotov, and T. Zylkin (2019). On the widely differing effects of free trade agreements: Lessons from twenty years of trade integration. Journal of International Eco- nomics 116, 206–226. Baldwin, R. and D. Jaimovich (2012). Are Free Trade Agreements contagious? Journal of International Economics 88 (1), 1–16. Berman, N., P. Martin, and T. Mayer (2012). How do different exporters react to exchange rate changes? Quarterly Journal of Economics 127 (1), 437–492. Bernard, A. B., J. B. Jensen, S. J. Redding, and P. K. Schott (2007, Summer). Firms in International Trade. Journal of Economic Perspectives 21 (3), 105–130. Chaney, T. (2008). Distorted gravity: the intensive and extensive margins of international trade. American Economic Review 98 (4), 1707–1721. Conley, T. G., C. B. Hansen, and P. E. Rossi (2012, February). Plausibly Exogenous. The Review of Economics and Statistics 94 (1), 260–272. Crowley, M., L. Han, and T. Prayer (2022). The pro-competitive effects of trade agreements. Working Paper 2240, University of Cambridge, Cambridge. De Chaisemartin, C. and X. D’Haultfoeuille (2020). Difference-in-differences estimators of in- tertemporal treatment effects. arXiv preprint arXiv:2007.04267 . De Chaisemartin, C. and X. d’Haultfoeuille (2020). Two-way fixed effects estimators with het- erogeneous treatment effects. American Economic Review 110 (9), 2964–96. Fajgelbaum, P. D., P. K. Goldberg, P. J. Kennedy, and A. K. Khandelwal (2019, 11). The Return to Protectionism*. The Quarterly Journal of Economics 135 (1), 1–55. Fajgelbaum, P. D., P. K. Goldberg, P. J. Kennedy, and A. K. Khandelwal (2020, February). The return to protectionism. The Quarterly Journal of Economics 135 (1), 1–55. Feng, L., Z. Li, and D. L. Swenson (2016). The connection between imported intermediate 21 inputs and exports: Evidence from chinese firms. Journal of International Economics 101, 86–101. Fernades, A., K. Lefebvre, and N. Rocha (2021, November). Heterogeneous Impacts of SPS and TBT Regulations : Firm-Level Evidence from Deep Trade Agreements. Policy Research Working Paper Series 9700, The World Bank. Fernandes, A., N. Rocha, and M. Ruta (2021). The Economics of Deep Trade Agreements. CEPR Press. Fernandes, A. M., C. Freund, and M. D. Pierola (2016, March). Exporter behavior, country size and stage of development: Evidence from the exporter dynamics database. Journal of Development Economics 119, 121–137. Fontagné, L., G. Orefice, R. Piermartini, and N. Rocha (2015). Product standards and margins of trade: Firm-level evidence. Journal of International Economics 97 (1), 29–44. Freund, C. and E. Ornelas (2010). Regional trade agreements. Annual Review of Economics 2 (1), 139–166. Gabaix, X. and R. Ibragimov (2011). Rank- 1/2: a simple way to improve the ols estimation of tail exponents. Journal of Business & Economic Statistics 29 (1), 24–39. Goldberg, P. K., A. K. Khandelwal, N. Pavcnik, and P. Topalova (2010, 11). Imported Interme- diate Inputs and Domestic Product Growth: Evidence from India*. The Quarterly Journal of Economics 125 (4), 1727–1767. Head, K. and T. Mayer (2014). Gravity equations: Workhorse, toolkit, and cookbook. In Hand- book of International Economics, Volume 4, Handbook of International Economics, Chapter 4. Gita Gopinath, Elhanan Helpman and Kenneth Rogoff editors. Head, K., T. Mayer, and M. Thoenig (2014, May). Welfare and trade without pareto. American Economic Review 104 (5), 310–16. Hofmann, C., A. Osnago, and M. Ruta (2017, February). Horizontal depth: A new database on the content of preferential trade agreements. Policy Research Working Paper 7981, World Bank Group, Washington, DC. Horn, H., P. C. Mavroidis, and A. Sapir (2010, November). Beyond the WTO? An Anatomy of EU and US Preferential Trade Agreements. The World Economy 33 (11), 1565–1588. Lee, W., A. Mulabdic, and M. Ruta (2019, November). Third-Country Effects of Regional Trade Agreements: A Firm-Level Analysis. Policy Research Working Paper Series 9064, The World Bank. 22 Limão, N. (2016). Preferential trade agreements. In Handbook of commercial policy, Volume 1, pp. 279–367. Elsevier. Mattoo, A., A. Mulabdic, and M. Ruta (2017, September). Trade creation and trade diversion in deep agreements. Policy Research Working Paper Series 8206, The World Bank. Mattoo, A., N. Rocha, and M. Ruta (2020). Handbook of deep trade agreements. World Bank Publications. Melitz, M. J. (2003, November). The impact of trade on intra-industry reallocations and aggre- gate industry productivity. Econometrica 71 (6), 1695–1725. Melitz, M. J. and S. J. Redding (2015). New trade models, new welfare implications. American Economic Review 105 (3), 1105–46. Orefice, G. and N. Rocha (2014, January). Deep Integration and Production Networks: An Empirical Analysis. The World Economy 37 (1), 106–136. Redding, S. J. and D. E. Weinstein (2019, May). Aggregation and the gravity equation. AEA Papers and Proceedings 109, 450–55. Santos-Silva, J. M. C. and S. Tenreyro (2006, November). The Log of Gravity. The Review of Economics and Statistics 88 (4), 641–658. Spearot, A. C. (2013). Variable demand elasticities and tariff liberalization. Journal of Inter- national Economics 89 (1), 26–41. van Kippersluis, H. and C. A. Rietveld (2018, October). Beyond plausibly exogenous. Econo- metrics Journal 21 (3), 316–331. 23 Tables and Figures Table 1: Number of countries with variation in RTAs’ depth in the period 2000-2020 At least one partner having : constant changing RTA depth RTA depth Number of countries (EDD) 48 31 Number of country-pairs (final sample) 1004 701 Notes: Authors’ calculation on World Bank Content of Deep Trade Agreement data. Our final sample contains only the 31 exporting countries with variation in RTA depth. Table 2: In-sample descriptive statistics. Mean Std Dev Min Max Export 279032 1.1e+06 0 1.4e+06 RTA depth 14.4 13.8 0 48 RTA depth legally enf. 8.3 7.5 0 43 RTA depth WTO+ 6.2 5.2 0 14 RTA depth WTOX 2.1 3.1 0 29 RTA depth core 7.4 6.3 0 18 ln(1+τ ) 0.04 0.09 0 2.40 Notes: Authors’ calculation on Export Dynamic Database and World Bank Content of Deep Trade Agreement data. 24 25 Figure 1: Exports value and RTAs, major treated countries (a) RTAs, country level (b) Depth, country level (c) RTAs, firm level (d) Depth, firm level Note: K-density graphs are realized compiling country’s exports for last year of available data. Table 3: The trade effect of deep RTA. Within estimations. Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) ∗∗∗ RTAijt 0.047 (0.015) DTAijt 0.003∗∗∗ (0.001) DTAijt leg. 0.003∗∗∗ (0.001) DTAijt WTO+ 0.003∗ (0.001) DTAijt WTO-X 0.007∗∗∗ (0.002) DTAijt Core 0.002∗∗ (0.001) Ln(1+τijt ) -0.671∗∗∗ -0.671∗∗∗ -0.671∗∗∗ -0.671∗∗∗ -0.672∗∗∗ -0.671∗∗∗ (0.045) (0.045) (0.045) (0.045) (0.045) (0.045) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ ✔ Observations 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 26 Table 4: The trade effect of deep RTA. Robustness check using the weighted sum of provision as proxy for the RTA depth. Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) ∗∗∗ Weigh. DTAijt 0.006 (0.001) Weigh. DTAijt leg. 0.005∗∗∗ (0.002) Weigh. DTAijt WTO+ 0.004 (0.004) Weigh. DTAijt WTO-X 0.009∗∗∗ (0.002) Weigh. DTAijt core 0.004 (0.003) Ln(1+τijt ) -0.671∗∗∗ -0.672∗∗∗ -0.671∗∗∗ -0.672∗∗∗ -0.671∗∗∗ (0.045) (0.045) (0.045) (0.045) (0.045) Firm-Year FE ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ Observations 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. Table 5: The trade effect of deep RTA. Within estimations by firm characteristics. Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ DTAijt leg. -0.042 -0.022 -0.006 -0.005 -0.011 -0.001 (0.002) (0.002) (0.001) (0.001) (0.002) (0.001) DTAijt leg. × (kf > 75th ) 0.045∗∗∗ (0.002) DTAijt leg. × (kf > 90th ) 0.026∗∗∗ (0.002) DTAijt leg. × (kf > 75th t0) 0.010∗∗∗ (0.001) DTAijt leg. × (kf > 90th t0) 0.011∗∗∗ (0.001) DTAijt leg. × GVC 0.016∗∗∗ (0.002) DTAijt leg. × GVC bil. 0.008∗∗∗ (0.001) GVC bil. 0.141∗∗∗ (0.014) Ln(1+τijt ) -0.670∗∗∗ -0.667∗∗∗ -0.670∗∗∗ -0.667∗∗∗ -0.665∗∗∗ -0.662∗∗∗ (0.045) (0.045) (0.045) (0.045) (0.045) (0.045) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ ✔ Observations 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 27 Table 6: The selection effect of deep RTA. # Survivors # Survivors Avg Exp Avg Exp (1) (2) (3) (4) ∗∗∗ ∗∗∗ ∗∗ DTAijt leg. -0.006 -0.005 0.003 0.006∗∗∗ (0.001) (0.001) (0.002) (0.002) Ln(1+τijt ) 0.518∗∗∗ -0.165∗ (0.092) (0.098) Origin-Year FE ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ Observations 38,045 25,308 49,772 30,212 Notes: The dependent variable in columns (1)-(2) is the number of firms in country i exporting to destination j at time t. The dependent variable in columns (3)-(4) is the average export sales per exporting firm (i.e. total export sales over total number of exporters). The dependent variable in columns (5)-(6) is the number of exporters that survive at destination. PPML estimates in columns (1)-(6). Origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. Table 7: Intensive vs extensive margin contribution to export response to deep RTAs. ˆ β Vi /V Aggregate Response Aggregate Response (% of total) Intensive 0.045∗∗∗ 0.336 0.015 40 Extensive 0.034∗∗∗ 0.664 0.023 60 Total 0.038 Notes: β ˆ is the estimated coefficient for RTA depth on a gravity type regression (PPML) having the total country-pair-year specific exports for incumbent and entry- exit exporters. Vi /V is the share of total aggregate exports by respectively incumbent and entry-exit exporters. The aggregate response is calculated as β ˆ × Vi /V . Table 8: The trade effect of deep RTA. The role of firm size distribution Extensive Extensive Intensive Intensive Exp Exp (1) (2) (3) (4) (5) (6) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ DTAijt leg. -0.008 -0.006 0.009 0.008 -0.005 -0.004∗∗ (0.002) (0.002) (0.001) (0.002) (0.001) (0.002) DTAijt leg. × Pareto shapea 0.010∗∗∗ -0.008∗∗∗ 0.007∗∗∗ (0.002) (0.002) (0.002) DTAijt leg. × Pareto shapeb 0.004∗ -0.005∗∗∗ 0.003∗ (0.002) (0.002) (0.002) Ln(1+τijt ) -0.021 -0.020 0.185 0.186 -0.005 -0.005 (0.115) (0.115) (0.130) (0.130) (0.105) (0.105) Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ ✔ Observations 30,182 30,182 12,537 12,537 30,212 30,212 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. (a) Pareto shape parameter estimated following Head et al. (2014). (b) Pareto share parameter estimated following Gabaix and Ibragimov (2011) 28 Table 9: The trade effect of deep RTA. Within estimations by firm characteristics. 2SLS approach Exp Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) (7) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ DTAijt leg. 0.007 -0.042 -0.024 -0.019 -0.008 -0.004 0.001 (0.003) (0.007) (0.005) (0.004) (0.004) (0.004) (0.003) DTAijt leg. × (kf > 75th ) 0.054∗∗∗ (0.007) DTAijt leg. × (kf > 90th ) 0.041∗∗∗ (0.006) DTAijt leg. × (kf > 75th t0) 0.037∗∗∗ (0.005) DTAijt leg. × (kf > 90th t0) 0.030∗∗∗ (0.004) DTAijt leg. × GVC 0.016∗∗∗ (0.004) DTAijt leg. × GVC bil. 0.015∗∗∗ (0.002) GVC bil. -0.008 (0.023) Ln(1+τijt ) -0.575∗∗∗ -0.572∗∗∗ -0.567∗∗∗ -0.574∗∗∗ -0.570∗∗∗ -0.574∗∗∗ -0.577∗∗∗ (0.040) (0.040) (0.040) (0.040) (0.040) (0.040) (0.040) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ ✔ ✔ Observations 2,280,704 2,280,704 2,280,704 2,280,704 2,280,704 2,280,704 2,280,704 IV DTAijt 0.064∗∗∗ 0.065∗∗∗ 0.065∗∗∗ 0.065∗∗∗ 0.065∗∗∗ 0.066∗∗∗ 0.064∗∗∗ ¯) IV DTAijt × I (kf > k 0.108∗∗∗ 0.111∗∗∗ 0.109∗∗∗ 0.110∗∗∗ 0.114∗∗∗ 0.084∗∗∗ Joint F-stat 318 159 159 159 159 163 161 Notes: 2SLS estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical signifi- cance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 29 Table 10: Deep trade agreements and the export of firms with plausibly exogenous instrument. Union of Confidence Interval estimations Dep Var ν Coeff. in Min Max tab 9 90% CI 90% CI Panel (a): Top-50 destinations. Firm Exports 0.0006 0.008*** 0.028 0.042 (0.0004) (0.003) Panel (b): market share above 0.5%. Firm Exports 0.0006 0.008*** 0.027 0.041 (0.0004) (0.003) Notes : UCI based on γ coefficients from a regression of firm exports on the IV. Standard errors in parenthesis cluster by origin-destination- year. To meet the maximum number of covariates allowed by plausexog STATA command we had to reduce the number of destinations (i.e. number of origin-destination dummies). In panel (a) we use top-50 des- tination countries. In panel (b) we keep destinations counting for at least the 0.5% of total exports of all origin countries in the sample. For the same computational reasons, the plausibly exogeneity test has been con- ducted by replacing destination-year fixed effects by destination-period fixed effects (each period covering a six-year windows). Figure 2: Firm Exports Event Study Note: Figure plots event time dummies for targeted firms relative to untargeted firms. Regression includes firm-year, destination-year and origin-destination fixed effects. Standard errors are clustered by origin-destination-year. Error bars show 90% confidence intervals. 30 Table 11: Robustness check using the estimator by Chaisemartin D’Haultfoeuille (2020) Dependent variable: Bilateral exports (ln) Time t t+1 t+2 t+3 DTAijt 0.094∗∗ 0.170∗∗ 0.270∗∗∗ 0.154 (0.045) (0.069) (0.082) (0.107) No. observ. 1.568.450 1.440.873 1.279.444 1.100.902 No. switchers 174.852 167.748 159.379 141.907 Dependent variable is the log of exports from the country i to the country j at time t in current million dollars. t is the year of first depth variation in the country-pair. Bootstrapped (100) standard errors clustered at the country-pair-time level are in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. A1 Appendix tables and figures Figure A1: Frequency of legally enforceable provisions in RTAs Note: Provisions are ranked from the less frequent to the most frequent. 31 32 Figure A2: Exports value and RTAs, all available countries (a) RTAs, country level (b) Depth, country level (c) RTAs, firm level (d) Depth, firm level Note: K-density graphs are realized compiling country’s exports for last year of available data. 33 Table A1: Time periods covered by country specific EDD data. Country (iso3) Period Country (iso3) Period ECU 2002-2019 ROU 2005-2011 EGY 2005-2016 RWA 2005-2016 ETH 2008-2017 SEN 2000-2020 GAB 2009-2015 SLV 2006-2020 GHA 2010-2019 SRB 2006-2019 GIN 2009-2012 STP 2014 GTM 2005-2013 SWZ 2012 HRV 2007-2015 TLS 2006-2012 IRN 2006-2010 TZA 2003-2017 MEX 2011-2016 UGA 2000-2010 JOR 2003-2012 URY 2001-2020 KEN 2006-2020 YEM 2008-2012 KGZ 2006-2012 ZAF 2009-2020 KHM 2016-2019 ALB 2007-2019 KWT 2009-2010 BDI 2010-2016 LBN 2008-2012 BEN 2016-2020 MDG 2007-2012 BFA 2005-2012 MKD 2008-2017 BGD 2005-2016 MLI 2005-2008 BGR 2001-2006 MMR 2011-2013 BOL 2006-2012 MUS 2010-2020 BWA 2003-2013 MWI 2005-2020 CHL 2000-2020 NER 2008-2010 CIV 2009-2019 NIC 2012-2014 COL 2000-2020 NPL 2011-2014 CPV 2010-2020 PAK 2015-2017 DOM 2006-2020 PER 2000-2020 GEO 2000-2020 PRY 2012-2020 Notes: World Bank Export Dynamics Database. Table A2: Firms’ export growth and change in RTAs’ depth. Firm’s average export growth before RTA depth change (1) (2) (3) Change in RTA depth 0.001 0.001 -0.000 (0.001) (0.001) (0.002) Origin FE ✔ ✔ Destination FE ✔ ✔ Observations 216,877 216,875 216,874 Notes: OLS estimates, robust standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 34 Table A3: The effect of Deep Trade Agreements. Robustness check using Linear OLS estimations. Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) RTAijt 0.054∗∗∗ (0.018) DTAijt 0.002∗∗∗ (0.001) DTAijt leg. 0.005∗∗∗ (0.001) DTAijt WTO+ 0.005∗∗ (0.002) DTAijt WTO-X 0.014∗∗∗ (0.002) DTAijt Core 0.004∗∗ (0.002) Ln(1+τijt ) -0.560∗∗∗ -0.560∗∗∗ -0.560∗∗∗ -0.561∗∗∗ -0.561∗∗∗ -0.561∗∗∗ (0.039) (0.039) (0.039) (0.039) (0.039) (0.039) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ ✔ Observations 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 Notes: OLS estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 35 Table A4: The trade effect of deep RTA. Cross-section identification. Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) RTAijt 0.266∗∗∗ (0.019) DTAijt 0.006∗∗∗ (0.001) DTAijt leg. 0.022∗∗∗ (0.001) DTAijt WTO+ 0.032∗∗∗ (0.002) DTAijt WTO-X 0.038∗∗∗ (0.004) DTAijt Core 0.025∗∗∗ (0.002) Distance (ln) -0.368∗∗∗ -0.406∗∗∗ -0.352∗∗∗ -0.351∗∗∗ -0.390∗∗∗ -0.355∗∗∗ (0.017) (0.016) (0.017) (0.016) (0.017) (0.016) Contiguity 0.392∗∗∗ 0.380∗∗∗ 0.381∗∗∗ 0.382∗∗∗ 0.380∗∗∗ 0.391∗∗∗ (0.027) (0.027) (0.024) (0.023) (0.026) (0.024) Language 0.168∗∗∗ 0.209∗∗∗ 0.214∗∗∗ 0.207∗∗∗ 0.222∗∗∗ 0.209∗∗∗ (0.019) (0.021) (0.019) (0.019) (0.021) (0.019) Colony 0.174∗∗∗ 0.199∗∗∗ 0.177∗∗∗ 0.170∗∗∗ 0.168∗∗∗ 0.184∗∗∗ (0.047) (0.045) (0.044) (0.044) (0.043) (0.044) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE No No No No No No Observations 3,488,011 3,488,011 3,488,011 3,488,011 3,488,011 3,488,011 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 36 Table A5: The trade effect of deep RTA. Cross-section identification controlling for applied tariffs. Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) ∗∗∗ RTAijt 0.164 (0.017) DTAijt 0.003∗∗∗ (0.001) DTAijt leg. 0.015∗∗∗ (0.001) DTAijt WTO+ 0.022∗∗∗ (0.002) DTAijt WTO-X 0.022∗∗∗ (0.003) DTAijt Core 0.017∗∗∗ (0.001) Ln(1+τijt ) -0.613∗∗∗ -0.661∗∗∗ -0.625∗∗∗ -0.612∗∗∗ -0.673∗∗∗ -0.617∗∗∗ (0.048) (0.048) (0.048) (0.048) (0.049) (0.049) Distance (ln) -0.425∗∗∗ -0.458∗∗∗ -0.403∗∗∗ -0.402∗∗∗ -0.439∗∗∗ -0.407∗∗∗ (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Contiguity 0.287∗∗∗ 0.268∗∗∗ 0.289∗∗∗ 0.292∗∗∗ 0.274∗∗∗ 0.291∗∗∗ (0.024) (0.024) (0.023) (0.022) (0.024) (0.023) Language 0.106∗∗∗ 0.124∗∗∗ 0.140∗∗∗ 0.135∗∗∗ 0.136∗∗∗ 0.136∗∗∗ (0.015) (0.015) (0.015) (0.015) (0.015) (0.015) Colony 0.186∗∗∗ 0.199∗∗∗ 0.187∗∗∗ 0.184∗∗∗ 0.183∗∗∗ 0.192∗∗∗ (0.049) (0.048) (0.047) (0.048) (0.047) (0.047) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE No No No No No No Observations 2,388,510 2,388,510 2,388,510 2,388,510 2,388,510 2,388,510 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 37 Table A6: The trade effect of deep RTA. Within estimations controlling for the presence of a RTA. Exp Exp Exp Exp (1) (2) (3) (4) DTAijt 0.005∗∗∗ 0.005∗∗ (0.001) (0.002) DTAijt leg. 0.004∗∗ 0.000 (0.002) (0.002) Ln(1+τijt ) -0.715∗∗∗ -0.714∗∗∗ -0.662∗∗∗ -0.662∗∗∗ (0.118) (0.118) (0.119) (0.119) RTA adjusted -0.042 0.022 -0.054 0.062 (0.039) (0.034) (0.059) (0.038) Depth non-coded RTA =1 =1 Closest Closest Firm-Year FE ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ Observations 2,670,797 2,670,797 2,447,547 2,447,547 Notes: PPML estimates, origin-destination-year cluster standard errors in paren- theses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. Table A7: The trade effect of deep RTA. Robustness check using sub-sample of country pairs that switch RTA status during the period. Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) ∗∗∗ RTAijt 0.059 (0.016) DTAijt 0.002∗∗∗ (0.001) DTAijt leg. 0.003∗∗∗ (0.001) DTAijt WTO+ 0.003∗∗ (0.001) DTAijt WTO-X 0.006∗∗∗ (0.002) DTAijt Core 0.003∗∗ (0.001) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ ✔ Observations 2,718,720 2,718,720 2,718,720 2,718,720 2,718,720 2,718,720 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 38 Table A8: The trade effect of deep RTA. Within estimations by exporting firm char- acteristics. Binned model. Exp Exp (1) (2) ∗∗∗ DTAijt leg. × Big 0.003 0.003∗∗∗ (0.001) (0.001) DTAijt leg. × Medium -0.045∗∗∗ -0.040∗∗∗ (0.002) (0.002) DTAijt leg. × Small -0.150∗∗∗ -0.156∗∗∗ (0.007) (0.010) Ln(1+τijt ) -0.670∗∗∗ (0.045) Firm-Year FE ✔ ✔ Destination-Year FE ✔ ✔ Origin-Destination FE ✔ ✔ Observations 3,898,746 2,388,213 Notes: PPML estimates, origin-destination-year clus- ter standard errors in parentheses. ***, ** and * in- dicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. Table A9: The heterogeneous effect of deep RTA. Robustness check using the weighted sum of provision as proxy for the RTA depth. Exp Exp Exp Exp Exp Exp (1) (2) (3) (4) (5) (6) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Weig. DTAijt leg. -0.077 -0.037 -0.009 -0.009 -0.022 -0.002 (0.006) (0.005) (0.003) (0.003) (0.004) (0.002) Weig. DTAijt leg. × (kf > 75th ) 0.082∗∗∗ (0.006) Weig. DTAijt leg. × (kf > 90th ) 0.044∗∗∗ (0.004) Weig. DTAijt leg. × (kf > 75th t0) 0.016∗∗∗ (0.003) Weig. DTAijt leg. × (kf > 90th t0) 0.018∗∗∗ (0.002) Weig. DTAijt leg. × GVC 0.031∗∗∗ (0.003) Weig. DTAijt leg. × GVC bil. 0.013∗∗∗ (0.002) GVC bil. 0.166∗∗∗ (0.013) Ln(1+τijt ) -0.671∗∗∗ -0.669∗∗∗ -0.671∗∗∗ -0.668∗∗∗ -0.666∗∗∗ -0.663∗∗∗ (0.045) (0.045) (0.045) (0.045) (0.045) (0.045) Firm-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Destination-Year FE ✔ ✔ ✔ ✔ ✔ ✔ Origin-Destination FE ✔ ✔ ✔ ✔ ✔ ✔ Observations 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 2,388,213 Notes: PPML estimates, origin-destination-year cluster standard errors in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1. 39 Table A10: Robustness check using the estimator by Chaisemartin D’Haultfoeuille (2020) with control for tariffs Dependent variable: Bilateral exports (ln) Time t t+1 t+2 t+3 DTAijt 0.090∗ 0.251∗∗ 0.162 -0.064 (0.053) (0.109) (0.141) (0.161) Tariffs control yes yes yes yes No. observ. 1.003.047 913.310 803.168 684.178 No. switchers 88.713 82.109 77.587 61.992 Dependent variable is the log of exports from the country i to the country j at time t in cur- rent million dollars. t is the year of first depth variation in the country-pair. Bootstrapped (100) standard errors clustered at the country-pair-time level are in parentheses. ***, ** and * indicate statistical significance levels for p-val. < 0.01, p-val. < 0.05, and p-val.< 0.1.