The real effects of finance: evidence from exports

Transcription

1 The real effects of finance: evidence from exports Bo Becker * and David Greenberg ** *** This version: November 2003 Abstract. In this paper, we investigate the link between financial development and exports. Using bilateral trade data, we find that having a better financial system increases exports. To explain this result, we propose that exporting firms face significant up-front fixed costs in product design, marketing, distribution etc. In an economy where outside financing for such investments is difficult to secure, exports suffer. Consistent with this explanation, we find that the marginal impact of financial development on exports is higher in those industries and country pairs where up-front investments are large, due to either product characteristics or economic distance between exporter and importer. Further supporting evidence comes from periods of bank crises, when exports of goods with high up-front costs fall more than other exports. Finally, we examine the effect of finance on the responsiveness of exports to exchange rates. We find that short-term elasticity of exports is considerably higher in countries with good finance, and that the allocation of exports across importers is much more sensitive to relative real exchange rates. * Graduate School of Business, University of Chicago; bo.becker@gsb.uchicago.edu ** Graduate School of Business, University of Chicago *** We wish to thank John Romalis, Luigi Zingales, Atif Mian, Per Strömberg, Raghuram Rajan, Jens Josephson, René Stulz, Ulf Axelson, Anil Kashyap, Marianne Bertrand as well as seminar participants at the CEPR/Banca D Italia conference on Financial Structure, Product Market Structure and Economic Performance and the University of Chicago for comments and suggestions.

2 There is a large empirical literature establishing the link between finance and growth (e.g. King and Levine (1994), Rajan and Zingales (1998)) and another distinct literature linking trade and growth (e.g. Frankel and Romer (1999)). In this paper, we propose that these two literatures are related. We find that one channel through which financial development may promote growth is exports. There is considerable evidence that exporting firms face large fixed costs. 1 Fixed costs for exporters include identifying potential markets, developing products to match foreign regulation and tastes, and development of marketing and distribution networks. Such fixed costs are intangible and difficult to observe, may have long gestation periods and are probably firm- or even person-specific. Hence, these fixed costs are a natural candidate for investments that are difficult to finance externally, especially when finance is underdeveloped. This suggests that export performance may depend on financial development. We test whether exporting performance is affected by financial development in a traditional gravity equation setting. Indeed, we find that good financial development increases merchandise exports significantly and by economically important magnitudes. Beyond standard trade determinants such as size, location and income, financial development helps explain the relatively poor trade performance of e.g. Austria, Egypt, Greece and Venezuela and the good performance of e.g. Finland, Norway, New Zealand and Malaysia. The effect is large: a one standard deviation increase in our preferred financial measure, accounting standards, corresponds to a 57% increase in exports. Naturally, exports vary in the amount of up-front costs they require. We identify four plausible proxies for fixed costs, based on country-pair differences and product characteristics. When a product is standardized, when it is exported to a neighboring or close by country, or to a country that shares a language with the exporter, fixed costs are likely to be lower. Hence fixed costs are easier to finance internally and require less outside finance. Financial development should then matter less. In further tests, we establish that this is indeed the case. Exports are more sensitive to financial development when fixed costs are high. For robustness, we also exploit time series variation in finance. In a group of tests, we use the classification of banking crises of Kaminsky and Reinhart (2001). We find that exports fall after (2004). 1 See e.g. Rogers and Tybout (1997), Bernard and Wagner (2001) and Bernard and Jensen 1

3 bank crises, and that those exports which require high fixed costs fall more than other exports. On average, exports fall by approximately 11% and the fall in differentiated (high fixed cost) exports is 14% larger than the fall in undifferentiated exports. In a final set of tests, we establish that finance affects how exports react to real exchange rate changes. Exports generally react slowly to exchange rates. Baldwin and Krugman (1989) and Dixit (1989) have highlighted fixed costs as an explanation for such sluggish exports. In fact, such sluggish responses to exchange rates can be exacerbated if access to external financing is limited. Even if a firm wants to enter, it may not always be able to secure sufficient funds to cover fixed costs. In panel data, we test whether the exchange rate elasticity of aggregate exports is higher in countries with well developed finance and find strong evidence that it is. In countries with better finance, export elasticities are approximately twice as large (at the one year horizon). To address endogeneity problems with exchange rates (i.e. shocks that affect both exports and exchange rates), we also study relative export growth to different importing countries. These regressions do not include the exporting country s exchange rate, and hence are robust to shocks that induce correlation between exports and the home exchange rate. In line with the aggregate results, we find that in countries with better finance, the allocation of exports across importing countries is much more responsive to exchange rates. The rest of the paper is organized as follows: section 1 provides some background on two key related literatures, finance and growth and sunk costs of trade. Section 2 introduces the theory underpinning our cross-sectional and time series tests. Section 3 describes the data and section 4 the empirical results from the basic cross-sectional and time series specifications. Section 5 addresses comparative advantage issues and other robustness tests. Section 6 presents the empirical methodology and results for exchange rate elasticity tests. Section 7 concludes. 1. Background and literature 1.1. Cross-country studies of growth and finance The recent wave of research on the growth implications of finance was initiated with crosscountry findings that growth rates are higher in countries with more developed financial sectors (King and Levine (1994)). Since it is difficult to control for all conceivable determinants of growth, the cross-country regressions used to establish these findings potentially suffer from omitted variable bias. 2

4 Recent work has circumvented the omitted variable problem by focusing on within country, between industry variation (e.g. Rajan and Zingales (1998), Claessens and Laeven (2003), Carlin and Mayer (2003), Fisman and Love (2003), Braun (2003)). 2 The cross-industry method has proven very successful, and the authors cited above have extended the empirical understanding of the links between finance and growth. For instance, Rajan and Zingales (1998) find that industries which are more in need of outside financing grow more quickly when the financial sector is more developed. By comparing industry rather than aggregate growth rates, they can include country fixed effects. One concern with the cross-industry empirical studies is that comparative advantage could explain the findings. For example, if plastics is growing faster than pottery in the US and slower (than pottery) in Turkey, this could be due to financial constraints, but also could be due to comparative advantage pulling the two countries toward specialization in different industries. Without a theory for which industries should grow where (i.e. a model of international trade and specialization), the conclusion that finance is restraining industry growth is open to some question. Several authors have addressed this issue. Rajan and Zingales (1998), for example, test explicitly whether human capital-driven comparative advantage can explain their findings. Whether or not this concern has been convincingly tackled, we are able to sidestep it. We first avoid comparative advantage concerns by looking at levels of exports (comparative advantage primarily affects the composition, not the level, of trade). Later, we introduce data on exports in different industries, and then carefully address comparative advantage issues. Our method has drawbacks. In particular, we require proxies for the up-front investment need of exporting firms. Some of the fixed cost proxies relate directly to trade: distance, for example, proxies for up-front investment needs, but is also related to transport costs. We attempt to address these potential problems in the robustness section. Our paper is not the first to investigate the effect financial development can have on trade rather than output. Two previous papers study the effect of finance on the composition of trade rather than its level. Vlachos and Svaleryd (forthcoming) suggest that finance may be a source of 2 Carlin and Mayer show that the growth rate of industries that are dependent on bank financing (as opposed to outside finance in general or equity finance) or skilled labor are sensitive to financial development. Braun finds that industries which make large investments in intangible assets grow relatively faster in a well developed financial system. 3

5 comparative advantage, and find that net exports in externally dependent industries are higher in countries with good finance. Beck (2002) finds that economies with better developed financial sectors export more in manufacturing, interpreting this as evidence for finance as a source of comparative advantage in manufacturing. He also finds that imports of manufactured products are strongly increasing in financial development, which at first glance seems at odds with the interpretation of finance as a source of comparative advantage. Beck s conclusion that finance confers a comparative advantage in manufacturing implies better finance should be a comparative disadvantage in non-manufacturing industries. By including a wider set of industries, we are able to show that finance increases the level of exports everywhere, including non-manufacturing industries, consistent with our proposed explanation based on fixed cost financing Fixed costs of trade Our explanation for why financial development plays an important role in facilitating exports hinges on the fact that exporting firms face large fixed costs. The existence of fixed costs for a potential exporter has long been recognized in the literature on international trade, even if the potential importance of financial development for financing them has not been emphasized. Fixed costs are key elements in theoretical work by Baldwin (1988), Baldwin and Krugman (1989) and Dixit (1989), who suggest that the sluggish response of import penetration to exchange rates may be due to such fixed costs. More recently, Melitz (2003) addresses the effect of fixed costs on firm composition in exporting industries. Hausmann and Rodrik (2002) point to the inefficiencies that may arise if private incentives to invest in export development are lower than the social benefits. There is firm level evidence of the fixed costs of exporting. Roberts and Tybout (1997) find that a firm s current exporting status is largely determined by its previous export experience. Prior exporting experience increases the probability of exports by up to 60% in their sample of Colombian firms. They infer that sunk costs must play an important role in a firm s decision to export. Similar findings have been reported for other countries (see e.g. Bernard and Wagner (2001) for Germany and Bernard and Jensen (forthcoming) for the US). Detailed information about the fixed costs faced by exporting firms in developing countries is also being gathered by the World Bank. The World Bank Standards and Trade Database is a survey of 690 exporting firms in developing countries across a wide range of industries that investigates the costs imposed on exporting firms by different impediments to trade. These costs include R&D expenditures, marketing costs, distance to markets, translation costs, and other fixed 4

6 exporting costs. The survey also inquires about product standards, government standards and other technical barriers to trade faced by exporting firms. The survey has not been finalized. 2. Theory and methodology: cross-sectional and time-series tests 2.1. The gravity equation We use the gravity equation for international trade as our empirical starting point (see e.g. Linneman (1966)). This equation captures an empirically robust relationship linking trade volumes to exporter GDP and importer GDP. It often includes one or several measures of distance. For an overview, see Frankel (1997). A typical empirical specification of the gravity equation states that total trade between two countries i and j will be lnt ij = α + β 1 y i + β 2 y j +β 2 lnd ij + e ij where T ij is trade, y i is the log of country i s income, and D ij is the distance between the two countries. If there is no particular prediction about balanced vs. unbalanced trade flows, the symmetric equation applies to total trade. Other geographic variables, such as country size, common borders and country population are sometimes included (see e.g. Frankel and Romer (1999) regarding geographical determinants of trade). Our regressions separate trade flows by their direction, i.e. the dependent variable in our regression is bilateral exports rather than total bilateral trade. We also include a variable measuring financial development in the exporting country. Our basic specification for the exports from country i to country j is log[1+e ij ] = α + β 1 y i + β 2 y j +β 3 lnd ij + γ F i + e ij (1) 2.2. Fixed costs to export and the need for outside finance We depart from the standard gravity equation by including financial development as an explanatory variable. We propose that financial development is important for exports because exporting activity involves up-front investment that cannot be financed internally. There are several reasons why these exporters sunk costs are large and difficult to finance. The investments are made long before any export revenue is collected and provide limited collateral compared to e.g. machine or real estate investments. To the extent that foreign investments are tangible, they 5

7 may still be harder to seize than domestic assets. Revenues from abroad may be difficult to verify for outsiders, and be more difficult to extract from the firm. Therefore, export revenues may be more difficult to pledge to outsiders than domestic revenues. Finally, export revenues may be volatile and difficult to predict, for both firm insiders and outsiders. For all of the above reasons, it may be very difficult to secure outside financing for export investment. A sophisticated financial system makes it easier to finance these intangible investments since lenders have access to more information, contracts are enforced more reliably, and financial intermediaries are more capable of assessing potential risks and rewards. Even if one accepts that certain kinds of exports require large up-front investments, it is possible that countries that have poor finance do not have a comparative advantage in these products. If poor countries do not export products with high up-front costs for reasons other than finance, the impact on welfare from lost trade due to poor finance would be minimal. The evidence from the trade data suggests this is not the case, however. Poor countries have many export opportunities that require high fixed costs. For example, several tropical agricultural products have large markets far away. Across industries, manufactured goods that correspond to early industrialization are often in our high fixed cost category. Textiles and apparel and simple electronics are cases in point. For poor countries, taking advantage of these opportunities is likely to be pivotal for good export performance. If finance is constraining these exports more than exports of e.g. paper and pulp, they should react more to financial improvements Variation in sunk costs We cannot control for all country-wide variables that might affect exports. Some of these variables could be correlated with finance and hence bias our results. We get around this problem by studying the interaction of finance with variables that measure when fixed costs are high (i.e. where the need for external financing is high). We exploit variation both across country-pairs (i.e. how distant and dissimilar are the two countries) and across industries (i.e. how standardized are the products) to identify exports where the sunk costs are expected to be large. Since well developed finance should ameliorate the problem of financing sunk costs, the coefficient on the interaction of financial development and proxies for sunk costs should have a positive sign, i.e. better finance should have a greater marginal effect on exports when sunk costs are high. When we use interaction terms, we include fixed effects for exporter, importer and industry. This eliminates many potential sources of omitted variable bias. Reverse causality also seems less likely to explain 6

8 results where we identify specific country pairs and industries where finance has a larger effect on exports. We use four different proxies for up-front investment. First, the physical distance between exporter and importer is a proxy for overall difference between the economies, regulatory environment, culture etc. The second proxy is the common border dummy. We also use proxies of sunk costs which are unrelated to geography. The third proxy for sunk costs is a dummy for whether the official language for exporter and importer are the same. A common language facilitates communication, eliminates translation costs, and may imply that certain products may be exported with minor adjustments to domestic output (e.g. software). Finally, we use a proxy based on Rauch s (1999) measure of whether industry output has standardized prices or consists of differentiated products. Rauch assigns each SITC industry to one of three categories, differentiated products (e.g. footwear), goods with reference prices 3 (e.g. Polymerization and Copolymerization Products, SITC 583) and exchange-traded goods (e.g. Lead). For the first category, he argues that the uninformativeness of prices prevents globally scanning traders from substituting for organized exchanges in matching international buyers and sellers. In industries with output of many different varieties and no established prices, it will be more difficult to identify and develop export opportunities. Hence, we predict that the costs of identifying and developing profitable trade opportunities are higher. None of our proxies is immune to criticism, and alternative explanations can be imagined for each one of them. We stress the consistent results they give, and attempt to carefully control for alternative explanations of our findings. 3. Data Bilateral data Data on international trade in merchandize is available at the industry level under the SITC-classification system. We use bilateral trade data from the Statistics Canada World Trade 3 Rauch (1999) describes why this category of goods are more transparent than differentiated goods: prices can be quoted for these products without mentioning the name of the manufacturer, and these reference prices are found to be sufficiently useful by industry actors to be worth listing in trade publications. For example, a price per pound of Polyoxyethylene Sorbitan Monostearate is quoted weekly in Chemical Marketing Reporter on the basis of surveys of suppliers. 7

9 Database, which provides data on annual bilateral trade flows from at the 4 digit SITC level and using the U.S. Bureau of Economic Analysis industry classification system (See Feenstra et al 1997). There are more than 170 countries, but we only have financial variables for at most a hundred countries, and for many regressions we have fewer exporters than importers. Most of the analysis in this paper uses total bilateral exports (across all industries), as in standard gravity equations. For cross-sectional regressions, we use 1995 as our base year, but have re-run regressions for 1985 with similar results. For exchange rate elasticity regressions, the whole panel from 1970 to 1998 is used. While country-pair differences are a good source of variation in investment needs, we also use industry variation as an alternative source of variation, based on Rauch (1999). He categorizes industries at the level of four-digit SITC codes based on whether the output is exchange-traded, reference-priced, or differentiated (i.e. no standard prices). We use a modified version of his measure in this paper. Of Rauch s three categories, only firms that produce differentiated products are likely to face significant up-front costs in adapting their product for export. Hence we define a dummy variable equal to one if a product is differentiated and zero otherwise. We calculate total bilateral trade in differentiated and undifferentiated goods. The (log) difference between these two is the dependent variable in some regressions. 4 We include several controls commonly used in estimation of gravity equations. For each country-pair, we use a dummy variable equal to one if they share a land border and a dummy equal to one if they share an official language. These data are from Jon Haveman s website. 5 We also use distance in kilometers, as measured between largest cities, reported in Fitzpatrick and Modlin (1986). These three variables are also used as proxies for up-front investment need. Finally, from Rose (2003) we get dummy variables for mutual membership of a regional trade agreement (these are the EU/EEC, US-Israel FTA, NAFTA, CARICOM, PATCRA, ANZCERTA, CACM, Mercosur, ASEAN and SPRTC), and data on colonial history: we use a dummy equal to one if exporter and importer were either involved in a colonial relationship or had the same colonizer. 4 To provide some intuition of differentiation, appendix table A1 presents the fraction of world trade in each of 34 BEA industry classifications that is differentiated. 5 ta.html 8

10 Country data For each exporting and each importing country in our sample, we need to measure financial development, and controls such as GDP, area and population. We use GDP and population numbers from the Summers-Heston data set. Country size (in million acres) comes from The Universal Almanac We get data on human and physical capital per capita from Hall and Jones (1999). They calculated a human capital measure which is piecewise log-linear in average years of schooling (from on Barro and Lee (1993)); their capital measure is based on the perpetual inventory method. There is no consensus on how to measure financial development across countries. As our primary variable, we use accounting standards, a measure of the quality of accounting in a country. The Center for International Financial Analysis and Research (CIFAR) created an index for different countries by rating the annual reports of at least three firms in every country on the inclusion or omission of 90 items. Comprehensive data on the measure dates to 1990, and is discussed in Rajan and Zingales (1998). The advantage of this measure is that it proxies for the amount of external financing available, rather than the amount actually given. An alternative to accounting standards is some measure of the stock of actual financing. We use the ratio of stock market capitalization to GDP, and the ratio of credit to the private sector to GDP, both reported by Beck, Demirgüç-Kunt and Levine (2001). 6 More countries report data for these measures than for accounting standards, but they are more likely to be subject to reverse causality problems. Of the different finance measures, we prefer accounting standards, and only report specifications with the alternative measures to establish robustness. Exporting country measures of financial development are reported in Table 1 with bilateral exporting volumes. Our instruments for financial development, dummies for legal origin (UK, French, German, and Scandinavian) and a measure of settler mortality in those exporting countries which are former colonies, come from La Porta et al (1997, 1998) and Acemoglu et al (2001) respectively. Time series variation in finance comes from banking crises as defined by Kaminsky and Reinhart (2001). They report banking crises during the period. The starting date is given for the full set of crises, and we use this to define a dummy equal to one when an exporter experiences the onset of a banking crisis (29 events). We also define a dummy variable equal to one sources. 6 These authors use raw data from the IMF s International Financial Statistics and country-specific 9

11 during the following three years, assuming that some crises were not resolved for a few years after the onset. Kaminsky and Reinhart also provide peak dates for a subset of crises when there was a simultaneous balance of payment crisis ( twin crises ). We define a dummy variable equal to one for all years when Kaminsky and Reinhart report an ongoing crisis (i.e. from the on-set to the peak). This alternative measure covers a smaller set of crises (about half) but contains more information on the duration, and takes on the value of one for 83 exporter-years. We use both measures. Table 2 reports summary statistics for our proxies of up-front investment costs for the whole sample. The sample actually used in the regressions varies somewhat depending on data availability for individual variables. 4. Results 4.1. Financial development and export volumes We now estimate the gravity equation of bilateral exports with different measures of financial development. Column one of Table 3 presents results with accounting standards as the measure of financial development. Exporter and importer controls, included but not reported, include GDP, population, and area. 7 Bilateral controls include log of distance, and dummy variables for common land border, common main language, membership in a common regional trading agreement and shared colonial history. 8 Accounting standards enter positively and significantly. The magnitude of the coefficient is economically important. An increase in accounting standards of one standard deviation predicts an increase in the log of export volume by 0.45, which corresponds to a 57% increase in exports. Moving from the 25 th percentile of accounting standards (54, Brazil) to the 75 th percentile (70, New Zealand and South Africa) predicts an increase in the log of export volume by 0.51 or an increase in exports by 67%. Columns two and three report the same regression with alternative measures of financial development: private credit over GDP and stock market capitalization over GDP. The coefficients 7 The coefficients are as expected for GDP (positive) and distance (negative). Larger and more populous countries trade less, similar to the results in Frankel and Romer (1999). 8 Because data on bilateral tariff levels in suitable format are not available, we did not include tariffs. Including average (unweighted) tariff levels for each country had no effect on any other coefficients but reduced the number of observations, and we do not include tariff data in any reported specifications. 10

12 for the alternative measures of financial development are positive and statistically significant at the 1% level, although the magnitudes vary somewhat. 9 The sample sizes differ depending on data availability for each measure of financial development. In subsequent regressions, we focus on accounting standards, since it is a priori more attractive and the results are similar across the various measures Fixed effect regressions using proxies for up-front costs Although the regressions in Table 3 suggest that finance may play an important role in promoting exports, they do not control for all possible country-specific variables that may be correlated with financial development (and might be the actual determinants of exports). We attempt to circumvent this problem by exploring the interaction of financial development and potential need for external finance. First, we test the hypothesis that financial development has a larger impact on exports when sunk costs are higher by including an interaction of finance and a proxy for sunk costs. All exporter and importer country specific measures are dropped in these specifications since they are absorbed by the country fixed effects. Second, we regress the log difference in differentiated and undifferentiated exports on accounting standards in order to test the hypothesis that countries with better finance are able to export a larger fraction of differentiated goods (which we propose require higher up-front investment). Implicitly, any fixed country effects that affect both types of trade similarly are differenced out. For each proxy of fixed costs, we expect the interaction with finance to enter in the following way: if the proxy is associated with higher dependence on external finance because of large sunk costs (distance) we predict a positive sign (high costs imply more need for finance). For the proxies that predict low sunk costs (common border, common language), we predict a negative sign. Table 4, column one reports the results using distance as the proxy for up-front costs. Our hypothesis predicts that the interaction coefficient should be positive, i.e. finance should be more important for trade when the two countries are farther apart (because up-front costs are higher). We find that the interaction coefficient is positive and significant at the 1% level. To judge the magnitude, the coefficient can be compared to the average coefficient on accounting standards (Table 3, column one). An increase of finance by one standard deviation (14.2) would increase 9 The implied increase in export volume for a one standard deviation increase in financial development is 44% for private credit and 100% for market capitalization. 11

13 exports from Egypt to the United States (distance 9042 km) relative to exports from Egypt to Sudan (distance 1606 km) by 94%. Common border and common language are used as the proxies for up-front costs in Table 4, column two and three, respectively. The interaction effect should be negative for both common border and common language since the impact of finance on exports is predicted to be lower when up-front costs are lower. The interaction coefficient is significantly negative in both cases. Finally, Table 4, column four, reports the results of the regression of the log of differentiated minus undifferentiated products on our measure of financial development. If differentiated products are more difficult to finance than undifferentiated products, we expect that countries with better finance should export a larger fraction of differentiated goods. Hence we expect the accounting standard coefficient to be positive. Fixed effects of the kind used in columns one to three, which capture country-wide effects on exports, would be cancelled out by the differencing, hence we exclude them. Bilateral variables are included, but all are insignificant (i.e. distance has a similar effect on differentiated and undifferentiated trade). As predicted, the coefficient of accounting standards is positive and significant. Exports of differentiated product categories are more sensitive to financial development than undifferentiated exports, consistent with the sunk cost prediction. These interaction and difference results could in theory be related to comparative advantage. We explore this possibility in the robustness section below Banking crises We use banking crises as a source of time series variation in finance, an approach also used by Laeven et al (2002). The source of crisis data is Kaminsky and Reinhart (2001), who document banking and balance of payments crises during the 1973 to 1997 period. Some of the bank crises included in the sample are those in Spain in 1978, Finland in 1991, Brazil in 1994, and Malaysia and the Philippines in During bank crises private credit growth falls, bank deposit growth is negative, real interest rates rise, stock markets fall and the availability of outside finance is likely to be severely constrained. Meanwhile, real exchange rates tend to drop, which should result in higher exports (and lower imports). For an overview of this macroeconomic performance, see 12

14 Kaminsky and Reinhart (2001). 10 We exploit the fact that diminished availability of external finance should disproportionately hurt exports in industries with differentiated products, which have larger up-front costs. Table 5 reports the effect of banking crises on export performance. We regress bilateral export growth on exporter-importer dummies (allowing country-pair specific trends in exports) over the period. Because export growth rates are strongly autocorrelated, we allow firstorder error autocorrelation, and estimate coefficients using GLS. We use two alternative bank crisis variables. Table 5, column one, reports the export growth rate in the year a banking crisis hits (the narrow measure). The effect is negative and significant (-11.5% average export growth above trends). In column two, we include a dummy for the following three years. The negative export growth continues, implying that the export performance worsens further for a few years after the crisis first hit. In column three, we use the alternative broad measure of banking crises: a dummy equal to one during the period from crisis on-set to peak. This dummy covers fewer crises, but does contain more information on their duration. The result is again negative and significant. The coefficient magnitude is not directly comparable, because the dummy captures a larger part of the whole banking crisis episode (ranging in duration from one to six years). The coefficient is negative in all specifications, suggesting that the direct impact of the banking crisis on exports is more important than any improvement in the terms of trade. In columns four to six, we repeat the above specifications for the difference between differentiated and undifferentiated export growth. If lack of fixed cost financing during a crisis affects differentiated exports more than undifferentiated, we expect to see negative and significant effects on the difference. Indeed, that is what we find. Except for the year one to three dummy, the effects are significant and negative. The growth rate of differentiated exports is 14% lower than the growth rate of undifferentiated exports in the year immediately following the onset of a crisis compared to non-crisis years. To the extent that a banking crisis captures worsening credit availability, but no commensurate change in other institutions, these results suggest that poor external finance may in 10 There are several studies of the firm level effects of a banking crisis. For one example, with evidence of the credit crunch in Thailand following the East Asian crisis, see Domaç and Ferri (1999). 13

15 fact be responsible for poor export performance. At the very least, this result reduces the set of institutions which can plausibly be argued to explain our findings Importer s finance While research has shown that much financing occurs locally (see e.g. Jayaratne and Strahan (1996)) and access to international financing may be limited for all but the largest of firms, some fixed costs can perhaps be financed where the investment activity takes place - in the importing country. We can test this prediction directly by including a measure of importer s finance in the gravity equation (as in Table 3). We expect a positive effect of importing country finance on exports. Table 6, column one and column two report the effect of exporter s and importer s finance respectively on the level of finance in the structural gravity equation. Importer s finance does have a positive effect, but it is much weaker than exporter s. This result seems plausible if most of the investment has to be undertaken by exporting firms, and if most investments must be financed locally. In column three, both importer s and exporter s accounting standards are included, and both remain significant. The sample of bilateral pairs for which we have data on both exporter s and importer s finance is relatively limited. We therefore rerun the regressions in column four and five with private credit, the most widely available measure of finance. The results are similar: there is a positive effect of importer s finance, but it is weaker than for exporter s finance. 11 These results suggest while importer s financial development can help exports, local finance in the exporting country is far more important. Incidentally, these results also suggest that comparative advantage is unlikely to explain the positive effect of finance across the board. If financial development is proxying for some comparative advantage (so that exports are increasing in exporter s financial development) then they should also be decreasing in the financial development of the importer. The next section explores the question of comparative advantage more directly. 11 We also interacted both importer s and exporter s finance with cost proxies and found that both importer s and exporter s finance enter positively and significantly (results not reported). 14

16 5. Robustness tests 5.1. Comparative advantage International trade theory has traditionally focused on comparative advantage as a determinant of the patterns of trade across countries. The standard explanation for why a country exports more in a certain industry is that it has a comparative advantage in that industry, not because financial development helps to finance sunk costs in exports. It is natural to ask whether our results can be explained by a more traditional comparative advantage story. One potential explanation for our results in the structural gravity equation (Table 3) that can be immediately ruled out is that countries with high levels of financial development have a comparative advantage in the industries represented in the sample data. Since our trade data covers most of world trade in goods, the comparative advantage explanation would require countries with high levels of financial development to have a comparative advantage in all goods, which is impossible. Our interaction specification (Table 4) is more susceptible to comparative advantage problems. We therefore repeat the fixed effects regressions and include interactions of our cost proxies with standard comparative advantage determinants: GDP per capita, human capital per capita, physical capital per capita, and land per capita. Table 7, Panel A reports the OLS regression of exports on exporter and importer fixed effects as well as distance, common border, common language, regional trading area and colonial history (these coefficients not reported) and interactions. The interaction of accounting standards with cost proxies is compared to interactions with the comparative advantage variables. Including these determinants reduces the significance of the distance interaction (columns one and two), but no other variable (except land per capita) interacts significantly with distance either, suggesting that colinearity rather than misspecification is the issue. The interactions with common border (columns three and four) and common language (columns five and six) are slightly changed (one up, one down) but still significant. Land per capita interacts with the border variable, but all other comparative advantage interactions are insignificant. Table 7, panel B, presents results for differentiated vs. undifferentiated exports (as in Table 4, column four). Including comparative advantage controls reduces the coefficient s magnitude and significance (it is still significant at the 5% level). In this case, all the controls have significant 15

17 effects on the explanatory variable. Some of the signs are surprising, however: both physical and human capital enter negatively. The comparative advantage results show that finance is essentially unaffected in two of the cases (border, language), has a smaller coefficient but remains significant in one case (differentiation), and loses significance entirely in the case of distance. In the problematic case, none of the added controls are significant, so omitted variable bias was unlikely to be a problem. Overall, the results suggest that while precise estimation is not entirely feasible, the overall case can be made that high fixed cost exports are more sensitive to finance and that this does not reflect standard sources of comparative advantage Instrumental variables estimates One possible problem with our results is reverse causality. Finance may improve in order to accommodate expected opportunities. Such opportunities could include exports. To rule this out, we use two instruments which have been shown to correlate with institutions over the long term. The first instrument is legal origin from La Porta et al (1997, 1998), which correlates with many measures of financial development. Secondly, we use a measure of European settler mortality in former colonies from Acmeoglu et al (2001), which has been shown to capture permanent features of the institutional environment. For all three interactions (distance, border, language) we instrument the interactions with the interaction of cost proxies with legal system or settler mortality. 12 The results from these IV regressions are reported in Table 8 next to the earlier OLS. In all cases, standard errors go up somewhat, but coefficients remain significant. These results rule out one possible reverse causality problem: that recent opportunities to export have driven the observed variation in finance Censored regressions Because some country pairs have zero bilateral trade (approximately 10% of the cases where we have all variables) it may be important to distinguish between the effect of trade determinants on the probability of having positive exports, and the effect on the level of exports. We have no variables that can plausibly be argued to explain only one of these aspects of trade data (which could allow separating effects), but we can specify the econometric model in a more rigorous manner. In order to do this, we assume that there is an underlying (unobserved) desired 12 We also tried IV specifications of the structural gravity equations, with similarly stable results. 16

18 trade level which ranges over the whole real line, and which is observed if it is positive. When the underlying dependent variable is negative, we observe zero. Our Tobit specification for the exports from country i to country j is log[1+e ij ] = α + β 1 y i + β 2 y j +β 2 lnd ij + γ F i + e ij if RHS > 0, otherwise log[1+e ij ] = 0 (1 ) The results of Tobit regressions in a structural gravity setting, allowing for heteroskedasticity, is reported in Table 9. Column one to three report the OLS estimates (from Table 3), and column four to six Tobit estimates. In all three cases, the coefficients are similar in magnitude and significant in both cases. Tobit FE specifications are not estimated, because of possible bias problems induced by dummies. 6. Finance and the exchange rate elasticity of exports 13 If exporting firms must incur fixed costs to export, the time-series pattern of trade should reflect this. Hence, looking at time-series evidence allows additional tests of the general theory that financing problems limit exports. Furthermore, the time-series pattern of exports is an object of study in its own right (see e.g. Obstfeld and Rogoff (1995)). We propose that the quality of financial intermediation could affect the speed and scope of adjustments of exports to shocks. We chose to focus on one of the most obvious shocks to the export opportunities: exchange rate movements. For individual firms, exchange rate swings should affect the profitability of exporting. If there are fixed costs, firms do not always respond to changes in exchange rates. For example, when the home exchange rate is low, it may be better to continue exporting at a negative profit in order to retain a foothold which is otherwise lost. This argument was developed by Baldwin and Krugman (1989) and Dixit (1989). In fact, financing problems are likely to make the problem worse. When exchange rates are high, even firms who wish to enter may be unable to do so if they cannot access sufficient financing. When exchange rate is low and exports unprofitable, firms may be even more reluctant to exit, since re-entering is difficult Export elasticities: empirical findings We now test the main time-series implication of fixed costs: countries with more developed financial systems should have larger responses to exchange rate changes. Differences in the speed of adjustment of exports to exchange rates may be more dramatic at short horizons, but there are 13 This section presents preliminary work. 17

19 seasonality problems with high-frequency data which are avoided with the annual trade data. We therefore use annual export numbers from the same Statistics Canada data set from which the cross-sectional data is taken. All countries for which there is data on accounting standards and annual trade numbers for the whole period are included in the sample. The dependent variable is annual merchandise export growth and the independent variable is exchange rate movements. Exchange rate movements are in real terms and trade-weighted. The period covered is 1980 to Table 10 reports three alternative specifications. In column one, export growth is regressed on lagged depreciation and lagged depreciation interacted with finance (lagged depreciation times demeaned accounting standards) and fixed country effects. The coefficient on depreciation is positive as expected and significant. The parameter estimate of 0.19 implies that a ten percent depreciation causes a 1.9 percent export increase in the following year. The interaction is also positive and significant, suggesting that countries with better finance respond more to short-term movements in the exchange rate. The magnitude is large: an increase in accounting standards of 10 units (e.g. from Greece, 55, to Japan, 65) increases the elasticity by 0.10, or about half of the average coefficient. For a country with accounting standards at the 25 th percentile (Brazil, 54) the one-year elasticity implied by the regression in column one is 0.09 whereas a country at the 75 th percentile (70, New Zealand and South Africa) the elasticity implied is We next estimate specifications more closely aligned with the model. Column two includes the previous year s depreciation as a control while column three reports a regression including lagged export growth. The coefficient on lagged depreciation is insignificant while lagged export growth has a negative and highly significant coefficient. The effects of depreciation and the finance interaction are still significant in both specifications. These results suggest that countries with better financial systems respond more to changes in export opportunities, as captured by exchange movements. However, we have not controlled for factors that may simultaneously affect both exports and exchange rates. We address the issue of endogenous exchange rates in the next subsection Cross-country allocation of exports Generally, how can we handle the fact that exchange rates are affected by e.g. productivity shocks which also affect export performance directly? We us a simple approach: looking at the relative growth of exports to different countries. This difference approach is easy to interpret. In 18

20 effect, we ask whether exports from country A to importer B grow faster than exports from A to C, when the exchange rate of B appreciates relative to the exchange rate of C. Let q i be exports to country i, and x i the home country s exchange rate vis-à-vis country i. Suppose that some productivity shock a t affects exports q t and that it is correlated with exchange rate movements x t. q it = α x + β a (2) it t Since a and x are correlated, omitting the shock a t causes a bias in the estimated coefficient for x t. However, differencing exports to i and to j will eliminate the shock: q it q jt it jt t t ( x x ) = α x α x + β a β a = α (3) it jt Equation (3) relates export growth to country i minus export growth to j to the difference in exchange rate changes. This difference between exchange rate changes is simply the change in the i-to-j exchange rate. Any general exporter-shocks are now eliminated. The intuition behind this test is very straightforward: if the Yen rises against the dollar, exports to Japan should grow faster than exports to the US (for any exporter). We examine how the elasticity of relative exports varies with finance. All export growth rates are normalized with export growth rate (from the same exporter) to the US, and we use importing country-us dollar exchange rate movement as independent variable. We use annual data on exports to OECD countries 14, a sample of importers whose real exchange rates are probably well measured (since consumption baskets are similar), who are big importers, and who have reasonably similar import composition. Focusing on these countries should minimize possible problems with demand elasticity-variation across industries and results driven by cases of dramatic export growth from a small base. 14 The Czech Republic, Poland, the Slovak Republic and Hungary are not included in the trade data and are excluded. We tried excluding Mexico and South Korea, who are the only new OECD members since the early seventies (apart from the Eastern Europeans), as well as being the two poorest OECD economies, but this did not change results. 19