Information frictions and the two margins of trade: Evidence from Slovenian manufacturing

Zaurino, Elena; Polanec, Saéso Working Paper Information frictions and the two margins of trade: Evidence from Slovenian manufacturing JRC Working Papers in Economics and Finance, No. 2023/7 Provided in Cooperation with: Joint Research Centre (JRC), European Commission Suggested Citation: Zaurino, Elena; Polanec, Saéso (2023) : Information frictions and the two margins of trade: Evidence from Slovenian manufacturing, JRC Working Papers in Economics and Finance, No. 2023/7, European Commission, Ispra This Version is available at: https://hdl.handle.net/10419/283096 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Information frictions and the two margins of trade: Evidence from Slovenian manufacturing Zaurino, E. and Polanec, S. 20XX JRC Working Papers in Economics and Finance, 2023/7 This publication is a Working Paper to provide evidence-based scientific support to the European policymaking process. Working Papers are pre-publication versions of technical papers, academic articles, book chapters, or reviews. Authors may release working papers to share ideas or to receive feedback on their work. This is done before the author submits the final version of the paper to a peer reviewed journal or conference for publication. Working papers can be cited by other peer-reviewed work. The contents of this publication do not necessarily reflect the position or opinion of the European Commission. Neither the European Commission nor any person acting on behalf of the Commission is responsible for the use that might be made of this publication. For information on the methodology and quality underlying the data used in this publication for which the source is neither Eurostat nor other Commission services, users should contact the referenced source. The designations employed and the presentation of material on the maps do not imply the expression of any opinion whatsoever on the part of the European Union concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Contact information Name: Elena Zaurino Address: Via Enrico Fermi, 2749 21027 Ispra (VA), Italy Email: [email protected] Tel.: + 39 0332 78 6258 EU Science Hub https://joint-research-centre.ec.europa.eu JRC132921 Ispra: European Commission, 2023 © European Union, 2023 The reuse policy of the European Commission documents is implemented by the Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents (OJ L 330, 14.12.2011, p. 39). Unless otherwise noted, the reuse of this document is authorised under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence (https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed provided appropriate credit is given and any changes are indicated. For any use or reproduction of photos or other material that is not owned by the European Union permission must be sought directly from the copyright holders. How to cite this report: Zaurino, E. and Polanec, S. Information frictions and the two margins of trade: Evidence from Slovenian manufacturing - JRC Working Papers in Economics and Finance, 2023/7, European Commission, Ispra, Italy, 2023, JRC132921. Information frictions and the two margins of trade: Evidence from Slovenian manufacturing∗ Elena Zaurino KULeuven & European Commission JRC Sašo Polanec University of Ljubljana June 23, 2023 Abstract We empirically investigate whether firms lower information frictions in foreign sourcing through prior exporting. Using a panel of Slovenian manufacturing firms in the period 1996-2011, we estimate the probability of import entry in a new market when the firm is already exporting to the same country and we find a positive and significant relation. To control for the endogeneity of the export decision, we implement an instrumental variable approach exploiting the notion of sequential exporting. Moreover, we rule out productivity growth as being the only predictor of entry in a foreign market through several falsification tests. These findings suggest information frictions play an important role for firms trading in international markets. JEL Classification: F14, L20, D22, D83. Keywords: Sourcing; Import entry; Information Frictions; Sunk costs. ∗Corresponding author: Elena Zaurino (email: [email protected]). Elena Zaurino is very grateful to her PhD supervisor Jo Van Biesebroeck for his invaluable support and guidance. We thank Paola Conconi, Ron Davies, Jan De Loecker, Keith Head, Sebastian Fleitas, Thierry Mayer, Mattia Nardotto, Marjana Subotic, Hylke Vandenbussche, Stijn Vanormelingen, Gonzague Vannoorenberghe, Patrick Van Cayseele, Frank Verboven, Daniele Verdini, and all participants to VIVES seminar, UCL PhD Workshop, KUL Summer Graduate Workshop, 25th Spring Meeting Young Economist, Conference on Globalization and Development, ETSG, and KULeuven IO internal seminars for valuable comments and discussion. We also thank the Slovenian Statistical Office (SURS) for providing the data. The views expressed are purely those of the authors and may not in any circumstances be regarded as stating an official position of the European Commission. Any errors or omissions are our own. 1 Executive Summary Nowadays, exchange in intermediate goods accounts for a significant portion of international trade flows, giving rise to the phenomenon of global value chains. As such, a firm’s production often spans across multiple countries, making it imperative to understand the sourcing decisions that firms make. When firms decide to import goods from foreign countries, they have to undergo a search process to find the right suppliers, which can be quite costly, especially when the buyers and sellers are geographically distant. This paper aims to provide evidence for a new channel through which firms can reduce the costs of collecting information when searching for foreign suppliers. We study the idea that a firm that is already exporting to a foreign market can indirectly learn about new potential suppliers (importing) and reduce search frictions as a result. We posit that the probability of a firm starting to import from a foreign country increases if it was already exporting to that country. The evidence provided suggests that export entry increases the probability of starting to import from the same country by about 0.05 percentage points, with the strongest effect being observed three years after the entry. We use two longitudinal datasets covering the activity of Slovenian manufacturing firms in the period 1996-2011. We include full company accounts, as well as records of Slovenian firms’ export and import flows reported at the firm-country level. Standard gravity variables to account for geographical factors are inlcuded to complete the dataset. We then implement a linear probability model with a large number of fixed effects to control for several unobservable factors that could bias the results. We exploit the high-dimensionality of the data to add firm-country-specific fixed effects to account for time-invariant characteristics specific to a given firm’s relationship with a specific country comparing the change in import probability between firms that export and those that do not, after controlling for firm-year and country-year fixed effects. Despite these controls, we acknowledge that there could still be unobserved firmcountry-specific shocks that could increase both the probability of import and export entry, making it difficult to determine the causal relationship between the two. To overcome this challenge, we introduce a novel instrumental variable based on the idea of sequential exporting among neighboring countries, i.e. the (lagged) export entry of the same firm in a country d6=cwhich is a neighbor of c. Our study supports the idea that firms with prior experience in exporting to a foreign market are more likely to also start importing from that same market, building 1 on the idea that exporting can help firms reduce search frictions when looking for new suppliers. We find a positive and significant relation that is consistent across a set of different specifications, both for the linear probability model using OLS and 2SLS. The results are robust to several robustness checks that enable us to rule out that firm’s productivity is the unobserved factor driving the relationship between import and export as it would be suggested by theories of the heterogeneous firms in internaitonal trade (Melitz (2003) on the export side and Antras et al. (2017) on the import side). Furthermore, we find that only larger firms are able to leverage their prior exporting experience to start importing, while smaller firms tend to access foreign markets indirectly through intermediaries and are unable to directly trade with foreign partners. 2 2 Introduction Nowadays exchange in intermediate goods accounts approximately for two-thirds of international trade flows giving origin to the phenomenon of global value chains (Johnson and Noguera, 2017). As global value chains become more and more important, a firm’s production will often span multiple countries and investigating how firms’ sourcing decisions take place is of first-order importance. When a firm decides whether and from which countries to import from, it will inevitably start with searching and selecting ex-ante unknown suppliers. The search process to acquire information about foreign partners is costly especially when buyers and sellers are geographically distant. Therefore, finding ways of reducing these information frictions is key for profitmaximizing firms. In this paper, we provide evidence for a new channel through which firms can reduce costs of information collection when searching for foreign suppliers: we study whether firm exporting to a foreign market can indirectly recover information about new potential suppliers (importing).1According to the mechanism we investigate, search frictions may be reduced by interactions between firms associated with trade flows already taking place in the opposite direction, i.e. firms that are already exporting to a country will recover more easily information about potential suppliers than non-exporting firms to the same market. The empirical question we want to address in this paper is: does the probability of importing increase when the firm was already exporting to the same country? Motivating evidence displayed in Figure 1 seems to suggest so. In particular, export entry increases by around 0.03 percentage points the probability of starting to import from the same country, with the effect being the strongest the first year after the entry. Unlike previous studies that have analyzed export and import jointly2, this is the first one to link these two activities in a dynamic framework3to explain the process of internationalization of firms. The timing is as follows: at t0the firm is endowed with an information set about foreign markets; at t1it starts to export and indirectly learns about possible suppliers for intermediate inputs in a destination market; eventually, at t2the firm adds new suppliers to its portfolio and starts to import back from the same country. Why is the exporting margin driving the relation and not the other way 1By margins of trade we mean the ways in which a firm can be active in international markets, either by exporting or importing, following Bernard et al. (2018a) nomenclature. 2See among others Bernard et al. (2018a); Damijan et al. (2014); Bas and Strauss-Kahn (2014); Kasahara and Lapham (2013). 3The only other study addressing a similar question is Albornoz and García Lembergman (2015) who use data on Argentinean firms. The main difference with their paper is that we use an IV approach to control for endogeneity issues. 3 Figure 1: Event-study on the effect of exporting on import decision from same country Note(s): Figure 1 plots the estimated event study coefficients from a regression where the dependent variable is an indicator variable equal to one if firms fstarts to import from country cat calendar year t, and the regressors are event time dummies equal to one in the x-th year after the first year in which firm fstarted to export to country cfor the first time. Firm ×country, firm ×year and country ×year fixed effects are included in the specification. We also normalize the coefficient on the year prior to the event equal to zero. We cluster standard errors at firm-country level. Source: Own calculations based on AJPES, SURS and FURS data. round? Because firms want to reach as many profitable export markets as they can in order to benefit from economies of scale (Chaney, 2008; Arkolakis, 2010). This hints at the fact that firms seek to increase the number of destinations as much as possible since an additional country leads to an increase in profit (assuming the size of additional market-specific fixed costs is lower than the marginal increase in revenues). On the other hand, when a firm decides whether to import a product, it is not interested in sourcing it from as many countries, but only from one, which is the cheapest possible source or the best in terms of quality. Figures 2a and 2b provide motivating evidence behind this intuition: the number of export destinations is higher than the number of import origins, either if we consider one product exported (imported) across multiple destinations (origins), or a firm. Moreover, if we look at number of origins and destinations by firm-product as depicted in Figure 8 in the Appendix, a similar pattern emerges, and we can also notice that numbers of sourcing origins is in line with Antras et al. (2017) where they find that the typical firm purchases the same input from one country only. Since differences in the ability to start sourcing from foreign markets have been ex- 4 Figure 2: Average number of destination and origin countries (1996-2011) (a) By CN8 product (b) By firm Source: Own calculations based on AJPES, SURS and FURS data. plained as mainly driven by heterogeneity in productivity (Antras et al., 2017)4we try to rule out productivity being the main (unobserved) determinant of import and export choices. In this sense, our study follows in spirit Armenter and Koren (2015) who point out that productivity differences can only account for a fraction of the exposure to international markets. We employ two longitudinal datasets covering the activity of Slovenian manufacturing firms during the period 1996-2011. The first dataset contains the full company accounts, including nominal measures of output and different inputs at the firm level. The second dataset includes records of export and import flows of Slovenian firms reported at firm-country level. Due to the focus of the exercise, we have selected all manufacturing firms that are engaged in import or export activities at least once throughout the observed period. A third dataset with standard gravity variables such as distance and contiguity between countries is added to account for geographical dimension in the construction of the instrumental variable. Our empirical strategy produces estimates of the effect of export entry on the probability of starting to import in the following year. We take advantage of the highdimensionality of the data to control for several unobservable factors that might bias the results, implementing a linear probability model with a plethora of fixed effects.The main level of variation we exploit is within firm-country over time so that we absorb time-invariant characteristics specific to the relation of a given firm with a specific country. For instance, the fact that a firm has stronger and more constant ties with Japan because it has a Japanese CEO. We also add firm-year fixed effects which ac- 4Similarly on the export side, Bernard et al. (2003) or Melitz (2003) assume that differences in the ability of firms to enter foreign markets are entirely driven by heterogeneous productivities. 5 Table 1: Numbers of registered and active firms, by year Year No. Firms No. Exporters No. Importers 1996 5,410 2,031 3,566 1997 5,526 2,006 3,486 1998 5,568 2,012 3,485 1999 5,564 2,057 3,469 2000 5,562 2,064 3,352 2001 5,504 2,102 3,332 2002 5,470 2,080 3,218 2003 5,455 2,141 3,370 2004 5,434 1,911 2,951 2005 5,406 1,481 2,027 2006 5,332 1,592 2,156 2007 5,356 1,553 2,402 2008 5,299 1,577 2,453 2009 5,272 1,556 2,285 2010 5,201 1,553 2,282 2011 5,055 1,565 2,293 Notes: The sample on which the statistics are computed includes only manufacturing sectors, i.e. from 10 to 33 of the NACE Revision 2 industry classes. Source: Own calculations based on AJPES, SURS and FURS data. and plastics products (22), and machinery (28). In Table 3, we look at the number of entries both in the export and import market divided by geographical area. 4 Framework, Specification and Instruments Export entry and foreign sourcing decisions are related through various channels. In this paper, we focus on the role information frictions play. For this purpose, we adapt a standard framework used to analyze entry decisions to foreign markets as a special case. Then, we depict the resulting specification and the instruments we propose to control for endogeneity in the estimation. 12 Table 2: Number of firms in manufacturing sector by industry (1996-2011) Sector Code No. Firms No. Exporters No. Importers Fabricated metals 25 12,233 4,629 6,159 Rubber & plastic 22 8,394 2,997 5,092 Machinery 28 5,591 3,013 3,664 Printing 18 5,090 820 1,662 Wood 16 5,047 2,151 1,723 Food 10 3,484 878 1,850 Furniture 31 3,204 1,461 1,613 Electrical equipment 27 3,024 1,379 2,070 Wearing apparel 14 2,987 715 1,578 Computers & electronics 26 2,807 1,282 2,013 Non-metallic minerals 23 2,635 940 1,678 Other (n.e.s.) 32 2,011 670 1,116 Textiles 13 1,917 746 1,255 Chemicals 20 1,615 850 1,182 Automotive 29 1,480 870 1,082 Paper 17 1,283 520 773 Basic metals 24 997 684 687 Leather 15 716 347 423 Beverages 11 595 183 267 Other transports 30 532 238 322 Notes: The sample on which the statistics are computed includes only manufacturing sectors, i.e. from 10 to 33 of the NACE Revision 2 industry classes. For confidentiality issues, we do not report information for sectors that are scarcely populated. Source: Own calculations based on AJPES, SURS and FURS data. 4.1 Theoretical Framework To motivate our empirical work, we start by adapting the multi-period model of entry decisions presented in Bernard and Jensen (2004)12 to the context of sourcing decisions. A firm fstarts to import from a foreign country cat time t( 1 imp fct ) if the increment in expected (gross) profits associated with importing, Πft, exceeds the sunk cost to start importing, Fimp ic . Since any import decision involves the trade-off between saving variable cost from the usage of foreign inputs and paying a sunk cost of importing13, we can simplify the change in expected profits to a change in expected costs, Cft, under the assumption that revenues remain constant.14 It follows that the participation condition 12The theoretical foundation for their empirical model comes from Roberts and Tybout (1997) where they develop and estimate a dynamic discrete choice model that enables them to separate the role played by firm heterogeneity and sunk entry costs in explaining export decision. 13We abstract from potential improvement in quality of inputs due to foreign sourcing. 14Given the revenues Rft =pft ×Qft where the price pft =f(mcft, µft)is a function of marginal cost and markup of firm fat time t, a change in marginal costs (mcft) due to foreign sourcing will lead to a reduction in price pft (more or less proportionally according to pass-through rate), which will in turn 13 Table 3: Number of entries by geographical area (1996-2011) Export Import Area No. Entries Share No. Entries Share Central America 153 1% 160 1% Central Asia 163 1% 24 0% Eastern Asia 633 3% 1,889 8% Eastern Europe 6,372 31% 5,372 23% Northern Africa 510 3% 132 1% Other more developed regions 1,198 6% 1,952 8% South America 334 2% 210 1% South-Eastern Asia 500 2% 726 3% Southern Asia 485 2% 441 2% Sub-Saharan Africa 169 1% 90 0% Western Asia 1,545 8% 960 4% Western Europe 8,235 41% 11,772 50% Total 20,297 100% 23,728 100% Note: The sample is made of manufacturing firms engaged in international trade at least once. An entry is defined as the fact that a firm starts to trade with a foreign country for the first time, i.e. the firm has not done so in the three previous years. This is why the baseline year now is 1999, since in the first three years of the sample (1996-1998) the number of entries are mechanically equal to zero. Source: Own calculations based on AJPES, SURS and FURS data. for importing is: 1 imp f,c,t =   1if Πf,t −Fimp f,c (1 − 1 imp f,c,t−1)>0 0otherwise (1) Similar to Das et al. (2007), the return to becoming an importer today includes the option value of being able to continue importing next period without incurring again the sunk costs. If sunk costs do matter, they appear directly in the firm’s participation condition (1) as the coefficients on binary variables that describe the past importing status of the firm, here simplified to the previous year only, Mf,c,t−115. The novelty of this theoretical framework is that the sunk cost of imports Fimp f,c can be interpreted as the sum of two components: (i) search cost, Simp c(ii) cost of access to foreign markets, Fimp c. Concerning the first one, each firm needs to incur it in order to gather information about new suppliers in the foreign market, e.g. price and quality of their goods, but also other characteristics related to delivery conditions like temporal availability of goods and logistics. One way of gathering this information could be to physically go in the foreign market to meet up with business partners (Eaton et al., impact output and revenues, and eventually profits given that Πft =Rft −Cft. Here, we abstract from the revenue channel and we focus on the cost channel only. Another simplification is that the setup of the model contains no real option value related to reduction in expected costs of entry to other markets. 15In our estimation, we implicitly allow the effect of exporting activity on subsequent entry costs to last only one period as in Roberts and Tybout (1997) and Morales et al. (2019) 14 2014). For instance, existing business partners (buyers of its export goods) can provide references about potential new suppliers regarding the prices and key characteristics of goods (and related services). If the firm is already exporting to that market, this search process might be facilitated such that Simp cmight be reduced by an αfraction. The second component of the sunk cost, Fimp c, represents all (non-search) costs related to the start of a trading activity with a foreign partner, such as understanding the regulation, and it is modeled as a function of standard gravity variables. In this respect, Antras et al. (2017) found that fixed costs of importing are 13 percent lower for countries with a common language, and increase in distance with an elasticity of 0.19. Hence, the sunk cost of imports can be re-written as: Fimp f,c =Simp c(α 1 exp f,c,t−1) + Fimp c(2) where the search cost Simp c16 is reduced by a share α∈[0,1] if the firm fwas already exporting to the same market cat time t−1( 1 exp fct−1= 1). Thus, equation (1) becomes: 1 imp f,c,t =   1if Πf,t −[Simp cα 1 exp f,c,t−1+Fimp c](1 − 1 imp f,c,t−1)>0 0otherwise (3) Adapting Roberts and Tybout (1997) to an import decision context, we estimate equation (3) as a reduced-form expression in exogenous plant and market characteristics at period t. In order to parametrize the model, we assume that variation in Πft −Fimp f,c = Πf,t −[Simp cα 1 exp f,c,t−1+Fimp c]comes from three different sources: a country-specific sunk costs common to all firms, which includes the start-up cost Fimp cand the search cost Simp c, a firm-country specific component of the search cost which depends on previous exporting activities of firm fin the same market c, and observable time-varying firmspecific characteristics. This intuition leads us to the empirical specification in Section 4.2. 4.2 Empirical Specification Our main estimating equation for import choice: 1 imp f,c,t =α0+α1 1 exp f,c,t−1+ϕf,c +ϕc,t +ϕf,t +εf,c,t (4) 16The time subscript is omitted from entry costs to keep the notion tractable. In the empirical section, however, we can test whether they vary over time. 15 where 1 imp f,c,t is an indicator variable equal to 1 if firm fis importing from origin country cin year tfor the first time (i.e. it was not importing in t−1and t−2), and 1 exp f,c,t−1 indicates whether the same firm fwas exporting for the first time to destination cin the preceding year (t−1).17 The main level of variation we are interested in is the temporal variation of a firm-country pair, captured by ϕf,c, which allows us to see whether a firm that started to export to a new destination in the past (i.e. the regressor switches from 0 to 1 over time) is more likely to start importing back from that same country (i.e. the dependent variable switches from 0 to 1 over time). Using firm-country FE allows to control for any unobserved factors specific to the firm and the country the firm is trading with. For instance the fact that a firm with a Japanese CEO will be more likely to have trading relations with Japanese firms as compared to a firm with a Belgian CEO. Thanks to the high granularity of the data, we can also control for unobserved time-varying shocks that may be firm-specific (through ϕf,t) or countryspecific (through ϕc,t). Even after including these controls, we are still left with a problem of endogeneity. There are, indeed, unobserved factors that affect the decision to import by the firm. These could be firm-specific permanent, or at least highly serially correlated, unobserved factors or time-varying shocks that increase the probability of import entry. A natural candidate is a firm-specific productivity shock that might increase the demand for inputs by the firm and thereby its imports of inputs. In order to control for this, we adopt several strategies that are illustrated in Section 5.1 and we are able to exclude that such within-firm productivity improvements are the main drivers behind the positive relation between exports and subsequent imports from the same country. Estimating equation (4) allows us to investigate whether firms are more likely to learn about new sourcing opportunities after they already had export activity in a given country such that we expect 0< α1<1. The logic is simple: firms should be able to find new potential suppliers with greater likelihood in a country when they are already present through exporting compared to firms without previous trading relationship with that same foreign market. That is because previous exporting activities reduces the cost of recovering information about suppliers. However, using a simple OLS estimator, we cannot verify whether there is a causal link from export to import entry. In order to do that, we need to solve the identification challenge arising from the fact that there could be unobserved shocks either in the foreign country, other than information spillovers, leading to the same sequential pattern between exporting and 17Since we are interested in testing whether previous exporting activity in a country increases the likelihood of starting to import from the same country, we define the left-hand side variable, 1 imp f,c,t, as import entry in country cat time tand by construction 1 imp f,c,t−1will be set equal to zero, which is why it disappears from equation4. 16 importing besides the unobserved within-firm shocks discussed in section 5.1. For instance, aggregate shocks in a foreign market could foster both Slovenian imports from that country (aggregate supply shock, i.e. foreign suppliers become more productive and therefore start to export) and could lead export from Slovenia towards the same country to grow (aggregate demand shock). Given that adjustments take time, either imports from or exports to Slovenia could come one before the other, but any inference on causality from Slovenian exports to Slovenian imports might be erroneous. Moreover, even assuming we are able to isolate the information channel, we cannot exclude the possibility that the mechanism works in the opposite direction, from imports to exports, giving rise to a simultaneity issue which is exacerbated by the presence of serial correlation in the unobserved error term. 4.3 IV identification Therefore, in order to interpret the effect of past export activity on import entry as causal, and excluding the reverse relation from importing to exporting, we implement an IV strategy on the endogenous variable, 1 exp f,c,t−1. Conceptually, we can think of our problem as similar to a problem of demand estimation where demand and supply constitute a simultaneous system of equations, and researchers use a supply shifter to identify demand parameters (Bresnahan, 1989). Analogously, in our case imports and exports may be simultaneously determined and we cannot disentangle which one affects the other one. In order to correctly estimate the coefficient of interest, we resort to an export shifter, which is a variable that exogenously affects the export decision. The export choice equation expressed as a function of the export shifter and other variables represents the first-stage of a 2SLS procedure. The second stage corresponds to the import choice equation (4) already discussed in Section 4.2.We have added the superscript imp and exp to variables that also appears in the export equation in order to distinguish the two. The equations for the two stages are: First stage: 1 exp f,c,t−1=β0+β1 1 imp f,c,t−2+β2Zf,c,t−1+ϕexp f,c +ϕexp f,t−1+ηf,c,t−1+εexp f,c,t−1 1 imp f,c,t =α0+α1Xf,c,t−1+ϕimp f,c +ϕimp f,t +α2ηf,c,t +εimp f,c,t The endogeneity problem arises because of a non i.i.d. component of the error term, what we call ηf,c,t, which is serially correlated over time creating a problem of simultaneity: either Xf,c,t−1is affecting Mf,c,t or Mf,c,t−2is affecting Xf,c,t−1through the unobserved impact of ηf,c,t−1on ηf,c,t. To avoid biased estimates of the coefficient of interest α1, we use Zf,c,t−1as instrument, which is a vector made of exogenous variables. Therefore, we introduce our novel instrument, which is (lagged) export entry into at 17 least one neighboring country dof the partner country c,exportf,d6=c,t−1(see Figure 3 for a graphical representation). It builds on the notion of extended gravity (Morales et al., 2019) according to which firms tend to enter foreign markets similar to previous destinations. The instrument is constructed using the gravity variable contiguity from the CEPII GeoDist dataset, which indicates whether a pair of countries are neighbors. 18 Figure 3: Entry in the neighboring country IV setup Two underlying assumptions should hold in order for the instrument to be valid: (1) the aggregate productivity spillovers do not travel across neighboring countries, e.g. the fact that firms in the foreign country become more productive does not make also firms in the neighboring country more productive causing higher trade with Slovenian firms; (2) information spillovers, on the other hand, do travel across contiguous and similar countries, meaning that information about the foreign customers come from the firms in the neighboring country who are in contact with Slovenian firms. They are quite strong assumptions and we cannot directly test them. However, we can adopt an alternative version of the instrument that imposes a weaker condition: (3) we allow for the existence of productivity spillovers across neighboring countries, but we assume information and productivity spillovers do not decay/spread across neighboring countries at the same rate (see Figure 4). The intuition behind the instrument relies on distinct type of information flows that characterize export and import choices. In the sequential exporting argument, the firm enters exporting markets sequentially because there is a positive option value of doing so (instead of doing it in one step). This option value is positive only because firms learn something about the demand parameters from exporting in a neighboring country. Therefore, within country spillovers are about the potential partners (search), whereas instruments exploit correlations between markets’ characteristics in terms of 18When we create this variable, we balance the dataset by year and country, which means we fill each pair firm-country with zeros not only along the time dimension but also for each potential country there could be trade flows from/to. Mainly due to computational reasons, we have restricted the analysis to the top 70 partner countries of Slovenian firms. 18 Figure 4: Spread of information vs other shocks (a) Information shocks (b) Other shocks demand, other standard determinants of trade like in Chaney (2008) distance, etc., and also firm-level drivers of trade flows (Melitz, 2003). As we cannot completely rule out that there are geographically correlated shocks inducing a positive temporal correlation, we use an alternative instrument which is not subject to the main problem: the weighted average of the world import demand of a product kconstructed using a “shift-share” approach proposed by Hummels et al. (2014).19 Worldwide imports of product kfor country cis an aggregate measure and therefore, by construction, exogenous to Slovenian firms. The weight, on the other hand, is built to vary across firms and time building on the fact that each firm has its bundle of imported products. Moreover, the weights used to build the firm-level average are the pre-sample share of product kpurchased over the total of imports such that we do not have to worry about contemporaneous shocks to technology that could affect the types of inputs used and the import decision. In Section 7, we provide the results using this alternative instrument and compare them with results using the main instrument. 5 Baseline Results There are several ways of estimating a binary-choice model, but given the high dimensionality of the fixed effects, the linear probability model is the only feasible option. Table 4 reports the estimation results of several models based on estimating equation 4 above using Linear Probability Model estimated with an OLS estimator. Regardless of the specification, entering a new export market increases the probability of sourcing from that same country in the following year. Since we are interested in import entry, 19There is a rich emerging literature on shift-share instruments which are used in a multitude of contexts as illustrated in Adao et al. (2019); Borusyak et al. (2019); Goldsmith-Pinkham et al. (2020). 19 we drop a pair firm-country after the first year the firm starts importing from a new country. Therefore, the pair is dropped from t+1 as soon as 1 imp fct becomes positive. If a firm starts to import more than once within the period considered (e.g. in 1999 and then again in 2004 where the definition of new import entry applies because the firm is not importing in t-1, t-2 and t-3 as it is required to have 1 imp fct−x= 0 for all x=1,2,3), then we consider only the first entry as a true entry. Table 4: Linear Probability Model for the Decision to Start Importing (OLS estimator) Source: Own calculations based on AJPES, SURS and FURS data. Column (1) reports the estimated coefficient of import entry indicator regressed on lagged export indicator in the specification without any fixed effects. The coefficient is around 3 percent and statistically significant, which implies that the lagged entry in a new market increases the likelihood of import entry from the same market by 3 percentage point. In the three following columns (columns (2)-(4)), we sequentially include firm, year and country FE, and we find that the coefficients are slightly lower, particularly when country fixed effects are included, although in all specifications statistically significant and around 2 percentage points. Then, we add to the specification the interacted fixed effects starting from firm-country fixed effects to take into account constant factors that are specific to a firm trading with a given destination (column (5)), constituting our preferred specification since we base our identification on within-firm- country temporal variation. The coefficients remain similar and point to the fact that export entry in a country leads to an increase of almost two percentage points in the probability of import entry the following year. In column (6), we also add firm-year FE 20 in order to take into account any potential shock that could hit the firm in a given year. Eventually, in column (7), we control for country-specific aggregate shocks (countryyear FE) and we still find a positive effect, even though the magnitude of the effect has been reduced by as much as one third in comparison to the first specification. To give a sense of the magnitude, we can look at results from the most complete specification in the last column of Table 4, which suggests that export incursion in a market increases the probability of starting to import the next year by slightly more than 1 percentage point. Given that the unconditional probability of import entry is 3%, it corresponds to a 33%increase in the probability of importing with respect to the unconditional probability which is quite remarkable. In the last two columns, we check for two factors potential affecting our estimates. We restrict the sample to 1996-2004 in order to eliminate the effect of reporting cutoffs to comply with EU rules: results remain the same. Eventually, we relax the assumptions that only a change in export behavior at time t−1produce an effect on the change in import behavior at time tadding to the specification also changes in export at time t−2and t−3. We find that coefficients for the three separate lagged regressors - jointly inserted - are very similar in magnitude hinting at the fact that there is not a specific timing for the effect to take place. Table 5: (Reverse) Linear Probability Model for the Decision to Start Exporting (OLS estimator) Source: Own calculations based on AJPES, SURS and FURS data. The relation could potentially also go the other way around: now it is import experience, and the associated interactions with new suppliers, that increases the likelihood 21 Figure 5: Heterogeneity by size, OLS Note(s): OLS linear probability regression where the main regressor is interacted with the decile of sales a firm belongs to. Firm-country, firm-year and country-year fixed effects are included in this specification. The dashed horizontal line represents the estimated coefficient for the full sample. Source: Own calculations based on AJPES, SURS and FURS data. chases of product kfrom the world (less its purchases from Slovenia) at time t,.22 An increase in WIDckt could result from shocks to demand either in consumer tastes or in firms’ uses of particular inputs, or reflect a reduction in comparative advantage by c in product k. Since this instrument has country-product-time variation, we get a single value for each firm-country-year by using aggregation similar to Hummels et al. (2014). 23 More specifically, we weight W IDckt for importing country cselling HS code at 6- digit product kat time tusing sharefkt0which is the share of value of imports of product kin total value of imports for firm fin the pre-sample year (1996) t0. For those firms which were not there in the pre-sample year (either because they entered or began to export within sample), we use information from their first year of exporting and use data from the second year onward for the regressions. Then we create a time varying instrument for firm ftrading with partner country c, namely shWIDfct =Pk∈Kcsharefkt0WIDckt where Kcrepresents the set of products exported by firm fto country cin the pre-sample year t0. The logic behind this demand-type instrument is as follows: over time there are shocks to the demand of product kfrom country cwhich are exogenous to firm f, and these are reflected in changing import demand to the world as a whole; since firm f 22The validity of this type of instrument has been extensively explored in recent papers (Adao et al., 2019; Borusyak et al., 2019; Goldsmith-Pinkham et al., 2020). 23Similar demand shock is used in Mayer et al. (2016). 28 Figure 6: Heterogeneity by size, IV export to a neighboring country Note(s): Linear probability regression where the main regressor is interacted with the decile of sales a firm belongs to and (lagged) export entry is used as instrumental variable. Firm-country, firm-year and country-year fixed effects are included in this specification. The dashed horizontal line represents the estimated coefficient for the full sample. Source: Own calculations based on AJPES, SURS and FURS data. exports product kmore than other firms (i.e. the share is larger), it benefits from these changes more than a firm which does not export the same product to that country or which exports it but in a relatively smaller share. A potential threat to identification might be that WIDckt is directly affecting Mfct. However, thanks to the high level of disaggregation of the data, it is reasonable to exclude this threat. Assume firm fis exporting and importing a single-product k. Due to a technological improvement, the production of this product k, say computers, becomes cheaper in several countries. Then, country c, e.g. Germany, starts to import more computers from all over the World (WIDckt >0). At the same time, the Slovenian firm fstarts to import a lot of computers from Germany too. It is unlikely that Germany both imports and exports computers.24 It could happen, however, that we capture intra-industry trade if products are reported at 2-digit level. For instance, Germany might be importing computer-related components from other countries and export computers to Slovenia. Since products are reported at 6-digit level, we can safely exclude we are capturing this intra-industry type of trade. Table 9 illustrates coefficient estimates when export entry is instrumented with the weighted average of World import demand: the coefficient moves from almost 3 percentage points of the OLS estimation displayed in the column (1) to 22 percentage 24Unless many firms in Germany are doing the so-called carry-along trade (Bernard et al., 2019) also called pass-on-trade in Damijan et al. (2013). 29 Table 9: Shift-share instrument: IV results Source: Own calculations based on AJPES, SURS and FURS data. points when using the shift-share, which remains positive and highly significant (column (3)). Given that these results are very similar in sign and size to those in Table 8 where the instrument was export to at least a neighboring country, we can conclude we do not need to worry about geographically correlated errors. First-stage results and some tests for validity and significance of the instrument, reported here in the same fashion as in Table 9, suggest that the shift-share instrument is working correctly. LATE and compliers’ characteristics Since the IV estimand corresponds to the local average treatment effect (LATE), which is the effect of the treatment on a specific subgroup of the population, the compliers, we need to investigate this group’s characteristics. The exogenous variation driven by the instrument is only a subset of the total variation in export entry. IV, thus, reduces the variation in the data and the variation we are left with comes only from the units which responded to the instrument in the first place (Cunningham, 2021). Therefore, the questions are: how is this sub-sample of units composed? What are its characteristics? To answer these queries, we describe the distribution of characteristics of the compliers as opposed to characteristics for the whole sample following the methodology in Pinotti (2017) for some variables of interest. This is based on the following formula: E(g(K)|compliers) = [E(g(K)Xf,c,t−1)|Zf,c,t−1= 1) −E(g(K)Xf,c,t−1)|Zf,c,t−1= 0)] [E(Xf,c,t−1|Zf,c,t−1= 1) −E(Xf,c,t−1|Zf,c,t−1= 0)] (7) where g(K) is the distribution of an individual characteristic K (see, e.g. Angrist et al. (2016)). The right-hand side of Equation (7) is easily estimated by the 2SLS regression of g(K )Xf,c,t−1on Xf,c,t−1using vector component of Zf,c,t−1as instruments. More specifically, we pick three characteristics of the firms, i.e. employment, sales and tan- 30 gible capital25. Therefore, we present the distribution of these characteristics for the whole sample vs distribution for compliers in order to answer the following question is: are the average characteristics of compliers similar to those of other firms? As shown in Figures 9-15 in the Appendix, compliers are, on average, larger, both in terms of employment and sales, and are also more capital intense. While the full sample distribution (blue bars) follows more or less a Pareto one, the compliers distribution (red bars) is more skewed to the right. In particular, red bars are always higher than blue bars for greater values of the variable considered confirming that compliers, namely firms that started to export in previous year(s), are systematically different from other firms. This holds across all possible definitions of the instrumental variable, either export entry or export dummy to a neighboring country - being lagged by one, two or three periods, or the shift-share instrument. We also perform the IV regression by Figure 7: Heterogeneity by size, IV shift-share Note(s): Linear probability regression where the main regressor is interacted with the decile of sales a firm belongs to and the shift-share variable is used as instrument. Firm-country, firm-year and countryyear fixed effects are included in this specification. Source: Own calculations based on AJPES, SURS and FURS data. deciles of size for the shift-share instrument (in its logarithmic version). Results from these regressions are reported in Figure 7 where estimates have a similar pattern as those depicted in Figure 6: coefficients are higher, the larger the firm is. Therefore, the positive effect of export on import entry is mainly driven by the activities of firms in the right tail of the distribution. 25They have been transformed from continuous to categorical variables divided in different categories as illustrated in each figure 31 8 Conclusions This paper shows a positive effect of export on the decision to start importing in the subsequent year. We attribute this effect to reduction of information frictions. We find that an export incursion in a market increases the probability of starting to import the next year by slightly more than 1 percentage point using the OLS specification. Given that the unconditional probability of import entry is 3%, it corresponds to an increase of 33%in the probability of importing with respect to the unconditional probability which is quite remarkable. However, OLS estimates are subject to endogeneity issues even after controlling for a plethora of fixed effects. Therefore, we introduce a novel instrument which builds on the notion of sequentiality in exporting: export entry in at least one neighboring country. The IV approach allows us to assert causality of the relation and argue that the direction of the sequentiality goes from export to import, and not the other way around. IV estimates do confirm the positive effect of export on import entry found using OLS. Moreover, the coefficients increase significantly in size across several specifications. This, at first, might seem suspicious given that we would have expected a reduction in magnitude for the IV estimates due to an over-estimation of OLS. Given that the IV estimand is providing a local average treatment effect (LATE), which is the effect on a sub-sample of the population, i.e. only firms that have responded to the treatment, we investigate whether the effect is heterogeneous across size categories. We find that the effects of lagged export entry on import entry are increasing in firm size, for OLS and, even more, for IV estimates, More specifically, coefficients for our main IV specification are around 10-20 percentage points for the last three deciles of the size distribution, while they become negative for smaller firms. We additionally perform several robustness checks that rule out the role of productivity as the main driver of this relation. Overall, we can interpret these findings to be consistent with a narrative of information frictions about suppliers that can be overcome by exploiting the firm’s presence in a foreign market through exporting. However, this positive effect holds only for firms that are larger and therefore more likely to be already importers, reinforcing even more their presence in international markets at the expense of smaller firms. 32 A Appendix Figure 8: Average number of destination and origin countries by firm-product (1996- 2011) Source: Own calculations based on AJPES, SURS and FURS data. 33 Table 10: Value (m. Euros) , by year sample: manufacturing firms engaged in international trade at least once Year Export value Export value (>0) Import value Import value (>0) 1996 662 1763 486 737 1997 721 1987 533 845 1998 821 2272 607 970 1999 880 2382 628 1007 2000 1,084 2920 775 1286 2001 1,303 3413 941 1554 2002 1,412 3713 984 1672 2003 1,479 3767 1027 1662 2004 1,572 4471 1078 1986 2005 1,811 6609 1215 3240 2006 2,068 6928 1356 3352 2007 2,313 7976 1493 3330 2008 2,342 7868 1504 3248 2009 1,893 6412 1113 2568 2010 2,233 7478 1371 3124 2011 2,507 8097 1476 3254 Source: Own calculations based on AJPES, SURS and FURS data. 34 Distribution of characteristics for the whole sample vs distribution for compliers (a) Employment (b) Sales (c) Capital Figure 9: Instrument: Export entry to dat time t−1 (a) Employment (b) Sales (c) Capital Figure 10: Instrument: Export entry to dat time t−2 (a) Employment (b) Sales (c) Capital Figure 11: Instrument: Export entry to dat time t−3 35 (a) Employment (b) Sales (c) Capital Figure 12: Instrument: Export dummy to dat time t−1 (a) Employment (b) Sales (c) Capital Figure 13: Instrument: Export dummy to dat time t−2 (a) Employment (b) Sales (c) Capital Figure 14: Instrument: Export dummy to dat time t−3 (a) Employment (b) Sales (c) Capital Figure 15: Instrument: shift-share log(shareWID) 36 References Daron Acemoglu and Joshua Angrist. How large are human-capital externalities? evidence from compulsory schooling laws. NBER macroeconomics annual, 15:9–59, 2000. Rodrigo Adao, Michal Kolesár, and Eduardo Morales. Shift-share designs: Theory and inference. The Quarterly Journal of Economics, 134(4):1949–2010, 2019. JaeBin Ahn, Amit K Khandelwal, and Shang-Jin Wei. The role of intermediaries in facilitating trade. Journal of International Economics, 84(1):73–85, 2011. Facundo Albornoz and Ezequiel García Lembergman. Importing after exporting. 2015. Facundo Albornoz, Héctor F Calvo Pardo, Gregory Corcos, and Emanuel Ornelas. Sequential exporting. Journal of International Economics, 88(1):17–31, 2012. Treb Allen. Information frictions in trade. Econometrica, 82(6):2041–2083, 2014. Joshua D Angrist and Alan B Keueger. Does compulsory school attendance affect schooling and earnings? The Quarterly Journal of Economics, 106(4):979–1014, 1991. Joshua D Angrist, Sarah R Cohodes, Susan M Dynarski, Parag A Pathak, and Christopher R Walters. Stand and deliver: Effects of boston’s charter high schools on college preparation, entry, and choice. Journal of Labor Economics, 34(2):275–318, 2016. Pol Antras, Teresa C Fort, and Felix Tintelnot. The margins of global sourcing: Theory and evidence from us firms. American Economic Review, 107(9):2514–64, 2017. Costas Arkolakis. Market penetration costs and the new consumers margin in international trade. Journal of political economy, 118(6):1151–1199, 2010. Costas Arkolakis, Sharat Ganapati, and Marc-Andreas Muendler. The extensive margin of exporting products: A firm-level analysis. American Economic Journal: Macroeconomics, 13(4):182–245, 2021. Roc Armenter and Miklós Koren. Economies of scale and the size of exporters. Journal of the European Economic Association, 13(3):482–511, 2015. Maria Bas and Vanessa Strauss-Kahn. Does importing more inputs raise exports? firm-level evidence from france. Review of World Economics, 150(2):241–275, 2014. Giuseppe Berlingieri, Emanuel Ornelas, and Luca Marcolin. Service offshoring and export experience. 2021. Andrew B Bernard and J Bradford Jensen. Why some firms export. Review of economics and Statistics, 86 (2):561–569, 2004. Andrew B Bernard, Jonathan Eaton, J Bradford Jensen, and Samuel Kortum. Plants and productivity in international trade. American economic review, 93(4):1268–1290, 2003. Andrew B Bernard, Marco Grazzi, and Chiara Tomasi. Intermediaries in international trade: Products and destinations. Review of Economics and Statistics, 97(4):916–920, 2015. 37