Policy Research Working Paper 11136 Industrial Policy under Constraints Evidence from Pakistan’s Export Subsidy Schemes Stefania Lovo Gonzalo J. Varela Economic Policy Global Department June 2025 Policy Research Working Paper 11136 Abstract This paper investigates the impact of an export promotion expense of non-eligible or lower-rated products. The paper policy consisting of ad-valorem subsidies for a set of tar- presents suggestive evidence indicating that the observed geted products, on the performance of Pakistani exports in effects are partially attributed to shifts in the composition the textile sector. The findings show that the policy had a of exporters, through exits and entries alongside capacity small overall impact on exports, while it induced substantial constraints. Finally, the evidence shows that the scheme reallocation of exports across products. The policy induced induced strategic misreporting at the border, which con- an increase in exports of traditional products, mostly gar- tributed marginally to the overall effects. ments, which were eligible for the highest rebate rates, at the This paper is a product of the Economic Policy Global Department. 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 gvarela@worldbank.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 Industrial Policy under Constraints: Evidence from Pakistan’s Export Subsidy Schemes Stefania Lovo∗ Gonzalo Varela† Keywords: Export promotion, export subsidies, industrial policy, Pakistan JEL codes: F13, F14, F63, O2 ∗ University of Reading. Email: s.lovo@reading.ac.uk † World Bank 1 Introduction Over the past decades, industrial policy has shifted from inward-looking and protec- tionist to outward-oriented and focused on promoting exports and export sophistication (Juh´ asz et al., 2023). The presence of positive externalities, such as learning from ex- porting, has offered theoretical justification for deviating from policy neutrality (Harrison and Rodr´ ıguez-Clare, 2010). Yet the impact of industrial policies geared towards export promotion is still highly debated among economists, and the few empirical evaluations available show ambiguous and context-specific results (Belloc and Di Maio, 2018). De- spite this ambiguity, governments across the world spend substantial public funds on export promotion interventions. Promotion schemes typically target specific products or destinations considered to be of key importance. Hence, while the schemes aim at increasing exports, they can also alter the allocation of resources to favor some products and destinations at the expense of others. Understanding the impact of these mea- sures on export growth and composition matters for better designed and more impactful interventions. In this paper, we study the impact of the “Duty Drawback of Taxes” (DDT) scheme, effectively an export subsidy. The scheme is considered a key export promotion policy adopted by the Pakistani government for the textile sector, which accounts for 55% of total exports. Prior to 2010, the subsidy scheme was implemented intermittently, with extremely limited budget allocations and frequent interruptions in disbursements.1 At the end of 2014, the government launched a trade policy aimed at boosting textile exports, which involved a substantial increase in the budget allocated to the scheme. Over the period 2017-2020, funds allocated to the scheme reached about 1% of the total federal budget. The scheme targeted specific products and adopted heterogeneous rebate rates across eligible products. We investigate the impact of the scheme, using product level data, which allows us to circumvent issues of self-selection into export promotion policies at the firm level. By comparing eligible and non-eligible products, we can also investigate export reallocation. We do so by using a synthetic control method approach, combined with event studies to test for pre-trends across different eligibility categories. Our results show that the DDT scheme had a small positive impact on aggregate tex- tile exports. The small aggregate effect, however, masks substantial reallocation across products. For each US$1 spent on the DDT schemes, only US$1.1 were generated as additional exports. The policy induced an increase in exports of products eligible for the highest rebate rates at the expense of non-eligible, and of lower-rebate rate products. We then investigate potential underlying mechanisms. Analysis of exporters’ transactions data reveals that the effects are partially explained by a shift in exporters composition, as entrants are less likely to be specialized in low-rate products while firms that exit the export market are those more likely to be specialized in lower-rate and non-eligible products. In addition, we provide suggestive evidence of supply constraints that prevent small and medium-sized firms from expanding exports of high-rate products without contracting exports of lower-rate products. We also show that the scheme induced an 1 https://www.brecorder.com/news/3881630 2 increase in strategic misreporting at the border for non-eligible products. This strategic behavior, however, had a small overall impact on recorded exports, and does not explain the larger reallocation we document. Results have broad implications and suggest that interventions targeted at specific products can lead to unintended reallocations, particu- larly if firms face capacity or financial constraints. The concentration of the export base into fewer products that increases the vulnerability of exports to product-specific shocks, should be weighed against the small aggregate effect of the scheme when evaluating the overall impact of this intervention. The contribution of this paper is twofold. First, it addresses a gap in the empirical literature by investigating the impact of export subsidies, which has received limited attention thus far (Defever et al. (2020a), Helmers and Trofimenko (2013)). Further- more, the broader topic of export promotion policies remains highly debated among economists, with evaluations yielding ambiguous results (Belloc and Di Maio, 2018). Second, previous studies on export promotions have predominantly focused on exten- sive margin effects, such as the entry of new firms, products, or destinations targeted by the interventions (Lederman et al. (2010), Martincus and Carballo (2010), Munch and Schaur (2018)). This paper adopts a more comprehensive perspective, assessing the ramifications of export subsidies not only on the intended target products but also on non-targeted exports. The remainder of this paper is structured as follows. Section 2 provides background on the intervention. Section 3 and 4 describe the data and methodological approach. Sections 5 and 6 discuss the main results and the analysis of underlying mechanisms, including exporters’ entry and exit dynamics, capacity constraints and misreporting. Section 7 concludes. 2 Export Promotion and Performance Export promotion exists in various forms. Most of the literature has focused on generic export assistance, which goes from the existence of export promotion agencies to techni- cal assistance provided to specific firms. Studies tend to investigate the impact of export promotion on the extensive margin, both in terms of new products and destinations, but also in terms of entry in the export market. At the cross-country level, Lederman et al. (2010) for example, look at the effect of export promotion agencies (EPA) and find that they help overcome foreign trade barriers, thereby increasing export participation. Mar- tincus and Carballo (2010) find that export promotion assistance in Chile has favored export growth and also the growth in the number of destinations and products. The ef- fects were more pronounced for smaller firms. Similar effects were also found for Peru, in Martincus and Carballo (2008). On export participation, Broocks and Van Biesebroeck (2017) found positive effects of Flemish export assistance on entry in the export market. Similar effects were found for Dutch firms, especially small firms (Munch and Schaur, 2018). Fewer studies have focused on intensive margin effects. Martincus and Carballo (2008), for example, used firm-level data on Peru, and found that assistance from export promotion agencies had no effect on the intensive margins of exports. Finally, besides 3 export participation and growth, there is evidence that export promotion measures help firms survive in export markets (Van Biesebroeck et al., 2016). Less attention has been given to export subsidies, especially when differential rates are applied, which seems to be common practice (Tokarick and Subramanian, 2003). One rationale for varying rebate rates is to reflect the different (indirect) tax burdens these products face, taking into consideration the structure of inputs necessary for their production (Tokarick and Subramanian, 2003). The challenge with this approach is to design a differentiated structure of subsidy rates is that the finer the differentiation, the greater will be the information requirements, e.g. input-output coefficients, rendering it more difficult and costly to implement. Another rationale for differential drawback rates relates to the externalities associated with exporting specific products. For example, if local discovery costs generate knowledge spillovers with implications on future economic growth, then it may be warranted to subsidize the production or exports of products more likely to generate more of these spillovers (Hausmann et al., 2007). Analogously, negative rates could be applicable to products with negative externalities (e.g., prod- ucts with high water content in a water scarce context, or inputs the policy maker may consider crucial for the development of a high spillover downstream sector). A similar challenge with this rationale arises in terms of information requirements. The identifica- tion of positive or negative externalities associated with the production and exports of a particular product is not without debate. Also importantly, the rebate rate differen- tiation increases incentives for rent-seeking and lobbying for advantageous rates, as well as for misreporting at the border. Few studies have investigated the effects of export promotion on non-targeted products. For instance, Defever et al. (2020b) found that subsidized firms in Nepal increased their exports of targeted product-destination combi- nations but observed no significant effects for non-targeted ones. In this paper, we aim to contribute to this limited literature by examining the impact of export subsidies tar- geting traditional export products, both on the targeted products and on non-targeted exports. 2.1 Export promotion policies in Pakistan Pakistan has long been implementing measures to support exports. Traditionally, ex- port promotion schemes were limited to five so-called “zero-rated sectors”: textiles, sports goods, surgical goods, leather and carpets (Naqvi et al., 2019). Over the last decade, however, there has been a growing consensus on the need for a level playing field, advocating for wider accessibility of these schemes. Hence, two main modalities have emerged. On the one hand, various schemes allow all exporters to avoid paying duties on inputs necessary to produce exportables. These include the regime of “Manufacturing under Bond”, the regime of “Export Oriented Units” and the regime of “Temporary Importation” (this one is sector specific). These schemes impose certain conditions on the types of imported inputs that are eligible, and provide the suspension or exemption of import duties, sales tax, federal excise tax and withholding taxes on the acquisition 4 of inputs required for the manufactures of exportable products.2 In addition, under the regime of “Duty and Tax Remission for Exporters” exporters who choose to import inputs pay the duties and then file for a refund, which takes a long time to be processed, according to anecdotal evidence. Thus, these systems are not widely used by exporters in Pakistan. On the other hand, there are schemes that allow exporters to claim a rebate regardless of whether they import inputs and pay duties. The rebate, however, is often conditional upon achieving positive export growth. Two main schemes are in place: the Duty Drawback of Taxes scheme (DDT), targeting the textile sector, and the Drawback of Local Taxes and Levies (DLTL), targeting non-textile products. Both schemes provide rebates to exporters of eligible products and are effectively used as export subsidies.3 In this paper, we focus on the DDT scheme. Table 1: Product coverage and rates of DDT scheme Products Category Rate Chapters 2015 2016 2017 2018 2019 2020 (HS8 digits) Garments High 260 61, 62 4% 4% 7% 7% 4% 4% Made-ups Medium 120 56, 57, 63 2% 2% 6% 6% 3% 3% Processed Low 161 52, 53, 54, 55, 1% 1% 5% 5% 2% 2% Fabrics 56, 58, 59, 60 Fabric-Yarn Very low 284 50, 51, 52, 53, - - 4% 4% - - 54, 55, 56, 58, 59, 60 Non-eligible None 106 50, 51, 52, 53, 54, 55, 56, 58, 59, 60, 63 Condition 10% growth None 10% growth Budget 55 53 275 410 360 124 Authors’ summary from various SROs. The budget is in ml USD. In 2018, 2019 and 2020, 50% of rebate was given unconditionally and the remaining 50% was subject to a 10% growth in previous year. The scheme has been characterized by administrative complexity and delays in the payment of refunds. While a version of the scheme was also in place before 2010, rebates were largely not paid. In 2019 the Government of Pakistan settled Rs17.6 billion worth of claims on account of both schemes that had been pending for over a decade.4 The government repurposed the scheme in 2010, but the allocated budget was very limited and again, there were major delays in the payments. It was only in 2015, that the scheme 2 In Lovo and Varela (2022) we investigated the role of duty exemptions and found that they are only imperfectly functional in helping exporters to be immune to trade frictions. 3 De jure the drawback rates should be calibrated based on actual duties incurred on imported inputs, averaged at the product level. The Input Output Coefficient Organization (IOCO) is mandated to provide information on input-output coefficients used across the industries, to determine eligibility and rates. De facto, inputs from IOCO are unrelated to rebate rates. The fact that only certain products are eligible for rebates and that rates vary among products, even within narrowly defined sectors, suggests that these schemes function as export promotion policies falling under the broad category of export subsidies, subject to specific conditions. Hence, the policy is likely to have an uneven impact across export items. 4 https://www.thenews.com.pk/print/586205-govt-clears-rs17-6bln-in-ddt-dltl-claims-of-exporters 5 gained importance, and was reviewed annually to adjust the product coverage, rebate rates and budget allocations. This was partly the consequence of an IMF intervention in late 2013, as the government committed to promote exports in order to reduce the current account deficit (IMF, 2017). We reviewed the relevant Statutory Regulatory Ordinances (SROs) that were used to introduce these changes, over the last decade, and characterized products based on their eligibility over time. Products were initially grouped under three main categories: Garments, Made-Ups (Home Textile) and Processed Fabrics. In 2017 and 2018 an addi- tional category was introduced: Fabric-Yarn (1). These categories have different rebate rates. While rates vary by periods, their ranking between categories remain the same, with Garments having the highest rate and Processed Fabrics (Fabric-Yarn in 2017-18) the lowest. This emphasis on Garments, the best established sector, is at odds with the role usually attributed to export promotion policies, associated with diversification or so- phistication of export bundles, or maximization of positive externalities associated with specific product exporting. For example, export promotion agencies are often expected to have larger impacts on the introduction of new products as well as new markets, than on selling more of an already exported product or to a country that is already a des- tination market for a firm (Martincus and Carballo, 2008). Rebates could be obtained only if exports for a specific product line had experienced a 10% growth in the previous year (as indicated at the bottom of Table 1). While, in principle, these conditions can promote export growth in general, the fact that they are focused on product-specific growth can also offer a strong incentive to focus on a particular product line, at the expense of other products. Table 1 also reports the 2-digit chapters included in each eligibility category, and shows substantial overlapping. This is because eligibility for Garments is determined at the chapter level (61-62), whereas for other categories, it is often determined at lower levels of aggregation. As an example, within chapter 56, all products within the 4-digit line 5608 are eligible for a medium rate (Made-ups), while all those within the line 5602 and 5603 are eligible for a low rebate rate (Processed Fabrics). At the 6-digit level for example, product line 6005.32 faces a low rebate rate (Processed Fabrics), line 6005.41 is eligible for the lowest rate and in 2017-18 only (Fabrics-Yarn), and line 6005.36 is not eligible. In some instances, eligibility also varies within a 6-digit classification. Depending on data availability and the scope of analysis (international versus within-Pakistan comparisons), we will adopt a more or less refined categorization of products into the different eligibility categories as further described below. 3 Data We use export data from several sources. For Pakistan, we have access to export data at 8-digit product level from the Ministry of Commerce of Pakistan, while for international comparisons we employ data at 6 digit product level (HS) from COMTRADE. 5 To 5 Because eligibility is sometimes determined at the 8-digit level, as indicated above, we assigned a 6-digit code to a given eligibility category if the majority of the sub-products are eligible for the given rate. This approximation is of limited concern, since it involves only 48 products of more than 5,000 product lines. 6 investigate differences among exporters, and entry and exit, we use exporter transactions’ data (custom-level data) for various periods. Data across periods are obtained from different sources and due to the absence of a common firm identifier, they cannot be linked. Data from 2005-2010 and 2015-2017 are from the Central Bank of Pakistan and data from 2018-2021 are from Ministry of Commerce. Figure 1: Exports by year and eligibility category Authors’ calculations based on data at 8 digit level from Ministry of Commerce of Pakistan. Categorisation into the difference eligibility groups is done at the 8-digit level. Exports in billion USD, constant. Exports are converted from Rupees to USD using an average exchange rate over the period. Vertical line indicates year in which DDTs were introduced. Figure 1 plots total exports by eligibility category and reveals that, pre-2015, gar- ments products, which enjoyed the highest rebate rate, had a total volume of exports similar to that of the Made-ups and Fabric-Yarn categories, hovering around a total of 1.5 billion USD for each category. It also shows a divergence in patterns between Gar- ments and the other two categories, emerging after 2015. This is especially pronounced with respect to the Fabrics-Yarn category, which was eligible for the lowest rate and only in 2017-18, when the DDT budget saw a substantial increase. This is a first indication that textile exports might have been influenced by the eligibility for DDTs. Yet, we cannot exclude the potential effects of varying global demands for the different product categories. In an attempt to provide some prima facie international comparisons, Figure 2 com- pares trends in Pakistani exports with global trends for the different eligibility categories. It shows that garments products, while aligned with the global average in the years pre- ceding the introduction of the DDT scheme, experienced an increase after 2015. The opposite is generally observed for the low, very low and non-eligible categories. This descriptive evidence reveals a striking diversion in export patterns across the different eligibility categories after the introduction of the DDT scheme. Yet, it also shows that simple global averages do not offer a robust counterfactual to assess the impact of the DDT scheme, since pre-trends are not always closely aligned for all categories. This motivates the use of the synthetic control method proposed below. 7 Figure 2: Comparison of Pakistani exports by eligibility category with global trends (a) Garments (high) (b) Made-ups (medium) (c) Processed (low) (d) Fabrics-Yarn (very low) (e) Non-eligible Authors’ calculations based on COMTRADE data at 6-digit level. The export index is obtained by dividing exports in any given period by exports in 2014, and then multiplying the result by 100. 4 Empirical Framework We start by testing whether exports across the different eligibility categories were on a similar trajectory before the roll-out of the DDT scheme. We provide support for this hy- pothesis by testing for pre-trends within an event-study approach. All our specifications are based on variations of the following equation: 2019 Eit = γt Dit + µi + υt + it (1) t=2008,t=2014 where E indicates exports of product i at time t, µ are product fixed effects, υ are time fixed effects, and is the idiosyncratic error term. Depending on the specification, D takes value one for a given eligibility category, and zero for the specified control group. Treatment variables are interacted with year dummies, γt , with the exclusion of 2014, the last pre-treatment period. For this analysis, i is defined at the 8-digit level. This specification could also serve to explore whether export patterns diverged after the introduction of the DDT scheme. Yet, it cannot be used to estimate the average treatment effect of the scheme, since spillover effects across treatment categories are expected to be large (a violation of SUTVA assumptions). This is the case, for example, if multi-product exporters intending to benefit from higher rates were to increase exports of garment products and reduce exports of other products due to constrained capacity. To estimate the magnitude of the impact of the DDT scheme on export performance, we adopt a comparative case study approach using a synthetic control estimator as in Abadie et al. (2015). This method produces an estimated counterfactual, the synthetic control, obtained by optimally weighting exports of other countries. The synthetic con- trol method relaxes the parallel trend assumption, and instead re-weights units to match 8 their pre-exposure trends. It also offers a dynamic estimate of the average treatment effects, providing insight into the dynamic effects of the scheme. As recommended in Abadie et al. (2015) we restrict the pool of donor countries, and consider only major textile exporters (top 40). This should reduce the risk of overfitting, which is further reduced by the fact that we have a long pre-treatment period at our disposal. We use pre-intervention exports (outcome variable) to optimally select weights. We also consid- ered using GDP and GDP per capita as additional pre-intervention variables, but their inclusion did not improve the fit over the pre-intervention period. Following Abadie et al. (2010) weights are chosen to minimize: k J +1 vm (X1m − wj Xjm )2 (2) m=1 j =2 where vm is a weight that reflects the relative importance assigned to the m-th variable to minimize the difference between the covariates for the treatment unit X0 and the units in the synthetic control Xj . For this analysis, we use 6-digit level data but determine eligibility and aggregate data at the 2-digit (chapter) level. The rationale for doing this is that the synthetic control method requires us to identify a suitable counterfactual for each category. For example, while the category “Processed Fabrics” considers a combination of product lines that are relevant in the Pakistani context, such combination of products might not be relevant to other countries, making direct international comparisons difficult.6 Thus, we proceed with the following steps. First, we determined whether a 2-digit chapter belongs to one of the five eligibility categories, based on the highest export share attributed to a specific rate within a chapter (as detailed in Table A.1 of the Appendix). For example, considering chapter 53, 82% of export value in this chapter is subject to a “Low” rate, 15% is subject to “Very Low” rates and 3% is not eligible. Hence, the entire chapter 53 is assigned to the category “Processed Fabrics (Low rate)”. Second, we employ the synthetic control approach separately for each 2-digit chapter (all results are shown in Figures A.2 - A.6 of the Appendix and are summarized in Table A.4 of the Appendix). This allows us to check consistency in behaviour across all chapters assigned to a given category. For example, considering chapters 61 and 62, which belong to the category “Garment (high rate)” in Figure A.2, we can observe that both chapters experienced a reduction with respect to the counterfactual in the post-DDT period, thus responding to the intervention is a similar way. Once consistency had been confirmed, we aggregated chapters into four eligibility categories and employ the synthetic method approach separately for each of the four categories.7 6 We initially employed a more detailed 6-digit category definition, but could not find a suitable synthetic control for some categories. Hence, we opted for the 2-digit level approach. 7 Products in the very-low rate and non-eligible are grouped together since all 2-digit chapters in these two categories respond in a similar way. 9 5 Results Before presenting estimates of the impact of the DDT scheme on exports, we investigate whether exports of products in different eligibility categories displayed similar trends before its implementation. This allows us to exclude the possibility that the scheme targeted products that were already on a different growth trajectory. Establishing this is crucial, as it demonstrates that, although eligibility across categories was not randomly assigned, exports in each category were on comparable paths prior to the reform. This mitigates concerns that eligibility might simply reflect pre-existing growth differences across product categories. 5.1 Targeting and pre-trends Figure 3: Event studies: testing for pre-trends (a) Garments vs non-eligible (b) Garments vs all others Note: In panel a), the control group includes only non-eligible products, which is a relatively small group of products. In panel b), the control group includes all non-garments products. Results are based using data at the 8-digit level from the Ministry of Commerce of Pakistan. Bars indicate 95% confidence intervals. Event studies for other categories also confirm that we cannot reject the hypothesis of parallel pre-trends and are shown in Figure A.1 of the Appendix. Our results on differences in pre-trends across eligibility categories are shown in Figure 3, which plots estimates of γt , obtained by estimating equation 1, for the cat- egory Garments (high-rate) compared to the non-eligible category (panel a), and all non-garment categories (panel b), respectively. Results confirm that we cannot reject the hypothesis of parallel trends between garment products and other product lines in the pre-2015 period. This is confirmed also for the other categories, as shown in Figure A.1. When considering post-2015 estimates, results show an increase in exports for gar- ment products with respect to other products (panel e). While this indicates that DDTs favored the expansion of the garment sector, with respect to other product lines, results cannot be used to determine the magnitude of average treatment effects for individual categories. As mentioned above, exporters facing capacity or financial constraints might be able to expand exports of a given product line, only at the expense of other product exports. This implies that all product lines are directly or indirectly affected by the rates applied to other categories, contaminating potential control groups and violating SUTVA assumptions. 10 5.2 Baseline results Results using the synthetic control method are summarized in Figure 4 for total textile exports and the four eligibility categories. Estimates show that textile exports expe- rienced an average annual increase of 1.9% over the period 2015-2019 (about US$0.38 billion). Considering that the budget allocated to the DDT scheme varies between 0.02% of GDP in 2015 to 0.14% of GDP in 2018 and 2019, back-of-the-envelope calculations indicate that for each US$1 spent on the DDT scheme, only US$1.1 was generated as additional exports. Figure 4: Synthetic control method results by eligibility category (a) Textile (overall) (b) Garments (high) (c) Made-ups (medium) (d) Processed (low) (e) Fabric-Yarn (very-low) and non-eligible Note: Authors’ calculations based COMTRADE data. Weights used for the synthetic control are reported in Table A.3. A sinthetic control group is fitted, separately, for each category. The overall effect is driven by a strong positive performance of the high-rate Garments category (panel b, +13%), and to a lesser extent of the medium-rate made-ups category (+ 5%, panel c). The effects are statistically significant at 10% (p-value 0.64) and 5% (p-value 0.25) using a one-sided test, as suggested by Abadie et al. (2015) and further described below. These effects are counterbalanced by the negative performance of the lower rate and non-eligible product lines (-16%, significant at 1%, panel e). This latter category is similar in size to the Garments category, accounting for an average of about US$4.3 billion in annual exports.8 Overall, results confirm that the positive impact on high and medium-rate products has been partially offset by the negative impact on lower rate products and non-eligible 8 It is important for the synthetic controls to reproduce closely the pre-intervention pattern of exports. This is achieved for most chapters, but with less precision in the case of chapters 53 and 60. Yet, these are relatively small chapters when compared to the others in the Processed Fabrics category. The discrepancies might not be easily visible in the graphs proposed in the Appendix, as we opt to focus on a common axis scale, within category, to reflect the relative importance of each chapter within its own category. For robustness, we also run the estimations without these two chapters and obtain very similar results. 11 products, leading to a small overall positive effect for the textile sector as a whole. Before exploring mechanisms and additional results, we provide some robustness checks in the next section. 5.3 Robustness We have conducted several robustness checks. First, we show placebo studies where we reassign our treatment in the data to each country in our donor pool. This implies obtaining synthetic control estimates for countries that were not affected by the treat- ment. Placebo results are shown in Figure A.7 of the Appendix, which confirm that the estimated effects for Pakistan (red line) are unusually large relative to the distribution of placebo effects. The only exception is the Processed Fabrics (low rate) category, where the placebo analysis is weaker. This is also the category for which our synthetic control provides a less precise match. Yet, given its relatively small size (2% of textile exports), this does not affect the overall interpretation of our results. These permutations can also be used to conduct inference and compute p-values based on one-sided. These are reported at the bottom of each graph and indicate that all effects, except the Processed Fabrics (low rate) category, are statistically significant using conventional confidence levels. We also report bias-corrected synthetic control gaps to adjust for discrepancies in predictor variable values between a treated unit and its donor pool, as proposed by Abadie and L’Hour (2021) and Ben-Michael et al. (2021). Results are shown in Figure A.8 of the Appendix, and are consistent with our previous estimates. Another set of robustness checks involves using a placebo intervention date that precedes the actual intervention. In Figure A.9 we impose a treatment date in 2012. It is reassuring to observe that the synthetic control groups remain aligned with actual Pakistan exports during the pre-treatment period. One last concern is that in December 2013, Pakistan obtained GSP plus status from the European Union, granting duty-free access to 96 percent of Pakistani exports to the EU. Evidence suggests that Pakistan only partially benefited from the GSP plus status, as growth in garment exports (10%) was slower than in neighboring countries such as Bangladesh (13%) and India (17%) (Hamid and Nabi, 2017). EU trade accounts for around 35% of textile exports. While duties were dropped to zero for all textile products, without differentiation, we can further exclude the possible influence of the acquired GSP plus status by excluding exports to the EU in our analysis. We produced synthetic control estimates considering non-EU trade and found similar results (Figure A.10). Garments exports to non-EU countries experienced an average annual increase of about 15% from 2015. All other products experienced a decline in exports, around -20%, leading to an overall null effect for total textile exports. 6 Mechanisms and Additional Results In this section, we provide additional suggestive results to shed light on the mechanisms that underlie the results presented earlier. To do so, we rely on exporter transactions 12 data for the period 2005-2010 and 2015-2021. Data from 2005-2010 and 2015-2017 are from Pakistan Customs, provided by the Central Bank of Pakistan, and data from 2018- 2021 are from the Ministry of Commerce. Unfortunately, while we are able to track exporters over time within the sub-periods, they cannot be linked across subperiods. Hence, our analysis below will be limited by data availability. 6.1 Entry and exit of exporters To shed some light on the contribution of entry and exit from the export market, we can define entrants and exiters within each sub-period, hence excluding the initial and final year of each sub-period to define entry and exit, respectively. Table 2: Analysis of entry and exit Entry Exit (1) (2) (3) (4) Higher rate Low-rate only High-rate only Low-rate only Entry 0.082*** 0.032*** Exit 0.085*** 0.043*** (0.009) (0.006) (0.008) (0.006) Post 0.179*** -0.038*** Post 0.252*** -0.033*** (0.008) (0.005) (0.008) (0.005) Entry x Post -0.002 -0.026*** Exit x post -0.055*** 0.029*** (0.012) (0.009) (0.013) (0.010) Year FE Yes Yes Yes Yes Exporters 66859 66859 66987 66987 Results obtained using exporter transaction data. The years 2005, 2015 and 2018, which are the first years of each sub-period, are excluded from the analysis of entry (columns 1 and 2). The years 2010, 2017, and 2021, which are the last year of each sub-period, are excluded from the analysis of exit (columns 3 and 4). Higher rates include garments (high) and made-ups (medium). Lower rates include Processed Fabric (low) and Fabrics-Yarn (very low). The dependent variable is binary and indicates whether an exporter specializes in the products indicated in the header. Results are obtained by regressing the binary output on a dummy taking value one if the exporter is an entrant (or exiter) and zero if incumbent, interacted with a Post variable taking value 1 for periods after the introduction of the DDT scheme. Robust standard errors in parentheses. *** p-value < 1%, ** p-value < 5%, * p-value < 10%. In Table 2, we present descriptive evidence concerning two binary outcome vari- ables. The first variable takes a value of one if an exporter exclusively exports high and medium-rate products, while the second variable takes a value of one if an exporter exclusively exports lower-rate products or non-eligible products. Columns 1 and 2 fo- cus on “entrants”, defined as exporters that were not present in previous years within a sub-period. They compare the propensity of entrants to be specialized in these two groups of products to that of incumbents, both before and after the introduction of the DDT scheme. The results indicate that in the post-DDT period, new exporters are less likely to specialize in lower-rate or non-eligible products. On the other hand, we find no statistically significant difference in terms of specialization in higher-rate products. For exiters (column 3 and 4), exporters that exit the custom dataset during the post-DDT period were less likely to be specialized in higher-rate products and more likely to be specialized in lower-rate or non-eligible products. Overall, these findings suggest that the aggregate changes in exports documented above have been partly driven by shifts in the composition of exporting firms, where new entrants are less inclined to focus on 13 lower-rate products, and exiting firms are more (less) likely to have specialized in (higher) lower-rate or non-eligible products. 6.2 Capacity constraints In this section we examine whether the limited overall effect and the reallocation induced by the DDT scheme, away from low rate products and into high rate products, can be explained by firms facing capacity constraints. While there can be synergies between product lines in terms of fixed costs, such as rental cost of premises, equipment or man- agerial supervision, firms can face diseconomies of scope due to constraints in expanding capacity. If firms were not constrained in their capacity, an export subsidy for a specific product could induce them to add that product to their export basket or expand their exports, without compromising the exports of other products (low or no-rate product). But capacity constraints may mean firms need to choose one or the other. Firms’ growth in Pakistan is notably constrained by a variety of factors, such as the shallow financial system. The latest World Bank Enterprise Survey (2022) reports that only 2.1% of firms in Pakistan had contracted loans or lines of credit. This compares to 23.7% of them across South Asia, or 25.7% across low-middle income countries. Large firms (those with more than 100 employees) fare substantially better than average, with 6.2% of them having contracted loans or lines of credit, but still lag comparators. Also, access to finance is reported as the most important constraint to doing business, after political instability. Hence, despite export incentives on specific products, firms might find it challenging to expand production without reducing output in other product lines. Figure 5: Exports change by type of exporter and eligibility criteria Results are based on exporter transactions data at the 6-digit level. The figure shows coefficients (βgt ) obtained by estimating the following specification: Eigt = 2020 t=2016 βgt Tt × Dig + θt + ηi + igt , where total exports (E ) at 6-digit level (i) are regressed on year dummies (Tt ) interacted with a binary indicator taking value one for medium and high-rate products (Di g ), including year (θ) and 2-digit (η ) fixed effects. Regressions are run separately for the 4 types of exporters indicated in the header. Effectively, results show changes in exports with respect to the baseline in 2015, for Medium-High rate and Lower rate products, respectively. Medium-high rate include garments and made-ups products. Lower rates include Processed, Fabric-Yarn and non-eligible products. Bars show 95% confidence interval based on standard errors clustered at the 2-digit level. . 14 Figure 5 provides some suggestive evidence of capacity constraints. The results focus on the post-DDT period and compare export growth for exporters of different sizes. The analysis focuses only on exporters who export both low or no-rate products, and medium or high rate products. Thus, differences in export growth between lower and higher rate products are more likely to reflect within-exporter reallocation, although we cannot test this directly as we do not have a complete panel of exporters for the entire post-reform period. We distinguish exporters into 4 groups. The top 5% exporters are the largest exporters in terms of total exports (including non-textile products) and account for about 40% of total textile exports. The second group includes exporters in the top quartile of the size distribution (excluding the top 5%), which account for another 40% of total textile exports. The other two groups include exporters in the second quartile of the size distribution and those in the bottom 50%. While we cannot measure capacity constraints directly, it is reasonable to expect large exporters to be less constrained by capacity limitations, as they are more likely to have access to credit and other financial resources that enable them to scale up production in response to increased demand or policy incentives. Results in Figure 5 displays growth patterns for both lower rates and medium-high rates products for the four categories of exporters who export both types of products. It shows that the Top 5% exporters increased their export of Medium-High rate products over time without reducing their exports of lower rate products. In contrast, smaller ex- porters exhibit different patterns. Those in the top quartile, for example, experienced an increase in exports of Medium-High rates and a corresponding decline in the exports of lower rate products. This indicates that medium-size exporters, potentially constrained by capacity or financial limitations, appear to reallocate resources from lower-rate prod- ucts to medium-high-rate products. Smaller exporters (bottom 50%), also show a decline in exports of lower-rate products but do not exhibit a corresponding increase in exports of higher-rate product. At this stage, we can only speculate that small exporters, after attempting to shift to high-rate products, face challenges due to increased competition from large exporters incentivized by the subsidy. Large exporters, taking advantage of economies of scale and established market presence, can capture a significant share of the high-rate market, leaving small exporters with limited room to expand. This intensified competition may result in pricing pressure, forcing small exporters to compete on lower margins just to maintain their market presence, which restricts their ability to grow their high-rate exports profitable. Overall, the reform seems to have primarily benefited larger exporters who can ex- pand without negatively impacting other product lines, possibly due to fewer capacity constraints. For smaller exporters, instead, the reform has led to a reduction in overall textile exports, driven by a decline in the export of lower-rate products, without a cor- responding increase in high-rate exports. This can potentially exacerbate disparities in export growth across firms. 15 6.3 Misreporting The above results show that the DDT scheme led to an increase in exports for high-rate products at the expense of low-rate and non-eligible ones. In this section, we investigate whether this effect could be the result of systematic misreporting rather than an actual change in the export composition. This would occur if firms misreport product codes at Customs in order to gain access to higher rebate rates. To investigate this matter, we compare data on exports reported by Pakistan with data on import from Pakistan reported by partner countries. The misreporting of trade flows has been investigated in the literature and linked to trade costs. Javorcik and Narciso (2017), for example, focus on imports and finds that higher tariffs lead to the under-reporting of imports. They argue that misreporting is more likely in the case of differentiated products, as it is more difficult to accurately assess the true price of differentiated products due to their intrinsic features and different qualities. On the other hand, Kee and Nicita (2022) investigate exports misreporting to evade non-tariff barriers (NTB). They argue that exporters have an incentive to under- report exports to avoid NTB-related costs, but at the same time, they face the risk of paying a penalty if caught. They find the effect to be more pronounced in the case of homogeneous products. Finally, Ferrantino et al. (2012) find evidence of under-reporting exports at the Chinese border to avoid paying value-added tax (VAT). Figure 6: Value discrepancies over time for non-eligible products (a) Top 10 destination countries (b) Top 10 non-EU destination countries Authors’ calculations based on matched exports and imports at 6 digit level from COMTRADE. Imports from Pakistan reported by destination country should equal what Pakistan reports as exports plus the insurance and freight costs (CIF) (as exports are typically reported on a ‘free-on-board’ (FOB) basis. Total exports and mirrored imports are normalized by taking ratios with respect to 2014 values. Top 10 destination countries are: USA, China, UK, Germany, Spain, Belgium, Italy, The Netherlands, United Arab Emirates, France. Top 10 non-EU destinations countries are: USA, China, Canada, Hong Kong, Egypt, Korea, Sri Lanka, Turkey, United Arab Emirates, South Africa. To investigate potential misreporting, we plot total exports and mirrored imports over time for the different treatment categories. Figure 6 focuses on non-eligible prod- ucts and shows an increase in discrepancies after the introduction of DDTs, especially for non-EU destinations. The post-2016 gaps are statistically significant, as shown in Figure A.11 of the Appendix. We do not find, however, significant differences for other product categories (Figure A.12). These results are suggestive of non-eligible products being misreported in order to benefit from the rebates offered by other categories. The finding is not unreasonable. Characteristics of eligible and non-eligible products can be confounded, which makes misreporting easier. For example, twine, cordage, ropes, 16 etc. of synthetic fiber (HS 5607) were not eligible while twine, cordage, ropes, etc. of textile materials (HS 5608) were eligible under the medium-rate (made-ups) category. Sometimes eligible and non-eligible products belong to the same 4 digit category, for example, unbleached or bleached warp knit fabrics of artificial fibers (HS 600541) were not eligible, while dyed warp knit fabrics of artificial fibers (HS 600542) were eligible under the low-rate category. Yet, while the results are suggestive of misreporting of non-eligible products, the magnitude of the misreporting is too low to have a visible impact on total exports in other categories. Hence, the fact that we do not observe an increase in discrepancies for higher-rate products indicates that while some misreporting took place, it is not driving the results presented above. 7 Conclusions This paper adds to the empirical literature on the impact that a specific form of in- dustrial policy - an export subsidy scheme - has on exports. It does so in the context of a low-income country like Pakistan. In this paper, we investigate the impact of the Duty Drawback scheme for the textile sector (DDT) adopted in Pakistan since 2015 on the performance of Pakistani exporters. Our study combines an event-study approach to exclude the presence of pre-trends with a synthetic control method to estimate the effects of the DDT scheme on Pakistani textile exports. Using product-level data, we show that products in different eligible categories were on similar trajectories before the introduction of the scheme. Synthetic control estimates show that while the DDT scheme had only a small positive overall impact on textile exports, it induced re-allocation across products within the textile sector. More specifically, the policies induced an increase in exports of products eligible for the highest rebate rates at the expense of non-eligible, or lower-rate products. The paper also sheds light on the mechanisms underlying these results. Using ex- porters’ transactions data, we find that the scheme contributes to shaping the composi- tion of exports by affecting entry and exit decisions. The subsidy scheme leads to firms being less likely to enter into exporting of low-rate products, while they are more likely to exit exporting of low-rate products. We also show suggestive evidence of capacity constraints partially accounting for the small overall effects found. Most exporters ex- pand exports of products subject to medium- and high- subsidy rates at the expense of decreasing exports of ineligible or low-rate products. Only the largest exporters, likely less affected by credit constraints, can expand exports of medium and high rate products without affecting their other export lines. We also present some evidence of increased under-reporting of non-eligible products; however, while statistically significant, this under-reporting is not economically meaningful enough to drive the results. The effects are driven by the product choices of both existing and new exporters, and do not seem to be driven by strategic misreporting. Altogether, these results have important policy implications. The fact that targeted export subsidies shape the export structure through entry and exit decisions points to the importance of product targeting in the schemes’ design. Targeting export subsidies 17 towards specific products might impact negatively other products, especially in a context where firms face capacity constraints. This could lead to an inefficient allocation of resources, as firms may divert attention and resources from non-subsidized to subsidized products, potentially harming overall export performance. These findings underscore the importance of monitoring the effects of such interventions. A careful assessment of the capacity constraints faced by firms, along with an assessment of the features of the products to target, is crucial in the design of export subsidy schemes to ensure balanced and sustainable growth in the export sector. References Abadie, A., A. Diamond, and J. Hainmueller (2010). Synthetic control methods for comparative case studies: Estimating the effect of california’s tobacco control program. Journal of the American statistical Association 105 (490), 493–505. Abadie, A., A. Diamond, and J. Hainmueller (2015). Comparative politics and the synthetic control method. American Journal of Political Science 59 (2), 495–510. Abadie, A. and J. L’Hour (2021). A penalized synthetic control estimator for disaggre- gated data. Journal of the American Statistical Association 116 (536), 1817–1834. Belloc, M. and M. Di Maio (2018). Survey of the literature on successful strategies and practices for export promotion by developing countries. International Growth Centre 14. Ben-Michael, E., A. Feller, and J. Rothstein (2021). The augmented synthetic control method. Journal of the American Statistical Association 116 (536), 1789–1803. Broocks, A. and J. Van Biesebroeck (2017). The impact of export promotion on export market entry. Journal of International Economics 107, 19–33. no, and G. Varela (2020a). All these worlds are yours, Defever, F., J.-D. Reyes, A. Ria˜ except india: The effectiveness of cash subsidies to export in nepal. European Economic Review 128, 103494. no, and G. Varela (2020b). All these worlds are yours, Defever, F., J.-D. Reyes, A. Ria˜ except india: The effectiveness of cash subsidies to export in nepal. European Economic Review 128, 103494. Ferrantino, M. J., X. Liu, and Z. Wang (2012). Evasion behaviors of exporters and im- porters: Evidence from the u.s.–china trade data discrepancy. Journal of International Economics 86 (1), 141–157. Hamid, N. and I. Nabi (2017). Implementing policies for competitive garments manu- facturing. International Growth Centre F-37211-PAK-1. 18 Harrison, A. and A. Rodr´ ıguez-Clare (2010). Trade, foreign investment, and industrial policy for developing countries. Handbook of development economics 5, 4039–4214. Hausmann, R., J. Hwang, and D. Rodrik (2007). What you export matters. Journal of Economic Growth 12 (1), 1–25. Helmers, C. and N. Trofimenko (2013). The use and abuse of export subsidies: evidence from colombia. The World Economy 36 (4), 465–486. IMF (2017). Article iv consultation: Press release. Technical report, IMF Country Report No. 17/212. Javorcik, B. S. and G. Narciso (2017). Wto accession and tariff evasion. Journal of Development Economics 125, 59–71. asz, R., N. J. Lane, and D. Rodrik (2023). The new economics of industrial policy. Juh´ Technical report, National Bureau of Economic Research. Kee, H. L. and A. Nicita (2022). Trade fraud and non-tariff measures. World Bank Policy Research Working Paper, 10112 . Lederman, D., M. Olarreaga, and L. Payton (2010). Export promotion agencies: Do they work? Journal of development economics 91 (2), 257–265. Lovo, S. and G. Varela (2022). Internationally linked firms and productivity in pakistan: a look at the top end of the distribution. The Journal of Development Studies 58 (10), 2110–2131. Martincus, C. V. and J. Carballo (2008). Is export promotion effective in developing countries? firm-level evidence on the intensive and the extensive margins of exports. Journal of International Economics 76 (1), 89–106. Martincus, C. V. and J. Carballo (2010). Beyond the average effects: The distributional impacts of export promotion programs in developing countries. Journal of Develop- ment Economics 92 (2), 201–214. Munch, J. and G. Schaur (2018). The effect of export promotion on firm-level perfor- mance. American Economic Journal: Economic Policy 10 (1), 357–87. Naqvi, W., D. Hayat, M. Javed, and V. Ahmed (2019). A review of export promotion and exemption schemes. Tokarick, S. and A. Subramanian (2003). Export financing and duty drawbacks: Note on issues raised by developing countries in the doha round–communication from the international monetary fund. Document WTTFCOH15, World Trade Organization, Geneva . Van Biesebroeck, J., J. Konings, and C. Volpe Martincus (2016). Did export promotion help firms weather the crisis? Economic Policy 31 (88), 653–702. 19 Table A.1: Percentage of total export value by chapter and category % of total export value (up to 2014 included) Processed HS2 Garments Made-ups Fabric-Yarn Non-eligible Assigned Category Fabrics Rate High Medium Low Very low None 50 0 0 0 86 14 Fabric-Yarn 51 0 0 0 3 97 Non-eligible 52 0 0 27 66 7 Fabric-Yarn 53 0 0 82 15 3 Processed Fabrics 54 0 0 55 44 1 Processed Fabrics 55 0 0 66 34 1 Processed Fabrics 56 0 29 2 44 24 Fabric-Yarn 57 0 100 0 0 0 Made-ups 58 0 0 30 29 41 Non-eligible 59 0 0 74 15 11 Processed Fabrics 60 0 0 73 27 0 Processed Fabrics 61 100 0 0 0 0 Garments 62 100 0 0 0 0 Garments 63 0 99 0 0 1 Made-ups Note: Percentages refer to exported value during the pre-reform period, from 2000 to 2014. Chapters 57, 61, 62, 63 are almost unequivocally assigned to their eligibility category. This is also true, to a lower extent, for chapters 50, 53, 59, and 60 where more than 70% of the exports in the chapter are subject to either a low (Processed Fabrics) or very low (Fabric-Yarn) rate. For the other chapters (52, 54, 55, 56, and 58) the assignment is less definitive, yet all these chapters seems to show similar results when analysed separately, hence their attribution to a given category does not pose concerns. Appendix: Tables 20 Table A.2: Weights used for synthetic controls by chapter country 50 51 52 53 54 55 56 57 58 59 60 61 62 63 ARG 0.428 AUS 0.217 0.193 BEL 0.044 BLR 0.625 0.016 BRA 0.218 CAN 0.026 0.059 0.196 CHN 0.185 0.008 0.007 0.058 CZE 0.045 DEU 0.044 0.165 0.304 EGY 0.088 0.992 FRA 0.059 GRC 0.273 0.276 GTM 0.086 IND 0.025 0.057 ISR 0.683 0.517 0.04 ITA 0.336 KHM 0.766 LTU 0.593 LUX 0.259 0.9 0.567 MEX 0.144 0.02 0.027 MYS 0.433 0.649 NLD 0.07 0.076 PER 0.219 0.054 0.037 THA 0.05 0.215 0.088 TUN 0.459 0.062 TUR 0.637 URY 0.1 USA 0.043 0.226 VNM 0.157 ZAF 0.008 0.159 0.103 0.605 The header indicates the 2-digit chapter to which the weights refer. These weights are used to produce the results in Figures A.2 -A.6. Table A.3: Weights used for synthetic controls by chapter country Garments Made-ups Processed fabrics Yarn - Fabrics and non-eligible ARG 0.611 AUS 0.174 BEL 0.646 BRA 0.095 CHN 0.003 0.053 0.064 DNK 0.683 IND 0.085 JPN 0.582 MYS 0.366 THA 0.029 0.023 VNM 0.038 0.302 ZAF 0.247 The header indicates eligibility category to which the weights refer. These weights are used to produce the results in Figure 4. 21 Table A.4: Impact and export value by chapter Chapter Annual exports in bl USD Effect (%) Average effect (1) (2) (3) Garments (high-rate) 61 2.043 7.5 18.2 62 1.619 31.7 Made-ups (medium-rate) 57 0.136 -23.8 4.7 63 3.399 5.8 Processed fabrics (low-rate) 53* 0.003 73.8 54 0.044 -8.7 55 0.427 -24.5 -34.3 59 0.010 -138.2 60* 0.053 -122.6 Fabrics-Yarn (very low) 50 0.001 -15.2 52 4.457 -16.2 -16.5 56 0.034 -51.1 Non-eligible 51 0.011 -51.6 -40.2 58 0.021 -34.4 Note: summary of the results presented in Figures A.2 - A.6. Exports in column 2 refer to the pre-2014 period. The average effect is the weighted average of the effect for individual chapters within a category, using annual exports as weight. *For these chapters the synthetic control method is unable to find a good fit for the pre-treatment period. 22 Appendix: Figures Figure A.1: Event study by eligibility category (a) Garments vs non-eligible (b) Made-ups vs non-eligible (c) Processed vs non-eligible (d) Fabric-Yarn vs non-eligible (e) Garments vs all others Note: In the first four graph, the control group includes only non-eligible products. In graph e) the control include all non-garments products. Categories are determined at the 8-digit level. Bars indicate 95% confidence interval. 23 Figure A.2: Synthetic control method by chapters: garments (high rate) Authors’ calculations based COMTRADE data at 6 digit level. Weights used for the are reported in Table A.2. The Chapter and corresponding average annual exports in the pre-2014 period are reported in the header. Dashed line represents the synthetic control. Figure A.3: Synthetic control method by chapters: made-ups (medium rate) Authors’ calculations based COMTRADE data at 6 digit level. Weights used for the are reported in Table A.2. The Chapter and corresponding average annual exports in the pre-2014 period are reported in header. Dashed line represents the synthetic control. 24 Figure A.4: Synthetic control method by chapters: processed fabrics (low rate) Authors’ calculations based COMTRADE data at 6 digit level. Weights used for the are reported in Table A.2. The Chapter and corresponding average annual exports in the pre-2014 period are reported in header. Dashed line represents the synthetic control. Figure A.5: Synthetic control method by chapters: fabrics - yarn (very low rate, only 2017-18) Authors’ calculations based COMTRADE data at 6 digit level. Weights used for the are reported in Table A.2. The Chapter and corresponding average annual exports in the pre-2014 period are reported in header. Dashed line represents the synthetic control. 25 Figure A.6: Synthetic control method by chapters: non-eligible Authors’ calculations based COMTRADE data at 6 digit level. Weights used for the are reported in Table A.2. The Chapter and corresponding average annual exports in the pre-2014 period are reported in header. Dashed line represents the synthetic control. Figure A.7: Placebo tests and inference (a) Garments (p-value = 0.64) (b) Made-ups (p-value = 0.25) (d) Fabrics-Yarn and non-eligible (p-value (c) Processed fabrics (p-value = 0.21) = 0) Red line show results when Pakistan is the treated country. P-values are computed by considering the proportion of times the effect in Pakistan is higher (or lower for negative effects) than other estimated effects. As recommended by Abadie et al. (2015) we exclude those countries whose pre-2014 synthetic control does not provide a good fit (i.e. countries with a pre-2014 average gap of more than 5 times the average gap of Pakistan). For example, for Garments, 2 out of 31 countries showing a good pre-2014 fit for they synthetic control, display a higher effect, which implies a p-value of 2/31. 26 Figure A.8: Bias corrected effects (a) Garments (b) Made-ups (c) Processed fabrics (d) Fabrics-Yarn and non-eligible Results are obtained by implementing the bias correction proposed by Abadie and L’Hour (2021) and Ben-Michael et al. (2021). Figure A.9: Placebo intervention date: 2012 (a) Garments (b) Made-ups (c) Processed fabrics (d) Fabrics-Yarn and non-eligible Results are obtained by using a placebo treatment in 2012. 27 Figure A.10: Comparison between Pakistani exports and synthetic control group: excluding exports to the EU (a) Garments (b) Made-ups (c) Processed fabrics (d) Fabrics-Yarn and non-eligible Authors’ calculations based COMTRADE data. Figure A.11: Differences in exports and mirrored imports over time: regression analysis (a) Top 10 destinations (b) Top 10 non-EU destinations Results are obtained by pooling exports and mirrored imports, where exports are considered as “treated”. Co- efficients are obtained by regressing the flow values on year dummies interacted with a treatment dummy, as in standard event studies, and controlling for product and year fixed effects. Vertical bars indicate 90% confidence interval. 28 Figure A.12: Value discrepancies over time by treatment categories and year (a) Garments (High-rate) (b) Madeups (Medium-rate) (c) Fabric-Yarn (Very low-rate) Authors’ calculations based on exports and mirrored imports at 6 digit level from COMTRADE. Total exports and mirrored imports are normalised by taking the ratio with respect to 2014 values. Top 10 destination countries are: USA, China, UK, Germany, Spain, Belgium, Italy, The Netherlands, United Arab Emirates, France. 29