Exchange Rate Volatility, Financial Constraints, and Trade: Empirical Evidence from Chinese Firms1 ´ ro Je ´ ricourt and Sandra Poncet ˆ me He In this paper, we study how firm-level export performance is affected by Real Exchange Rate (RER) volatility and investigate whether this effect depends on existing financial constraints. Our empirical analysis relies on export data for more than 100,000 Chinese exporters over the 2000–6 period. We confirm a trade-deterring effect of RER volatility. We find that firms’ decision to begin exporting and the exported value decrease for desti- nations with a higher exchange rate volatility and that this effect is magnified for finan- cially vulnerable firms. As expected, financial development seems to dampen this negative impact, especially on the intensive margin of export. These results provide micro-founded evidence suggesting that the existence of well-developed financial markets allows firms to hedge exchange rate risk. The results also support a key role of financial constraints in determining the macro impact of RER volatility on real out- comes. JEL codes: F14, F31, L25 INTRODUCTION The increasing volatility of exchange rates after the collapse of the Bretton Woods agreements has been a source of concern for both policymakers and aca- demics. In a context where firms are risk averse, exchange rate risk increases trade costs and reduces the gains from international trade (Ethier 1973). However, initial macroeconomic evidence on the effect of exchange rate volatili- ty on trade has been quite mixed, identifying an effect that is either significant but small or insignificant (see Greenaway and Kneller (2007) and Byrne et al. 1. Je ˆ me He ´ ro ´ ricourt is an assistant professor at the University of Lille 1, a researcher at EQUIPPE-Universite ´ s de Lille & CES (U. Paris 1), and a research fellow at CEPII; his email address is jerome.hericourt@univ-lille1.fr. Sandra Poncet (corresponding author) is a professor at the Paris School of Economics (University of Paris 1) and a scientific advisor at CEPII; her email address is sandra.poncet@ cepii.fr. We thank two anonymous referees and the editor for very insightful comments. We are also grateful to Raphael Auer, Agne ´ nassy-Que ` s Be ´ , Kenza Benhima, Nicolas Berman, Nicolas Coeurdacier, ´ re Fabrice Defever, Peter Egger, Sarantis Kalyvitis, Sergey Nigay, Katrin Rabitsch, Glenn Ryap, and participants at several seminars and conferences for very useful comments and discussions on earlier drafts of the paper. Part of this research was funded by the French Agence Nationale de la Recherche (ANR) under grant ANR-11-JSH1 002 01. A supplemental appendix to this article is available at http://wber. oxfordjournals.org/. Any remaining errors are ours. THE WORLD BANK ECONOMIC REVIEW, VOL. 29, NO. 3, pp. 550– 578 doi:10.1093/wber/lht035 Advance Access Publication October 24, 2013 # The Author 2013. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 550 ´ ricourt and Poncet He 551 (2008) for a survey). Even Rose (2000), who finds a very large effect of currency union on international trade, finds a small effect of nominal exchange rate vola- tility. More recent works have emphasized that these results may occur because of an aggregation bias2 (Byrne et al. (2008) study the impact of the nominal ex- change rate volatility; Broda and Romalis (2011)3 focus on real exchange rate volatility)4 and because of an excessive focus on richer countries with highly de- veloped financial markets. More substantial negative effects of the real exchange rate volatility on trade are found for developing countries (Grier and Smallwood 2007). There is a considerable lack of firm-level evidence on the impact of exchange rate volatility on exporting behavior and how this relationship may be influenced by financial constraints, which are likely to be much stronger and more binding in developing countries. A careful firm-level study of these relationships may provide clearer evidence regarding the exacerbating role of exchange rate volatili- ty for export costs and the role of financial development in alleviating these addi- tional costs. This paper aims to fill these gaps. We study the impact of Real Exchange Rate (RER) volatility on exporting behavior and the role of financial constraints, together with financial development, in shaping this relationship at the firm level. Our empirical estimations rely on export data for more than 100,000 Chinese exporters over the 2000–6 period. China is a highly relevant case for several reasons. First, the country displays an especially high export rate given its size, leading to substantial exposure to exchange rate fluctuations. Second, China is interesting because it is characterized by low financial development but rather high regional heterogeneity, which isuseful to identify a non-linear effect of ex- change rate volatility depending on credit constraints. Finally, the Chinese Yuan was strongly pegged to the US dollar during nearly the entire period considered, implying that the volatility we identify is truly exogenous to Chinese economic developments over the considered period. More precisely, the Chinese exchange rate policy over the period is best de- scribed as a fixed peg versus the US dollar until July 2005. At that time, the Chinese government switched to a reference to a basket of other currencies. However, Frankel and Wei (2007) find that the de facto regime remained a peg to a basket that put virtually all of the weight on the dollar. Subsequently, some weight was shifted to a few non-dollar currencies. In any case, the peg remained 2. Because the effects of RER volatility differ across firms or sectors and countries, aggregating across these disparate units can produce weaker or insignificant results. 3. Broda and Romalis (2011) also address the issue of reverse causality between exchange rate volatility and trade. Once the problem is controlled for, they still find a negative, albeit reduced, impact of volatility on trade. 4. Although the volatility of the real exchange rate differs conceptually from that of the nominal exchange rate, as shown by Clark et al. (2004), they do not differ much in reality. In the literature, volatility indicators based on real or nominal exchange rates are used similarly, but with a strong preference for the former. 552 THE WORLD BANK ECONOMIC REVIEW fairly strong in 2006.5 The Chinese exchange rate policy is also characterized by limited convertibility and misalignment over the period. Although impediments to convertibility may have decreased as China made tentative attempts to inter- nationalize the yuan by the end of the 2000s, charges of undervaluation devel- oped over the period. Nevertheless, these two dimensions apply to all exporters independently of the destination country; hence, they are not country-specific. Our econometric approach, which focuses on the repercussions of RER volatili- ty, exploits cross-country variations for firm exports and accounts for these common specificities through time fixed effects. We expect a negative impact of exchange rate volatility on trade through an in- crease in the variable and sunk costs of exporting. The former effect is implicitly addressed by Ethier (1973) and is the most intuitive one: exchange rate risk creates an uncertainty for the exporter’s earnings in her own currency, which is similar to an increase in variable costs. However, exchange rate volatility may also increase the sunk costs of exports, which can be seen as a form of investment in intangible capital. In practice, most investment expenditures are at least partly irreversible; they are made of sunk costs that cannot be recovered if market con- ditions turn out to be worse than expected. The combination of investment irre- versibility and asymmetric adjustment costs induces a negative relationship between price volatility and investment (Pindyck 1988, 1991), especially in de- veloping economies (see Pindyck and Solimano 1993). In such a context, high volatility has consistently proven to reduce growth and investment, especially private investment (Ramey and Ramey (1995), using the volatility of output growth; Aizenman and Marion (1999), focusing on RER volatility; Schnabl (2007), relying on nominal exchange rate volatility). Bloom et al. (2007) find similar results within a firm-level framework with partial irreversibility: higher uncertainty reduces the responsiveness of investment to a firm-level demand shock. Only recently, however, has the macro literature explicitly identified a rela- tionship between credit constraints and the size of the impact of volatility. Aghion et al. (2009) show that the local financial development plays a key role in the magnitude of the repercussions linked to exchange rate volatility. Relying on a panel of 83 countries over the 1960-2000 period, these authors show that the negative impact of RER volatility on productivity growth decreases with a coun- try’s financial development. Within an identical framework, but focusing on foreign currency (dollar) liabilities, Benhima (2012) shows for a panel of 76 emerging and industrial countries between 1995 and 2004 that the higher the share of foreign currency in external debt, the more detrimental RER volatility is to growth. This finding tends to support the idea that the effect of RER volatility depends critically on the existence of credit constraints. 5. In any case, our results are unchanged when excluding the years 2005 and 2006. More details on this robustness check are available upon request. ´ ricourt and Poncet He 553 The link between volatility and export performance has usually been investi- gated using macro and, less frequently, disaggregated data at the sectoral level.6 Some papers examine the impact of the RER on exporting firms (e.g., Berman et al. (2012) on France; Li et al. (2012) and Park et al. (2010) on China), but they focus on the impact of the exchange rate level rather than its volatility, and they do not account for the role of financial constraints. Firm-level studies of the impact of exchange rate volatility on economic or trade performance for develop- ing countries are scarce. Carranza et al. (2003) find a negative impact of RER volatility on a sample of 163 Peruvian firms, and Cheung and Sengupta (2012) study the impact of real effective exchange rate variations and volatility on the share of exports-to-sales ratio for a sample of a few thousand Indian non- financial sector firms and find support for a negative effect of volatility. With regard to the role of credit constraints in modeling the impact of RER volatility, especially on export performance, research is almost nonexistent. To our knowledge, Caglayan and Demir’s (2012) study is the only firm-level study connecting firm productivity, RER movements, and the issue of access to exter- nal finance. Based on a data set of 1,000 private Turkish firms, their results support a negative impact of exchange rate volatility on productivity growth, which is downplayed by better access to external finance. We depart from these previous works by using a much wider data set of firms, by examining whether firms move their exports away from partners character- ized by higher exchange rate volatility, and more importantly, by investigating the presence of a non-linear effect of exchange rate volatility on performance de- pending on the level of financial constraints in the Chinese context. The latter issue is considered through two complementary dimensions. First, we infer firm- level financial vulnerability from the financial dependence of firms’ activities. This approach was pioneered by Rajan and Zingales (1998) and has proven to be a robust methodology to detect credit constraints and assess their evolution (Kroszner et al. 2006, Manova et al. 2011). Second, we exploit Chinese cross- provincial heterogeneity to study how financial development may mitigate both credit constraints and exchange rate volatility. This paper contributes to the existing literature on various levels. First, we provide a micro-founded investigation of Aghion et al.’s (2009) prediction that exchange rate volatility is especially harmful to firms that have high liquidity needs when local financial development is low. Second, our methodology allows us to circumvent a number of endogeneity problems that may have flawed some related studies. The use of firm-level data mitigates the issue of reverse causality from trade to exchange rate volatility (cf. Broda and Romalis 2011) and the well- known simultaneity bias between exporting behavior and financial proxies for credit constraints at the firm level. It is very unlikely that a Chinese firm shock would impact exchange rate volatility or measures of financial dependence based 6. Some papers examine the impact of RER variations on Chinese trade, including Marquez and Schindler (2007), Ahmed (2009), Freund et al. (2011), and Cheung et al. (2012). 554 THE WORLD BANK ECONOMIC REVIEW on data from US firms. Furthermore, using cross-regional data within a single country instead of cross-country data reduces the risk of confusion between fi- nancial development and other macro characteristics.Third, our results provide insight into the main sources of the apparent lack of a macro impact of exchange rate volatility: the level of financial constraints and financial development appears to be more important than the aggregation bias to explain this puzzle. We find that the repercussions of RER volatility are not unconditional, even at the micro level, and are mainly related to financial factors. Our results are consistent with the aforementioned macro studies, especially that of Aghion et al. (2009). Both the decision to begin exporting and the export- ed value decrease for destinations with higher exchange rate volatility. This export-deterring effect is magnified for financially vulnerable firms. For firms that are most dependent on external finance, a 10 percent increase in RER vola- tility decreases the value exported by 11 percent and the probability of entering by 3 percent. As expected, financial development seems to dampen this negative impact, especially on the intensive margin of export. These results are robust to various definitions of trade margins, measures of exchange rate volatility and fi- nancial dependence, subsamples, and the inclusion of additional controls. We therefore provide micro support to the macro literature suggesting that financial development is a key determinant in identifying the impact of RER volatility on real outcomes. In the next section, we survey the different theoretical mechanisms underlying our approach before discussing our general methodology and presenting our da- tabase in section 2. In section 3, we begin by presenting the results on the inten- sive margin and then on the extensive margin before introducing some robustness checks and a general discussion of our findings. Section 4 concludes. E X C H A N G E R AT E VO L AT I L I T Y, F I N A N C I A L CO N S T R A I N T S , AND EXPORTS: THEORETICAL UNDERPINNINGS Related Literature Our approach stands at the crossroads of two strands of the literature. First, there is a rapidly increasing number of papers that consider the behavior of firms that manufacture and export several products to several destinations. It is widely known that aggregate exports are concentrated in a small number of major players (Eaton et al. 2004) and that large exporters are involved in exporting more than one product (Bernard et al. 2011; Eckel et al. 2011). Bernard et al. (2011) show that the proportion of multi-product firms that export, the number of destinations for each product, and the range of products they export to each market all increase in response to reduced variable trade costs. Even closer to our work is that of Berthou and Fontagne ´ (2013), who document the impact of the introduction of the euro on the export decisions of French firms, the number of products exported, and average sales per product. Their results point to a ´ ricourt and Poncet He 555 heterogeneous trade creation effect across euro area destinations: for those firms exporting to destinations characterized by lower monetary policy coordination (that is, higher exchange rate volatility) before 1999, exports grew by 12.8 percent following the introduction of the euro, with 20 percent of the effect due to an increase in the number of products exported. By contrast, no effect arises regarding the decision to export. Conversely, they find a negative effect on all three definitions of trade margins for euro area destinations with closer monetary policy coordination before 1999, indicating that the additional competitive pres- sure more than offset the benefits of zero volatility. Secondly, there is growing empirical evidence that credit constraints impact exporting behavior (Manova 2013; Greenaway et al. 2007; Berman and He ´ ricourt 2010; Minetti and Zhu 2011). The first paper on this topic by Manova (2013) incorporates financial frictions into a heterogeneous-firm model before applying it to aggregate trade data. She finds that 20 percent to 25 percent of the impact of credit constraints on trade are driven by reductions in total (domesti- cally sold and exported) output. Of the additional trade-specific effect, one-third reflects limited firm entry into exporting, and two-thirds is due to contractions in the sales of exporters. Both extensive and intensive margins are therefore affected by credit constraints. All subsequent papers consistently find that the effect is magnified when firms belong to industries that rely more on external finance (Minetti and Zhu 2011), and in developing countries (Berman and He ´ ricourt 2010) compared to developed ones (Greenaway et al. 2007). Our paper explores the possibility of a negative impact of exchange rate vola- tility on trade that is proportionally stronger for financially vulnerable firms and, consequently, weaker with high levels of financial development. This impact can be generated by several mechanisms. One can think of exchange rate risk creating uncertainty for the earnings of the exporter, which is equivalent to uncertainty on variable trade costs. The results of Bernard et al. (2011) and Berthou and Fontagne ´ (2013) show that all trade margins are potentially involved. The exis- tence of well-developed financial markets should allow agents to hedge the ex- change rate risk, thus dampening or eliminating its negative effect on trade. This effect has not been clearly established either empirically (Dominguez and Tesar 2001) or theoretically (Demers 1991). Therefore, it is interesting to examine whether micro data can provide clearer insights. Another mechanism that is more focused on the sunk costs of exports and therefore is especially appropriate for the decision to export to new markets may also be at work. On the one hand, export capacity may be considered a type of investment in intangible capital (such as R&D); on the other hand, exchange rate movements give rise to additional sunk costs (Greenaway and Kneller 2007). The negative impact of exchange rate volatility on exports can be rationalized through the asymmetry of adjustment costs leading to investment irreversibility. When facing a real depreciation of its own currency, the current earnings of a firm rise. The firm may use this additional income to fund the sunk costs of enter- ing new markets. However, once these investments are made, it is impossible to 556 THE WORLD BANK ECONOMIC REVIEW back out and recover what they cost, even in the case of an abrupt subsequent currency appreciation. If firms are credit constrained, they will face additional difficulties in funding new investments and will be even more reluctant to take the chance of engaging in exports to markets characterized by highly volatile ex- change rates. Several approaches may theoretically rationalize this mechanism. In the study of Aizenman and Marion (1999), the introduction of credit rationing leads to a nonlinearity in the intertemporal budget constraint. In their framework, the supply of credit facing a developing country is bounded by a credit ceiling inde- pendently of the level of demand. The credit ceiling hampers the expansion of in- vestment in the high-demand state without moderating the drop in investment in the low-demand state. Thus, this asymmetric pattern implies that higher volatili- ty reduces the average rate of investment, and this effect is magnified with credit constraints. An alternative mechanism is proposed by Aghion et al. (2009). Suppose that an exporter faces fixed wage costs in the local currency. When the bilateral exchange rate vis-a` -vis that of the exporting market fluctuates, the ex- porter cannot completely pass the cost change through to the exporting market because of competitive pressures, for example. Then, exchange rate volatility leads to fluctuations in profits, which can lower investments in an environment where external finance is more costly than internal finance. Following an ex- change rate appreciation, the current earnings of firms decline. This reduces their ability to borrow to survive idiosyncratic liquidity shocks and invest in the longer term. Depreciations have the opposite effect. However, the existence of a credit constraint implies that, in general, the positive effects of a depreciation will not fully compensate for the negative effects of an appreciation. By reducing the cost of external finance, financial development relaxes credit constraints and should decrease the impact of volatility on the sunk cost activity–in our case, exports. Key Testable Predictions We can summarize the testable predictions from these models for export perfor- mance–that is, both the intensive (export value) and extensive (decision to begin exporting) margin. Testable Prediction 1. Export performance decreases with exchange rate vola- tility. We therefore expect the link between volatility, on the one hand, and the exported value and the decision to begin exporting to a market, on the other hand, to be negative. Testable Prediction 2. The negative impact of exchange rate volatility on export performance is magnified for financially vulnerable firms. Export perfor- mance is disproportionately decreased by exchange rate volatility for those firms. Testable Prediction 3. By relaxing credit constraints, financial development decreases the impact of exchange rate volatility on export performance propor- tionally more for financially vulnerable firms. ´ ricourt and Poncet He 557 DATA S OURCES AND EM PI R I CA L ME TH O D O LO GY Exchange Rate Volatility Exchange rate volatility is computed as the yearly standard deviation of monthly log differences in the real exchange rate. Because we rely on an indirect quotation (that is, one unit of foreign currency equals X units of yuan), we compute the real exchange rate as the nominal exchange rate of the yuan with respect to the partner’s currency multiplied by the partner’s consumer price (CPI) level.7 Thus, we do not divide by Chinese prices because of the likely mediocre quality of Chinese CPI, which would produce useless additional noise in the estimates. Because our empirical specification includes year dummies (see subsection II – “Empirical Specification” below), the impact on our estimates should be negligi- ble; the Chinese CPI is common to all exporters, so most of its variance is ab- sorbed in these time fixed effects. To ensure that our results are not dependent on a specific definition of volatility, we perform several robustness checks in which alternative definitions of the ex- change rate are used to build our volatility indicator (still using the yearly standard deviation of monthly log differences): the nominal exchange rate, the real exchange rate computed as the nominal exchange rate of the yuan with respect to the part- ner’s currency multiplied by the partner’s CPI and divided by Chinese CPI or the HP (Hodrick and Prescott 1997) detrended real exchange rate. We also consider a specification in which the standard deviation of the log-level of the real exchange rate is considered instead of our benchmark measure of volatility.8 Because our empirical specification includes firm-destination fixed effects to mitigate the endogeneity issue, the repercussions of RER volatility on firm export performance is identified from the variation within a firm-destination over time. Thus, our results reflect how firms allocate resources to a given market over time. In unreported checks (available upon request), we verify that our findings are not sensitive to the source of variation we exploit for the RER volatility. When esti- mating a specification that concentrates on the variation across countries, we find that RER volatility is a significant determinant of how firms allocate resourc- es across markets. We obtain a negative effect of RER volatility that is magnified for financially vulnerable firms. Trade Data The main data source is a database collected by the Chinese Customs. It contains Chinese firm-level yearly export flows by year, HS6 product, and destination 7. Monthly data on nominal exchange rates and prices are taken from the International Financial Statistics (IFS). 8. Our specification assumes that firms respond rather quickly to changes in RER volatility. This assumption is consistent with the unreported results, available upon request, that indicate that when introducing both the contemporaneous volatility and the one-year lagged volatility to explain export decisions, the former is associated with greater statistical significance. 558 THE WORLD BANK ECONOMIC REVIEW country over the 2000–6 period. It covers 113,368 exporting firms and 158 destinations Financial Vulnerability and Financial Development We compute firm-level financial vulnerability as the weighted average of the fi- nancial vulnerability of a firm’s activities. The weights are the average share of the sector in the exports by the firm over our sample period.9 0 1 X ExportsF FinVulnF ¼ @P s  FinVulns A ð1Þ ExportsF s s s We use three different measures of the financial vulnerability of a sector FinVulns, in line with other studies on the same topic. These variables are meant to capture the technological characteristics of each sector that are exogenous to the financial environment of firms and to determine the degree of reliance of the firms in each sector on external finance. Although firms in all industries may face liquidity con- straints, there are systematic differences across sectors in the relative importance of up-front costs and the lag between the time when production expenses are incurred and revenues are realized. We capture these differences with a measure of the ex- ternal finance dependence in a sector (referred to hereafter as “financial depen- dence”), constructed as the share of capital expenditures not financed out of cash flows from operations. For robustness, we also use an indicator of the asset intangi- bility of firms. This measure is the ratio of intangible assets to fixed assets. Thus, it captures another dimension of the dependence of a firm on access to external fi- nancing: the difficulty of using assets as collateral in obtaining financing. As a third indicator, we follow Manova et al. (2011), who use the share of R&D spend- ing in total sales (R&D) based on the fact that as a long-term investment, research and development often implies greater reliance on external finance. As is standard practice in the literature, these indicators are computed using data on all publicly traded US-based companies from Compustat’s annual indus- trial files. The value of the indicator in each sector is obtained as the median value among all firms in each sector. Indicators of the financial vulnerability of a sector are available for 27 three-digit ISIC sectors.10 We borrow the values 9. In the unreported results, which are available upon request, we also verify that our results hold when measuring the financial vulnerability of a firm as the financial vulnerability of its main (ISIC) sector of activity, identified as the one with the greatest export share over the period. Furthermore, our findings hold when the main sector of activity or the weights are based on the first year for which the firm reports exports instead of the average over 2000– 6. In subsection III, “Additional Robustness Tests and General Discussion”, we discuss the results of Table S.8 reported in the online appendix (available at http://wber. oxfordjournals.org/), in which margins of trade are defined at the firm-sector-destination level, and the measure of financial vulnerability is a sector characteristic as in the prior literature. 10. We use a correspondence table between the international trade nomenclatures and the ISIC Rev. 2 categories, developed at the CEPII to match the Chinese HS 6-digit product codes with the ISIC three-digit sector categories. ´ ricourt and Poncet He 559 computed by Kroszner et al. (2006). As explained by Manova et al. (2011), the use of US data is not only motivated by the lack of data for most other countries, including China, but it also has several advantages. Rajan and Zingales (1998) note thatthe United States has one of the most advanced and sophisticated finan- cial systems, so the values for US firms reflect the technology-specific component of external finance needs, or what can be called the finance content of an indus- try. Measuring these indices in the Chinese context would likely lead to different values, reflecting the fact that firms organize production differently in a credit- constrained environment. Thus, such measures would be endogenous to financial development in China, whereas measures based on data from US firms can be seen as exogenous in this respect. In addition to these firm-sector indicators of financial vulnerability, we use the level of financial development at the regional level. We thus adapt the methodol- ogy first used by Rajan and Zingales (1998), which consists of filtering the impact of financial liberalization by financial vulnerability to isolate its direct finance-related causal effect. We measure local financial development as the share of total credit over GDP in the province.11 Finally, descriptive statistics of key variables are given in Tables 1 and 2 below. Empirical Specification We estimate the following specification: F ExportPerfijt ¼ a RERVolatility jt þ b RERVolatility jt  FinVulnF þ g RERVolatility jt  FinDev jt þ d RERVolatility jt  FinVulnF  FinDevit þ t FinVulnF  FinDevit þ h FinDevit þ fZ jt þ lF F j þ ut þ 1ijt ð2Þ where ExportPerfF ijt is a measure of the export performance of firm F in province i for export destination j in year t. We use two alternative measures of export per- formance to capture the intensive and extensive margin of exports, respectively, the log of the total free-on-board export sales towards destination j in year t, and the decision to begin exporting to market j in year t. The latter is constructed as a change of export status at the firm-country level; it takes the value 1 when a firm exports to country j at time t but did not at time t –1.12 Our regressions ( performed with the linear within estimator for the intensive margin and the conditional logit model for the extensive margin) include firm- country fixed effects lF j and year dummies ut. Firm fixed effects capture the 11. In robustness checks, we verify that our results are similar when using the ratio of deposits over GDP. 12. In that set of regressions, our sample consists of a firm-country series of zeros followed by a decision to begin exporting. For a given firm-country, we can have several beginnings. For example, the subsequent export statuses 011001 become . 1 . . 01 in our sample, with . denoting a missing value. 560 THE WORLD BANK ECONOMIC REVIEW T A B L E 1 . Summary Statistics: Key Variables Variable Mean Std. Dev. Min Max Firm-country export value (million US $) 0.75 11.9 0.1 7,440 Nb of products exported (firm-country) 4.66 13.95 1 1329 RER volatility 0.02 0.02 .01 0.44 GDP (trillion US $) 1.54 2.98 0.1 13.7 Price index (effective exchange rate) 234.4 309.8 0.003 3549 Country-sector imports (billion US $) 14.0 28.8 0.1 271 External dependence 0.37 0.26 2 0.45 1.14 Intangibility 0.08 0.05 0 0.43 R&D 0.02 0.02 0 0.09 Financial development (total credit/GDP, %) 1.14 0.47 0.58 3.31 Export start dummy (firm-country) 0.226 0.42 0 1 Notes: The summary statistics are computed on the 3,731,351 firm-country-year observations that make up our final regression sample used in Table 3 to study the intensive margin. The only ex- ception is the statistics for the start dummy which are computed for the sample (8,801,335 observa- tions) used in Table 7. Source: Authors’ calculations based on Chinese customs and other data described in the text. T A B L E 2 . Descriptive Statistics for Financial Vulnerability Indicators Distribution External dependence Intangibility R&D 5% 0.01 0.01 0.004 10% 0.061 0.019 0.009 50% 0.326 0.074 0.019 90% 0.770 0.149 0.065 95% 0.838 0.160 0.070 Notes: The summary statistics are computed on the 3,731,351 firm-country-year observations that make up our final regression sample used in Table 3 to study the intensive margin. Source: Authors’ calculations based on Chinese customs and other data described in the text. impact of both local endowments and sector-specific characteristics (including fi- nancial vulnerability). Our conditioning set Z consists of destination-year specif- ic variables. In standard models of international trade, exports depend on the destination country’s market size and price index. We use country j’s GDP13 and effective real exchange rate.14 We also account for country j’s demand for goods from the main sector of the firms (identified as the one with the highest export share over the period). Following Berman et al. (2012), we use the log of the total import value for the country-sector in the year taken from BACI.15 13. GDP data come from the World Development Indicators. 14. The effective exchange rate is computed using CEPII and IFS data as an average of the real exchange rates of destination country j toward all of its trade partners, weighted by the share of each trade partner in country j’s total imports. 15. This data set, which is constructed using COMTRADE original data, provides bilateral trade flows at the product level (Gaulier and Zignago 2010). BACI is downloadable from http://www.cepii.fr/ anglaisgraph/bdd/baci.htm (last accessed: October 10, 2013). Trade flows are aggregated to the 27 three-digit ISIC sectors for which our indicators of the financial vulnerability of a sector are available. ´ ricourt and Poncet He 561 We first focus on the unconditional effect of volatility on export performance, i.e., on a benchmark specification with b, g, d, and t all restricted to 0. Consistent with prediction 1 from section 2, we expect a to be negative. In a second step, we condition the impact of volatility on the financial vulnerability of a firm by introducing an interaction term between these two variables: prediction 2 leads us to expect b to be negative. Note that the financial vulnerability vari- able alone does not appear because it is captured by the firm-country fixed effects. We further modify our empirical specification in a third and final step to allow a and b to vary depending on the development of the local financial sector. In this case, our main parameters of interest are those on the double interaction between RER volatility and financial development ( g) and on the triple interac- tion between RER volatility, financial vulnerability, and financial development (d). Following prediction 3, both parameters should be positive. Note also that the relative size and significance of a in comparison with the other parameters provides interesting insight into the respective roles of the aforementioned aggregation bias and heterogeneity in terms of financial develop- ment. More precisely, a non-significant a compared to b, g, and d suggests that the impact of exchange rate volatility on exports is not unconditional but emerges mainly because of the credit constraints of firms and low financial development. Finally, Moulton (1990) shows that regressions with more aggregate indica- tors on the right-hand side could induce a downward bias in the estimation of standard errors. All regressions are thus clustered at the province level16 using the Froot (1989) correction. R ES ULT S We study the joint effects of exchange rate volatility and financial constraints on both margins of trade separately: the size of exports by firm (the intensive margin) and the decision to begin exporting (the extensive margin).17 Intensive Margin Table 3 presents the estimations of the impact of RER volatility on the value ex- ported by firms. Column (1) reports the estimates of a specification based only on the two proxies for the destination countries’ market size and price index (which are significant and display the expected positive signs), and column (2) presents the unconditional relationship between RER volatility and export performance. Column (3) includes an alternative measure of market size, the country-sector imports, which appears positive and significant. The following columns add a 16. Because the province level is the most aggregated one (i.e., with the smallest number of clusters) in our case, it gives the most possible conservative standard errors and appears to be the safest choice. Note that our results are mostly unchanged when standard errors are clustered at the destination country level. 17. Robustness checks relying on alternative definitions for both margins are presented in the Appendix. 562 THE WORLD BANK ECONOMIC REVIEW T A B L E 3 . Intensive Margin, Exchange Rate Volatility and Financial Constraints Dependent variable Log export value (firm-destination-year) (1) (2) (3) (4) (5) (6) Financial indicator Ext dep Intang. R&D RER volatility (a) 2 0.439a 2 0.305a 0.402 0.123 0.153 (0.119) (0.106) (0.246) (0.183) (0.172) Ln country GDP 0.321a 0.312a 0.061 0.061 0.060 0.061 (0.068) (0.066) (0.068) (0.068) (0.068) (0.068) Ln country price index 0.027c 0.027c 0.050c 0.050c 0.050c 0.050c (0.014) (0.014) (0.014) (0.014) (0.014) (0.014) Ln country-sector imports 0.357a 0.356a 0.357a 0.356a (0.014) (0.014) (0.014) (0.014) RER volatility  Fin. vulnerability (b) 2 1.900a 2 5.686a 2 18.574a (0.478) (1.466) (4.379) Fixed effects Firm-country and year R-squared 0.03 0.03 0.03 0.03 0.03 0.03 Observations 3,731,351 Nb of firm-country pairs 1,128,873 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. Standard errors are clustered at the province level; a, b, and c, respectively, denote significance at the 1 percent, 5 percent, and 10 percent levels. Source: Authors’ calculations based on Chinese customs and other data described in the text. variable interacting RER volatility with a measure of firm-level financial depen- dence. Columns (2) and (3) show that exchange rate volatility appears to be nega- tively associated with export performance (i.e., the a parameter of Equation 2 is significant and negative). We check the robustness of this negative relationship when volatility is com- puted based on the yearly standard deviation of monthly log differences of various definitions of the exchange rate. The results reported in Table S.1 in the online appendix (available at http://wber.oxfordjournals.org/) confirm that the unconditional impact of exchange rate volatility on the intensive margin is nega- tive and significant (and quantitatively very close to our main definition of vola- tility, i.e., the standard deviation of the RER defined as the nominal exchange rate of the yuan with respect to the partner’s currency multiplied by the partner’s CPI) whether we consider a “full” RER in which the Chinese CPI is introduced as the denominator (columns (1) and (2)), the nominal exchange rate (columns (3) and (4)), the log-level of RER (columns (5) and (6)), or the HP detrended version of our benchmark RER (columns (7) and (8)).18 The subsequent results suggest that the magnitude of this effect depends on the extent of the financial constraints. Columns (4) to (6) of Table 3 show that the interaction with financial vulnerability enters with a negative and significant 18. These results are also robust in specifications based on variables measured using two-year windows. This additional set of results is available upon request. ´ ricourt and Poncet He 563 coefficient regardless of the indicator of financial dependence used: external de- pendence in column (4), asset intangibility in column (5), and R&D intensity in column (6). Across our three indicators, we consistently observe that the negative impact of RER volatility on exports grows with financial vulnerability. These results suggest that the negative impact of exchange rate volatility on export per- formance is not unconditional but is instead proportional to the degree of finan- cial vulnerability. These results are robust to various robustness checks. First, Table S.1 confirms an export-deterring effect of RER volatility that rises with financial vulnerability regardless of the definition of volatility that is used. Second, in the unreported results (available upon request), we check that the estimates of Equation 2 are robust to the inclusion of sector-year fixed effects, where the sector corresponds to the firm’s main sector of activity, identified as the one with the greatest export share over the period. This allows us to verify that although a large component of the variance in exchange rate volatility may be year-specific, our results do not solely reflect the sector-specific trends. The results are qualitatively identical.19 To illustrate these results, we can compare the decrease in the export perfor- mance due to RER volatility for firms at the 10th and 90th percentiles of the dis- tribution of financial vulnerability. Table 2 above reports summary statistics on the distribution of the three indicators of financial vulnerability. Using coeffi- cients from column (4) in Table 3 for the intensive margin, all things being equal, the effect of a 10 percent increase in RER volatility on export value is 0.1  a þ 0.1  b  FinVuln. Hence, our results (a ¼ 0.402 and b ¼ 2 1.90) suggest that the export value is reduced by 10.6 percent [0.1  0.402 2 1.9  0.1  0.770] at the 90th percentile of financial dependence. At the 10th percentile, the export value seems slightly increased, by 2.9 percent [0.1  0.402 2 1.9  0.1  0.061]. The key point is that the differential impact between the 90th and 10th percentiles of financial dependence is equal to 2 13.5 percent [0.1  ( 2 1.9)  (0.77 2 0.061)], which is strongly negative and significant. In Table 4, we check the robustness of our results to the inclusion of additional controls. Financial vulnerability is measured using external dependence. The first five columns check that our measured impact of RER volatility does not simply capture the impact of the RER level. In column (1), the explanatory variables are restricted to RER volatility and RER level. Because we rely on an indirect quota- tion, an increase in the level of the exchange rate, implying a depreciation, is ex- pected to have a positive impact on export performance. This intuition is confirmed: RER volatility and RER level enter with reverse signs, negative and positive, respectively, which are significant in both cases. In column (2), the posi- tive impact of the level of RER becomes insignificant once we adopt the bench- mark specification from column (4) in Table 3 and add the macroeconomic variables for the destination country (GDP, import price, demand). In column 19. In other unreported checks, we show that our results hold when adding interactions between year dummies and our proxy for financial vulnerability. 564 T A B L E 4 . Intensive Margin: Including RER in Level and Income Volatility Dependent variable Log export value (firm-destination-year) THE WORLD BANK ECONOMIC REVIEW Financial indicator External dependence (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) RER volatility (a) 2 0.636a 2 0.308a 2 0.036 0.399 0.223 2 0.548a 2 0.238c 0.272 0.520c 0.504c (0.151) (0.103) (0.255) (0.243) (0.217) (0.184) (0.125) (0.314) (0.282) (0.278) Ln country GDP 0.054 0.054 0.057 0.064 0.063 0.063 (0.075) (0.075) (0.075) (0.077) (0.077) (0.077) Ln country price index 0.048a 0.048a 0.048a 0.037b 0.037b 0.037b (0.013) (0.013) (0.013) (0.017) (0.017) (0.017) Ln country-sector imports 0.357a 0.356a 0.355a 0.407a 0.406a 0.406a (0.014) (0.014) (0.014) (0.017) (0.017) (0.017) RER Volatility  Fin. vulnerability (b) 2 1.612a 2 1.901a 2 1.427a 2 2.187a 2 2.025a 2 1.981a (0.391) (0.479) (0.400) (0.494) (0.537) (0.523) Ln RER  Fin. vulnerability 0.477a 0.465a (0.137) (0.141) Ln RER 0.316a 0.013 0.142c 0.014 2 0.158a (0.037) (0.020) (0.081) (0.020) (0.046) GDP volatility 2 2.453a 2 1.721a 2 2.004a 2 1.721a 2 1.338a (0.226) (0.234) (0.303) (0.234) (0.316) GDP Volatility  Fin. vulnerability 2 1.232b 2 1.057c (0.532) (0.565) Fixed effects Firm-country and year R-squared 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 Observations 3,731,351 3,158,760 Number of firm-country pairs 1,128,873 952,132 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. Standard errors are clustered at the province level; a, b, and c, respectively, denote significance at the 1 percent, 5 percent, and 10 percent levels. Source: Authors’ calculations based on Chinese customs and other data described in the text. ´ ricourt and Poncet He 565 (3), we add the interactive terms between financial vulnerability and both the level of RER and the volatility of RER. The former interactive term attracts a positive and significant coefficient. The reasoning is symmetrical to the one con- cerning RER volatility: financially constrained firms disproportionately take ad- vantage of a depreciating exchange rate. However, this is contrary to the findings of Desai et al. (2008), which suggest that lower financial constraints increase firms’ ability to expand activity during currency crises. Our results in columns (3) to (5) confirm that including the level of RER does not affect our main result of a negative b. In the remaining columns (6 to 10), we verify that RER volatility does not act as a mere proxy for economic fluctuations. We consider the repercussions of the volatility of the partner’s GDP, which is computed as the standard deviation of year-to-year changes in quarterly GDP taken from the IFS. As argued by Baum et al. (2004) and Grier and Smallwood (2007), foreign income uncertainty may equally matter for trade. Consistent with their findings, GDP volatility enters with a negative sign: income volatility has a significant deterrent effect on the value exported. However, this inclusion does not affect our benchmark result of a negative impact of RER volatility that grows with financial vulnerability. In columns (8) and (10), we further include the interactive term between GDP volatility and financial dependence. In column (10), it is significant only at the 10 percent level (the negative impact of income volatility seems to vary, but only weakly, with the level of credit constraints for a firm), whereas our main finding on the impact of RER volatility is not altered; the interaction between RER volatility and financial dependence remains nega- tive and significant. Table 5 verifies that our results are robust to various changes in the sample. Again, financial vulnerability is measured using external dependence. Column (1) restricts the sample to firms exporting to more than one country, whereas column (2) concentrates on multi-product firms. The point estimates are virtually unaffected. In column (3), we exclude observations for Macao and Hong Kong because we are concerned that RER volatility may have different implications in the case of these two “Greater China” territories than for other international partners. Once again, the negative coefficient on the interactive term between RER volatility and financial vulnerability remains. In columns (4) to (7), we in- vestigate whether our results vary across firm-level productivity, proxied as the number of products or the number of product-country pairs that a firm exports. We investigate this by regressing our main specification on subsamples divided around the median of our productivity proxies. Our main findings remain un- changed in all specifications, indicating that they apply to both low- and high- productivity firms. We now ask whether recent developments in China’s financial system have helped reduce the export losses from real exchange rate uncertainty. As previous- ly mentioned, Aghion et al. (2009) suggest that the effect of RER volatility depends critically on the level of local financial development. We modify our 566 THE WORLD BANK ECONOMIC REVIEW T A B L E 5 . Intensive Margin: Controlling for Various Subsamples Dependent variable Log export value (firm-destination-year) Financial indicator External dependence (1) (2) (3) (4) (5) (6) (7) Country Product No HK High Nb Low Nb High Nb Low Nb Nb . 1 Nb . 1 or Macao products products prod-dest prod-dest RER volatility (a) 0.384 0.359 0.435c 0.799c 0.179 0.507 0.391 (0.244) (0.270) (0.228) (0.394) (0.204) (0.336) (0.250) Ln country GDP 0.051 0.101c 0.031 0.170b 0.004 0.201a 0.057 (0.064) (0.058) (0.079) (0.066) (0.085) (0.071) (0.068) Ln country price index 0.048a 0.035b 0.032b 0.040b 0.056a 0.043b 0.048a (0.015) (0.014) (0.013) (0.017) (0.014) (0.018) (0.015) Ln country-sector imports 0.355a 0.333a 0.342a 0.312a 0.409a 0.313a 0.355a (0.013) (0.013) (0.015) (0.013) (0.020) (0.012) (0.014) RER volatility  2 1.866a 2 1.722a 2 1.921a 2 3.314a 2 0.968b 2 2.545a 2 1.892a Fin. vulnerability (b) (0.467) (0.602) (0.466) (0.927) (0.382) (0.722) (0.478) Fixed effects Firm-country and year R-squared 0.03 0.04 0.03 0.02 0.04 0.03 0.03 Observations 3,659,052 2,019,033 3,472,215 1,836,309 1,895,042 1,862,175 3,719,937 Number of firm-country pairs 1,106,403 781,138 1,059,036 532,927 595,946 527,300 1,128,139 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. Standard errors are clustered at the province level; a, b, and c, respectively, denote significance at the 1 percent, 5 percent, and 10 percent levels. Source: Authors’ calculations based on Chinese customs and other data described in the text. ´ ricourt and Poncet He 567 empirical specification to allow b in Equation 2 to vary depending on the devel- opment of the local financial sector. Our main parameter of interest is that of the triple interaction between RER volatility, financial vulnerability, and financial development (d in Equation 2). We first split the provinces into two groups depending on whether their financial development is below or above the national median or the national mean in 2000 (the initial year of our sample). The corresponding results are reported in columns (1) and (2) of Table 6. The positive coefficient attracted by the T A B L E 6 . Intensive Margin: The Role of Financial Development Dependent variable Log export value (firm-destination-year) Financial indicator External dependence (1) (2) (3) (4) (5) RER volatility (a) 0.455c 0.467c 0.312 0.292 0.299 (0.259) (0.272) (0.248) (0.238) (0.228) Ln country GDP 0.059 0.059 0.057 0.059 0.049 (0.069) (0.069) (0.068) (0.068) (0.069) Ln country price index 0.050a 0.050a 0.050a 0.049a 0.050a (0.014) (0.014) (0.014) (0.014) (0.015) Ln country-sector imports 0.357a 0.357a 0.356a 0.354a 0.358a (0.014) (0.014) (0.014) (0.014) (0.013) RER Volatility  Fin. vulnerability (b) 2 2.824a 2 2.875a 2 1.718a 2 1.622a 2 1.614a (0.433) (0.462) (0.611) (0.475) (0.462) RER Volatility  Financial vulnerability  2.062a High Fin. Devt (above median) (0.589) RER Volatility  Financial vulnerability  2.177a High Fin. Devt (above mean) (0.568) RER Volatility  High Fin. Devt (above median) 2 0.015 (0.271) RER Volatility  High Fin. Devt (above mean) 2 0.047 (0.260) RER Volatility  Financial vulnerability  7.069a 3.034b 2.878b Fin. Devt (d) (1.981) (1.234) (1.160) RER Volatility  Fin. Devt ( g) 2 2.170a 2 0.666 2 0.770 (0.658) (0.457) (0.572) Financial vulnerability  Fin. Devt 0.263c 0.260c (0.146) (0.138) Financial Development 0.087 2 0.016 (0.061) (0.056) Province-year fixed effects no no no no yes Fixed effects Firm-country and year R-squared 0.03 0.03 0.03 0.03 0.03 Observations 3,731,351 Number of firm-country pairs 1,128,873 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. Standard errors are clustered at the province level; a, b, and c, respectively, denote significance at the 1 percent, 5 percent, and 10 percent levels. Source: Authors’ calculations based on Chinese customs and other data described in the text. 568 THE WORLD BANK ECONOMIC REVIEW interactive terms between RER volatility and financial vulnerability in the case of provinces thatare highly financially developed indicates that the negative effect of RER volatility on the export value of firms is less present when credit is abun- dant. In the following columns, we use the time-varying proxy for financial de- velopment and interact it directly with RER volatility and financial dependence. The interaction between local financial development and financial dependence is also included. We also add the level of financial development and its interaction with RER volatility (the g parameter) in columns (4) and (5). In column (5), we include province-year fixed effects to account for the time-varying characteristics of the local economy (including financial development, which drops as a conse- quence). In this way, any variable correlated with financial development that could impact the export performance of firms are captured by these fixed effects but should not affect our coefficients of interest (b, g, and d), unless its effect runs through a financial channel. The results confirm our previously measured negative interaction between RER volatility and financial vulnerability but suggest that the losses are mitigated by high local financial development. In all columns, we find that financial devel- opment dampens the negative impact of real exchange rate volatility on exports and that the relaxation effect increases with the level of sectoral financial depen- dence of firms. The triple interaction among RER, financial dependence, and fi- nancial development is positive and significant. In other words, the positive offsetting effect of financial development on RER volatility is magnified by the fi- nancial constraints for firms. This result is in line with Aghion et al.’s (2009) ob- servation that financial development reduces the magnitude of performance deterioration induced by RER volatility. Conversely, there is no evidence of an unconditional effect on financial constraints; the interaction between RER vola- tility and financial development ( g) is insignificant. As an additional check, we verify in Table S.3 in the online appendix that our main results hold when measuring the intensive margin based on the average export value for the firm-country pair, computed as the ratio of total export value over the number of products exported (expressed in natural logarithms). All of our key results remain: the negative impact of RER volatility on the inten- sive margin increases with the credit constraints for firms regardless of the defini- tion of financial vulnerability that is used (columns (2) to (4)). Finally, the relaxing effect of financial development also persists (columns (5) to (8)), with an even stronger significance compared to our preferred specification. Extensive Margin In this section, we assess the joint effect of RER volatility and financial con- straints on the extensive margin of trade at the firm-country level (i.e. how they affect entry decisions). Columns (1) to (6) of Table 7 replicate Table 3. The ex- plained variable is now the decision for a firm to begin exporting to market j. It is constructed as a change of export status at the firm-country level; it takes the value 1 when a firm exports to country j in year t but did not in year t 2 1. Once T A B L E 7 . Extensive Margin, Exchange Rate Volatility, and Financial Constraints Dependent variable Pr(XF F i,j,t . 0j Xi,j,t- 1 ¼ 0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Financial indicator Ext dep Intang. R&D External dependence RER volatility (a) 2 0.864a 2 0.735a 0.094 0.019 2 0.454a 2 0.779a 2 0.197 2 0.702a 0.024 (0.099) (0.080) (0.226) (0.190) (0.153) (0.079) (0.209) (0.130) (0.230) Ln country GDP 0.072 0.051 2 0.219a 2 0.218a 2 0.220a 2 0.219a 2 0.267a 2 0.237a 2 0.252a 2 0.252a (0.055) (0.055) (0.057) (0.057) (0.057) (0.057) (0.070) (0.072) (0.072) (0.072) Ln country price index 0.099a 0.102a 0.125a 0.124a 0.125a 0.124a 0.109a 0.108a 0.077a 0.077a (0.020) (0.020) (0.021) (0.021) (0.021) (0.021) (0.019) (0.019) (0.029) (0.029) Ln country-sector imports 0.379a 0.378a 0.379a 0.379a 0.379a 0.372a 0.395a 0.394a (0.033) (0.033) (0.033) (0.033) (0.033) (0.033) (0.053) (0.053) RER volatility  Fin. vulnerability (b) 2 2.233a 2 9.852a 2 11.731a 2 1.462a 2 1.923a (0.431) (1.973) (3.612) (0.374) (0.370) Ln RER  Fin. vulnerability 1.252a (0.231) Ln RER 0.101a 2 0.377a (0.036) (0.100) GDP volatility 0.076 0.950c (0.193) (0.574) GDP volatility  Fin. vulnerability 2 2.433b He (1.178) ´ ricourt and Poncet Fixed effects Firm-country and year Pseudo-R-squared 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 0.20 Observations 8,801,335 8,801,335 6,996,782 Nb of firm-country pairs 1,867,840 1,867,840 1,492,028 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. Standard errors are clustered at province level; a, b, and c, respectively, denote significance at the 1 percent, 5 percent, and 10 percent levels. Source: Authors’ calculations based on Chinese customs and other data described in the text. 569 570 THE WORLD BANK ECONOMIC REVIEW again, the unconditional impact of RER volatility (a parameter) appears negative and significant (columns (2) and (3)), but adding interactive terms to each of our measures of firm-level financial dependence shows that the magnitude of this effect is usually conditioned by the extent of the financial constraints (columns (4) to (6)): the b parameter appears negative and highly significant, and a becomes insignificant except when the financial dependence indicator is the share of R&D spending in total sales. Quantitatively, the impact of an uncondi- tional 10 percent increase in exchange rate volatility (the a parameter in column (3)) decreases the probability of beginning to export by 1.29 percent.20 Similarly, if we distinguish between firms at the 10th and 90th percentiles of the distribution of financial vulnerability, we can compare the decrease in the extensive margin due to RER volatility conditioning on financial vulnerability. Using coefficients a ¼ 0.094 and b ¼ 2 2.233 from column (4), this means that, all things being equal, the negative effect of an additional 10 percent in RER volatility on the probability of entering is 2 2.8 percent [(0.094  0.1)  (0.226)  (1 2 0.226) þ (0.1  ( 2 2.233)  0.77)  0.226  (1 2 0.226)] at the 90th percentile of finan- cial dependence. The effect is practically 0 [ 2 0.07 percent ¼ (0.094  0.1)  (0.226)  (1 2 0.226) þ (0.1  ( 2 2.233)  0.061)  0.226  (1 2 0.226)] at the 10th percentile. The net differential effect on the 90th percentile relative to the 10th percentile is thus equal to 2 2.8 percent. As before, in Table S.2 in the online appendix, we check the robustness of these results using the yearly standard deviation of monthly log differences from various definitions of the exchange rate (with the RER deflated by the Chinese CPI in columns (1) and (2), the NER in columns (3) and (4), and the HP-filtered RER in columns (7) and (8)). In columns (5) and (6), we verify that similar qualitative results are obtained when volatility is computed as the yearly standard deviations of the log-level of RER. In unreported additional checks, we show that our results also hold when adding interactions between year dummies and our proxies for financial vulnera- bility.21 Overall, the negative impact of RER volatility on the probability of be- ginning to export is magnified by financial vulnerability. In columns ((7) to (10)) of Table 7, as before, we check the robustness of our results to the inclusion of additional macro controls, namely the log of RER and GDP volatility. The RER level enters positively and significantly (column (7)), and its interaction with 20. This figure is obtained from the derivative of the choice probabilities (Train 2003). The change in the probability that a firm F will choose alternative X (begin exporting) given a change in an observed factor ZF,X entering the representative utility of that alternative (and holding the representative utility of other alternatives (no exporting) constant) is bZ  PF,X(1 – PF,X), with PF,X being the average probability that firm i will choose alternative X (begin exporting). Based on an average probability to begin exporting of 22.6 percent, our estimates suggest that the derivative of beginning to export with respect to an additional 10 percent in RER volatility is 2 1.29 percent ¼ 0.1  – 0.735  0.226  (1 – 0.226). 21. We were not able to implement regressions using sector-year dummies to control more systematically for sector-specific trends because the latter are too numerous to allow the maximization of the log-likelihood function. ´ ricourt and Poncet He 571 financial vulnerability is also positive and significant (column (8)); financially constrained firms disproportionately take advantage of a depreciating exchange rate to enter the export market. In columns (9) and (10), GDP volatility fails to enter significantly, but its interaction with financial dependence is negative and significant; financially constrained firms are more harmed by the instability of foreign demand. In any case, these additional estimates do not affect our bench- mark result of a negative impact of RER volatility that grows with financial vul- nerability. Table 8 checks the robustness of these results across various subsamples. Financial vulnerability continues to be measured using external dependence. The results are unchanged for multi-destination (column (1)) and multi-product (column(2)) firms and when the observations for Macao and Hong Kong are ex- cluded (column (3)): the b parameter remains negative and significant, and entry into the export market is still disproportionately more harmed by exchange rate volatility in the case of financially constrained firms. This result also holds when we divide the sample by the median of our proxies for firm-level productivity, the number of products exported (columns (4) and (5)), or the number of product-destinations by firm (columns (6) and (7)). Interestingly, the uncondi- tional impact of RER volatility on entry (coefficient a) also remains negative and significant for firms with a low number of products or a low number of product- destinations. The probability that low-diversified firms will begin exporting is also harmed by RER volatility, even for zero financial vulnerability. We complete this overview by examining the impact of local financial develop- ment heterogeneity on these results. Once again, we measure local financial de- velopment as the share of total credit over GDP in the province, and we perform estimations replicating the ones presented in Table 6.22 We find that the triple in- teraction among exchange rate volatility, financial dependence, and financial de- velopment (the d parameter) is positive and significant in most specifications, whether we consider groups above the national mean/median of financial devel- opment in 2000 (columns (1) and (2)) or use the time-varying proxy for financial development (column (3)). The entry into export markets of financially con- strained firms is less hampered by RER volatility when financial development is high. However, in column (4), the significance switches from the d to the g pa- rameter. Financial development still reduces the negative impact of RER volatili- ty but does so independently of the level of financial constraints for firms. Overall, the evidence seems less strong than for the intensive margin, but the pre- sumption that financial development reduces the magnitude of performance deterioration induced by RER volatility remains, along the lines of Aghion et al. (2009). We check how our results behave when considering the export status at the firm-country level instead of the decision to begin exporting to understand the 22. However, we cannot provide estimations including province-year fixed effects; the maximization of the log-likelihood function proved to be impossible. 572 THE WORLD BANK ECONOMIC REVIEW T A B L E 8 . Extensive Margin: Controlling for Various Subsamples Dependent variable Pr(XF F i,j,t . 0j Xi,j,t21 ¼ 0) Financial indicator External dependence (1) (2) (3) (4) (5) (6) (7) Country Product No HK High Nb Low Nb High Nb Low Nb Nb . 1 Nb . 1 or Macao products products prod-dest prod-dest RER volatility (a) 2 0.295 2 0.067 2 0.274 2 0.145 2 0.616b 2 0.137 2 0.570b (0.198) (0.317) (0.194) (0.287) (0.278) (0.295) (0.226) Ln country GDP 0.297a 0.308a 0.305a 0.352a 0.475a 0.444a 0.413a (0.052) (0.049) (0.077) (0.070) (0.053) (0.076) (0.040) Ln country price index 0.064a 0.063a 0.056a 0.054a 0.020 0.043a 0.028 (0.014) (0.016) (0.012) (0.016) (0.019) (0.012) (0.020) Ln country-sector imports 0.417a 0.356a 0.403a 0.335a 0.491a 0.384a 0.451a (0.036) (0.041) (0.036) (0.040) (0.026) (0.039) (0.033) RER volatility  2 1.622a 2 2.086b 2 1.607a 2 1.904a 2 1.067a 2 2.041a 2 1.167a Fin. vulnerability (b) (0.378) (0.814) (0.384) (0.594) (0.367) (0.578) (0.410) Fixed effects Firm-country and year Pseudo-R-squared 0.10 0.11 0.10 0.07 0.13 0.07 0.12 Observations 4,617,726 1,684,176 4,496,413 2,276,599 2,341,127 2,304,527 2,313,199 Number of firm-country pairs 1,193,670 489,613 1,159,777 559,590 634,080 546,015 647,655 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. Standard errors are clustered at province level; a, b, and c, respectively, denote significance at the 1 percent, 5 percent, and 10 percent levels. Source: Authors’ calculations based on Chinese customs and other data described in the text. ´ ricourt and Poncet He 573 extensive margin. Our dependent variable is therefore defined as a dummy vari- able taking the value 1 when a firm exports to country j at time t. The results, which are still based on a conditional logit specification with firm-country fixed effects, are reported in Table S.4 in the online appendix. These results are quali- tatively identical to those presented in Tables 7 and 9. We find some evidence of an unconditional negative impact of RER volatility (column (1)). This negative impact is again magnified by firm-level financial dependence (columns (2) to (4)). Finally, there is still some evidence that financial development produces a signifi- cant relaxation effect in this context (columns (5) to (8)). T A B L E 9 . Extensive Margin: The Role of Financial Development Dependent variable Pr(XF F i,j,t . 0j Xi,j,t21 ¼ 0) Financial indicator External dependence (1) (2) (3) (4) RER volatility (a) 2 0.506 2 0.577 0.029 2 0.067 (0.466) (0.492) (0.232) (0.215) Ln country GDP 2 0.225a 2 0.226a 2 0.222a 2 0.220a (0.052) (0.051) (0.053) (0.053) Ln country price index 0.122a 0.122a 0.124a 0.124a (0.021) (0.021) (0.021) (0.021) Ln country-sector imports 0.380a 0.380a 0.379a 0.375a (0.032) (0.032) (0.033) (0.032) RER Volatility  Fin. vulnerability (b) 2 4.762a 2 4.885a 2 2.137a 2 1.777a (1.268) (1.277) (0.724) (0.360) RER Volatility  Financial vulnerability  4.385b High Fin. Devt (above median) (2.060) RER Volatility  Financial vulnerability  4.487b High Fin. Devt (above mean) (2.025) RER Volatility  1.556 High Fin. Devt (above median) (1.091) RER Volatility  1.633 High Fin. Devt (above mean) (1.087) RER Volatility  Financial vulnerability  6.503b 2 0.072 Fin. Devt (d) (3.000) (1.679) RER Volatility  Fin. Devt ( g) 2 0.866 1.552c (0.981) (0.813) Financial vulnerability  Fin. Devt 0.590 (0.383) Financial Development 0.358 0.127 (0.230) (0.186) Fixed effects Firm-country and year Pseudo-R-squared 0.20 0.20 0.20 0.20 Observations 8,801,335 Number of firm-country pairs 1,867,840 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. Standard errors are clustered at province level; a, b, and c, respectively, denote significance at the 1 percent, 5 percent, and 10 percent levels. Source: Authors’ calculations based on Chinese customs and other data described in the text. 574 THE WORLD BANK ECONOMIC REVIEW Finally, Table S.5 in the Appendix reports the results of an alternative definition of the extensive margin, namely the (log) number of HS6 products shipped to a country, in the spirit of Manova et al. (2011). We still find a negative impact of RER volatility on export performance, which is magnified for financially vulnerable firms. The evidence is much weaker regarding the relaxing impact of financial de- velopment; the d coefficient is correctly signed (positive) but fails to be significant. Additional Robustness Tests and General Discussion Our empirical work so far has exploited the variation in export performance over time and across destinations for firms of different sectors. Because a great propor- tion of the firms in our sample export goods to more than one ISIC three-digit sector, in what follows, we also use the variation across sectors within firms. Our proxy for the intensive margin becomes the (log) export value of the firm for a given sector/country pair in a year. The extensive margin is defined as the (log) number of HS6 products for a given sector/country pair in a year. Otherwise iden- tical to Equation 2, these regressions include firm-sector-country fixed effects, so the coefficients are identified from the time-series variation within firm-sector- country triplets over time. Therefore, our estimates consider the way in which firms choose to allocate their limited financial resources in the various sector- country export markets in which they operate over time.23 The results are reported in Tables S.6 and S.7 in the online appendix for the intensive and extensive margins, respectively. In both cases, exchange rate volatility impacts export perfor- mance negatively, disproportionately more for financially vulnerable firms. There is still a relaxing impact of financial development for this specific definition of the intensive margin. However, no evidence of such an effect of financial development can be identified for the range of products exported. Defining the margins of trade at the firm-sector-country level also allows us to check how our results behave when we define financial vulnerability relying on pure sectoral indicators. That is, a firm’s financial dependence is identified without using any kind of weighting schemes based on firm-level exports; the measure of financial vulnerability is therefore a pure sector characteristic and is hence exogenous to firm-level developments. The results are reported in Table S.8 in the online appendix. The results are generally qualitatively similar to the ones presented above, with a somewhat weaker significance (especially for the intangibility indicator). This not surprising because we are considering a quite aggregated sectoral level with limited variance compared to our firm-level variables. Overall, however, the reading of Tables S.6 and S.7 is not fundamen- tally altered by this modification. In additional unreported checks (available upon request), we assess the robust- ness of our results to the exclusion of the US as an export destination in the 23. In unreported results that are available upon request, we verify that our main message holds when including both firm-sector-year fixed effects and country fixed effects– that is, when focusing on how firms allocate resources across countries for a given sector-year. ´ ricourt and Poncet He 575 sample. This allows us to ensure that our results are not biased by the presence of the country toward which volatility is reduced by construction during most of the period considered. Similarly, we perform additional estimates excluding the years 2005 and 2006 to verify that the switch from pegging the US dollar only to a basket of several currencies in July 2005 does not impact our results. In both exercises, our results remain qualitatively identical. Moreover, we verify that our results hold for exporters irrespective of their ownership structure (whether domestic or foreign) and irrespective of the export regime (whether ordinary or processing). We also perform estimations on a sub- sample excluding intermediary firms. Our measure of financial constraints may be less relevant for those firms that do not produce the goods they sell because it is computed from information based on production technology. We follow Ahn et al.’s (2011) approach to identify these firms based on Chinese characters in the name of the firm that mean “importer”, “exporter”, and/or “trading” in English.24 We also estimate specifications adding firm-country level imports from the countries where the firm is also exporting. In all of these checks, once again, the negative impact of exchange rate volatility appears magnified for fi- nancially vulnerable firms and relaxed by a high level of financial development. Finally, we verify that the differentiated impact of RER volatility depending on financial development does not simply reflect a correlation between financial development and trade costs. It may be that provinces with a more developed fi- nancial system also benefit from easier and cheaper international access. In this case, our results may identify an uncertainty related to distance. We replicate our benchmark result by examining the double interaction between RER volatility and financial dependence (column (4) of Tables 3 and 7) and the triple interac- tion depending on financial development (columns (3) and (4) of Tables 6 and 9) when adding interactive terms with three proxies for the geographical trade ad- vantages of coastal location, western location, and distance to partner country,25 respectively. Our findings of a trade-deterring effect of RER volatility that is pro- portional to financial constraints and that is relaxed by financial development appear fully robust to these controls for geography. Together, Tables 3 to 9 shed new light on the joint role of exchange rate vola- tility and financial constraints on exporting behavior. Our results suggest that ex- change rate volatility negatively impacts both the intensive (total value exported by firm and destination) and extensive (decision of a firm to begin exporting to destination) margin but that this impact is mainly conditioned on the extent of firm-level financial constraints. Our findings also support the idea that higher fi- nancial development offsets this negative impact, both for the intensive margin and the probability of entering a new export market,but not for the range of 24. In pinyin (Romanized Chinese), these phrases are “jin4chu1kou3”, “jing1mao4”, “mao4yi4”, “ke1mao4”, and “wai4jing1”. 25. We use the GeoDist dataset (Mayer and Zignago 2011), available at http://www.cepii.fr/ francgraph/bdd/distances.htm (last accessed: October 10, 2013). 576 THE WORLD BANK ECONOMIC REVIEW products exported. Overall, these results provide insight into the main sources for the apparent lack of macro impact of exchange rate volatility: the level of fi- nancial constraints and financial development clearly dominate the aggregation bias hypothesis because b and d are consistently more significant than a. By doing so, we provide micro support to the macro literature that points to finan- cial development as a key determinant of the impact of RER volatility on real outcomes. CONCLUSION This paper relies on a firm-level database covering exporters from China to study how export performance is affected by real exchange rate volatility. Our empiri- cal strategy investigates how RER volatility affects the extensive and intensive margins of firm-level exports to their international partners. The features of the Chinese exchange rate system that are common to all exporters and all destina- tion markets, such as limited convertibility and misalignment, are controlled for through fixed effects. Our results suggest that even in the specific context of China’s restricted and misaligned ER regime, volatility is a significant barrier to Chinese exporters’ performance. We find a trade-deterring effect of RER volatili- ty, the magnitude of which depends mainly on the extent of financial constraints. Although firms tend to export less and to reduce their entry into destinations with higher exchange rate volatility, this negative effect is even stronger for finan- cially vulnerable firms. Furthermore, financial development appears to dampen this negative impact, especially on the intensive margin of export. These results suggest that the development of credit markets would help firms overcome the additional export (both variable and sunk) costs related to RER volatility. This finding could support the expansion of exports by firms, particu- larly to those destinations characterized by RER-related uncertainty. More gen- erally, our study emphasizes that emerging countries should be careful when relaxing their exchange rate regime. Hard-fixed pegs for developing countries are certainly not always a panacea, but moving to a fully floating regime without an adequate level of financial development could also prove to be very hazardous for trade performance. REFERENCES ` re, and K. Rogoff. 2009. “Exchange Rate Volatility and Productivity Aghion, P., P. Bacchetta, R. Rancie Growth: The Role of Financial Development.” Journal of Monetary Economics 56: 494– 513. Ahmed, S. 2009. “Are Chinese Exports Sensitive to Changes in the Exchange Rate?” International Finance Discussion Papers 987. Board of Governors of the Federal Reserve System, Washington, DC. Ahn, J., A. K. Khandelwal, and S.-J. Wei. 2011. “The Role of Intermediaries in Facilitating Trade.” Journal of International Economics 84: 73– 85. Aizenman, J., and N. Marion. 1999. “Volatility and Investment: Interpreting Evidence from Developing Countries.” Economica 66 (262): 157– 79. ´ ricourt and Poncet He 577 Baum, C. F., M. Caglayan, and N. Ozkan. 2004. “Nonlinear Effects of Exchange rate Volatility on the Volume of Bilateral Exports.” Journal of Applied Econometrics 19 (1): 1– 23. Benhima, K. 2012. “Exchange Rate Volatility and Productivity Growth: The Role of Liability Dollarization.” Open Economies Review 23(3): 501– 29. Berman, N., and J. He´ ricourt. 2010. “Financial Factors and the Margins of Trade: Evidence from Cross-Country Firm-Level Data.” Journal of Development Economics 93 (2): 206–17. Berman, N., P. Martin, and T. Mayer. 2012. “How do Different Exporters React to Exchange Rate Changes? Theory, Empirics and Aggregate Implications.” Quarterly Journal of Economics 127 (1): 437–92. Bernard, A. B., S. J. Redding, and P. K. Schott. 2011. “Multiproduct Firms and Trade Liberalization.” The Quarterly Journal of Economics 126 (3): 1271–318. Berthou, A., and L. Fontagne´ . 2013. “How do Multi-Product Exporters React to a Change in Trade Costs?” Scandinavian Journal of Economics 115 (2): 326– 53. Bloom, N., S. Bond, and J. V. Reenen. 2007. “Uncertainty and Investment Dynamics.” Review of Economic Studies 74 (2): 391–415. Broda, C., and J. Romalis. 2011. “Identifying the Relationship Between Trade and Exchange Rate Volatility.” In T. Ito, and A. K. Rose, eds., Commodity Prices and Markets, East Asia Seminar on Economics, 20. Chicago, IL: University of Chicago Press. Byrne, J. P., N. Fiess, and R. MacDonald. 2008. “US Trade and Exchange Rate Volatility: A Real Sectoral Bilateral Analysis.” Journal of Macroeconomics 30 (1): 238– 59. Caglayan, M., and F. Demir. 2012. “Firm Productivity, Exchange Rate Movements, Sources of Finance and Export Orientation.” MPRA Paper 37397. University Library of Munich, Munich, Germany. ´ n-Sa Carranza, L. J., J. M. Cayoa, and J. E. Galdo ´ nchez. 2003. “Exchange Rate Volatility and Economic Performance in Peru: A Firm Level Analysis.” Emerging Markets Review 4: 472–96. Cheung, Y.-W., M. D. Chinn, and X. W. Qian. 2012. “Are Chinese Trade Flows Different?” NBER Working Papers 17875. National Bureau of Economic Research, Cambridge, MA. Cheung, Y.-W., and R. Sengupta. 2012. “Impact of Exchange Rate Movements on Exports: an Analysis of Indian Non-Financial Sector Firms.” MPRA Paper 43118. University Library of Munich, Munich, Germany. Clark, P., N. Tamirisa, and S.-J. Wei. 2004. “A New Look at Exchange Rate Volatility and Trade Flows.” IMF Occasional Paper 235. International Monetary Fund, Washington, DC. Demers, M. 1991. “Investment under Uncertainty, Irreversibility and the Arrival of Information over Time.” Review of Economic Studies 58 (2): 333–50. Desai, M. A., C. F. Foley, and K. J. Forbes. 2008. “Financial Constraints and Growth: Multinational and Local Firm Responses to Currency Depreciations.” Review of Financial Studies 21 (6): 2857–88. Dominguez, K. M. E., and L. L. Tesar. 2001. “A Reexamination of Exchange-Rate Exposure.” American Economic Review 91 (2): 396–99. Eaton, J., S. Kortum, and F. Kramarz. 2004. “Dissecting Trade: Firms, Industries, and Export Destinations.” American Economic Review 94 (2): 150–54. Eckel, C., L. Iacovone, B. Javorcik, and P. Neary. 2011. “Multi-Product Firms at Home and Away: Cost Versus Quality-Based Competence.” CEPR Discussion Papers No. 8186. Center for Economic Policy Research, London, UK. Ethier, W. 1973. “International Trade and the Forward Exchange Market.” American Economic Review 63 (3): 494– 503. Frankel, J., and S.-J. Wei. 2007. “Assessing China’s Exchange Rate Regime.” NBER Working Paper 13100. National Bureau of Economic Research, Cambridge, MA. Freund, C., C. Hong, and S.-J. Wei. 2011. “China Trade Response to Exchange Rate.” Mimeo. Froot, K. 1989. “Consistent Covariance Matrix Estimation with Cross-sectional Cependence and Heteroskedasticity in Financial Data.” Journal of Financial and Quantitative Analysis 24: 333– 35. 578 THE WORLD BANK ECONOMIC REVIEW Gaulier, G., and S. Zignago. 2010. “BACI: International Trade Database at the Product-Level. The 1994-2007 Version.” CEPII Working Paper 2010-23. Centre d’Etudes Prospectives et d’Information Internationales, Paris, France. Greenaway, D., and R. A. Kneller. 2007. “Firm Heterogeneity, Exporting and Foreign Direct Investment.” Economic Journal 117 (517): 134 –361. Greenaway, D., A. Guariglia, and R. A. Kneller. 2007. “Financial Factors and Exporting Decisions.” Journal of International Economics 73 (2): 377–95. Grier, K. B., and A. D. Smallwood. 2007. “Uncertainty and Export Performance: Evidence from 18 Countries.” Journal of Money, Credit and Banking 39 (4): 965–79. Hodrick, R., and E. Prescott. 1997. “Post-war U.S. Business Cycles: An Empirical Investigation.” Journal of Money, Credit and Banking 29 (1): 1 – 16. Kroszner, R., L. Laeven, and D. Klingebiel. 2006. “Banking Crises, Financial Dependence, and Growth.” Journal of Financial Economics 84 (1): 187 –228. Li, H., H. Ma, Y. Xu, and Y. Xiong. 2012. “How do Exchange Rate Movements Affect Chinese Exports? A firm-level investigation.” Mimeo, School of Economics and Management, Tsinghua University, Beijing, China. Manova, K., S.-J. Wei, and Z. Zhang. 2011. “Firm Exports and Multinational Activity Under Credit Constraints.” NBER Working Paper 16905. National Bureau of Economic Research, Cambridge, MA. Manova, K., 2013. “Credit Constraints, Heterogeneous Firms, and International Trade.” Review of Economic Studies 80: 711–44. Marquez, J., and J. W. Schindler. 2007. “Exchange-rate Effects on China’s Trade.” Review of International Economics 15 (5): 837– 53. Mayer, T., and S. Zignago. 2011. “Notes on CEPII’s Distances Measures: The GeoDist Database.” CEPII working paper 2011-25. Centre d’Etudes Prospectives et d’Information Internationales, Paris, France. Minetti, R., and S. C. Zhu. 2011. “Credit Constraints and Firm Export: Microeconomic Evidence from Italy.” Journal of International Economics 83 (2): 109–25. Moulton, B. R. 1990. “An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Unit.” The Review of Economics and Statistics 72 (2): 334 –38. Park, A., D. Yang, X. Shi, and Y. Jiang. 2010. “Exporting and Firm Performance: Chinese Exporters and the Asian Financial Crisis.” The Review of Economics and Statistics 92 (4): 822–42. Pindyck, R. S. 1988. “Irreversible Investment, Capacity Choice, and the Value of the Firm.” American Economic Review 78 (5): 969– 85. ———. 1991. “Irreversibility, Uncertainty, and Investment.” Journal of Economic Literature 29 (3): 1110– 48. Pindyck, R. S., and A. Solimano. 1993. “Economic Instability and Aggregate Investment.” In O. Blanchard, and S. Fischer, eds., NBER Macroeconomics Annual 1993. Cambridge, MA: MIT Press. Rajan, R. G., and L. Zingales. 1998. “Financial Dependence and Growth.” American Economic Review 88 (3): 559–86. Ramey, G., and V. A. Ramey. 1995. “Cross-Country Evidence on the Link between Volatility and Growth.” American Economic Review 85 (5): 1138–51. Rose, A. 2000. “One Money, One Market: the Effect of Common Currencies on Trade.” Economic Policy 15 (30): 7– 45. Schnabl, G. 2007. “Exchange Rate Volatility and Growth in Small Open economies at the EMU Periphery.” ECB Working Waper 773. European Central Bank, Frankfurt, Germany. Train, K. 2003. Discrete Choice Methods with Simulation. Cambridge, MA: Cambridge University Press.