Financial Frictions, Trade Credit, and the 2008-09 Global Financial Crisis

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1020r July 2012 Financial Frictions, Trade Credit, and the 2008-09 Global Financial Crisis Brahima Coulibaly Horacio Sapriza Andrei Zlate NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Financial Frictions, Trade Credit, and the 2008-09 Global Financial Crisis Brahima Coulibalya Horacio Saprizaa Andrei Zlatea; (cid:3) aBoard of Governors of the Federal Reserve System, Division of International Finance y July 3, 2012 Abstract This paper studies the role of the credit crunch in the severe contraction of economic activity during the 2008-09 global (cid:133)nancial crisis, using (cid:133)rm-level data from six emerging Asian economies. After controlling for the e⁄ect of falling demand, we (cid:133)nd that sales declined by less for (cid:133)rms with better pre-crisis (cid:133)nancial conditions. Amid the decline in external (cid:133)nancing opportunities, some (cid:133)rms relied more on trade credit from suppliers during the crisis, which allowed them to post relatively better sales. Export-intensive (cid:133)rms resorted less to trade credit as an alternative source of (cid:133)nance, which contributed to their larger declines in sales. JEL classi(cid:133)cation: F14, F23, G32 Keywords: trade credit, 2008-09 (cid:133)nancial crisis, emerging Asia, international trade. (cid:3)Corresponding author. Tel. +1-202-452-3542, Fax +1-202-736-5638. yE-mail addresses: Brahima.Coulibaly@frb.gov (B. Coulibaly), Horacio.Sapriza@frb.gov (H. Sapriza), Andrei.Zlate@frb.gov (A. Zlate). Mail Stop 24, 20th and C Streets N.W., Washington, D.C. 20551, U.S.A. The views in this paper are solely the responsibility of the authors and should not be interpreted as re(cid:135)ecting the views of the Board of Governors of the Federal Reserve System or of any other person associated with the Federal Reserve System. 1

1 Introduction The 2008-09 global (cid:133)nancial crisis had dramatic e⁄ects on the economic activity in both advanced and emerging market economies (EMEs). The contraction in EMEs was surprisingly large given that the crisis originated in the advanced economies. There are two main known channels through which the crisis could have spilled over to EMEs: the reduction in demand for these countries(cid:146) exports (the trade channel) and the deterioration in (cid:133)nancing conditions (the (cid:133)nancial channel). ForthesetofEMEsthatwestudyinthispaper(China,India,Indonesia,Malaysia,Taiwan,and Thailand), the data suggest that both channels were at play during the crisis, a pattern discernible in both the aggregate and (cid:133)rm-level data. As shown in Figure 1, the downturn of activity in EMEs coincided with a worsening of (cid:133)nancial conditions there. At the aggregate level (the top panels), the decline in real activity (real GDP, industrial production, exports) coincided with a marked slowdown in private credit growth. Similarly, at the (cid:133)rm level (the bottom panels), the decline in sales for both domestic and export-oriented (cid:133)rms coincided with a drop in their external (cid:133)nancing. Notably, the decline in exports was sharper than that of total output. [LOCATE FIGURE 1 ABOUT HERE] Motivated by these observations, our study explores the e⁄ect of (cid:133)nancial frictions in general, and trade credit in particular, on economic activity in our sample of EMEs during the global (cid:133)nancial crisis. Our focus on trade credit as an alternative source of (cid:133)nancing during the recent crisis is guided by the (cid:133)ndings of existing studies, which document the role of trade credit in mitigating (cid:133)nancing constraints during past EME crises (see for example Love et al., 2007).1 Morespeci(cid:133)cally,weuse(cid:133)rm-leveldatafromthesixemergingmarketcountriesmentionedabove to explore: (1) Whether (cid:133)nancial constraints, in the form of reduced access to external (cid:133)nancing through bank loans and bond issuance, adversely a⁄ected (cid:133)rm-level sales during the crisis, after 1Love et al. (2007) (cid:133)nd that during the Mexican devaluation in 1994-95 and the Asian crisis in 1997, trade credit facilitated the redistribution of credit from (cid:133)nancially viable (cid:133)rms to the less viable ones. 2

controlling for the deterioration in global demand. (2) Whether the ability of (cid:133)rms to partially replaceexternal(cid:133)nancewithtradecreditfromsuppliersenhancedtheirrelativeperformanceduring the crisis. (3) Whether the relative inability of export-intensive (cid:133)rms to use trade credit as an alternative source of (cid:133)nance contributed to the larger decline in sales experienced by these (cid:133)rms. As in some of the previous studies, we de(cid:133)ne trade credit as the (cid:133)nancing that (cid:133)rms receive from their upstream suppliers in the form of delayed payments for the transfer of goods and services.2 To disentangle the e⁄ect of (cid:133)nancial constraints from the demand-driven reduction in sales during the crisis, we use two types of explanatory variables in our regression analysis. First, we use (cid:133)rms(cid:146)pre-crisis degree of (cid:133)nancial vulnerability and reliance on various sources of (cid:133)nancing, including trade credit, to explain their sales performance during the crisis. Second, we construct (cid:133)rm-speci(cid:133)c measures of global demand, a novel approach that allows us to document the e⁄ect of (cid:133)nancial frictions on sales while controlling for the variation in demand. Our results can be summarized as follows: (1) Financial conditions contributed to the decline in sales for all (cid:133)rms, but sales declined by less for (cid:133)rms with better (cid:133)nancial conditions prior to the crisis, such as those with more liquid assets and less exposure to external (cid:133)nance. Moreover, after controlling for pre-crisis (cid:133)nancial characteristics and demand conditions during the crisis, the export-intensive (cid:133)rms recorded larger declines in sales than their domestically-oriented counterparts. (2) Trade credit declined by less for the (cid:133)nancially-vulnerable (cid:133)rms, especially (cid:133)rms with more exposure to short-term debt before the crisis, suggesting that some (cid:133)rms relied more on trade credit to cope with the dire (cid:133)nancial conditions during the crisis. In addition, (cid:133)rms that were able to replace external (cid:133)nance with trade credit during the crisis (cid:150)predominantly the domesticoriented (cid:133)rms (cid:150)experienced smaller declines in sales than (cid:133)rms that did not. (3) Exporters with comparable (cid:133)nancial vulnerabilities had limited access to trade credit as an alternative source of 2Compared with "trade credit," the literature uses "trade (cid:133)nance" to refer to a broader range of short-term (cid:133)nancing related to the international trade activities of (cid:133)rms, such as working capital loans, letters of credit and trade insurance provided by (cid:133)rms, banks or government agencies (see U.S. Department of Commerce, 2007). 3

(cid:133)nancing, which contributed to the larger decline in their sales relative to non-exporters. These results highlight the interaction between (cid:133)nancial constraints and the real sector in propagating the e⁄ects of the global (cid:133)nancial crisis. In addition, the (cid:133)nding that trade credit was more scarce for exporters than for non-exporters points indirectly to the presence of (cid:133)nancial frictions among the factors that contributed to the disproportionately large decline in exports during the crisis.3 Our results are not driven by di⁄erences in demand for domestic and export-oriented (cid:133)rms, or by di⁄erences in the (cid:133)rms(cid:146)pre-crisis levels of inventories. We control for the e⁄ect of demand on (cid:133)rm-level sales by constructing an index of global demand from (cid:133)rm-level data on export reliance, sector-level data on exports by destination, and real GDP growth across destinations as a proxy for the change in demand. We also control for the role of (cid:133)rms(cid:146)inventories in o⁄setting the impact of (cid:133)nancial constraints on sales, since (cid:133)rms may draw on inventories when their production is restrained by (cid:133)nancial constraints. 1.1 Literature Review Our study adds to the existing evidence (cid:150)which is somewhat mixed (cid:150)on the e⁄ect of trade credit and (cid:133)nancial constraints more generally on economic activity. For trade credit and trade (cid:133)nancing, Chor and Manova (2012) use (cid:133)rm level data to show that the decline in U.S. imports during the 2008-09 global crisis was larger for countries of origin and sectors with adverse credit conditions, including limited reliance on trade credit. Similarly, Amiti and Weinstein (2011) document that trade (cid:133)nancing from banks played a key role in the transmission of (cid:133)nancial shocks to the Japanese exporting (cid:133)rms during the crisis that a⁄ected the country in the 1990s. However, using disaggregated U.S. imports and exports data, Levchenko et 3The extent to which our results document the e⁄ect of (cid:133)nancial frictions on international trade is restricted by data availability, since our dependent variable is the quarterly change in (cid:133)rm-level sales rather than in exports. However, we include the (cid:133)rms(cid:146)reliance on exports (available at the annual frequency only) in the set of explanatory variables. 4

al. (2010, 2011) (cid:133)nd no evidence that trade credit played a role in the collapse of trade during the 2008-09 crisis. Other studies explore the role of broadly-de(cid:133)ned (cid:133)nancial constraints on economic activity, including on international trade. Manova et al. (2009) use (cid:133)rm-level data for Chinese exporters from 2005 to show that multinational a¢ liates and joint ventures had better export performance than private domestic (cid:133)rms, especially in sectors with greater reliance on external (cid:133)nance and fewer hard assets to be used as collateral.4 Kolasa et al. (2010) use Polish (cid:133)rm-level data to show that foreign-owned (cid:133)rms proved more resilient during the 2008-09 crisis, which they argue was due to intra-group lending mechanisms supporting the credit-constrained a¢ liates. Along the same line, Rappoport et al. (2011) use matched customs and (cid:133)rm-level bank credit data from Peru to document the adverse e⁄ect of credit shortages on trade during the recent crisis. However, Bricongne et al. (2012) (cid:133)nd that (cid:133)nancial constraints played little role in explaining the decline in French exports. Compared with the papers mentioned above, we study the role of trade credit received from suppliers (rather than trade (cid:133)nance received from banks or foreign a¢ liates) measured at the (cid:133)rm level (rather than at the sector level) as an alternative source of external (cid:133)nancing during the 2008-09 crisis. In addition to trade credit, we also examine the relationship between (cid:133)rms(cid:146)(cid:133)nancial vulnerability prior to the crisis and their sales performance during the crisis, while also taking into account (cid:133)rms(cid:146)reliance on exports and exposure to global demand. Due to our focus on (cid:133)rms from emerging Asia, the 2008-09 global (cid:133)nancial crisis (cid:150)which originated in advanced economies (cid:150)can be viewed as an exogenous event in our analysis. Our (cid:133)nding that (cid:133)nancially-vulnerable (cid:133)rms (those with higher exposure to short-term debt before the crisis) substituted toward trade credit during the crisis is consistent with the literature 4For another study, see also Kalemli-Ozcan, Kamil and Villegas-Sanchez (2010). 5

on bank and trade credit channels. In an early study, Meltzer (1960) concludes that when liquidity conditions were tight, "(cid:133)rms with relatively large cash balances increased the average length of time for which [trade] credit was extended. And this extension of trade credit appears to have favored those (cid:133)rms against whom credit rationing is said to discriminate." More recently, Kohler et al. (2000) use a panel of publicly-traded (cid:133)rms from the United Kingdom and (cid:133)nd that, during recessions, (cid:133)rms with direct access to capital markets extended more trade credit and received less in return, thus making credit available to other (cid:133)rms that rely more on bank credit. In line with Meltzer (1960), they argue that there is a "trade credit channel" that o⁄sets the traditional bank credit channel in the monetary economics literature.5 The rest of the paper is organized as follows. Section 2 describes the data, Section 3 describes the empirical speci(cid:133)cations, and Section 4 discusses the (cid:133)ndings and robustness analyses. Section 5 concludes. 2 Data Description Our empirical analysis uses annual and quarterly data for almost 6,000 publicly-traded manufacturing (cid:133)rms from six emerging Asian countries (cid:150)China, India, Indonesia, Malaysia, Taiwan and Thailand (cid:150)obtained from the Worldscope database. The choice of the six emerging market countries is driven by data availability. Given our interest in export status as one of the determinants of (cid:133)rm performance in EMEs, we work with the countries where a reasonable number of (cid:133)rms report both exports and sales for the pre-crisis years. Thus, about one quarter of the (cid:133)rms in our sample with sales data for 2007 also report exports for the same year (about 1,600 (cid:133)rms).6 5In Kohler, Britton and Yates (2000), an important assumption behind the idea of the o⁄setting trade credit channel is that the adverse (cid:133)nancial shock must cause the external (cid:133)nance premium to rise by more for bankdependent (cid:133)rms than for (cid:133)rms with access to capital markets that provide trade credit. While a tightening of liquidity conditions may worsen (cid:133)rms(cid:146)access to bank credit, those (cid:133)rms that can directly fund themselves in capital markets may step in to (cid:133)ll the (cid:133)nancing gap, thus reducing the e⁄ect of the credit tightening on the economy. 6A similar fraction of (cid:133)rms reported exports on average between 2005 and 2007. 6

The (cid:133)rm-level data display contours similar to those of the aggregate data (Figure 1). First, aggregate activity measured by real GDP or industrial production contracted signi(cid:133)cantly between 2008:Q3 and 2009:Q1 (top-left panel). The median (cid:133)rm-level output, measured by sales, displays a similar pattern (bottom-left panel). Second, global trade fell sharply during the global (cid:133)nancial crisis: The top-middle panel shows the signi(cid:133)cant decline in exports between 2008:Q3 and 2009:Q1 forthecountriesinoursample. Atthe(cid:133)rmlevel, wemeasureexportsasthemediansalesofexportintensive (cid:133)rms whose exports represent more than 50 percent of total sales (bottom-middle panel). The aggregate and (cid:133)rm-level exports data display similar contours, with signi(cid:133)cant declines at the height of the global (cid:133)nancial crisis. Third, another notable feature of the crisis was the signi(cid:133)cant deterioration in credit provision and an attendant run-up in the cost of capital. This feature is captured in the panels to the right by the growth of credit extended to the private sector at the aggregate level (top), and by the median external (cid:133)nancing at the (cid:133)rm level (which is available at the annual frequency only, bottom panel). The similarity in the patterns of aggregate and (cid:133)rm-level data provides reassurance for the use of micro data to understand the linkages between (cid:133)nancial conditions, trade, and economic activity during the global (cid:133)nancial crisis. The richness of the micro data allows us to conduct the analysis while controlling for other factors that otherwise would have confounded estimation. [LOCATE TABLE 1 ABOUT HERE] Usingthesemicrodata, ourcross-sectionaldatasetisconstructedasfollows. (Seethesummary statistics in Table 1.) The dependent variable is the contraction in (cid:133)rm-level activity during the crisis, measured by the percent decline in quarterly sales from peak (2008:Q3) to trough (2009:Q1). Among the explanatory variables, the exports-to-sales ratio is constructed using data available at theannualfrequencyfor2007(thepre-crisisyear). Forthebaselineresults, wetreat(cid:133)rmsreporting salesbutnotexportsasnon-exporters,andassignthemanexports-to-salesratioequaltozero. This 7

approach is supported by the property that export-reporting (cid:133)rms in our sample have larger sales and larger total assets on average than (cid:133)rms with missing exports, a pattern which is consistent with previous studies documenting that exporting (cid:133)rms are larger than their domestically-oriented counterparts (Bernard et al., 2007).7 In the robustness analysis, we relax this assumption by exploring two additional methods to construct the export status of these (cid:133)rms, and obtain similar results. Also in the set of explanatory variables, the pre-crisis measures of (cid:133)nancial vulnerability consist of working capital (the di⁄erence between current assets and current liabilities, as an indicator of liquidity) and short-term debt, each normalized by total assets. There are also standard measures ofexternalandinternal(cid:133)nancingpriortothecrisis,computedastotalexternal(cid:133)nanceandretained earnings normalized by total assets. In addition, the (cid:133)rms(cid:146)pre-crisis reliance on trade credit from suppliers is measured as the amount of accounts payable normalized by the cost of goods sold, following Love et al. (2007) and Levchenko et al. (2010).8 In addition to (cid:133)rm-level data, we use annual data on exports by destination detailed at the two-digit sector level (provided by Comtrade) and quarterly real GDP data to construct a measure of global demand conditions during the crisis, as described in the next section. One advantage of the Worldscope data is the availability of (cid:133)rm-level sales at the quarterly frequency, even if the data on exports, working capital, short-term debt, external (cid:133)nance and retained earnings are available at the annual frequency only. Thus, the data allow us to study the link between (cid:133)rms(cid:146)sales performance during the crisis measured quarterly (as the percent change in sales between the peak and trough quarters) and a set of (cid:133)rm characteristics measured with annual data from the pre-crisis period. Given that the downturn and subsequent recovery of 7After controlling for country and industry e⁄ects, the export-reporting (cid:133)rms in our sample had on average $177 million larger sales, and $128 million larger total assets than (cid:133)rms not reporting exports for 2007. These di⁄erences are statistically signi(cid:133)cant at the 1 percent level. 8We multiply the ratio between accounts payable and the cost of goods sold by 360, and interpret the product as the number of days for which trade credit is received, as in Love et al. (2007). 8

economic activity in EMEs occurred over just a few quarters, the use of annual data to measure the peak-to-trough decline in sales would have understated the e⁄ect of the crisis on sales and the corresponding variation across (cid:133)rms. Firms covered by the Worldscope database report their (cid:133)nancial indicators according to each country(cid:146)s (cid:133)scal year (FY), which coincides with the calendar year for all countries in our sample except for India and Thailand. To match the (cid:133)rm-level data with the period marked by the crisis, the (cid:133)scal years are converted into calendar years by re-aligning the quarterly data for India (where FY 2009 started in April 2008) and Thailand (where FY 2009 started in October 2008).9 For the same reason, for India and Thailand, the annual data reported for FY 2009 is assigned to calendar year 2008. Out of the initial 6,000 (cid:133)rms, our econometric analysis is con(cid:133)ned to the sub-sample of (cid:133)rms for which data are simultaneously available for the dependent and explanatory variables. The sample size is further reduced by the removal of outliers; we replace observations in both the top and bottom percentiles for external (cid:133)nance (which can be either positive or negative) with missing values, and those in the bottom percentiles for retained earnings and working capital. For the exports-to-sales ratio, short-term debt, and accounts payable (which have a lower zero bound), we replace the outliers with missing values for observations in the top percentiles only. 3 Empirical Methodology To study the cross-sectional behavior of (cid:133)rm-level sales and trade credit received during the crisis, this paper uses three alternative econometric speci(cid:133)cations, described by Models 1-3 below. 9For India, the (cid:133)rst quarter of FY2009 became the second quarter of calendar year 2008. For Thailand, the (cid:133)rst quarter of FY2009 became the fourth quarter of calendar year 2008. 9

3.1 Model 1: Determinants of Firm Performance The (cid:133)rst model studies the determinants of (cid:133)rm performance during the crisis, which is expressed as the peak-to-trough percent change in (cid:133)rm sales between 2008:Q3 and 2009:Q1, and constitutes the dependent variable in the regression. The set of explanatory variables consists of (cid:133)rm-speci(cid:133)c characteristics, including (cid:133)nancial vulnerability, reliance on di⁄erent sources of (cid:133)nancing, and export status all measured in 2007 (the pre-crisis year), as well as the change in the (cid:133)rm-speci(cid:133)c global demand conditions during the crisis. The econometric speci(cid:133)cation is as follows: %(cid:1)Sales = (cid:11)+(cid:12) FinVuln +(cid:12) FinSource +(cid:12) Exp/Sales +(cid:12) %(cid:1)Demand + i 1 2007;i 2 2007;i 3 2007;i 4 isc + (cid:14) Country + (cid:14) Industry + (cid:14) Size +" ; (1) c ci s si v vi i c s v X X X where the explanatory variables are: 1. Indicators re(cid:135)ecting the degree of (cid:133)nancial vulnerability across (cid:133)rms at the onset of the crisis (FinVuln ). These are working capital as a measure of (cid:133)nancial liquidity, and the stock 2007;i of short-term debt normalized by total assets. 2. Measures of (cid:133)rms(cid:146)reliance on external and internal sources of (cid:133)nance prior to the crisis (FinSource ). These include the total external (cid:133)nance and retained earnings, each nor- 2007;i malized by total assets, and the amount of trade credit received from suppliers, measured as the stock of accounts payable normalized by the cost of goods sold in 2007. 3. Firms(cid:146)export status (Exp/Sales ), measured as the exports-to-sales ratio in 2007. 2007;i 4. A(cid:133)rm-speci(cid:133)cmeasureforthechangeinglobaldemandconditionsduringthecrisis(%(cid:1)Demand ), isc constructed as described in Section 3.1.1 below. 10

5. Dummy variables to isolate the country, industry, and (cid:133)rm size e⁄ects. Firms are assigned to 22 industry groups provided by the Worldscope database, after excluding non-manufacturing (cid:133)rms from the sample. Firms are also ranked in three size categories (top, mid, and bottom) based on their total assets in 2007. 6. In the robustness analysis, we include additional variables such as the initial level of sales in 2008:Q3 to control for convergence e⁄ects, two alternative measures of export status, as described in Section 4.2, and indices for domestic and external demand. 3.1.1 Demand Index We construct the (cid:133)rm-speci(cid:133)c index of global demand as a function of the (cid:133)rms(cid:146)exports-to-sales ratio, the sector-speci(cid:133)c exposure to demand from various foreign destinations, and the real GDP growthacrossdestinationsasaproxyforthechangeindemandduringthecrisis. Sincethe(cid:133)rm-level export data are not detailed by destination, we use the sector-level data on exports by destination detailed at the 2-digit level (provided by Comtrade SITC rev.3) for each country of origin to approximate the (cid:133)rms(cid:146)exposure to foreign destinations.10 For each sector and country of origin, wecomputethesharesof31exportdestinationsfor2007. Onaverage,the31destinationscomprised morethan90percentofourcountries(cid:146)exports. Thus,thesector-speci(cid:133)crelianceonforeignmarkets isassignedtothecorresponding(cid:133)rmsaccordingtotheirprimarysectorofactivity,countryoforigin, and (cid:133)rm-speci(cid:133)c degree of export reliance. Using the method just described, the demand index (%(cid:1)Demand ) for (cid:133)rm i in sector s and isc country of origin c is a weighted average of the change in domestic and external demand between 2008:Q3and2009:Q1,withtheweightsgivenbythe(cid:133)rm-speci(cid:133)cexports-to-salesratio(Exp/Sales ) i 10For Taiwan, since Comtrade does not provide exports data, we compute the demand index using the sectorspeci(cid:133)c export shares by destination from Malaysia, a country in our sample whose exports structure is most similar tothatofTaiwan. OurresultsarerobusttotheexclusionofTaiwan,suggestingthatthisassumptionisnotmaterially a⁄ecting our results. 11

in 2007: %(cid:1)Demand = (1 Exp/Sales ) %(cid:1)GDP +Exp/Sales (w %(cid:1)GDP ): (2) isc (cid:0) i (cid:2) c i(cid:2) dsc (cid:2) d d X In equation (2), the real GDP growth between 2008:Q3 and 2009:Q1 in country of origin c represents a proxy for the change in domestic demand conditions. Similarly, the average real GDP growth between 2008:Q3 and 2009:Q1 across the 31 foreign destinations is a proxy for the change external demand conditions, weighted by the shares w of each destination d in the exports of dsc sector s from country of origin c in 2007.11 3.2 Model 2: Substitution Across Sources of Financing In the second model, we explore the (cid:133)rms(cid:146)ability to substitute across various sources of (cid:133)nancing during 2008, the crisis year, as a strategy to relax their (cid:133)nancial constraints. To this end, the set of explanatory variables includes the dynamic trade-o⁄between the trade credit and the external (cid:133)nancing that (cid:133)rms received during the crisis as a new dimension of (cid:133)rm heterogeneity, in addition to the pre-crisis (cid:133)nancial indicators in Model 1. Figure 2 illustrates the trade-o⁄between external (cid:133)nance and new trade credit received in 2008 by the (cid:133)rms in our sample. On the horizontal axis, the amount of external (cid:133)nance (normalized by total assets) measures the (cid:135)ow of (cid:133)rm (cid:133)nancing from outside sources in 2008, such as the issuance and/or retirement of stock and debt. Thus, negative values of external (cid:133)nance correspond to (cid:133)rms that repurchased equity or experienced declines in their outstanding debt during the crisis.12 On the vertical axis, the di⁄erence in the stock of accounts payable between 2007 and 2008 normalized 11IntherobustnessanalysisdiscussedinSection4.2,theglobaldemandindexissplitintoitsdomesticandexternal demandcomponents,namely(1 Exp=Sales ) %(cid:1)GDP andExp=Sales w %(cid:1)GDP respectively,which (cid:0) i (cid:3) c i (cid:3) d dsc (cid:3) d enter separately in an alternative speci(cid:133)cation for Model 1. 12ForFigures1and2,theWorldscope(cid:133)rm-leveldataonexternal(cid:133)nanceis P availableattheannualfrequencyonly. 12

by the cost of goods sold shows the change in trade credit received from suppliers during the crisis. Positive values on the vertical axis correspond to (cid:133)rms that obtained more trade credit in 2008 relative to the previous year. [LOCATE FIGURE 2 ABOUT HERE] BasedonFigure2,(cid:133)rmsareclassi(cid:133)edacrossthefourquadrantsde(cid:133)nedbythezerolines,labeled as 1-4, starting with quadrant 1in the North-West and moving clockwise towards quadrant4 in the South-West. The classi(cid:133)cation of (cid:133)rms is as follows: (1) Firms in quadrant 1 posted an increase in trade credit received from suppliers but negative external (cid:133)nancing, thus replacing external (cid:133)nance with trade credit during the crisis. (2) Firms in quadrant 2 experienced both an increase in trade credit and positive external (cid:133)nancing, thus becoming less constrained along both dimensions. (3) Quadrant 3 includes (cid:133)rms with declines in trade credit but positive external (cid:133)nancing, thus substituting trade credit with external (cid:133)nance in the crisis year. (4) Finally, quadrant 4 consists of (cid:133)rms with reduced access to both sources of (cid:133)nance in 2008. In order to study the extent to which the substitution between external (cid:133)nance and trade credit a⁄ected sales during the crisis, we add a set of dummy variables (Quad , for q = 1;2;3) to the q;i speci(cid:133)cation described by equation (1). The dummy variables re(cid:135)ect the (cid:133)rms(cid:146)distribution across the (cid:133)rst three quadrants: %(cid:1)Sales = (cid:11)+(cid:12) FinVuln +(cid:12) FinSource +(cid:12) Exp/Sales +(cid:12) %(cid:1)Demand +(cid:12) Prod + i 1 2007;i 2 2007;i 3 2007;i 4 isc 5 i +(cid:12) TotalAssets + (cid:18) Quad + (cid:14) Country + (cid:14) Industry +" (3) 6 i s qi c ci s si i (cid:3) q=1;2;3 c s X X X We also include measures of (cid:133)rm productivity (Prod , including return on assets, gross pro(cid:133)ts i normalizedbytotalassets, andsalesnormalizedbytotalassets)and(cid:133)rmsize(TotalAssets , replaci ing the size dummy) measured prior to the crisis, in order to address a potential endogeneity issue 13

related to omitted variables.13 Firms(cid:146)productivity and size prior to the crisis are characteristics that may a⁄ect both their sales and their access to trade credit (or external (cid:133)nancing) during the crisis. For instance, this would be the case if the less productive and smaller (cid:133)rms experienced a larger decline in both sales and trade credit received during the crisis. On average, after controlling for (cid:133)nancial characteristics, export status, demand, country, industryandsize, weexpect(cid:133)rmsinquadrant1tooutperformtheircounterpartsinquadrant4, since their improved access to trade credit should o⁄set, at least partially, the reduced access to external (cid:133)nance. We also expect (cid:133)rms in quadrant 2 to outperform those in other quadrants. Finally, (cid:133)rms in quadrant 3 should fare better than those in quadrant 4. In the robustness analyses, we include the initial level of sales in 2008:Q3 to control for convergence e⁄ects, and use two alternative measures of export status. 3.3 Model 3: Determinants of Trade Credit The third model studies the characteristics of (cid:133)rms that received more trade credit from suppliers during the crisis. In particular, it explores whether the use of trade credit from suppliers as an alternative source of (cid:133)nance di⁄ered across exporters and non-exporters. Payables (cid:1) = (cid:11)+(cid:12) FinVuln +(cid:12) FinSource +(cid:12) Exp/Sales +(cid:12) %(cid:1)Demand + CGS 1 2007;i 2 2007;i 3 2007;i 4 isc (cid:18) (cid:19)i + (cid:14) Country + (cid:14) Industry + (cid:14) Size +" (4) c ci s si l li i c s l=2;3 X X X In the speci(cid:133)cation described by equation (4), we use quarterly data to construct the dependent variable as the peak-to-trough change in accounts payable normalized by the four-quarter sum of the cost of goods sold between 2008:Q3 and 2009:Q1. The explanatory variables consist of annual 13For the use of these measures as proxies for (cid:133)rm productivity, see Barber and Lyon (1996) and Glen and Singh (2004)forreturnonassets;LoveandZicchino(2006)forpro(cid:133)tsnormalizedbytotalassets;HanandRousseau(2009) and Nahata (2008) for sales normalized by total assets. 14

indicators of (cid:133)nancial vulnerability, dependence on various sources of (cid:133)nance other than trade credit, export reliance in 2007, exposure to demand conditions, and dummy variables for country, industry and size. In the robustness analysis, we include the initial level of trade credit to control for convergence e⁄ects, use two alternative measures export status, and construct an alternative measure of trade credit using accounts payable normalized by total assets (rather than the cost of goods sold), as in Fisman and Love (2003). 4 Results This section discusses the baseline results for Models 1-3 described above. 4.1 Baseline Results 4.1.1 Model 1: Determinants of Firm Performance The (cid:133)rst column of Table 2 shows the baseline results for Model 1. The results suggest that, in addition to the deterioration in global demand, the (cid:133)rm-speci(cid:133)c (cid:133)nancial conditions prior to the crisis a⁄ected sales performance. Thus, greater (cid:133)nancial liquidity ex-ante enhanced sales performance, as evidenced by the positive and statistically signi(cid:133)cant coe¢ cient for working capital.14 The coe¢ cient estimate shows that, for (cid:133)rms with working capital (normalized by total assets) one standard deviation above the mean, sales growth was 2.8 percentage points higher. In contrast, greater reliance on external (cid:133)nance ex-ante a⁄ected (cid:133)rms(cid:146)sales negatively during the crisis. Firms with external (cid:133)nance (normalized by assets) one standard deviation above the mean su⁄ered a 1.7 percentage points larger decline in sales. The coe¢ cient on the demand index is positive and statistically signi(cid:133)cant as we would expect. Those (cid:133)rms for which speci(cid:133)c demand conditions deteriorated by one standard deviation (4.4 14The results are similar when the quick ratio (the sum of cash, cash equivalents and net receivables divided by current liabilities) is used instead of working capital as a measure of (cid:133)nancial liquidity. 15

percentage points) more than the mean, sales growth was 6.8 percentage points lower.15 It is notable, however, that even after controlling for the decline in global demand, the (cid:133)rm-speci(cid:133)c indicators of (cid:133)nancial liquidity and reliance on external (cid:133)nance prior to the crisis are statistically signi(cid:133)cant with the expected signs. The results suggest that, in addition to demand, (cid:133)nancial conditions contributed to the decline in (cid:133)rm-level sales during the crisis. Finally, export-oriented (cid:133)rms experienced a relatively more severe deterioration in sales than their domestic-oriented counterparts. Thus, sales for (cid:133)rms with one standard deviation more in exports as a fraction of sales before the crisis fell by 1.4 percentage points more during the crisis. The results are preserved when we include the initial level of sales as an explanatory variable to control for convergence e⁄ects (see column 2 of Table 2). [LOCATE TABLE 2 ABOUT HERE] 4.1.2 Model 2: Substitution across Sources of Finance Although greater reliance on external (cid:133)nance prior to the crisis disrupted sales, the (cid:133)rms(cid:146)ability to substitute external (cid:133)nancing with trade credit during the crisis enhanced their sales performance. In Table 3, the (cid:133)rst column shows the baseline results for Model 2, which includes the substitution quadrants along with controls for (cid:133)rm productivity and size. The coe¢ cient on quadrant 1 is positive and statistically signi(cid:133)cant, suggesting that (cid:133)rms that were able to obtain more trade credit from suppliers experienced a smaller decline in sales (by about 8 percentage points less) relative to (cid:133)rms in quadrant 4 (which experienced a reduction in both trade credit and external (cid:133)nance during the crisis). The coe¢ cient on quadrant 3 is not statisticallysigni(cid:133)cant, i.e. (cid:133)rmswithlesstradecreditbutmoreexternal(cid:133)nancedidnotfarebetter than (cid:133)rms with less access to both sources of (cid:133)nance, a result which highlights the importance of 15Thechangesin salesand globaldemand areexpressed in thesameunitsforthesameinterval. Fortheformer,it is the percent change in (cid:133)rm-level sales. For the latter, it is the percent change in the weighted average of domestic and foreign GDP, weighted by each (cid:133)rm(cid:146)s export reliance and sector-level export shares across 31 destinations. 16

thetradecreditdimensioninunderstandingthesalesperformanceacross(cid:133)rms. Thus, theabilityto substitute away from external (cid:133)nance towards trade credit enhanced sales, but the reverse was not true, possibly owing to the more onerous nature of external (cid:133)nance at times of (cid:133)nancial turmoil. Finally, as expected, (cid:133)rms in quadrant 2 (which obtained more trade credit and more external (cid:133)nancing) experienced smaller declines in sales relative to the (cid:133)rms in quadrant 4 (by almost 11 percentage points less). We do not believe that the results for the quadrants are driven by reverse causality. The peakto-trough change in sales is measured over 2008:Q3-2009:Q1, whereas external (cid:133)nancing and trade credit (i.e. the variables used to construct the quadrants) are measured over 2008. As such, the overlap between the change in sales and the (cid:133)nancing variables used to construct the quadrants is only one quarter, namely 2008:Q4. Moreover, it seems implausible that the (cid:133)rms(cid:146)lack of access to external (cid:133)nancing was due exclusively to the anticipation of the decline in sales, but not related to the exogenous deterioration in global funding conditions emanating from the shock in advanced economies, following the collapse of Lehman Brothers in September 2008. Interestingly, once we control for the substitution between external (cid:133)nancing and trade credit in Model 2, the coe¢ cient on the export-to-sales ratio becomes statistically insigni(cid:133)cant (it was signi(cid:133)cant in Model 1). This is consistent with the notion that exporters and non-exporters di⁄ered in their ability to substitute external (cid:133)nancing with trade credit. Indeed, in Figure 2, out of the (cid:133)rms in quadrants 1 and 2 ((cid:133)rms that received more trade credit during the crisis), only 6.1 percent were export-intensive (with exports representing at least half of total sales), compared with 9.2 percent for the (cid:133)rms in quadrants 3 and 4.16 The pre-crisis (cid:133)nancial indicators are statistically signi(cid:133)cant and have the expected sign, in line with our previous results. Firms that were more (cid:133)nancially liquid before the crisis performed 16Overall, 7.9 percent of the (cid:133)rms in Figure 2 had exports accounting for more than half of sales in 2007. 17

better during the crisis; (cid:133)rms with ex-ante working capital one standard deviation above the mean enjoyed 3.1 percentage points more in sales growth. Similarly, (cid:133)rms with greater reliance on trade credit from suppliers before the crisis fared better; (cid:133)rms with trade credit one standard deviation above the mean experienced 2.9 percentage points more in sales growth. These results are preserved when we control for the initial level of sales, as shown in column 2 of Table 3. The pre-crisis (cid:133)nancial variables and the substitution quadrants preserve their sign and statistical signi(cid:133)cance. In addition, the control variables for productivity (pro(cid:133)ts and sales normalized by assets) and size (total assets) are positive and statistically signi(cid:133)cant as expected. [LOCATE TABLE 3 ABOUT HERE] 4.1.3 Model 3: Determinants of Trade Credit Given that increased access to trade credit improved sales performance, the next set of results shed light on the characteristics of (cid:133)rms that were able to use more trade credit as an alternative source of (cid:133)nance during the crisis, based on the speci(cid:133)cation in Model 3. Table 4 ((cid:133)rst column) shows the baseline results for the link between the change in trade credit and (cid:133)rm-speci(cid:133)c indicators of (cid:133)nancial vulnerability and export reliance measured prior to the crisis. First, the more (cid:133)nancially-vulnerable (cid:133)rms increased their use of trade credit during the crisis, as shown by the coe¢ cient on short-term debt, which is positive and statistically signi(cid:133)cant. The (cid:133)nding indicates that some of the (cid:133)nancially-vulnerable (cid:133)rms used more trade credit as an alternative form of (cid:133)nance, likely due to the fact that their short-term debt matured and became di¢ cult to rollover during the crisis. Firms with short-term debt one standard deviation above the mean received trade credit for 3.6 additional days from suppliers at the height of the crisis. This result is economically signi(cid:133)cant, since for the average (cid:133)rm, trade credit from suppliers declined by about 4 days from the peak to the trough quarters of the crisis (see Table 1). 18

Second,themoreexport-oriented(cid:133)rmswerelessabletousetradecreditasanalternativesource of(cid:133)nance. Theresultsshowanegativeandstatistically-signi(cid:133)cantrelationbetweentheexportshare ofsalesin2007andthepeak-to-troughchangeintradecreditduringthecrisis. Firmswithexportsto-sales ratio one standard deviation above the mean received trade credit from suppliers for 1.2 less days. This result is consistent with the uneven distribution of export-oriented (cid:133)rms across the four quadrants in Figure 2, as discussed above. The (cid:133)nding suggests that export-intensive (cid:133)rms experienced less access to trade credit as an alternative source of (cid:133)nance, which likely contributed to the decline in their sales. The negative link between the ex-ante export status and the use of trade credit during the crisis is preserved in the robustness analysis when the initial level of trade credit is added to control for convergence e⁄ects (column 2 of Table 4). It is also robust to alternative measures of export reliance (columns 3 and 4, also see Sections 4.2.1 and 4.2.2), and trade credit measured as accounts payable normalized by total assets rather than the cost of goods sold (column 5). In addition, the robustness analysis con(cid:133)rms that (cid:133)rms that were more vulnerable and less liquid ex-ante (i.e. were more reliant on external (cid:133)nancing and had less working capital) used more trade credit from suppliers during the crisis (see columns 4 and 5). [LOCATE TABLE 4 ABOUT HERE] 4.1.4 Discussion Our main results indicate that (cid:133)nancially-distressed (cid:133)rms received more trade credit during the crisis, which provided relief from the credit crunch and allowed them to maintain relatively better sales. However, exporters were less able to use trade credit as an alternative source of (cid:133)nancing, consistent with our view that their more binding (cid:133)nancial conditions contributed to the disproportionate decline in their sales during the crisis. These results raise two important questions. 19

First, why would (cid:133)nancially-distressed (cid:133)rms be able to use more trade credit from suppliers as analternativeformof(cid:133)nancing? Thetradecreditliteraturearguesthatsuppliershaveamonitoring advantage over banks. In the course of business, suppliers obtain information about the borrower which other lenders can only obtain at a cost (see Schwartz and Whitcomb, 1978 and 1979, or Emery, 1987). As such they are able to extend credit to (cid:133)rms that otherwise could not secure bank loans. Moreover, suppliers have an advantage over banks in enforcing debt repayments. They can credibly threaten to cut o⁄future supply of inputs, and also have industry knowledge that allows them to liquidate the collateral in case of default. This is consistent with (cid:133)ndings by Demirguc- Kunt and Maksimovic (2001) that trade credit is relatively more prevalent in countries with worse legal institutions. Second, why would exporters have less access to trade credit during the crisis? Studies (cid:133)nd that exporting (cid:133)rms are generally less (cid:133)nancially vulnerable than domestically-oriented ones (see Minetti and Zhu, 2011, or Muuls, 2008), whereas trade credit is a source of (cid:133)nancing used mostly by the more (cid:133)nancially-vulnerable (cid:133)rms, as shown in the previous paragraph. The results support our view that, since exporters did not have a need to develop the trade credit channel in normal times, theygenerallycouldnotuseitasanalternativesourceof(cid:133)nancingattheheightofthecrisis. 4.2 Additional Robustness Analysis The baseline results described above are subject to some caveats, which are addressed in the next set of analyses. 4.2.1 Exports-to-sales ratio computed from logit model The analysis so far assumed that (cid:133)rms reporting sales but not exports for 2007 were non-exporters, and thus assigned a value of zero to their exports-to-sales ratio, as discussed in Section 2. There 20

is a possibility that some exporters may fail to report exports data, in which case assigning zero values to their exports-to-sales ratio could be problematic. To ensure that our results are not driven by this assumption, we re-estimate Models 1, 2 and 3 with an alternative measure of export reliance, using the following approach. First, we estimate a logit model for the sub-sample based on (cid:133)rms that reported either positive or zero exports for 2007. The dependent variable is the exporting status (exporter vs. non-exporter), and the predictor variable is the log of the sales-to-total assets ratio in 2007 as a proxy for (cid:133)rm productivity, along with dummy variables to control for industry and country of origin.17 This approach follows the well-established empirical result from previous studies that exporting (cid:133)rms arelargerandmoreproductivethantheirdomestically-orientedcounterparts(Bernardetal.,2007). The results indicate a positive and statistically signi(cid:133)cant slope coe¢ cient for the log of the salesto-total assets ratio, suggesting that the probability of the (cid:133)rm being an exporter increases with (cid:133)rm productivity. Second, using the logit estimate, we compute the probability of exporting as a function of productivity for all the (cid:133)rms in our sample, including for (cid:133)rms that did not report exports but reported sales and total assets for 2007.18 Third, the resulting export probabilities are used as a proxy for export status in Models 1, 2 and 3. In addition, the global demand index is re-computed using the new proxy for the exports-to-sales ratio. The results, presented in the third column for each of Tables 2, 3 and 4, are largely similar to the baseline results for each model, con(cid:133)rming that the assumption of zero exports for the missing observations does not materially a⁄ect our baseline results. In Table 2 (column 3), the coe¢ cient for working capital is still positive and statistically signi(cid:133)cant, while those of external (cid:133)nance and export intensity are negative and statistically signi(cid:133)cant. In Table 3 (column 3), the coe¢ cients 17See Han and Rousseau (2009) and Nahata (2008) for the use of sales normalized by total assets as an indicator of (cid:133)rm productivity. 18Usingthelogitestimates,wecomputetheprobabilitythata(cid:133)rm isanexporterasfollows: prob(i=exporter)= exp(X (cid:12))=[1+exp(X (cid:12))]: i i 21

on quadrants 1 and 2 are still positive and statistically signi(cid:133)cant. In Table 4 (column 3), the coe¢ cient on export status is negative and statistically signi(cid:133)cant. Constructing the export status for non-reporting (cid:133)rms from the logit model has the advantage that it generates variation in the exports-to-sales ratio using a reasonable economic assumption (i.e. export status depends on (cid:133)rm productivity). One possible limitation of this approach arises from the inability to determine the properties of the second-stage estimator, which depend on the extent to which the sub-sample of (cid:133)rms reporting exports (used in the logit, (cid:133)rst-stage estimation) is representative of the entire sample of (cid:133)rms used in the second-stage regression. However, we do not believe that this limitation drives our results. To be sure, we perform an additional robustness check by imputing the (cid:133)rm-level export status from sector-level averages as described below. 4.2.2 Exports-to-sales ratio from sector averages On a country-by-country basis, we assign the average exports-to-sales ratio computed at the 3-digit sector level to the (cid:133)rms in that sector that have missing export observations for 2007. We use sector-level average ratios to impute exports at the (cid:133)rm level for each country if at least three (cid:133)rms in that sector and country reported exports. However, we keep the original exports-to-sales ratio for (cid:133)rms that report either zero or positive exports. The results for the determinants of sales performance, presented in the fourth column of Tables 2 and 3, are similar to the baseline results for each model. Similarly, in the fourth column of Table 4, the negative link between export status and trade credit during the crisis is consistent with the baseline results from the (cid:133)rst column. 22

4.2.3 Domestic vs. External Demand In order to assess the impact of domestic vs. external demand conditions separately on sales performance during the crisis, the global demand index in equation (2) is split into its domestic andexternaldemandcomponents,namely(1 Exp=Sales ) %(cid:1)GDP andExp=Sales w i c i dsc (cid:0) (cid:2) (cid:2) (cid:2) P %(cid:1)GDP ,respectively. ThetwocomponentsareincludedseparatelyintheestimationofModel1.19 d Theresults,reportedincolumn5ofTable2,showpositiveandstatisticallysigni(cid:133)cantlinksbetween (cid:133)rm sales and each of the domestic and external components of demand. In addition, column 6 showstheresultsforthemeasuresofdomesticandexternaldemandinteractedwithcountrydummy variables. To save space, only the country interactions with statistically signi(cid:133)cant coe¢ cients are reported.20 For domestic demand, the interacted terms are positive and statistically signi(cid:133)cant for Malaysia and Thailand, economies which su⁄ered large real GDP contractions during the crisis.21 Notably, for external demand, the interacted terms are positive and statistically signi(cid:133)cant for China, Malaysia, and Thailand, suggesting that sales of (cid:133)rms in these countries were more sensitive to the deterioration in external demand. 4.2.4 Inventories One potentially important variable missing from the headline analysis is the level of (cid:133)rms(cid:146)inventories. Inventories play an important role in meeting demand when production is disrupted. During the (cid:133)nancial crisis, (cid:133)rms with production constrained by the dire (cid:133)nancial conditions could have drawn on inventories to ful(cid:133)ll some or all of the demand for their products. In this case, (cid:133)nancial conditions would have a smaller e⁄ect on sales. We re-estimate Models 1 and 2 for the determinants of sales performance while controlling for 19The export-to-sales ratio is excluded due to its perfect correlation with the domestic demand component within each country. 20The complete set of results for the interacted variables are available upon request. 21Incontrast,China,IndiaandIndonesiaareamongthefewEMEsthatdidnotcontractduringthe2008-09crisis. 23

the inventories-to-sales ratio measured in 2007. Indeed, the results in Table 5 show positive and statistically signi(cid:133)cant coe¢ cients on the inventories-to-sales ratio (columns 1 and 3), indicating that(cid:133)rmswithhigherlevelsofinventoriesrelativetosalespriortothecrisisexperiencedarelatively smaller decline in sales during the crisis. In addition, the speci(cid:133)cations in columns (2) and (4) interacttheinventories-to-salesratiowiththe(cid:133)nancialvariablesalreadydiscussed. Thecoe¢ cients on the interacted terms with short-term debt and external (cid:133)nance are statistically signi(cid:133)cant and have the expected signs. Thus, the sales of (cid:133)rms with higher pre-crisis levels of inventories relative to sales were less constrained by their vulnerability position or reliance on external (cid:133)nance. These results point to an important role of inventories in alleviating, but not eliminating, the e⁄ect of (cid:133)nancial constraints on performance, as our benchmark results indicate. [LOCATE TABLE 5 ABOUT HERE] 5 Conclusions We explore the extent to which (cid:133)nancial conditions contributed to the decline in (cid:133)rms(cid:146)sales at the height of the 2008-09 global (cid:133)nancial crisis using micro data from six emerging market economies in Asia. Even after controlling for demand, we (cid:133)nd that (cid:133)nancial conditions adversely a⁄ected sales during the crisis, and that the use of trade credit played an important role in the relative performance of (cid:133)rms. In particular, when (cid:133)nancing conditions deteriorated, the more (cid:133)nanciallyvulnerable (cid:133)rms turned to trade credit from suppliers as a supplement to other forms of (cid:133)nancing. In addition, (cid:133)rms that were able to replace external (cid:133)nance with trade credit had better sales. In contrast to domestic-oriented (cid:133)rms, export-intensive (cid:133)rms with comparable (cid:133)nancial vulnerability relied less on trade credit as an alternative source of (cid:133)nancing, and experienced sharper declines in sales. Our (cid:133)ndings have implications for the design of policy to cushion the e⁄ect of future (cid:133)nancial 24

crises. Policy makers and (cid:133)rms would be well-advised to explore the development of trade credit as an additional source of (cid:133)nancing, which might not be as desirable in normal times, but could prove useful during crises when credit markets become impaired. 6 Acknowledgements We are grateful to the editors Paul Bergin and Fabio Ghironi, two anonymous referees, Shaghil Ahmed, Martin Bodenstein, Aitor Erce, Julian di Giovanni, Neil Ericsson, Benjamin Mandel, Patrice Robitaille, Katheryn Russ, Robert Vigfusson, and seminar participants at the 2011 North American Meeting of the Econometric Society, the XV CEMLA meeting, the 2010 European Trade Study Group meeting, and the 2010 conference on the Global Financial Crisis at the Federal Reserve Board for helpful comments. We thank Quoctrung Bui, Kavita Patel and Peter Weyand for excellent research assistance. References [1] Amiti M. & Weinstein D. (2011). Exports and (cid:133)nancial shocks. Quarterly Journal of Economics, 126(4), 1841-1877. [2] Barber, B. M. & Lyon J. D. (1996). Detecting abnormal operating performance: the empirical power and speci(cid:133)cation of test statistics. Journal of Financial Economics, 41(3), 359-399. [3] Bernard,A.B.,Jensen,J.B.,Redding,S.J.,&Schott,P.K.(2007).Firmsininternationaltrade. Journal of Economic Perspectives, 21(3), 105-130. [4] Bricongne, J.C., Fontagne, L., Gaulier, G., Taglioni, D. & Vicard, V. (2012). Firms and the global crisis: French exports in the turmoil. Journal of International Economics, 87(1), 134- 146. 25

[5] Chor, D.&Manova, K.(2012).O⁄thecli⁄andback: creditconditionsandinternationaltrade during the Global Financial Crisis. Journal of International Economics, 87(1), 117-133. [6] Demirguc-Kunt, A.&Maksimovic, V.(2001).Firmsas(cid:133)nancialintermediaries: evidencefrom trade credit data. World Bank Policy Research Working Paper 2696. [7] Emery, G. (1987). An optimal (cid:133)nancial response to variable demand. Journal of Financial and Quantitative Analysis, 22(2), 209-225. [8] Fisman, R. & Love, I. (2003). Trade credit, (cid:133)nancial intermediary development, and industry growth. Journal of Finance, 58(1), 353-374. [9] Glen, J. & Singh, A. (2004). Comparing capital structures and rates of return in developed and emerging markets. Emerging Markets Review, 5(2), 161-192. [10] Han, L. & Rousseau, P. L. (2009). Technology shocks, Q, and the propensity to merge. Vanderbilt University Department of Economics, Working Paper 09-W14. [11] Kalemli-Ozcan, S., Kamil H., & Villegas-Sanchez, C. (2010). What hinders investment in the aftermath of (cid:133)nancial crises: insolvent (cid:133)rms or illiquid banks? NBER Working Paper 16528. [12] Kohler, M., Britton, E. & Yates, T. (2000). Trade credit and the monetary transmission mechanism. Bank of England Working Paper 115. [13] Kolasa, M., Rubaszek, M. & Taglioni, D. (2010). Firms in the great global recession: the role of foreign ownership and (cid:133)nancial dependence. Emerging Markets Review, 11(4), 341-357. [14] Levchenko, A., Lewis L., & Tesar, L. (2010). The collapse in international trade during the 2008-2009(cid:133)nancialcrisis: insearchofthesmokinggun.IMF Economic Review,58(2),214-253. 26

[15] Levchenko,A.,LewisL.,&Tesar,L.(2011).Theroleoftrade(cid:133)nanceintheU.S.tradecollapse: a skeptic(cid:146)s view. In J.P. Chau⁄our & M. Malouche (Eds.), Trade (cid:133)nance during the great trade collapse. The World Bank. [16] Love, I., Preve L., & Sarria-Allende, V. (2007). Trade credit and bank credit: evidence from recent (cid:133)nancial crises. Journal of Financial Economics, 83(2), 453-469. [17] Love, I. & Zicchino, L. (2006). Financial development and dynamic investment behavior: evidence from panel VAR. Quarterly Review of Economics and Finance, 46(2), 190-210. [18] Manova, K., Wei, S.J. & Zhang, Z. (2009). Firm exports and multinational activity under credit constraints. mimeo, Stanford University. [19] Meltzer, A. (1960). Mercantile credit, monetary policy, and size of (cid:133)rms. Review of Economics and Statistics, 42(4), 429-437. [20] Minetti, R. & Chun Zhu, S. (2011). Credit constraints and (cid:133)rm export: microeconomic evidence from Italy. Journal of International Economics, 83(2), 109-125. [21] Muuls, M. (2008). Exporters and credit constraints: a (cid:133)rm level approach. mimeo, London School of Economics. [22] Nahata, R. (2008), Venture capital reputation and investment performance. Journal of Financial Economics, 90(2), 127-151. [23] Rappoport, V., Paravisini, D., Wolfenzon, D. & Schnabl, P. (2011). Dissecting the e⁄ect of credit supply on trade: evidence from matched credit-export data. NBER Working Paper 16795. 27

[24] Schwartz, R. A. & Whitcomb, D. (1978). Implicit transfers in the extension of trade credit. In K.E. Boulding & T.F. Wilson (Eds.), The Channels of Redistribution through the Financial System (pp. 191-208). New York: Praeger. [25] Schwartz, R. A. & Whitcomb, D. (1979). The trade credit decision. In J.L. Bicksler (Ed.), Handbook of Financial Economics pp. 257-73. Amsterdam: North-Holland. [26] U.S. Department of Commerce (2007). Trade (cid:133)nance guide: a quick reference for U.S. exporters. U.S. International Trade Administration. 28

Table 1. Summary statistics Variable Frequency No. obs. Mean St. dev. Min. Max. Unit Dependent variables %Δ Sales Quarterly 4,590 -19.09 40.41 -100.00 261.15 % 2008:Q3-2009:Q1 Change in Acc. Payable/CGS Quarterly 2,077 -4.03 30.73 -210.82 239.46 days 2008:Q3-2009:Q1 Change in Acc. Payable/Tot. Assets Quarterly 1,597 -2.45 6.26 -49.62 29.01 %-point 2008:Q3-2009:Q1 Explanatory variables Working Capital/Assets Annual 5,897 0.19 0.27 -1.24 1.00 - 2007 Short-term debt/Assets Annual 5,815 0.15 0.14 0.00 0.82 - 2007 External Finance/Assets Annual 5,825 0.06 0.13 -0.24 0.66 - 2007 Retained Earnings/Assets Annual 4,096 0.01 0.36 -3.50 0.90 - 2007 Acc. Payable/CGS Annual 5,679 65.84 49.41 0.00 359.73 days 2007 Export status and demand index Exports/Sales Annual 5,940 0.08 0.21 0.00 1.00 - 2007 Exports/Sales logit Annual 5,844 0.88 0.13 0.07 1.00 - 2007, Exports/Sales 3-digit sector Annual 4,351 0.35 0.28 0.00 0.99 - 2007, %Δ Demand Annual 5,562 -1.16 4.37 -6.80 3.30 % %Δ Demand, logit Annual 5,472 -3.41 1.16 -7.56 2.04 % %Δ Demand, 3-digit sector Annual 4,031 -1.13 3.22 -6.71 3.30 % Firm productivity, size and inventories Return on Assets Annual 5,658 6.81 8.65 -33.21 42.83 % 2007 Gross Profits/Total Assets Annual 5,648 0.34 0.59 -1.91 3.99 - 2007 Sales/Total Assets Annual 5,956 0.92 0.68 0.00 10.32 - 2007 Total Assets Annual 5,968 383.25 1,448.17 0.00 43,609.22 US$ mil 2007 Inventories/Sales Annual 5,801 0.26 0.37 0.00 3.55 - 2007 Convergence variables (initial levels) Sales Quarterly 4,798 61.87 133.03 -2.45 1310.87 US$ mil 2008:Q3 Acc. Payable/CGS Quarterly 1,642 56.82 47.29 0.09 346.99 days 2008:Q3 Acc. Payable /Total Assets Quarterly 1,703 11.69 10.87 0.01 77.58 % 2008:Q3 2007 Data sources: Worldscope (for firm-level data), Haver Analytics (for macro data used in the demand index) and authors' calculations. See Section 2 for details. 29

Table 2. Determinants of firm sales, Model 1 Dependent variable: % Change in Sales, 2008:Q3-2009:Q1 (1) (2) (3) (4) (5) (6) Robustness, Robustness, Robustness, Robustness, Robustness, Assumptions: Baseline control for exp. status from exp. status from dom. vs. ext. dom. vs. ext. initial sales productivity sector averages demand demand Working Capital/Assets 10.24** 9.845** 10.99** 17.55*** 9.845** 9.887** 2007 (4.708) (4.756) (4.728) (5.631) (4.756) (4.776) Short-term debt/Assets -3.629 -4.377 -1.910 -7.583 -4.379 -4.612 2007 (7.372) (7.475) (7.373) (8.613) (7.475) (7.518) External Finance/Assets -13.22** -12.98** -14.65*** -0.577 -12.95** -13.12** 2007 (5.223) (5.238) (5.318) (7.370) (5.238) (5.257) Retained Earnings/Assets -0.889 -0.768 -0.0510 -1.399 -0.763 -0.897 2007 (3.715) (3.723) (3.637) (4.271) (3.722) (3.802) Acc. Payable/CGS 0.0144 0.0118 0.00473 0.0163 0.0118 0.0115 2007 (0.0199) (0.0201) (0.0194) (0.0257) (0.0201) (0.0202) Exports/Sales -6.491** -5.851** -91.07*** -13.59*** 2007 (2.532) (2.617) (31.04) (4.085) %Δ Demand 1.548** 1.466** 1.317 3.574*** (0.666) (0.678) (1.047) (0.964) %Δ Domestic demand 1.295** (0.629) %Δ Domestic demand × Malaysia 17.40** (8.312) %Δ Domestic demand × Thailand 34.82*** (3.004) %Δ External demand 2.739** (1.083) %Δ External demand × China 18.46** (8.806) %Δ External demand × Malaysia 31.39** (15.26) %Δ External demand × Thailand 83.37*** (7.528) Sales -0.00683* -0.00331 -0.00274 -0.00682* -0.00679* 2008-Q3 (0.00394) (0.00416) (0.00449) (0.00394) (0.00395) Constant -30.11*** -30.32*** 52.29* -31.88*** -31.45*** -55.65*** (8.580) (8.637) (30.56) (6.922) (8.482) (18.39) Observations 3,063 3,020 3,024 1,757 3,020 3,020 R-squared 0.107 0.106 0.113 0.152 0.106 0.108 Country dummies Yes Yes Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes Yes Yes Firm size dummies Yes Yes Yes Yes Yes Yes Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Notes: Heteroskedasticity-robust standard errors are reported in parentheses. To preserve space, only the country interactions with statistically significant coefficients are reported in column 6. The demand index for firm i in sector s, country c is: %ΔDemand = (1–Exp/Sales ) × %ΔGDP + Exp/Sales × Σ ( Weightd × %ΔGDPd ), where: (1) Exp/Sales is the isc 2007,i c 2007,i d sc 2007,i exports/sales ratio obtained as follows: (a) In columns 1, 2, 5 and 6, it is equal to the exports/sales ratio for firms that report either zero or positive exports, and zero for firms that report no exports for 2007. (b) In column 3, it is the probability that firms were exporters in 2007, computed from a logit estimation of export status as a function of productivity (sales/total assets) for the sub-sample of firms that report either zero or positive exports for 2007. (c) In column 4, it is equal to the 3-digit sector average ratio for firms that do not report exports for 2007, and to the firm-specific ratio for firms that report either zero or positive exports for 2007. (2) %ΔGDP is real GDP growth for firms’ country of origin c between 2008:Q3 and 2009:Q1, as a c proxy for the change in domestic demand. (3) Weightd is the share of destination d in the exports of sector s from country c; sc for Taiwan, since Comtrade does not provide trade data, we use Malaysia’s export shares by destination. (4) %ΔGDPd is the real GDP growth for the destination country d between 2008:Q3 and 2009:Q1, as a proxy for the change in external demand. 30

Table 3. Determinants of firm sales, Model 2 Dependent variable: % Change in Sales, 2008:Q3-2009:Q1 (1) (2) (3) (4) Robustness, Robustness, Robustness, Assumptions: Baseline control for exp. status from exp. status from initial sales firm productivity sector averages Working Capital/Assets 11.38** 10.98** 11.63** 12.27* 2007 (4.661) (4.728) (4.702) (6.736) Short-term debt/Assets -7.530 -7.604 -6.929 -13.28 2007 (6.829) (6.944) (6.844) (9.236) External Finance/Assets -6.725 -6.223 -7.655 2.974 2007 (5.931) (5.977) (6.023) (7.907) Retained Earnings/Assets 4.185 4.304 4.153 6.536* 2007 (2.837) (2.866) (2.747) (3.572) Acc. Payable/CGS 0.0592*** 0.0578*** 0.0564*** 0.0513** 2007 (0.0198) (0.0198) (0.0194) (0.0250) Exports/Sales -3.355 -2.796 -97.24*** -15.12*** 2007 (2.570) (2.686) (35.02) (4.243) %Δ Demand 1.919*** 1.921** 1.286 3.503*** (0.739) (0.750) (1.060) (1.037) I_Quad1 8.385*** 8.226*** 8.457*** 7.434*** (1.853) (1.865) (1.853) (2.538) I_Quad2 10.61*** 10.43*** 10.06*** 9.705*** (2.286) (2.309) (2.293) (3.161) I_Quad3 0.326 0.442 0.0666 -0.126 (1.472) (1.496) (1.476) (1.823) Return on Assets -0.135 -0.126 -0.0897 -0.184 2007 (0.0987) (0.0997) (0.0990) (0.127) Gross Profits/Total Assets 4.554*** 4.700*** 4.634*** 4.647** 2007 (1.684) (1.710) (1.693) (2.173) Sales/Total Assets 1.555 2.591** 5.844*** 2.298 2007 (1.124) (1.291) (1.473) (1.406) Total Assets 7.75e-05 0.00284* 0.00782*** 0.00139 2007 (0.000328) (0.00157) (0.00244) (0.00191) Sales -0.0129** -0.0249*** -0.00249 2008-Q3 (0.00601) (0.00767) (0.00689) Constant -35.59*** -36.50*** 45.61 -35.81*** (9.836) (9.891) (34.30) (7.887) Observations 2,501 2,462 2,464 1,481 R-squared 0.152 0.151 0.164 0.191 Country dummies Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Notes: See the notes to Table 2, including for the construction of Exp/Sales and %ΔDemand . 2007,i isc 31

Table 4. Determinants of trade credit, Model 3 Dependent variable: Change in Acc. Payable normalized by the Cost of Goods Sold (columns 1-4) and by Total Assets (column 5) in the interval 2008:Q3-2009:Q1 (1) (2) (3) (4) (5) Robustness, Robustness, Robustness, Robustness, Assumptions: Baseline control for initial exp. status from exp. status from Acc. Payable acc. payable firm productivity sector averages norm. by assets Working Capital/Assets 6.641 -3.098 -1.118 -9.116 -1.671** 2007 (8.054) (7.896) (8.009) (6.841) (0.739) Short-term debt/Assets 25.94*** 10.16 12.66 18.52 -1.857 2007 (9.300) (9.454) (9.481) (11.90) (1.363) External Finance/Assets 13.81 10.53 7.909 29.09*** 0.566 2007 (9.883) (10.38) (10.60) (10.95) (1.159) Retained Earnings/Assets -0.491 -4.619 -4.213 -0.437 0.00617 2007 (6.927) (6.526) (6.270) (4.228) (0.613) Exports/Sales -5.777* -5.195* -61.27* -6.360* -1.142** 2007 (3.270) (3.139) (36.79) (3.596) (0.555) %Δ Demand 0.124 0.788 2.184 1.927** 0.0724 (0.884) (0.883) (1.400) (0.976) (0.130) Acc. Payable/CGS -0.233*** -0.237*** -0.160*** 2008:Q3 (0.0397) (0.0398) (0.0480) Acc. Payable /Total Assets -0.325*** 2008:Q3 2007 (0.0224) Constant 6.382 16.06 75.40** 11.16 0.473 (11.26) (12.61) (36.00) (9.311) (1.822) Observations 1,246 1,187 1,161 744 1,254 R-squared 0.054 0.148 0.157 0.142 0.468 Country dummies Yes Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes Yes Firm size dummies Yes Yes Yes Yes Yes Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Notes: See notes to Table 2, including for the construction of Exp/Sales and %ΔDemand . 2007,i isc 32

Table 5. Determinants of firm sales, robustness to inventories Dependent variable: % Change in Sales, 2008:Q3-2009:Q1 (1) (2) (3) (4) Models: Model 1 Model 1 Model 2 Model 2 Working capital/Assets 6.787 8.301 6.746 7.689 2007 (4.932) (5.337) (4.908) (5.559) Short-term debt/Assets -9.177 -21.47** -15.44** -20.69** 2007 (7.561) (8.532) (7.121) (8.483) External Finance/Assets -13.70*** -26.29*** -6.703 -20.69*** 2007 (5.207) (6.330) (5.912) (7.227) Retained Earnings/Assets -0.229 0.442 4.167 7.067* 2007 (3.717) (4.708) (2.887) (4.198) Acc. Payable/CGS -0.000992 -0.00532 0.0460** 0.0429* 2007 (0.0201) (0.0210) (0.0200) (0.0230) Exports/Sales -5.699** -5.885** -3.439 -3.798 2007 (2.611) (2.618) (2.736) (2.776) %Δ Demand 1.541** 1.648** 1.802** 1.875** (0.675) (0.679) (0.753) (0.772) I_Quad1 8.494*** 8.710*** (1.871) (1.862) I_Quad2 10.39*** 10.27*** (2.297) (2.293) I_Quad3 0.303 0.460 (1.485) (1.485) Return on Assets -0.128 -0.137 2007 (0.0998) (0.102) Gross Profits/Total Assets 5.558*** 5.270** 2007 (2.121) (2.114) Sales/Total Assets 3.082** 2.968** 2007 (1.239) (1.225) Inventories/Sales 5.462* 6.310* 2007 (3.188) (3.412) Inv/Sales × Working capital/Assets -10.58 -1.861 2007 2007 (10.15) (9.587) Inv/Sales × Short-term debt/Assets 40.04* 19.83 2007 2007 (23.33) (16.72) Inv/Sales × Ext. Finance/Assets 47.16*** 47.43*** 2007 2007 (16.15) (16.43) Inv/Sales × Ret. Earnings/Assets -1.267 -12.81 2007 2007 (11.79) (11.95) Inv/Sales × Acc. Payable/CGS -0.000323 0.00660 2007 2007 (0.0311) (0.0306) Sales -0.00479 -0.00468 -0.00593 -0.00635 2008-Q3 (0.00395) (0.00394) (0.00433) (0.00432) Constant -29.10*** -26.31*** -37.61*** -35.50*** (8.698) (8.475) (10.45) (10.25) Observations 3,000 3,000 2,449 2,449 R-squared 0.109 0.117 0.155 0.160 Country dummies Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes Firm size dummies Yes Yes Yes Yes Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Notes: Columns (1) and (3) add the inventory/sales ratio for 2007 as an explanatory variable to Models 1 and 2, respectively. Columns (2) and (4) add interactions of the inventory/sales ratio with each of the financial variables on rows 1-5. Exp/Sales is constructed as in columns 1, 2, 5 and 6 of Table 2. See the additional notes to Table 2. 2007,i 33

Figure 1. Aggregate vs. firm-level data Asian GDP and Ind. Production Asian Exports Asian Private Credit Growth Index: 2008Q1 = 100 Index: 2008Q1 = 100 Annual Rate 110 110 20 100 100 15 90 90 10 80 80 5 GDP Industrial Production 70 70 0 2008 2009 2008 2009 2007 2008 2009 Firm-level Sales Firm-level Sales by Export Status Firm-level External Finance Index: 2008Q1 = 100 Index: 2008Q1 = 100 Index: 2007 = 100 110 110 120 Exports >= 50% sales 100 100 100 90 90 80 80 80 60 70 70 40 60 60 20 50 50 0 2008 2009 2008 2009 2007 2008 2009 Data sources: Haver Analytics (aggregate data) and Worldscope (firm-level data). Notes: For the aggregate data (top panels), we use real GDP (in local currencies, seasonally-adjusted) and industrial production (IP); private credit growth (computed as the q/q annualized growth rate of outstanding private credit in local currencies, non seasonally-adjusted); and exports (in nominal US$, seasonally-adjusted) for six emerging Asian economies (China, India, Indonesia, Malaysia, Thailand and Taiwan). We normalize the GDP, IP and exports series relative to 2008:Q1, and take non-weighted averages of the resulting indices for the six countries. For private credit, we use: (1) China: Uses of credit funds of Financial Institutions, 100 Mil. Yuan, NSA; (2) India: Domestic Credit: Commercial Sector, NSA, Millions Rupees; (3) Indonesia: Commercial Bank Credit, NSA Bil. Rupiahs; (4) Malaysia: Banking Sector: Claims on Private Enterprises, NSA, Mil. Ringgit; (5) Taiwan: Loans/Investments of Major Financial Institutions: Claims on Private Sector, NSA, 100 Mil. NT$; (6) Thailand: Depository Corporations Survey: Claims on Other Sectors, NSA, Mil. Baht. For the firm-level data (bottom panels), we report medians computed across the full sample of firms from the six emerging Asian countries (China, India, Indonesia, Malaysia, Thailand and Taiwan), and report the values relative to 2008:Q1. For firm-level external finance (bottom-right panel), the data is available at the annual frequency only. 34

Figure 2. External financing and trade credit during the 2008-09 crisis Quad. 1 Quad. 2 Quad. 4 Quad. 3 Data source: Worldscope firm-level data. The sample includes firms from China, India, Indonesia, Malaysia, Taiwan and Thailand. Note: On the horizontal axis, the amount of external finance (available at the annual frequency only) normalized by total assets measures the flow of firm financing from outside sources in 2008, such as the issuance and/or retirement of stock and debt. Negative values of external finance correspond to firms that repurchased equity or experienced declines in their outstanding debt during the crisis. On the vertical axis, the difference in the stock of accounts payable between 2007 and 2008 normalized by the cost of goods sold shows the change in trade credit received from suppliers during the crisis. Positive values correspond to firms that obtained more trade credit in 2008 relative to the previous year. 35 80−7002 ,SGC/elbayap .cca ni egnahC 004 002 0 002− 004− External Finance vs. Trade Credit −.2 0 .2 .4 .6 External financing 2008/Total assets