Policy Research Working Paper 10601 Quality Regulation Creates and Reallocates Trade Lucas Zavala Ana Fernandes Ryan Haygood Tristan Reed Jose-Daniel Reyes Development Research Group Development Impact Group & Macroeconomics, Trade and Investment Global Practice November 2023 Policy Research Working Paper 10601 Abstract Quality regulation has become the dominant instrument Sanitary and phytosanitary and technical barriers to trade of trade policy. Panel evidence shows that regulations clas- measures increase the sales concentration of exporting firms sified as sanitary and phytosanitary measures and technical from lower-income countries, but do not affect the concen- barriers to trade both increase trade on average. Other tration of exporting firms from higher-income countries or non-tariff measures like quotas decrease trade. Sanitary and importing firms. The costs of quality regulation are primar- phytosanitary measures reallocate trade from lower-income ily borne by exporting firms, especially in lower-income exporting countries to higher-income exporting countries, countries. while technical barriers to trade measures do the opposite. This paper is a product of the Development Impact Group, the Development Impact Group, Development Economics and 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 lzavala@worldbank.org, afernandes@worldbank.org, treed@worldbank.org, reyes@worldbank.org, and ryan.haygood@yale.edu. 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 ∗ Quality Regulation Creates and Reallocates Trade Lucas Zavalaa , Ana Fernandesa , Ryan Haygoodb , Tristan Reeda, Jose-Daniel Reyesa JEL Codes: F13 F14, L11, O19. Keywords: Quality Regulation, Trade Policy, Reallocation, Market Concentration. ∗ We thank Penny Goldberg, Hiau-Looi Kee, and Ariel Weinberger for comments, as well as seminar participants at the World Bank, UNCTAD, Princeton, and Yale. This paper was supported by the Devel- opment Research Group Research Support Budget, the Knowledge for Change Program III for funding, the Umbrella Facility for Trade trust fund (financed by the governments of the Netherlands, Norway, Sweden, Switzerland and the United Kingdom), the Structural Transformation and Economic Growth Small Research Grant, and the Economic Growth Center SYLFF Research Grant. Jan Gabriel Oledan and Paula Suarez provided excellent research assistance. Computational reproducibility was verified by DIME Analytics. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank or its affiliated organizations, or those of the Executive Directors of the World Bank, their Managements, or the governments they represent. a: World Bank; b: Yale University 1 Introduction Over the past three decades, a decline in import tariffs has coincided with a rise in quality regulation (Figure 1). Quality regulations, which are codified in international trade agree- ments, specify minimum standards a product must conform with to be sold in the market, and apply to domestic as well as foreign firms. The most common are safety certification and labeling requirements. These regulations can affect firms’ capabilities, defined by Sut- ton (2012) as their cost of production, which shifts supply, and their product quality, which shifts demand. For example, conforming with food safety certification could increase firms’ fixed costs, given the investment required to bring the production process into conformity. Yet once consumers are assured by the certification that the food product is safe, demand could increase. The overall effect of quality regulation on average trade volumes and the allocation of trade across different countries and firms is ambiguous. The question of how quality regulation affects the allocation of trade has important im- plications for economic development and inequality between countries. As more importing countries adopt quality regulation, lower-income countries could benefit if regulation reas- sures customers that their products are of high quality. On the other hand, quality regulation could be a barrier to entry for exporting firms from lower-income countries, if their competi- tors in higher-income countries have a lower cost of conformity due to greater experience with quality regulation. This paper quantifies the heterogeneous effects of quality regulations and other non-tariff measures (NTMs) on trade, focusing on how they shape the allocation of trade between lower- and higher-income economies, and across firms within these economies. Using the UNCTAD (2019) nomenclature, we classify NTMs into five clusters according to the date of entry into force: (i) Sanitary and Phytosanitary measures (SPS), (ii) Technical Barriers to Trade (TBT), (iii) simultaneously SPS and TBT, (iv) pre-shipment inspections (PSI), and (v) non-technical measures. The first three clusters are quality regulations. Examples of SPS are a requirement limiting the use of hormones and antibiotics in the production of meat, or a sample test on imported oranges to check for the residue level of pesticides. Examples of TBT are restrictions on toxins in children’s toys, or a requirement that refrigerators need to carry a label indicating their size, weight and electricity consumption level. Trade agreements often oblige countries to enforce common standards and to recognize other countries’ conformity assessments to avoid duplicate inspections. Unlike the first three clusters, the fourth and fifth clusters of NTMs do not provide con- sumers with information on quality. An example of a PSI is a requirement that goods must be shipped directly from the country of origin, without stopping at a third country. Non- 2 technical measures include regulations that are explicitly protectionist like import quotas or minimum import prices. We employ a difference-in-differences research design that flexibly identifies heterogeneity in treatment across all five clusters of NTMs. The econometric approach uses measures imposed by other countries at the same time to control for potential endogeneity in the timing of measures with respect to trade flows into a given country as in Kee and Nicita (2022). The analysis yields results about average trade and the allocation of trade across countries and firms. On average, SPS, TBT, and simultaneous SPS-TBT measures (i.e., SBS and TBT mea- sures in the same law) increase imports, while PSI and non-technical measures reduce im- ports. These results are consistent with a model in which quality regulation provides con- sumers with useful information about quality and increases demand, and this effect outweighs the negative effect on trade of any increase in cost. As expected, non-technical measures and PSI reduce imports, because they make trade more costly but do not provide consumers with additional information. Although quality regulation increases trade, it reallocates trade between exporting coun- tries in different ways. SPS measures imposed by importing countries increase imports from higher-income origin countries relative to lower-income origin countries.1 This finding is consistent with the fact that higher-income countries have historically imposed stringent SPS measures on local producers (Ederington and Ruta, 2016), so their exporting firms’ products already conform when these standards are imposed by new export destinations. Lower-income countries, on the other hand, lose market share because they are unable to comply. SPS on average make lower-income countries less competitive. This is a troubling result given the comparative advantage of many lower-income countries in food production, which is generally subject to SPS measures. The patterns are reversed for TBT measures, which increase imports from lower-income countries relative to higher-income countries. TBTs are either relatively easier for exporting firms in lower-income countries to comply with, or they provide relatively more information about these exporting firms’ products. The reallocation effects wash out for laws that include both SPS and TBT measures imposed simultaneously on the same product.2 Analysis of firm-level data suggests that conformity with SPS and TBT standards can be a mechanism for creating “export superstars” in certain countries. In lower-income exporting countries, SPS and TBT measures increase the concentration of product markets measured 1 We employ the income classifications of the World Bank and group low- and lower-middle income coun- tries together as “lower-income” and upper-middle income and high-income countries as “higher-income” countries. 2 PSI and non-technical measures deter trade similarly across all exporting countries. 3 by the Herfindahl index of export sales (the sum of squared firm market shares). High concentration is consistent with the empirical regularity of “export superstars”, or a very small number of large firms that account for most exports of a single product from a country (Freund and Pierola, 2015). If a firm must specialize in conformity to quality regulation, this may become a barrier to entry in the market. Consistent with this idea, SPS and TBT measures do not change the concentration of firms in higher-income exporting countries. In these countries, where quality regulation has a longer legacy, all firms can conform to regulations and there is no specialization. Quality regulation has a negligible effect on the concentration of importing firms in the country that imposes the regulation. This result is consistent with the cost of conformity being faced primarily by exporting firms. If importing firms began specializing in quality regulation conformity, one would expect import markets to become more concentrated, sim- ilarly to export markets of lower-income countries. These results suggest that the costs of quality regulation are borne by exporting firms, especially from lower-income countries. Together, these findings on the heterogeneous impacts of quality regulation have impor- tant implications for the industrial organization, trade, and economic development litera- tures. Quality regulation increases trade on average, so it is not appropriate to call these regu- lations “non-tariff barriers” for all firms. This result provides support for the microeconomic model of Sutton (2012) in which quality, in addition to cost, determines market share and the export success of countries. Market shares behave differently in models with quality than in models of trade with Dixit-Stiglitz preferences: high-quality producers can retain large market share even as many firms enter. Macedoni and Weinberger (2022) provide a model of quality regulation and estimate the structural parameters of supply and demand of quality using domestic data from Chile. In their setting, quality regulation creates welfare by reallocating production to more efficient firms, particularly through the exit of inefficient firms. An avenue for future research is to study the supply and demand for quality in import markets, and whether regulation creates welfare in these markets. Our results provide lessons for the design of programs to help firms upgrade quality in response to regulation. To show this, it is helpful to compare our results to two recent experiments that provide quality certification to firms. Bai (2021) provides melon vendors in local markets in China with laser cut branding labels and shows they increase demand and profits. However, after the intervention is withdrawn, firms stop using the labels, as the incremental profits were not sufficient to justify the fixed cost of the technology in that market. Deutschmann, Bernard and Yameogo (2021) provide groundnut farmers in Senegal with a technology to screen for aflatoxin, which must be below a certain level for exports 4 to conform with an SPS measure in the European Union and other export markets. The experiment also provides a guaranteed price premium for conforming groundnuts. Similarly to the melons experiment, farmers adopt the quality certification technology initially, but stop using it when the price premium is withdrawn. Both studies demonstrate that, in equilibrium, producers may choose not to invest in a quality signal and lose access to some demand. Our finding that exporting firms from lower-income countries lose market share in import markets following SPS regulation is consistent with those firms choosing not to adopt a new signal given the costs and benefits. On average, this is not the case for TBTs, where the costs are lower relative to the benefits. Ensuring conformity with SPS measures and export success by developing country pro- ducers will require government support, including subsidies for adoption of quality screening and signaling technologies. Subsidies can be efficient, if consumers take time to adjust de- mand after receiving new quality signals, as demonstrated by Bai (2021), or if there remains a robust domestic market for non-conforming goods, as shown by Deutschmann et al. (2021). In this context, domestic quality regulation that prohibits low-quality goods from being sold in the domestic market (e.g., a ban in Myanmar on toxic pesticides or a food safety law in Ukraine conforming with the European Union) is another policy tool to move exporting firms to the high-quality equilibrium. Without such interventions, the increase in quality regulation could increase global inequality, as producers of low-quality goods in developing countries are excluded from larger markets in wealthier countries. The emergence of “ex- port superstars” in developing countries creates the potential for market power, additionally magnifying inequality within those countries (Zavala, 2023). Interventions to enhance quality should have very different effects depending on the nature of the quality regulation and the location of the firm. Our evidence suggests that developing country producers do not have trouble conforming with TBT measures, suggesting interventions are less needed in this area. This is because TBT conformity is less costly. SPS conformity typically requires important modifications to the production process (e.g., screening for aflatoxin) and potential destruction of non-conforming inventory (e.g., with excess aflatoxin). TBT, on the other hand, typically involves labelling of the product, with minor changes to the production process. 5 2 Background on Quality Regulation and Other NTMs 2.1 The Rise of Quality Regulation Import tariffs have been progressively lowered over the last 30 years as part of multilateral, bilateral, and unilateral trade liberalization episodes, leading to a decline in average tariffs from almost 20% in 1996 to less than 10% in 2020 for our sample (Figure 1). Simultaneously, the prevalence of quality regulation has increased. The average number of quality regulations per importing country-harmonized system (HS) 6-digit product increased from almost zero in 1996 to almost 4 in 2020. NTMs are separated broadly into technical and non-technical measures. Quality regu- lation – SPS and TBT – together with other technical measures such as PSI, regulate the appearance of imported products (such as a nutritional label), the production process (such as a pesticide restriction), or the logistics of importing (such as inspections at specific ports). NTMs comprise a tremendously heterogeneous set of policies. We classify each law in our sample based on the NTMs it imposes and the products that are affected. Figure 2 dis- plays the number of products affected by new regulations over time in the 5 largest NTM categories for two of the sample countries. An SPS could affect one product in one year and another product several years later. A given product may face an SPS in one year, followed by a TBT several years later. A single law might even impose an SPS and a TBT simultaneously. Finally, the timing and number of products regulated in a category varies widely across countries in our sample. 2.2 Literature on NTMs A large body of literature estimates the impacts of NTMs on trade, generally obtaining negative impacts, particularly for agricultural goods, often leading them to be called “non- tariff barriers”.3 This literature suffers from two caveats: (i) impacts are identified based on a cross-section of products subject to different NTMs at a fixed point in time and (ii) impacts are restricted to either NTMs as a whole (bundled) or to subsets of specific NTMs, either SPS or TBT (Disdier et al., 2008; Fontagn´ e, Orefice, Piermartini and Rocha, 2015) or border NTMs (Kee and Nicita, 2022). Our paper addresses these caveats and improves the 3 See Deardorff and Stern (1997) and Ederington and Ruta (2016) for surveys of the literature. A direct approach uses explicit measures of NTMs as regressors in a gravity model (Kee, Nicita and Olarreaga, 2009; Disdier, Fontagn´ e and Mimouni, 2008; Bao and Qiu, 2012) while an indirect approach estimates the trade- impeding effects of NTMs from smaller than predicted trade flows (Chen and Novy, 2011) or from larger than predicted price gaps cadot2016non. Chen, Hsieh and Song (2022) infer NTMs imposed on Chinese imports from the United States from the change in US imports relative to imports from other countries after controlling for the effect of tariffs. 6 understanding of the role of NTMs for trade along three dimensions. First, by considering specifications that allow for heterogeneity in the impacts of different types of NTMs—including the joint impact of clusters of NTMs imposed simultaneously— and across higher-income and lower-income exporting countries, we uncover trade-enhancing effects of certain types of NTMs and for certain types of countries, combined with the expected trade-deterring effects for other NTMs and countries. We rationalize the estimated impacts through cost versus information mechanisms discussed in Section 1. Specifically, we argue that SPS impose fixed costs that reduce supply and reallocate trade from lower- income to higher-income exporting countries, whereas TBT can signal quality that increases demand and reallocates trade from higher-income to lower-income exporting countries. This reallocation of trade across exporting countries complements Essaji (2008), Bao and Qiu (2012), and Bratt (2017), who find that developing countries’ exports are disproportionately harmed by NTMs.4 Second, our paper provides novel evidence on the differential impact of SPS versus TBT on the concentration of firm market shares across exporting countries, consistent with het- erogeneous costs of conformity with NTMs. Specifically, we show that SPS measures increase sales concentration for exporting firms from lower-income countries, implying that few firms there can comply with NTMs. SPS and TBT have no impact on sales concentration for ex- porting firms from higher-income countries, suggesting that most of their firms can comply with NTMs. This evidence adds to a growing literature on the impacts of NTMs on concen- tration, quality, and import substitution (Disdier, Gaign´ e and Herghelegiu, 2020; Macedoni and Weinberger, 2022; Ghodsi and Stehrer, 2022). We bring firm-level data across many lower-income and higher-income exporting countries to this literature.5 Third, we exploit time variation in NTMs for importing countries, which allows us to estimate specifications with a stringent set of fixed effects that control for unobserved het- erogeneity at the product-origin-destination level. Moreover, we explicitly tackle the endo- geneity of NTMs by predicting each country’s clusters of NTMs using the clusters of NTMs imposed by neighboring countries, following the approach proposed by Kee et al. (2009) and Kee and Nicita (2022). This methodology is appropriate for the Latin American importing countries in our sample, which share a common history and many common legal and cultural traditions and may thus introduce similar clusters of NTMs on similar products concurrently. 4 These three studies do suffer from important limitations: Essaji (2008) considers a cross-section of technical regulations in a single importing market (the USA in 1999); Bao and Qiu (2012) focuses on WTO notifications on TBTs (which are often incomplete); Bratt (2017) utilizes a cross-section of NTMs in 2003 without correcting for endogeneity. 5 Macedoni and Weinberger (2022) focus on domestic Chilean firms. Disdier et al. (2020) focus on French exporting firms. Ghodsi and Stehrer (2022) infer product varieties since they do not observe firms directly. 7 Furthermore, these countries are signatories to many regional trade agreements that either explicitly or implicitly foster regulatory convergence. Below we show strong correlations in the clusters of NTMs imposed on similar products across the Latin American countries. 3 Data This paper uses a panel of NTMs in 16 Latin America countries collected annually by the Latin American Integration Association (ALADI) from 2012-2017.6 For each country, the data set includes information on the inventory of active multilateral NTMs governing imports each year, classified according to the highest level of disaggregation of the 2019 international classification of NTMs (UNCTAD, 2019).7 Importantly, we observe the year in which the NTM was introduced and the products that it regulates. Products are classified using 6- digit HS codes. The initial 2012 data collection allows us to construct a panel of NTMs by product prior to 2012. Annual data collection then allows us to update this panel each year since 2012. This frequency is particularly important given the acceleration of NTMs after 2010 (Figure 1). Our novel data allow us to observe the implementation of NTMs over time between 1996 and 2017, for different types of NTMs and different countries in Latin America. To explore the relationship between NTMs and imports at the country level, we merge the NTM panel with import flows at the product level from 1996 to 2017. These were ob- tained from the World Integrated Trade Solution (WITS), the World Bank portal for trade data. This process involves the conversion of HS codes in the NTM database and in WITS to a consolidated version of the HS following the procedure described in Fernandes, Freund and Pierola (2016). For each importing country in Latin America, denoted d, the result- ing database includes information about the import value yodpt from each exporting/origin country o of each product p in each year t, along with the different N T Mdpt in force in the importing country. We abstract from input-output linkages and focus the analysis on final goods, as defined by the UN’s Broad Economic Categories (BEC). Finally, to understand the role of concentration in import and export markets, we use firm-level import and export customs transaction data sets obtained as part of the expansion of the Exporter Dynamics Database described in Fernandes et al. (2016). These data sets provide the universe of firm-level export and import transactions in each country. We use these data to construct a summary measure of export and import market structure. Since 6 The countries included are Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Ecuador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Peru, Paraguay, El Salvador, and Uruguay. 7 The information also includes NTMs applied to exports and NTMs applied in a bilateral fashion. These represent a negligible share of measures, so we do not consider them in our analysis. 8 competitive dynamics may differ across products and across countries, we treat each country- HS 6-digit product-partner country-year as a distinct market and measure the concentration of transactions across firms within that market. The concentration measure we use is the Herfindahl–Hirschman Index (HHI), equal to the sum of squared market shares for firms within a country-HS 6-digit product-partner country-year. Import market concentration is measured for the 10 Latin American importing countries covered by the customs transaction data (and within them, the universe of products and exporting partner countries) for the period 1996-2017 and for which we have NTM data.8 For example, in Chile, we observe the concentration across firms of imports of boneless beef (HS 020130) from Paraguay and, distinctly, the concentration across firms of imports of boneless beef from Brazil. Export market concentration is measured for all countries covered by the customs transaction data in the period 1996-2017 that export to the Latin American countries in our sample.9 Since Bangladesh is in the customs data, we observe the concentration across firms of exports of cotton t-shirts (HS 610910) to Chile and, distinctly, the concentration across firms of exports of cotton t-shirts to Colombia.10 4 Empirical Model and Results 4.1 Heterogeneous Impacts on Trade We begin by measuring the average impact of NTMs on bilateral trade. Our main specifi- cation combines a gravity equation with a modified Difference-in-Differences (DID) design: yodpt = exp(αodp + αot + αdt + βc N T Mcdpt + δτodpt + κc λcdpt ) + ϵodpt (1) c c where o indicates the exporting (origin) country, d indicates the importing (destination) country, p indicates the HS-6 digit product, and t indicates the year. The primary outcome variable is the value of imports, yodpt . The treatment indicator, N T Mcdpt , equals one if an NTM of type c is in force in destination d on product p during year t. We define five types 8 The countries included are Chile, Colombia, Costa Rica, Ecuador, Guatemala, Mexico, Peru, Paraguay, El Salvador and Uruguay. 9 The countries included are Albania, Burkina Faso, Bangladesh, Bulgaria, Bolivia, Botswana, Chile, China, Cˆ ote d’Ivoire, Colombia, Costa Rica, Dominican Republic, Ecuador,the Arab Republic of Egypt, Ethiopia, Gabon, Georgia, Guinea, Guatemala, Croatia, Indonesia, India, the Islamic Republic of Iran, Jordan, Kenya, Cambodia, the Lao People’s Democratic Republic, Lebanon, Sri Lanka, Morocco, Madagas- car, Mexico, North Macedonia, Myanmar, Mauritius, Malawi, Nicaragua, Nepal, Pakistan, Peru, Paraguay, Senegal, El Salvador, Thailand, Tanzania, Uganda, Uruguay, South Africa, and Zambia. 10 See Table A1 for summary statistics on our final data set. 9 of NTMs corresponding to the clusters in Figure 2:11 c = {SPS, TBT, SPS & TBT, Pre-Ship, Non-Tech} We include the average bilateral tariff rate, τodpt , to control for the correlation between NTMs and tariffs (Figure 1). Fixed effects at the exporting country-importing country- product level, αodp , control for market characteristics, so that the coefficients are estimated based on variation within markets over time. Fixed effects at the exporting country-year level, αot , and the importing country-year level, αdt , control for market trends and can be interpreted as the multilateral resistance terms in a structural gravity equation. The final control variables are the Inverse Mills Ratios of each NTM type, λcdpt , discussed in detail below. The coefficients of interest, βc , measure the percent change in import value following the implementation of a type-c measure (first difference), within a “treated” market (exporting country-importing country-product) relative to within an “untreated” market (second difference).12 Two key challenges arise when estimating equation (1). First, not all exporting country- importing country pairs trade a given product in every year of our sample. A logarithmic outcome variable excludes zero trade flows, creating a selection bias in favor of larger markets. This is particularly problematic in our setting, as NTMs cause trade in certain markets to disappear. This concern motivates the exponential specification in equation (1). We estimate this equation using a PPML estimator after adding zeros to the trade data along the time dimension (Santos Silva and Tenreyro, 2022).13 Second, NTMs may be implemented in a way that is correlated with the unobserved component of a country’s imports, ϵodpt , generating omitted variable bias. This is less likely for quality regulation, as such policies are often implemented in the context of multilateral trade agreements that are plausibly exogenous to local import trends (conditional on origin and destination trends, αot and αdt ). For example, concerns about pesticide safety may lead multiple countries in Latin America to adopt SPS measures simultaneously, independent of food import trends. Omitted variable bias is more of a concern for NTMs that are explicitly protectionist, like non-technical measures. For instance, a country may impose a quota in response to import growth, which creates competition for domestic producers. In this case, 11 If a destination-product market is treated by the same type of NTM multiple times through different laws, we consider it treated after the first treatment. 12 We exclude always-treated observations, i.e. exporting country-importing country-product cells subject to a type of NTM throughout the entire period, from the analysis. 13 For every origin-destination-product for which trade value is ever positive, we fill in zero for all years where trade is missing. OLS estimates of equation (1) using an indicator of positive trade value at the exporting country-importing country-product-year level as an outcome are shown in Table A2. These confirm that some markets disappear following new regulations. 10 the estimated effect of the non-technical measure on trade may be biased in the positive direction. To account for potential omitted variable bias, we adopt the approach of Kee and Nicita (2022) and estimate a control function for each type of NTM: c N T Mcdpt = Φ γc ′ N T M c′ dpt + ucdpt (2) c′ where N T M c′ dpt is the average treatment indicator for three neighboring countries of desti- nation d. The trade regulations of these neighbors are unlikely to be driven by the political economy concerns of country d, and therefore capture the exogenous component of its trade regulations. Since the outcome variable is an indicator variable, we use a Probit estimator: c Φ(·) indicates the Standard Normal distribution function. A positive estimate for γc ′ implies that a country is more likely to implement an NTM of type c if its neighbors implement an NTM of type c′ .14 c Given the estimates, γ ˆc′ , we compute the Inverse Mills Ratio for each type of NTM: c ϕ( c′ ˆc γ ′ N T M c′ dpt ) λcdpt = c Φ( ˆc c′ γ ′ N T M c′ dpt ) where ϕ(·) is the Standard Normal density function. The Inverse Mills Ratio (IMR) measures the hazard of non-selection: the higher the IMR for an NTM, the less likely an importing country is to have imposed the NTM, based on the NTMs imposed by its neighbors. Intu- itively, if the IMR is high but the importing country imposes an NTM anyway, endogeneity is more of a concern, and the regression should adjust for this correlation. Formally, we include the IMRs as controls in equation (1).15 The regression effectively compares treated and un- treated units with similar likelihoods of being treated, based on the behavior of neighboring countries. Column 1 of Table 1 presents our baseline estimates for equation (1). SPS and TBT measures enhance trade on average, whether implemented independently or simultaneously. This is consistent with the hypothesis that SPS and TBT increase product quality and stimulate demand. In contrast, pre-shipment inspections and non-technical measures deter trade, as expected given their often protectionist motives. Reassuringly, higher tariffs are also associated with lower imports, but we do not interpret this effect causally as we did not estimate a control function for tariffs. The estimated treatment effects are large, ranging 14 Estimates of the probit control functions are shown in Table A3. As expected, implementation of SPS and TBT is positively predicted by neighbors’ implementation of SPS and TBT. 15 The control function approach is preferred to an instrumental variables approach for two reasons: (1) the endogenous variables are binary and (2) we use a PPML estimator with high-dimensional fixed effects. 11 from a 21.1% increase in imports following the introduction of an SPS measure, to a 18.8% decrease in imports following the introduction of a non-technical measure. 4.1.1 Reallocation of Trade across Exporting Countries Quality regulation increases imports on average. Is this effect similar across exporting part- ners? To examine this question, we add interaction terms to our main specification: yodpt = exp(αodp +αdt +αot + βc N T Mcdpt + ζc N T Mcdpt ×higho +δτodpt + κc λcdpt )+ϵodpt c c c (3) where higho equals one if the exporting country is a higher-income country in 2010, according to the World Bank’s classification. Other variables are defined as before. The coefficients βc measure the impact of a type-c measure on imports from lower-income exporting countries, while the coefficient ζc measures the differential impact on imports from higher-income ex- porting countries, relative to lower-income exporting countries. The sum of βc and ζc gives the impact of a type-c measure on imports from higher-income exporting countries. Column 2 of Table 1 shows the results. SPS measures reallocate trade from lower-income exporting countries to higher-income exporting countries, while TBT measures reallocate trade from higher-income exporting countries to lower-income exporting countries. These reallocation effects wash out when SPS and TBT measures are implemented simultane- ously, indicating the importance of carefully classifying NTMs. In contrast, pre-shipment inspections and non-technical measures have a consistent negative effect across all exporting countries.16 SPS and TBT measures can stimulate demand by providing information about product quality, but they can also reduce supply by imposing costs of conformity. Heterogeneity in information benefits and conformity costs between higher- and lower-income exporting countries could explain the reallocation in Table 1. SPS measures will reallocate trade away from lower-income exporting countries if they are relatively harder to comply with, or if they provide relatively less information about the quality of products from these countries. This could be the case for environmental regulations. Conversely, TBT measures will reallocate trade toward lower-income exporting countries if they are relatively easier to comply with, or if they provide relatively more information about the quality of products. This could be 16 In Table A5, we report estimates of the regressions for differentiated and non-differentiated products separately, with differentiated products being those where the scope for quality information is higher (Rauch, 1999). The impacts of quality regulation disappear for non-differentiated goods, but non-technical measures still have negative effects. 12 the case for labelling requirements. In Table A4, we separately estimate the impacts of quality regulation on unit values and quantities.17 The effects of SPS measures are driven by unit values, while the effects of TBT measures are driven by quantities. This is consistent with SPS regulations being relatively more costly and TBT regulations providing relatively more information. In contrast, pre-shipment inspections and non-technical measures may be similarly dif- ficult for higher- and lower income exporting countries to comply with. For example, pre- shipment inspections may require passage through a particular port, which is costly for all exporting countries, and non-technical measures may impose similar quotas on all exporting countries. In these cases, there will be no reallocation across exporting countries. These measures can impose costs without improving quality information, leading to a net negative effect in Table 1. 4.2 Effects on Firms and Market Structure Quality regulation reallocates trade across countries. How does it affect the allocation of trade across firms within countries and, as a result, market structure? We look at this question from two angles: effects on exporter concentration and importer concentration. The effect on exporting and importing firms is theoretically ambiguous. On the one hand, measures that increase trade like SPS and TBT could enlarge the market, facilitating entry and reducing concentration. On the other hand, conformity with these measures requires technical skills and relationships with customs authorities, so they could serve as barriers to entry and increase concentration. 4.2.1 Concentration of Exporting Firms We measure the concentration of exporting firms’ sales at the origin-destination-product- year level using the HHI ∈ (0, 10, 000], where 10,000 indicates the product is exported by a monopolist, with market share equal to 100% of bilateral exports. These data are available for a subset of countries exporting to Latin America, so we cannot measure the concentration of exporting firms from all origin countries to a given destination. Column 1 of Table 2 replicates the impacts of NTMs on bilateral trade flows interacted with countries’ income classification for the sample with available exporting HHI data. The estimates are consistent with the reallocation shown in Table 1, although some precision is lost in this reduced sample. 17 Since unit values are undefined when trade flows equal zero, the sample is restricted to positive trade flows. To avoid comparing different units, we further restrict to quantities measured in kilograms, which cover over 70% of the sample. 13 Column 2 reports OLS estimates of equation (3) where the left-hand-side is the HHI of exporting firms from each origin-destination-product-year.18 SPS and joint SPS-TBT measures increase the concentration of firms exporting from lower-income countries. In contrast, there is no net change in the concentration of firms exporting from higher-income countries. We interpret these results as indicating that, relative to firms from higher-income countries, very few firms from lower-income countries can comply with NTMs. The average HHI is 7,349, indicating that products are effectively exported from a given origin to a given destination by only one or two firms. The additional increase in concentra- tion is meaningful in the context of the US Department of Justice’s guidelines for horizontal mergers, which state that HHI increases of more than 100 points in already concentrated markets are likely to enhance market power. 4.2.2 Concentration of Importing Firms Data on the concentration of importing firms are available at two levels of aggregation. First, we measure the HHI at the origin-destination-product-year level. Here, an HHI of 10,000 implies the product is imported by a monopsonist, with market share equal to 100% of bilateral imports. These data are available for a subset of importing countries but for all of their origin countries. Second, by summing imports across origins, we measure the HHI of importing firms at the destination-product-year level, which corresponds to a single market for each product in each importing country. Column 1 of Table 3 shows the impact of NTMs on bilateral trade flows for the available importing countries. The estimates are consistent with our baseline specification in Table 1. Column 2 reports OLS estimates of equation (1) where the left-hand-side is the HHI of importing firms in each origin-destination-product-year.19 Importantly, the measures that enhance trade in Column 1 also reduce the concentration of importing firms on average in Column 2. In contrast with exporting firms, importing firms enter the market and conform to NTMs on average. In Columns 3-4, we sum across origins and consider a single market for each product 18 Formally: HHIodpt = αodp + αdt + αot + βc N T Mcdpt + ζc N T Mcdpt × higho + δτodpt + κc λcdpt + ϵodpt c c c 19 Formally: HHIodpt = αodp + αdt + αot + βc N T Mcdpt + δτodpt + κc λcdpt + ϵodpt c c We do not estimate this specification for exporting HHI because not all origins are represented. 14 in each importing country. The average HHI is much lower at this level of aggregation: a value of 2,779 implies that products are effectively imported in a destination by four firms. Column 3 shows the impact of NTMs on multilateral trade flows.20 The net impact of NTMs on trade flows is negligible. Consequently, the overall concentration of importing firms does not increase, as shown in Column 4.21 Together, these results suggest that NTMs primarily impact trade flows through reallocation across exporting countries and across exporting firms within those countries. 5 Conclusions Quality regulation measures—either sanitary and phytosanitary standards (SPS), technical barriers to trade (TBT), or laws including both SPS and TBT measures—are sometimes classified with other NTMs as “non-tariff barriers” imposed with protectionist intent. Yet quality regulations are treated distinctly in international trade law, which recognizes such regulations can create trade by improving the quality of products. Our panel analysis con- firms that quality regulation is indeed trade-enhancing, unlike other NTMs like quotas or pre-shipment inspections. Sustained poverty reduction in lower-income countries is positively associated with access to larger markets through international trade agreements including the WTO (Goldberg and Reed, 2023). Standardizing quality regulation across countries has been a significant focus and achievement of the WTO. Under the WTO Agreement on the Application of Sanitary and Phytosanitary Measures and WTO Agreement on Technical Barriers to Trade, this process coincided with more countries adopting quality regulation similar to those of higher- income countries. TBT have increased exports from lower-income countries, creating the potential for poverty reduction. On the other hand, SPS have reduced exports from lower-income coun- tries, as their firms do not conform with production process standards from higher-income countries. SPS restrict entry of new varieties, while TBT, rather than being technical “bar- riers” to trade, have facilitated the entry of exporters from lower-income countries. Given the significant heterogeneity in treatment effects of quality regulation, further country- and regulation-specific research is required to design policy to upgrade quality. 20 Note that it is no longer possible to include origin-destination-product or origin-year fixed effects. In- stead, we include destination-product fixed effects. 21 TBT regulations are an exception, but the precise positive coefficient is economically small. 15 References Bai, Jie, “Melons as lemons: Asymmetric information, consumer learning and seller repu- tation,” CID Faculty Working Paper Series, 2021. Bao, Xiaohua and Larry D Qiu, “How do technical barriers to trade influence trade?,” Review of International Economics, 2012, 20 (4), 691–706. Bratt, Michael, “Estimating the bilateral impact of nontariff measures on trade,” Review of International Economics, 2017, 25 (5), 1105–1129. Chen, Natalie and Dennis Novy, “Gravity, trade integration, and heterogeneity across industries,” Journal of international Economics, 2011, 85 (2), 206–221. Chen, Tuo, Chang-Tai Hsieh, and Zheng Michael Song, “Non-tariff barriers in the US-China trade war,” Working Paper, National Bureau of Economic Research 2022. Deardorff, Alan V and Robert M Stern, “Measurement of nontariff barriers,” Working Paper, OECD 1997. Deutschmann, Joshua W, Tanguy Bernard, and Ouambi Yameogo, “Contracting and quality upgrading: Evidence from an experiment in Senegal,” 2021. Disdier, Anne-C´ elia, Carl Gaign´e, and Cristina Herghelegiu, “Do standards improve the quality of traded products?,” 2020. , Lionel Fontagn´ e, and Mondher Mimouni, “The impact of regulations on agricul- tural trade: evidence from the SPS and TBT agreements,” American Journal of Agricul- tural Economics, 2008, 90 (2), 336–350. Ederington, Josh and Michele Ruta, “Nontariff measures and the world trading sys- tem,” Handbook of Commercial Policy, 2016, 1, 211–277. Essaji, Azim, “Technical regulations and specialization in international trade,” Journal of International Economics, 2008, 76 (2), 166–176. Fernandes, Ana, Caroline Freund, and Martha Denisse Pierola, “Exporter behavior, country size and stage of development: Evidence from the exporter dynamics database,” Journal of Development Economics, 2016, 119 (C), 121–137. Fontagn´e, Lionel, Gianluca Orefice, Roberta Piermartini, and Nadia Rocha, “Product standards and margins of trade: Firm-level evidence,” Journal of international economics, 2015, 97 (1), 29–44. 16 Freund, Caroline and Martha Denisse Pierola, “Export superstars,” Review of Eco- nomics and Statistics, 2015, 97 (5), 1023–1032. Ghodsi, Mahdi and Robert Stehrer, “Non-Tariff Measures and the quality of imported products,” World Trade Review, 2022, 21 (1), 71–92. Goldberg, Pinelopi K. and Tristan Reed, “Demand Side Constraints in Development: The Role of Market Size, Trade, and (In)Equality,” Econometrica, 2023. Kee, Hiau Looi, Alessandro Nicita, and Marcelo Olarreaga, “Estimating trade re- strictiveness indices,” The Economic Journal, 2009, 119 (534), 172–199. and , “Trade fraud and non-tariff measures,” Journal of International Economics, 2022, 139, 103682. Macedoni, Luca and Ariel Weinberger, “Quality heterogeneity and misallocation: The welfare benefits of raising your standards,” Journal of International Economics, 2022, 134, 103544. Rauch, James E, “Networks versus markets in international trade,” Journal of interna- tional Economics, 1999, 48 (1), 7–35. Silva, JMC Santos and Silvana Tenreyro, “The log of gravity at 15,” Portuguese Economic Journal, 2022, 21 (3), 423–437. Sutton, John, Competing in capabilities: the globalization process, Oxford University Press, 2012. UNCTAD, “International Classification of Non-tariff Measures - 2019 version,” Technical Report 2019. Zavala, Lucas, “Unfair trade? Monopsony power in agricultural value chains,” 2023. 17 6 Tables and Figures Figure 1: Evolution of Tariffs and Quality Regulation over Time Note : Light blue line shows average applied tariff rate from the TRAINS/WITS database. Dark blue line shows average quality regulation at 1-digit level from ALADI database. Averages are computed across HS 6- digit products and the importing countries in our sample: Chile, Colombia, Costa Rica, Ecuador, Guatemala, Mexico, Peru, Paraguay, El Salvador and Uruguay. 18 Figure 2: Clustered Implementation of NTMs Note : Figure displays number of HS-6 digit products facing a new regulation over time in the top 5 categories of NTMs. Each dot represents a regulation, and the size of the dot corresponds to the number of products affected. Left panel shows regulations imposed by Mexico. Right panel shows regulations imposed by Brazil. Source: ALADI database. 19 Table 1: Effects of NTMs on Trade Flows and Reallocation Dependent variable: V V (1) (2) SPS 0.211∗∗∗ -0.164∗ (0.0525) (0.0959) SPS × Higher-Income 0.397∗∗∗ (0.107) TBT 0.0912∗∗ 0.244∗∗∗ (0.0375) (0.0835) TBT × Higher-Income -0.165∗ (0.0884) SPS & TBT 0.136∗∗ 0.150 (0.0535) (0.102) SPS & TBT × Higher-Income -0.0165 (0.102) Pre-Ship -0.0712 -0.163 (0.107) (0.220) Pre-Ship × Higher-Income 0.0981 (0.199) Non-Tech -0.188∗∗∗ -0.177 (0.0561) (0.111) Non-Tech × Higher-Income -0.00942 (0.104) Tariff -0.00116 0.000848 (0.000925) (0.00215) Tariff × Higher-Income -0.00211 (0.00237) Destination × origin × product FE Yes Yes Destination × year FE Yes Yes Origin × year FE Yes Yes Observations 7,127,263 7,127,263 ∗ Note: p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 Column 1 presents PPML estimates of Equation (1). Column 2 presents PPML estimates of Equation (3). FE stands for fixed effects. Specification controls for Inverse Mills Ratios for each NTM type following estimation of probit model of NTM selection based on NTM intensity in neighboring countries. Source: WITS and ALADI databases. 20 Table 2: Effects of NTMs on Concentration of Exporting Firms Dependent variable: V HHI (1) (2) SPS -0.219∗∗ 302.1∗∗∗ (0.0939) (111.5) SPS × Higher-Income 0.138 -270.5∗∗ (0.113) (122.4) TBT 0.110∗ 12.92 (0.0587) (61.73) TBT × Higher-Income 0.0226 122.0∗ (0.0680) (65.53) SPS & TBT -0.105 234.1∗∗ (0.0743) (105.2) SPS & TBT × Higher-Income 0.0692 -112.5 (0.0827) (106.3) Pre-Ship 0.215 -1,047∗∗∗ (0.160) (183.2) Pre-Ship × Higher-Income -0.391∗∗∗ 158.1 (0.111) (147.4) Non-Tech 0.00206 -339.4∗∗∗ (0.100) (97.97) Non-Tech × Higher-Income 0.0130 81.24 (0.0916) (85.93) Tariff 0.00279 0.524 (0.00212) (2.030) Tariff × Higher-Income -0.0114∗∗∗ 3.031 (0.00332) (2.168) Destination × origin × product FE Yes Yes Destination × year FE Yes Yes Origin × year FE Yes Yes Mean HHI 7,349 Observations 1,326,583 372,718 ∗ Note: p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 Column 1 presents PPML estimates of Equation (3). Column 2 presents OLS estimates of Equation (3) with HHI of exporting firms as the outcome. FE stands for fixed effects. Specification controls for Inverse Mills Ratios for each NTM type following estimation of probit model of NTM selection based on NTM intensity in neighboring countries. Sample includes exporting countries in EDD database (see text). Source: EDD, WITS, and ALADI databases. 21 Table 3: Effects of NTMs on Concentration of Importing Firms Dependent variable: V HHI V HHI (1) (2) (3) (4) SPS 0.313∗∗∗ -90.77∗∗ 0.0527 -168.9∗∗ (0.0608) (41.08) (0.0630) (82.77) TBT 0.0409 -11.56 -0.0793∗ 89.03∗∗ (0.0466) (19.37) (0.0421) (39.78) SPS & TBT 0.120 46.58 0.0263 134.1 (0.0980) (58.61) (0.0874) (115.4) Pre-Ship -0.103 61.12 0.0320 63.42 (0.198) (65.30) (0.171) (142.3) Non-Tech -0.182∗ -94.18∗∗∗ -0.128 -419.0∗∗∗ (0.107) (32.80) (0.0980) (76.12) Tariff -0.000890 1.281∗ -0.00229 -3.736∗∗ (0.00104) (0.715) (0.00142) (1.515) Summed across origins? No No Yes Yes Destination × origin × product FE Yes Yes No No Destination × product FE No No Yes Yes Destination × year FE Yes Yes Yes Yes Origin × year FE Yes Yes No No Mean HHI 6,746 2,779 Observations 3,750,449 1,081,675 155,787 96,349 ∗ Note: p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 Columns 1 and 3 present PPML estimates of Equation (1). Column 2 and 4 present OLS estimates of Equation (1) with HHI of importing firms as the outcome. Out- comes are defined at the origin-destination-product-year level in Columns 1-2, and at the destination-product year level in Columns 3-4 (summing across origins). FE stands for fixed effects. All specifications control for Inverse Mills Ratios for each NTM type following estimation of probit model of NTM selection based on NTM intensity in neighboring countries. Sample includes importing countries in EDD database (see text). Source: EDD, WITS, and ALADI databases. 22 Online Appendix ∗ Quality Regulation Creates and Reallocates Trade Lucas Zavalaa , Ana Fernandesa , Ryan Haygoodb , Tristan Reeda , Jose-Daniel Reyesa November 2023 ∗ a: World Bank; b: Yale University 1 A Appendix Figures and Tables Table A1: Summary Statistics Sample Table 1 Table 2 Table 3 Outcome variables: Import Value Exporter HHI Importer HHI Mean 752, 248 7, 369 6, 718 Std Dev 26, 618, 469 3, 024 3, 167 25th Pctile 641 4, 983 4, 004 Median 5, 532 9, 027 6, 916 75th Pctile 49, 120 10, 000 10, 000 Explanatory variables: % Avg % Avg % Avg SPS 6 9 6 TBT 19 22 22 SPS & TBT 8 11 9 Pre-Ship 3 4 3 Non-Tech 11 12 15 Tariff 13 10 11 Observations 9, 329, 157 1, 543, 813 4, 508, 116 % Positive Trade 35 23 27 Note: Panel A presents various statistics of outcome variables. Panel B presents percent averages of explanatory variables. Statistics and averages calculated conditional on positive trade flows. Each column corresponds to the estimating sample from a table in the main text. Observations do not match exactly because some categories are absorbed by fixed effects. Source: EDD, WITS, and ALADI databases. 2 Table A2: Reallocation: Extensive Margin Dependent variable: 1{V > 0} (1) SPS -0.122∗∗∗ (0.00567) SPS × Higher-Income 0.0324∗∗∗ (0.00592) TBT -0.0328∗∗∗ (0.00252) TBT × Higher-Income -0.0188∗∗∗ (0.00245) SPS & TBT -0.0921∗∗∗ (0.00428) SPS & TBT × Higher-Income 0.0531∗∗∗ (0.00374) Pre-Ship 0.0606∗∗∗ (0.00622) Pre-Ship × Higher-Income -0.0260∗∗∗ (0.00475) Non-Tech 0.000841 (0.00341) Non-Tech × Higher-Income -0.0256∗∗∗ (0.00291) Tariff 0.000974∗∗∗ (8.65e-05) Tariff × Higher-Income -0.000955∗∗∗ (9.26e-05) Destination × origin × product FE Yes Destination × year FE Yes Origin × year FE Yes Observations 7,677,096 ∗ Note: p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 Table presents linear probability model estimates of Equation 3 with an indicator for positive trade flows as the outcome. FE stands for fixed effects. Specification controls for Inverse Mills Ratios for each NTM type following estimation of probit model of NTM selection based on NTM intensity in neighboring countries. Source: WITS and ALADI databases. 3 Table A3: Control Function Estimates Dependent variable: SPS TBT SPS & TBT PSI Non-Tech (1) (2) (3) (4) (5) Neighbor SPS 0.858∗∗∗ 0.421∗∗∗ 0.624∗∗∗ −0.055∗∗∗ −0.155∗∗∗ (0.001) (0.001) (0.001) (0.001) (0.001) Neighbor TBT 0.165∗∗∗ 0.429∗∗∗ 0.096∗∗∗ 0.296∗∗∗ 0.303∗∗∗ (0.001) (0.001) (0.001) (0.001) (0.001) Neighbor SPS & TBT 0.396∗∗∗ −0.052∗∗∗ 0.061∗∗∗ 0.191∗∗∗ 0.056∗∗∗ (0.002) (0.001) (0.001) (0.001) (0.001) Neighbor PSI 0.123∗∗∗ 0.075∗∗∗ 0.038∗∗∗ −0.144∗∗∗ −0.278∗∗∗ (0.003) (0.001) (0.002) (0.002) (0.002) Neighbor Non-Tech −0.273∗∗∗ 0.315∗∗∗ 0.241∗∗∗ 0.319∗∗∗ 0.531∗∗∗ (0.002) (0.001) (0.001) (0.001) (0.001) Observations 10,592,129 10,592,129 10,592,129 10,592,129 10,592,129 ∗ Note: p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 Table presents Probit estimates of Equation 2. Each column represents a different type of NTM. Source: WITS and ALADI databases. 4 Table A4: Reallocation: Price vs. Quantity Dependent variable: log P log P log Q log Q (1) (2) (3) (4) SPS 0.0247∗∗∗ 0.0116 0.0101 0.0731 (0.00738) (0.0186) (0.0210) (0.0544) SPS × Higher-Income 0.0144 -0.0709 (0.0194) (0.0561) TBT 0.0396∗∗∗ 0.00144 0.216∗∗∗ 0.264∗∗∗ (0.00625) (0.0137) (0.0162) (0.0352) TBT × Higher-Income 0.0430∗∗∗ -0.0545 (0.0139) (0.0354) SPS & TBT 0.00558 0.0331∗ 0.165∗∗∗ 0.0312 (0.00910) (0.0194) (0.0270) (0.0543) SPS & TBT × Higher-Income -0.0309 0.153∗∗∗ (0.0196) (0.0530) Pre-Ship -0.124∗∗∗ -0.0854∗∗ -0.231∗∗∗ -0.0506 (0.0273) (0.0369) (0.0664) (0.0916) Pre-Ship × Higher-Income -0.0420 -0.204∗∗∗ (0.0280) (0.0711) Non-Tech 0.0335∗∗ 0.0397∗ 0.0319 0.131∗∗ (0.0154) (0.0232) (0.0354) (0.0551) Non-Tech × Higher-Income -0.00725 -0.112∗∗ (0.0189) (0.0470) Tariff -0.000321 0.000425 -0.00976∗∗∗ -0.00658∗∗∗ (0.000254) (0.000604) (0.000616) (0.00153) Tariff × Higher-Income -0.000839 -0.00363∗∗ (0.000645) (0.00163) Destination × origin × product FE Yes Yes Yes Yes Destination × year FE Yes Yes Yes Yes Origin × year FE Yes Yes Yes Yes Observations 1,380,125 1,380,125 1,380,125 1,380,125 ∗ Note: p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 Table presents OLS estimates of Equation 3 with log unit value and log quantity in kilograms as outcomes. FE stands for fixed effects. Specification controls for Inverse Mills Ratios for each NTM type following estimation of probit model of NTM selection based on NTM intensity in neighboring countries. Source: WITS and ALADI databases. 5 Table A5: Reallocation: Heterogeneity by Product Dependent variable: V Non-differentiated Non-differentiated Differentiated Differentiated (1) (2) (3) (4) SPS -0.0449 -0.0235 0.252∗∗∗ -0.271∗∗ (0.0866) (0.146) (0.0795) (0.117) SPS × Higher-Income -0.0236 0.548∗∗∗ (0.158) (0.136) TBT -0.0715 -0.0824 0.190∗∗∗ 0.362∗∗∗ (0.0717) (0.0926) (0.0410) (0.103) TBT × Higher-Income 0.0118 -0.186∗ (0.109) (0.105) SPS & TBT -0.0893 0.00754 0.224∗∗∗ 0.240∗ (0.0783) (0.103) (0.0784) (0.144) SPS & TBT × Higher-Income -0.106 -0.0215 (0.113) (0.136) Pre-Ship -0.0986 0.0255 -0.0189 -0.224 (0.237) (0.216) (0.120) (0.239) Pre-Ship × Higher-Income -0.137 0.222 (0.147) (0.214) Non-Tech -0.150 -0.398∗∗∗ -0.154∗∗ -0.0403 (0.139) (0.146) (0.0618) (0.129) Non-Tech × Higher-Income 0.286∗∗ -0.120 (0.128) (0.122) Tariff -0.00203 -0.00350 -0.000823 0.00346 (0.00126) (0.00298) (0.00142) (0.00289) Tariff × Higher-Income 0.00156 -0.00453 (0.00322) (0.00327) Destination × origin × product FE Yes Yes Yes Yes Destination × year FE Yes Yes Yes Yes Origin × year FE Yes Yes Yes Yes Observations 699,062 699,062 6,417,268 6,417,268 ∗ Note: p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01 Columns 1 and 3 present PPML estimates of Equation 1 separately for differentiated and non-differentiated products. Columns 2 and 4 present PPML estimates of Equation 3 separately for differentiated and non- differentiated. Products are classified following Rauch (1999). Observations do not match Table 1 because fixed effects absorb more categories when splitting the sample. FE stands for fixed effects. Specification controls for Inverse Mills Ratios for each NTM type following estimation of probit model of NTM selection based on NTM intensity in neighboring countries. Source: WITS and ALADI databases. 6