The Private Export Credit Insurance E¤ect on Trade Koen J.M. van der Veera a

De Nederlandsche Bank, The Netherlands, P.O. Box 98, 1000 AB, Amsterdam, [email protected], tel: +31(20) 524 5836, fax: +31(20) 524 2506. February, 2013

Abstract International trade relies on trade …nance (credit or insurance) by …nancial institutions. Evidence on the link between speci…c forms of trade …nance and trade is scarce, however, because detailed data on trade …nance is hard to come by. This paper …nds evidence of a link between private export credit insurance and exports. I use a unique bilateral data set which covers the value of exports insured – by one of the world’s leading private trade credit insurers –from 25 exporting countries to 183 destination countries in the period from 1992 to 2006. This panel data set allows me to use well-known econometric techniques to address the endogeneity of insured exports with respect to total exports. Applying a variety of trade models, I consistently …nd a positive and statistically signi…cant e¤ect of private export credit insurance on exports. The results suggest that the impact of private export credit insurance on international trade is larger than the value of exports insured.

JEL codes: F10, F14, G01, G20, G22. Keywords: trade …nance, export credit insurance, international trade, trade credit

Acknowledgements: I am especially grateful to Andrew Rose for his advice and numerous discussions throughout this project. I thank Martin Admiraal and Henk van Kerkho¤ for their help with collecting the data. For helpful suggestions on an earlier draft, I thank Peter Egger and Christoph Moser. I also thank Paul Becue, Gabriele Galati, Marco Hoeberichts, Eelke de Jong, Pierre Lafourcade, Iman van Lelyveld, Jeromin Zettelmeyer and seminar participants at the World Bank DECTI Trade Seminar, EBRD, ECB Workshop on Trade and Competitiveness, the 16th International Conference on Panel Data and De Nederlandsche Bank for comments. I thank the private trade credit insurer for kindly providing data and their sta¤ for fruitful discussions. I bear full responsibility for any remaining errors. The views expressed in this paper are those of the author and do not necessarily represent those of the Dutch central bank.

1

Introduction

Financial institutions play an important role in facilitating international trade. According to estimates by the International Monetary Fund, about 40 to 50 percent of world trade relies on some form of bank-intermediated trade …nance, public or private export credit insurance (see Asmundson et al., 2011). The importance of banks in supporting international trade has been well-established in recent studies. For example, in a key contribution on bank-intermediated trade …nance, Amiti and Weinstein (2011) relate …rms’ export performance to the health of the banks providing trade …nance and show that …nancial shocks are transmitted from banks to exporters.1 Likewise, there is also growing evidence that public export credit agencies are e¤ective in stimulating exports (Egger and Url, 2006; Moser, Nestmann and Wedow, 2008; Felbermayr and Yalcin, 2011). However, except for a recent contribution by Auboin and Engemann (2012), who …nd a positive e¤ect of a combined measure of publicly and privately insured trade credits on trade, little is known about the role of private trade credit insurers in world trade.2 This paper contributes to …ll this gap in the literature on trade …nance and examines the tradepromoting role of private trade credit insurance, a speci…c form of trade …nance that covered an estimated EUR1.84 trillion of international and domestic trade in 2011 (ICISA, 2012). I exploit a unique bilateral data set on the worldwide activities of a leading private trade credit insurer and …nd direct evidence of a positive e¤ect of private export credit insurance on exports. Importantly, the results suggest that there is a trade multiplier of private export credit insurance; every euro of insured exports generates more than one euro in total exports. Private export credit insurance is a useful tool for an exporter selling goods on credit to insure against the risk of nonpayment by an importer.3 It is distinct from freight insurance that covers the risk of loss or damage to goods in transit. A private insurance policy generally covers commercial and political risk. "Commercial risk" refers to nonpayment due to default or insolvency and "political risk" relates to nonpayment as a result of action by the importer’s government (i.e. intervention to prevent the transfer of payments, cancellation of a license, or acts of war or civil war). Jones (2010, p.9) explains the need for private trade credit insurance as follows:

"The need for trade credit insurance arises from the common practice of selling on credit and the demand by buyers to trade on open account, where they only pay for the goods and services after having on-sold them and are not willing to provide any form of 2

security, for example by way of full or partial advance payment, bank guarantee or letter of credit."4 I use a gravity model to test whether private export credit insurance stimulates trade and consistently …nd a positive and statistically signi…cant e¤ect. The data set includes yearly observations on the bilateral value (aggregated at the country level) of insured exports from 25 exporting countries to 183 destination countries covering the period from 1992 to 2006.5 The bilateral dimension of the data is an advantage compared to the destination country level dataset used by Auboin and Engemann (2012). It allows me to estimate an empirical gravity model consistent with a theoretical gravity equation as derived by, e.g., Anderson and van Wincoop (2003). Hence, I include time-varying multilateral resistance terms and account for unobserved bilateral heterogeneity simultaneously, avoiding the estimation biases highlighted by Anderson and van Wincoop (2003) and Baldwin and Taglioni (2006). The benchmark model adjusts econometrically for the endogeneity of insured exports with respect to total exports, but I also show results using the insurer’s claims over premium income as an instrumental variable for insured exports. The data covers the insurance provided by one of the "Big Three" private trade credit insurers, that together covered 87 percent of the world market in 2010: Euler Hermes (36%), Atradius (31%) and Coface (20%). As the data does not include information on insurance supplied by other insurers, the magnitude of the estimated trade multiplier of private export credit insurance must be interpreted with some caution. For a variety of samples, the results suggest a trade multiplier of private export credit insurance in the range of 1.3. This trade multiplier would imply that every euro of privately insured exports generates about 1.3 euro of total exports. The …nding of a trade multiplier above one indicates that the role of private trade credit insurers in supporting world trade is larger than the value of exports they cover. Apparently, they do not only facilitate exports covered by export credit insurance (in that case, the trade multiplier would be one), but also seem to stimulate non-insured exports. Several hypotheses from the literature could explain this outcome. First, the trade multiplier supports the idea that the reduction in risk due to a trade credit insurance policy increases exports to markets where a …rm would not sell otherwise (Funatsu, 1986). Also, following the recently developed theory of trade …nance by Antràs and Foley (2011), export credit insurance allows exporters to learn about the creditworthiness of importers, reducing the need to use insurance coverage after repeated transactions. Furthermore, export credit insurance

3

gives importers access to supplier credit, which reduces their transaction costs (Ferris, 1981) and could stimulate their import demand. This channel would further support the trade multiplier if, as Becue (2008) suggests, other suppliers use private insurance cover as a signal on the creditworthiness of an importer, improving the importer’s overall access to supplier credit. Finally, if insuring accounts receivable gives exporters better access to external …nance (Becue, 2008; Jones, 2010), this could add to a higher export level. This article builds on the literature on the trade-promoting e¤ect of public export credit guarantees. Two important contributions are Egger and Url (2006) and Moser, Nestmann and Wedow (2008) who …nd that Austrian and German public export credit guarantees stimulate trade in the long run. Private export credit insurance, however, di¤ers from the guarantees provided by public export credit agencies. A key di¤erence is that private export credit insurance mostly covers shortterm credits with a tenure of 60 to 120 days, while public guarantees generally cover projects with a duration between 2 and 5 years, where the actual shipment of the good usually follows a few years after the public provision of insurance cover. As a result, the trade multiplier of public guarantees found in previous studies needs some time to take place, whereas the results in this paper show that private export credit insurance stimulates trade in the short run.6 Another di¤erence relates to country coverage. Private insurers traditionally cover risks on trade between …rms in OECD countries, whereas governments mainly cover risks on exports to high risk countries. The results in this paper thus show that private export credit insurance is also, and in particular, important in stimulating short-term trade ‡ows between …rms in developed countries. In what follows, I discuss the empirical methodology and how the benchmark model deals with possible sources of endogeneity bias (Section 2). In Section 3, I present the main results and go into the mechanisms that could explain the trade multiplier. Section 4 provides sensitivity checks, and Section 5 concludes.

4

2 2.1

Empirical Methodology Speci…cation and Data

To estimate the private export credit insurance e¤ect on exports, I rely on the standard "gravity" model of bilateral trade. The gravity model explains trade between a pair of countries with the distance and their economic "masses". I augment the basic speci…cation with a number of conditioning variables that might also a¤ect bilateral trade, such as currency unions (Glick and Rose, 2002) and trade agreements (Rose, 2004). I initially assume that all regional trade agreements have the same e¤ect on trade, but relax this assumption below. I employ the following speci…cation:

ln(Expijt ) =

0

+

1 ln(Dij )

+

7 (Langij )

+

+

13 (Colonyijt )

+

2 ln(P opit )

8 (RT Aijt )

+

+

+

3 ln(P opjt )

9 (Borderij )

14 (EverColij )

+

+

+

4 ln(GDP pcit )

10 (Islandsij )

15 (SameCtryijt )

+

+

+

5 ln(GDP pcjt )

11 ln(Areaij )

1 ln(InsExpijt )

+

+

6 (CUijt )

12 (ComColij )

+ "ijt :

where i denotes the exporting country, j denotes the importer, t denotes time, ln(:) denotes the natural logarithm operator, and the variables are de…ned as: Exp denotes real FOB exports from i to j, measured in euro, D is the distance between i and j, P op is population, GDP pc is annual real GDP per capita, CU is a binary dummy variable which is unity if i and j use the same currency at time t, Lang is a binary variable which is unity if i and j have a common language, RT A is a binary variable which is unity if i and j have a regional trade agreement at t, Border is a binary variable which is unity if i and j share a land border, Islands is the number of island countries in the pair (0/1/2), Area is the log of the product of the areas of the countries, ComCol is a binary variable which is unity if i and j were both colonized by the same country, Colony is a binary variable which is unity if i colonizes j at time t (or vice versa), EverCol is a binary variable which is unity if i ever colonized j (or vice versa), SameCtry is a binary variable which is unity if i is part of the same country at time t,

5

InsExp denotes real privately insured exports from i to j, measured in euro, " represents the omitted other in‡uences on bilateral exports, assumed to be well behaved. The parameter of interest is

1.

This represents the private export credit insurance e¤ect on

exports holding other export determinants constant through the gravity model. I estimate the equation with OLS, using a robust covariance estimator (clustered by country-pair dyads) to handle heteroskedasticity, adding year-speci…c …xed e¤ects. I also adjust this speci…cation in two important ways. First, I add a comprehensive set of country-pair …xed e¤ects (i.e., a mutually exclusive and jointly exhaustive set of {

ij }

intercepts) to absorb any time invariant characteristics that are

common to a pair of countries. Second, I also add comprehensive sets of time-varying exporter and importer …xed e¤ects (i.e., sets of {

it }

and {

jt })

to take account of any time variant country-speci…c

factors. This …nal model accounts for multilateral resistance and unobserved bilateral heterogeneity simultaneously (see i.e. Baldwin and Taglioni, 2006). Moreover, I argue below that these …xed e¤ects also adjusts for endogeneity bias particular to estimating the private export credit insurance e¤ect on trade. I also show that the key results are robust to the use of other estimation strategies. The sources of the bilateral data set are described in Appendix Table A1. This data set includes annual observations between 1992 and 2006 (though with many missing observations) for some 183 territories and localities (I refer to these as "countries" below). The countries are listed in Table A2. A correlation matrix for the variables used in the regression analysis is presented in Table A3. 2.1.1

Data on Private Export Credit Insurance

The data on privately insured exports is the novel part of the data set and measures the real value of exports insured (InsExpijt ) by one of the "Big Three" private trade credit insurers.7 Summary statistics for exports insured by this private insurer are presented in Table 1. Several features regarding the data on insured exports are worth mentioning. First, the data set on insured exports is constrained in two ways. That is, the number of exporters is limited to 25 countries (all OECD members, except for Hong Kong) in which the private insurer is active and data is available. Second, the number of observations per exporter varies considerably (Table 1, Column 2). This re‡ects i) the entrance of the private insurer into new countries over the years and ii) di¤erences in the number of destination countries of each exporter. Finally, the insurance data su¤er from some measurement issues. Possible measurement errors arise because i) 6

clients of the insurer declare their turnover at di¤erent frequencies; monthly, quarterly or yearly, ii) the amounts are allocated to periods when they were invoiced by the insurer which does not always coincide with the period when the shipments took place, and iii) data is migrated from systems used by acquired companies. Part of the measurement errors is reduced by the yearly frequency of the data. I also show below that the results are robust to numerous sample changes, reducing the risk of obtaining biased results, and show results based on the method of instrumental variables.

2.2

Endogeneity Bias

Two issues regarding the empirical set-up need to be addressed: omitted variable bias and reverse causality. Omitted variable bias might be an issue in the benchmark speci…cation, especially as it does not include information on other insurers. Basically, the speci…cation estimates what happens to a country’s exports when the value of exports insured by one large private insurer increases, while the trading countries’GDP per capita, population size, transportation costs, and trade costs related to various institutional settings, do not change. Also, the total value of exports could explain the value of insured exports instead of the other way around. Having information on only one large private trade credit insurer could bias the estimate for the private export credit insurance e¤ect on trade, although the direction of the bias is unclear. For example, an increase of coverage could simply re‡ect an increase in the insurer’s share of the export credit insurance market. It is unclear why this would stimulate trade, creating a negative bias in the estimate of the private export credit insurance e¤ect. Yet, a positive bias arises if the change in the insurer’s coverage of exports is smaller than the change in coverage of the whole export credit insurance market. Then, part of the change in trade is wrongly contributed to the insurer. Although a bias in the estimates in either direction may not be ruled out completely, the benchmark speci…cation that controls for any time variant country-speci…c factors (i.e. the speci…cation with the comprehensive sets of {

it }

and {

jt })

is likely to adjust for changes in the supply of other

insurers. This is because supply decisions by private trade credit insurers, while in‡uencing the "bilateral dimension" (ijt), are taken at the "country dimension" (it and jt). The local business unit in a destination country determines the maximum exposure –the credit limit –on a particular …rm for the whole insurance group (Becue, 2008, p. 191-194).8 As a result, the decision by the local business unit to grant cover on a particular importing …rm in country "j" is generally independent of the

7

home country "i" of the exporting …rm. Indeed, insurers have an incentive to create "Chinese walls" between the risk units that determine credit limits and the commercial units that sell insurance, in order to assure that risks are objectively determined.9 Reverse causality might be a second source of endogeneity bias. Instead of some exogenous factor leading the insurer to extend more coverage (i.e. better marketing of products, improvements in risk management practices, e¢ ciency gains etc.), growth in trade could also explain growth in insured exports. Clearly, this would bias the estimate of the private export credit insurance e¤ect on trade upwards. In order to con…rm that there are no feedback e¤ects from changes in exports to changes in insured exports, I test for "strict exogeneity" by including the future level of insured exports to the benchmark model (see Wooldridge, 2002, p. 285, and Baier and Bergstrand, 2007).

3

Benchmark Results

The results of estimating the default speci…cation are presented in Table 2. The model is estimated with three di¤erent sets of …xed e¤ects (year, country-pair and year, and country-pair and timevarying country). Before I discuss the private export credit insurance e¤ect on trade, I brie‡y discuss the other determinants of trade ‡ows. The model …ts the data well. I obtain a high R-squared which is typical for gravity models and the estimates are sensible. For instance, exports between a pair of countries fall with distance and increase when countries share a currency, language, trade agreement or colonial heritage. In addition, countries with a higher GDP per capita import more. The sign of the coe¢ cient for the importer’s population and exporter’s real GDP per capita changes, however, when including country-pair …xed e¤ects. Thus, larger and richer countries trade more (cross-sectional variation), but importers with high population growth or exporters with high GDP per capita growth trade less, ceteris paribus. Turning to the estimates of greatest interest; private export credit insurance seems to stimulate exports. All estimates for insured exports are positive and statistically distinguishable from zero. As explained in previous Sections, only the speci…cation with country-pair and time-varying country …xed e¤ects accounts for bilateral heterogeneity and multilateral resistance simultaneously, and reduces possible estimation bias related to missing information on other insurers. Hence, I focus on the results of this preferred model to discuss the size of the private export credit insurance e¤ect. The model estimated with the full sample, Table 2 Column 3, shows that an increase of insured 8

exports by 1 per cent causes additional exports by about .01 per cent. To get a sense of the economic magnitude of this e¤ect, I compute the average trade multiplier of private export credit insurance. In the full sample, insured exports per observation (ijt; exporter-importer-year) average EUR83 million and the average amount of exports is about EUR2.2 billion. Subsequently, a 1 per cent increase in insured exports (EUR0.83 million on average) leads to an increase of exports of EUR0.22 million. The result for the full sample thus suggests that private export credit insurance induces a less than proportionate increase in exports. This …nding is not robust, however. Indeed, it turns out to be quite fragile. For example, when I exclude observations with insured to total exports below 1%, I …nd a statistically signi…cant elasticity of exports to insured exports of .07 (Table 2, Column 4), and a trade multiplier of 1.2. This trade multiplier results from the higher estimated elasticity, in combination with a higher average value of insured exports (EUR136 million) and total exports (EUR2.3 billion) in this subsample. Notably, the observations excluded from this estimation have a particularly low value of insured exports, more than half of which even below EUR0.5 million. A drawback of including these observations in which the insurer insures only a tiny share of exports is that idiosyncratic shifts in the behaviour of a single exporter or importer, or shipment, may dominate the estimated link between insured and total exports. In Columns 5 to 7 of Table 2, I further examine the magnitude of the private export credit insurance e¤ect by successively excluding observations with insured to total exports below 2, 5, and 10 percent. The estimates for insured exports are signi…cant at the 1 percent level for every subsample, and show that the elasticity of exports to insured exports increases up to .21 for the sample of observations with insured to total exports above 10 percent (Table 2, Column 7). In turn, the estimates in combination with the average values of insured and total exports reveal trade multipliers of respectively 1.2, 1.3, and 1.4 for the subsamples with insured to total exports above 2, 5 and 10 percent. I also …nd statistically signi…cant trade multipliers ranging from 1.2 to 1.4 for the subsamples in between these arbitrarily chosen thresholds.10 Overall, these results seem to suggest that a trade multiplier of private export credit insurance in the range of 1.3 is a reasonable estimate of the magnitude of the private export credit insurance e¤ect. This trade multiplier implies that every euro of privately insured exports generates about 1.3 euro of total exports. Private export credit insurance could increase exports through a number of ways. These mechanisms contribute to a trade multiplier above one, however, only if exports that are not covered by

9

private export credit insurance are stimulated. If insured exports only generate more insured exports, the multiplier would be simply one. For example, this would be the case if private export credit insurance only increases trade with importers that are covered under a whole turnover insurance policy. Keeping this in mind, there are several possible explanations for the …nding of a trade multiplier of private export credit insurance. The …rst explanation relates to the well-established fact that when …rms face substantial entry costs, previous export experience to a country increases the probability for a …rm to export again (Dixit, 1989), by as much as 60 percentage points (Roberts and Tybout, 1997). So, if the reduction in risk due to a trade credit insurance policy increases exports to markets where a …rm would not sell otherwise, as Funatsu (1986) proves, this makes additional exports of the …rm more likely.11 A second explanation for the trade multiplier of private export credit insurance follows from the recently developed theoretical model of trade …nance by Antràs and Foley (2011). Although they do not explicitly consider the role of trade credit insurance, their dynamic model shows that exporters can learn about the creditworthiness of importers by repeated transactions, and over time become more willing to …nance transactions through open accounts.12 This argument can be easily extended to the case of private export credit insurance, where an exporter may decide to export without costly insurance as a relationship with an importer develops. A third channel for the trade multiplier results from the …nancial advantage for importers of having access to supplier credit, which could stimulate their demand for imports. Supplier credit reduces an importer’s transaction costs by allowing bills to accumulate for periodic payment, enabling the importer to better forecast cash out‡ows, and to separate the payment cycle from the delivery cycle (Ferris, 1981). Also, because of the time value of money, supplier credit allows importers to increase other purchases (Schwartz, 1974). Notably, an industry insider provides an interesting argument why the importer’s reduction in transaction and …nancing costs might be an important driver of the trade multiplier of private trade credit insurance. That is, Becue (2008) argues that the news of a change in a private insurer’s credit policy on an importer –externalized by the "credit limits" –tends to travel fast among all suppliers of the importing …rm, providing valuable information on the creditworthiness of the importer. As a result, an upgrade generally improves the importer’s overall access to supplier credit (and vice versa), further limiting the importer’s transaction and …nancing costs, and potentially stimulating

10

other trade transactions by the importer. Due to this side e¤ect, Becue (2008) nicknamed private trade credit insurers the "invisible banks"; while they do not provide funding, their actions in‡uence …rms’overall access to supplier credit. Finally, private export credit insurance might facilitate the exporter’s access to bank credit and improved credit terms from lending institutions, because it avoids exceptional losses on trade receivables and is seen as a sign of good management (Becue, 2008; Jones, 2010). Essentially, exporters can increase their collateral value by insuring their accounts receivable. If better access to external …nance allows the exporter to increase its exports, this could also add to the trade multiplier of private export credit insurance. Overall, the …nding of a trade multiplier of private export credit insurance is important for several reasons. First, it shows that private export credit insurance not only facilitates insured exports (i.e. a trade multiplier of 1), but also seems to stimulate non-insured exports. The importance of private trade credit insurers in supporting international trade is thus larger than the value of exports they cover. Second, it shows that the trade multiplier of export credit insurance is not limited to the long run e¤ect of public guarantees found by Egger and Url (2006) and Moser, Nestmann and Wedow (2008). Finally, and more generally, it provides direct evidence of a link between a privately supplied form of trade …nance and exports. In the remainder of this paper I test for the exogeneity of insured exports and extensively check the sensitivity of the private export credit insurance e¤ect on trade.

3.1

Strict Exogeneity

The Section on endogeneity bias suggested that growth in trade could also explain growth in insured trade. In order to con…rm that there are no feedback e¤ects from changes in exports to changes in insured exports in the benchmark results, I test for "strict exogeneity" as suggested by Wooldridge (2002, p. 285).13 Subsequently, I add the future level of insured exports (Log Insured Exportsij;t+1 ) to the preferred benchmark model. If changes in insured exports are strictly exogenous to changes in exports in this speci…cation, then the future level of insured exports should be uncorrelated with contemporaneous exports. The results in Table 3 con…rm that insured exports are exogenous. The e¤ect of Log Insured Exportsij;t+1 on exports is economically small and not signi…cantly di¤erent from zero in the full sample and all subsamples for the various shares of insured exports.

11

4

Sensitivity Analysis

4.1

Sample Changes

I start the sensitivity analysis with a battery of robustness checks based on reasonable changes to the sample. The purpose of this exercise is to show that the results are not caused by some small subset of the sample. The results are presented in Table 4, Rows 1 to 11. Each of the rows in the table corresponds to a di¤erent sensitivity check, while the columns correspond to the benchmark model estimated for various subsamples of observations with insured to total exports above a certain level. I check the sensitivity of the results by selectively dropping di¤erent sets of observations. Since I am interested in exporter e¤ects, I begin by dropping di¤erent sets of importer observations. First, I drop all observations for importers that are industrial, and then successively delete observations for developing countries from Latin America or the Caribbean, the Middle East, Asia, Africa, or for (formerly) centrally managed economies.14 These robustness checks leave the basic results largely unchanged. The same goes when dropping small importers (de…ned as a country with fewer than one million people) or poor importers (those with real GDP per capita of less than EUR1000 per annum). I then check the sensitivity of the results for some sets of exporter observations. Successively, I drop non-European exporters and exporters not in the sample before 1995. Again, none of these changes to the sample undermine the …ndings. Finally, I check the sensitivity of the results by dropping the observations before 1999; the results remain resilient. I conclude that the …nding of a positive and statistically signi…cant e¤ect of private export credit insurance on trade is not due to some subset of the sample and is robust to reasonable changes in the sample. Countries with a higher level of insured exports seem to have higher trade than others.

4.2

Estimation Changes

Following the sample changes, I now examine the sensitivity to model changes. Successively, I test whether the main results hold when accounting for disaggregated regional trade agreements, zero (insured) trade, and trade dynamics. Finally, I show results using the method of instrumental variables; an alternative strategy to deal with endogeneity issues. The results in Appendix B also show that the private export credit insurance e¤ect on exports is robust to two alternative ways to account for "multilateral resistance" to trade in the gravity model.

12

4.2.1

Disaggregated Regional Trade Agreements

The benchmark speci…cation assumes that all regional trade agreements have the same e¤ect on trade. Recent advances in the literature on bilateral trade ‡ows, however, show that excluding individual free trade agreement e¤ects can generate biased results (see i.e. Eicher and Henn, 2011). Therefore, I decompose the dummy for regional trade agreements and estimate the model including separate dummies for APEC, EEA, EFTA EU, NAFTA, and one for all other bilateral trade agreements.15 I …nd evidence of a positive and statistically signi…cant e¤ect of EEA, EFTA and EU in the full sample, but these e¤ects disappear or even turn negative and signi…cant in some of the subsamples. However, the results in the Row 12 of Table 4 show that the …nding of a positive and statistically signi…cant e¤ect of private export credit insurance on exports is insensitive to this alternative speci…cation. 4.2.2

Zero (Insured) Trade

All the results above are generated from a linear-in-logs speci…cation that converts observations with zero (insured) exports to missing and these observations drop out of the sample, potentially introducing selection bias. The data set has 64752 observations including 48754 observations with zero insured exports of which 5082 also correspond to zero exports.16 As zero exports imply zero insured exports I examine the sensitivity of the results when correcting for sample selection due to zero insured exports. I follow Wooldridge (1995) and apply a sample selection model that is suitable for panel data with …xed e¤ects.17 Accordingly, for each year, I estimate a probit model where the dependent variable equals one if insured exports are positive. I derive the linear prediction of this model for each year and calculate the inverse Mills ratio, which I include as a regressor in the benchmark model. Table 4 Row 13 shows there is no impact on the main result. Again, the private export credit insurance e¤ect is positive and statistically signi…cant for all samples.18 A drawback of this approach is that it relies on di¤erences in functional form between the …rst and second stage when both equations include the same covariates. Therefore, I reestimate the system adding to the annual probit models a dummy variable equal to 1 for country-pairs that share the same legal origin, and a variable with an index for common religion (see also Helpman, Melitz and Rubinstein, 2008).19 Once more, the key …nding of a positive private export credit insurance e¤ect on exports persists (Table 4 Row 14).

13

4.2.3

Trade Dynamics

All results presented are based on static speci…cations of the gravity model. These models allow only for contemporaneous e¤ects of regressors on trade. Past trade patterns could, however, a¤ect current trade ‡ows in the presence of sunk costs (Dixit, 1989; Roberts and Tybout, 1997). Therefore, some authors propose to extend the gravity model with lags of trade (Eichengreen and Irwin, 1998; Bun and Klaassen, 2002). I examine whether the results hold up in a dynamic speci…cation of the benchmark model by including one lag of the dependent variable. The main result is robust to this inclusion of trade dynamics (see Table 4 Row 15), and none of the lagged dependent variables are signi…cantly di¤erent from zero. Again, insured exports seem to stimulate total exports. 4.2.4

Instrumental Variables

The benchmark model accounts for the endogeneity of insured exports econometrically. An alternative approach would be to use an instrument for insured exports and apply the method of instrumental variables. In this Section, I use the private insurer’s claims ratio – de…ned as claims (ijt) over premium income (ijt) –as an instrument for insured exports and examine the sensitivity of the previous estimates. In passing, I note that Hausman tests cannot reject the null that insured exports may be treated as exogenous (see Table 5). The claims (or loss) ratio is a key performance indicator for insurers, and an important determinant of an insurer’s decision to adjust its supply of insurance. For example, Winter (1994) formalizes the idea that an increase in claims – a loss shock – negatively a¤ects an insurer’s capital, and temporarily constrains the capacity to write coverage until higher premium rates allow capital to be built up again from retained earnings.20 Alternatively, Lai et al. (2000) emphasize the supply reducing e¤ect of increasing expectations about future losses that result from an increase in actual claims. Intuitively, the instrumental variable estimation uses only part of the variability in insured exports – the part that is correlated with the claims ratio – to estimate the relationship between insured exports and total exports. So, in case of a shock, i.e. a credit crisis or sovereign default, claims increase. The claims ratio also increases, as the private insurer can (in the short run) only raise the premia of new contracts, which are generally …xed for one year.21 However, in order to limit the rise in claims, the private trade credit insurer reduces its exposure by using its right to cancel a credit limit on any buyer at any given time (Swiss Re, 2006; Jones, 2010).22

14

A drawback of this approach is that the claims ratio might not be truly exogenous to trade, since both claims and trade are simultaneously in‡uenced by the risk environment. Still, I …nd no evidence indicating that this source of omitted variable bias is a serious issue in this model and underlying data. For example, I estimate the benchmark model including the claims ratio but …nd no statistically signi…cant correlation between the claims ratio and trade. Also, I …nd that none of the models is under-, weakly, or overidenti…ed (see Table 5). Moreover, I show results with various lags of the claims ratio as instrument and …nd longer lags to increase the estimated private export credit insurance e¤ect (see Table 5). And …nally, I show below that the results are robust to the inclusion of various possibly omitted variables to the model. The results for the First Stage regression on insured exports are presented in the …rst column of Table 5. I …nd a negative and statistically signi…cant e¤ect of the claims ratio on insured exports up to two years ahead. The point estimates indicate that a 1 percent increase in the claims ratio reduces insured exports by .11 percent in the same year, exports are .06 percent lower in the following year, and .04 percent lower the year thereafter. So, past claims ratios also in‡uence the current value of insured exports. The results for the Second Stage regression on exports are presented in Columns 2 to 5 of Table 5. I estimate the system with various lags of the claims ratio as instrument. The point estimate for the instrumented insured exports ranges between .02 and .09. Again, I examine whether the private export credit insurance e¤ect on trade is robust when successively dropping observations with insured to total exports below a certain threshold, and test the sensitivity of the benchmark IV model (see Table 6).23 First, I try to capture opportunity costs of providing insurance by estimating three variants of the IV model; dividing the claims ratio by the 12-month Libor rate, Euribor rate, and the dividend yield in the EMU market. Then, I add a measure of importer country risk –the inverse of the composite risk indicator from the International Country Risk Guide. Subsequently, I include a lag of the dependent variable to capture trade dynamics. I continue with the inclusion of a variable measuring domestic credit by the banking sector and a measure for domestic credit to the private sector, both at home and abroad. These variables from the World Bank might serve as proxies to capture the structure of the banking sector. And …nally, I estimate the system including all these controls to account for the combined in‡uence of these possibly omitted variables. None of the sensitivity checks, however, signi…cantly alter the results. Overall, the instrumental variable estimates support the …nding of a positive e¤ect of private

15

export credit insurance on trade. The coe¢ cients for insured exports are consistently positive and statistically signi…cant. Indeed, the point estimates are, if anything, somewhat larger than those associated with the benchmark model.

5

Conclusion

This paper examines whether private export credit insurance stimulates trade using a unique data set on the insurance provided by one of the world’s largest private trade credit insurers. Accounting for endogeneity issues, I consistently …nd a positive e¤ect of private export credit insurance on exports. The …ndings reveal a trade multiplier of private export credit insurance; every euro of insured exports generates more than one euro in total exports. The estimated magnitude of this trade multiplier should be interpreted with some caution, however, as the data covers information from one private trade credit insurer only. For a variety of samples, the results suggest a trade multiplier of private export credit insurance in the range of 1.3. This trade multiplier would imply that every euro of privately insured exports generates about 1.3 euro of total exports. The …nding of a trade multiplier above one indicates that the impact of private export credit insurance on international trade is larger than the value of exports insured. A number of hypotheses from the existing literature could explain this outcome. First, the trade multiplier supports the idea that the reduction in risk due to a trade credit insurance policy increases exports to markets where a …rm would not sell otherwise. Second, export credit insurance allows exporters to learn about the creditworthiness of importers, reducing the need to use insurance coverage after repeated transactions. Third, export credit insurance gives importers access to supplier credit, which reduces their transaction and …nancing costs and could stimulate their import demand. This channel would further support the trade multiplier if other suppliers use private insurance cover as a signal on the creditworthiness of an importer, improving the importer’s overall access to supplier credit. Finally, if insuring accounts receivable gives exporters better access to external …nance, this could increase their export level and generate additional exports on top of insured exports. A natural next step would be to examine to what extent these alternative channels contribute to the trade multiplier of private export credit insurance. Data containing information on insured and non-insured trade at the …rm-level could be used to examine the trade multiplier in more detail. I leave this for future research. 16

Table 1: Summary Statistics for Insured Exports (InsExpijt )a Exporting Countries (i) All United Kingdom The Netherlands France Australia Germany Belgium Denmark United States Sweden Spain Italy Norway Mexico Ireland Luxembourg Finland Switzerland New Zealand Austria Czech Republic Poland Hungary Greece Slovak Republic Hong Kong

First Year (t) in Sample

Obs.b

Number of Destination Countries (j)

1992 1992 1994 1992 1993 1994 1997 1999 1997 1998 1994 1998 1994 1993 1997 1997 1999 2003 2004 2003 2004 2005 2005 2004 2004 2006

14256 2327 1455 1230 1098 962 865 762 663 644 553 564 638 658 370 393 329 194 141 152 56 47 48 51 44 12

183 181 169 130 165 153 124 148 142 117 95 118 90 64 64 63 68 73 74 48 26 31 28 25 26 12

Insured Exports (ijt; millions) Mean Std.Dev. Min. Max.

83 151 138 24 19 244 84 95 50 87 15 25 46 24 15 16 27 74 12 25 50 2 3 22 10 12

342 459 514 72 68 692 284 288 176 290 60 79 118 147 89 42 72 238 35 67 151 4 6 30 28 20

:001 :001 :001 :001 :001 :001 :001 :002 :001 :001 :001 :001 :001 :001 :001 :001 :001 :002 :001 :002 :007 :005 :003 :019 :012 :045

a Data on insured exports from one of the "Big Three" private trade credit insurers. b Number of destination-year

data points involving a particular exporting country in the dataset.

17

6220 6220 5760 553 793 6030 2220 2220 1890 3650 762 922 1150 1990 1410 405 569 2010 270 473 894 24 24 108 135 70

Table 2: E¤ect of Private Export Credit Insurance on Exports in Gravity Model Fixed E¤ects: Insured to Total Exports

Year

Country-Pair & Year

( Expijtijt ) in Percent

All

All

All

Log Insured Exportsijt

(1) :10

(2) :03

(3) :01

InsExp

Log Distanceij

(:01)

(:00)

Log Imp Populationjt

:83

1:02

Log Exp Real GDP p/cit

1:04

:98

Log Imp Real GDP p/cjt

1:13

:48

Currency Unionijt

:18

:17

Common Languageij

:45

No. Islandsij Log Product Areaij

(:01)

> 2% (5) :09

> 5% (6) :13

:06

:14

(:02)

(:02)

> 10% (7) :21 (:05)

(:03)

:81

Common Borderij

(:00)

> 1% (4) :07

:97

Log Exp Populationit

RTAijt

Country-Pair & Time-Varying Country

(:02)

(:02)

(:09)

(:03)

(:08)

2:02

(:59)

(:22) (:28) (:08)

(:04)

:01

(:06)

:06

(:07)

(:07)

(:08)

:11 (:08)

(:07)

:04 (:06)

:15

(:04)

:06 (:08)

:11 (:11)

:01 (:11)

:10

(:14)

:05

(:59)

:06

(:10)

:27

(:06)

:05 (:01)

Common Colonizerij

1:58

Currently Colonyijt

:52

Ever Colonyij

:50

Common Countryijt

1:42

R2 RMSE Observations

:85 :96 14256

(:17) (:13)

:04

:26

:05

:24

:38

8:17

(:03)

(:15)

(:27)

(:18)

(:17)

(8:68)

:98 :35 14256

:99 :27 14256

:995 :23 8356

:996 :21 6869

:997 :18 4724

:998 :13 2815

(:10) (:09)

Data set includes bilateral annual observations covering 25 exporting countries and 183 importing countries, 1992 - 2006. Robust standard errors (clustered by country-pairs) in parentheses. Signi…cance: ***1%, **5%, *10%.

18

Table 3: Testing for Strict Exogeneity of Insured Exports Fixed E¤ects: Country-Pair & Time-Varying Country InsExp

Insured to Total Exports ( Expijtijt ) in Percent Log Insured Exportsijt

All

Log Insured Exportsij;t+1

:01

> 1% :06

> 2% :08

:00

:00

:00

(:00)

R2 RMSE Observations

(:01)

(:02)

> 5% :12 (:03)

:02

> 10% :19 (:05)

:02

(:00)

(:01)

(:01)

(:02)

(:03)

:992 :25 12171

:995 :21 7264

:996 :19 6014

:997 :17 4180

:999 :12 2503

Robust standard errors (clustered by country-pairs) in parentheses. Signi…cance: ***1%, **5%, *10%. Regressors included but not recorded: Log Distance; Log Exporter Population; Log Importer Population; Log Exporter Real GDP p/c; Log Importer Real GDP p/c; Currency Union dummy; Common Language dummy; Regional Trade Agreement dummy; Common Border dummy; # Islands; Log Product Area; Common Colonizer dummy; Currently Colony dummy; Ever Colony dummy; and Common Country dummy.

19

Table 4: Sensitivity Analysis of Private Export Credit Insurance E¤ect on Exports Fixed E¤ects: Country-Pair & Time-Varying Country InsExp

Insured to Total Exports ( Expijtijt ) in Percent

> 1%

> 2%

> 5%

> 10%

:01

:07

:09

:14

:21

:01

:08

:10

:15

:22

3. Drop Middle Eastern Importers

:01

:07

:09

:14

:22

4. Drop Asian Importers

:01

:06

:08

:13

:18

:07

:09

:13

:20

:06

:08

:12

:19

All

Sample Changes 1. Drop Industrial Importers

(:01)

2. Drop Latin America, Caribbean Importers

(:00)

(:00)

(:00)

:01

5. Drop African Importers

(:00)

6. Drop (Formerly) Centrally Managed Importersa

:01

(:01)

(:02)

(:01)

(:01)

(:01)

(:01)

(:01)

(:02)

(:02)

(:02)

(:02)

(:02)

(:02)

(:03)

(:03)

(:03)

(:03)

(:02)

(:03)

(:06)

(:06)

(:05)

(:04)

(:05)

(:06)

7. Drop Small Importers (Population 5% > 10% :16 :20 :25 :35

2. IV Model with Claims Ratio Relative to Libor

:02

:16

:20

:25

:35

3. IV Model with Claims Ratio Relative to Euribor (Sample Since 1999)

:03

:10

:12

:20

:26

4. IV Model with Claims Ratio Relative to Dividend Yield EMU Market

:02

:16

:20

:25

:35

5. Benchmark IV Model with Importer Country Risk (jt)

:02

:13

:16

:27

:43

6. Benchmark IV Model with Lagged Dependent Variable

:03

:18

:22

:32

:42

7. Benchmark IV Model with Domestic Credit by Banking Sector (it, jt)

:02

:12

:16

:23

:35

8. Benchmark IV Model with Domestic Credit to Private Sector (it, jt)

:02

:14

:17

:25

:36

9. Benchmark IV Model with All Four Controls Above

:02

:12

:17

:33

:49

(:01)

(:01) (:01)

(:01)

(:01) (:01)

(:01)

(:01) (:01)

(:05)

(:05)

(:05) (:05)

(:05) (:04)

(:05)

(:05)

(:04)

(:07)

(:07)

(:06) (:07)

(:07) (:05)

(:06)

(:06)

(:05)

(:10)

(:10)

(:09)

(:10)

(:12) (:09)

(:10)

(:10) (:10)

(:13)

(:13)

(:15) (:13)

(:16)

(:12)

(:14) (:14) (:16)

Robust standard errors (clustered by country-pairs) in parentheses. Signi…cance: ***1%, **5%, *10%. Regressors included but not recorded: Log Distance; Log Exporter Population; Log Importer Population; Log Exporter Real GDP p/c; Log Importer Real GDP p/c; Currency Union dummy; Common Language dummy; Regional Trade Agreement dummy; Common Border dummy; # Islands; Log Product Area; Common Colonizer dummy; Ever Colony dummy; and Common Country dummy.

22

References [1] Abraham, F. and G. Dewit, 2000. Export promotion via o¢ cial export insurance. Open Economies Review 11(1), 5-26. [2] Ahn, J., 2011. A theory of domestic and international trade …nance. IMF Working Paper 262, Washington, D.C.: International Monetary Fund. [3] Amiti, M. and D.E. Weinstein, 2011. Exports and …nancial shocks. Quarterly Journal of Economics 126(4), 1841-1877. [4] Anderson, J.E. and E. Van Wincoop, 2003. Gravity with gravitas: a solution to the border puzzle. American Economic Review 93(1), 170-192. [5] Antràs, P. and C.F. Foley 2011. Poultry in motion: a study of international trade …nance. NBER Working Paper 17091. [6] Asmundson, I., T. Dorsey, A. Khachatryan, I. Niculcea and M. Saito, 2011. Trade and trade …nance in the 2008-2009 …nancial crisis. IMF Working Paper 16, Washington, D.C.: International Monetary Fund. [7] Auboin, M. and M. Engemann 2012. Testing the trade credit and trade link: evidence from data on export credit insurance. Economic Research and Statistics Division Working Paper 18, World Trade Organization. Geneva: WTO. [8] Baier, S.L. and J.H. Bergstrand, 2007. Do free trade agreements actually increase members’ international trade? Journal of International Economics 71(1), 72-95. [9] Baier, S.L. and J.H. Bergstrand, 2009. Bonus vetus OLS: a simple method for approximating international trade-cost e¤ects using the gravity equation. Journal of International Economics 77(1), 77-85. [10] Baldwin, R. and D. Taglioni, 2006. Gravity for dummies and dummies for gravity equations. CEPR Discussion Paper 5850. [11] Becue, P., 2008. Handbook of credit insurance. The invisible bank. (in Dutch) Antwerpen: Intersentia. 23

[12] Berman, N. and P. Martin, 2012. The vulnerability of sub-saharan Africa to the …nancial crisis: the case of trade. IMF Economic Review 60(3), 329-364. [13] Bun, M. and F. Klaassen, 2002. The importance of dynamics in panel gravity models of trade. Discussion Paper: University of Amsterdam. [14] Dixit, A., 1989. Entry and exit decisions under uncertainty. Journal of Political Economy 97(3), 620-638. [15] Egger, P. and Nelson, D., 2011. How bad is antidumping?: Evidence from panel data. The Review of Economics and Statistics 93(4), 1374-1390. [16] Egger, P. and T. Url, 2006. Public export credit guarantees and foreign trade structure: evidence from Austria. The World Economy 29(4), 399-418. [17] Eichengreen, B. and D.A. Irwin, 1998. The role of history in bilateral trade ‡ows, in J.A. Frankel (Ed.), The regionalization of the world economy, Chicago: University of Chicago Press. [18] Eicher, T.S. and C. Henn, 2011. In search of WTO trade e¤ects: preferential trade agreements promote trade strongly, but unevenly. Journal of International Economics 83(2), 137-153. [19] Felbermayr, G. and E. Yalcin 2011. Export credit guarantees and export performance: an empirical analysis for Germany, The World Economy, forthcoming. [20] Ferris, J. S., 1981. A transactions theory of trade credit use. Quarterly Journal of Economics 96(2), 243-270. [21] Funatsu, H., 1986. Export credit insurance. Journal of Risk and Insurance 53(4), 680-692. [22] Glick, R. and A.K. Rose, 2002. Does a currency union a¤ect trade? The time series evidence. European Economic Review 46(6), 1125–1151. [23] Head, K., T. Mayer and J. Ries, 2010. The erosion of colonial trade linkages after independence. Journal of International Economics 81(1), 1-14. [24] Helpman, E., M. Melitz and Y. Rubinstein, 2008. Estimating trade ‡ows: trading partners and trading volumes. Quarterly Journal of Economics 123(2), 441-87.

24

[25] Iacovone, L. and V. Zavacka, 2009. Banking crises and exports: lessons from the past. World Bank Policy Research Working Paper 5016, Washington, D.C.: World Bank. [26] ICC, 2010. Rethinking trade …nance 2010. Global survey. ICC Banking Commission Market Intelligence Report, April. [27] ICISA, 2012. International Credit Insurance and Surety Association Yearbook 2012. Available at http://www.icisa.org. [28] Jones, P.M., 2010. Export credit insurance. Primer series on insurance, 15, Washington, D.C.: World Bank. [29] Kleibergen, F. and R. Paap, 2006. Generalized reduced rank tests using the singular value decomposition. Journal of Econometrics 127(1), 97-126. [30] Lai, G.C., R.C. Witt, H. Fung, R.D. MacMinn and P.L. Brockett, 2000. Great (and not so great) expectations: an endogenous economic explication of insurance cycles and liability crises. Journal of Risk and Insurance 67(4), 617-652. [31] Levchenko, A.A., L. Lewis and L.L. Tesar, 2010. The collapse of international trade during the 2008–2009 crisis: in search of the smoking gun. IMF Economic Review 58(2), 214-253. [32] Moser, C., T. Nestmann and M. Wedow, 2008. Political risk and export promotion: evidence from Germany. The World Economy 31(6), 781-803. [33] Petersen, M. and R. Rajan, 1997. Trade credit: theories and evidence. The Review of Financial Studies 10(3), 661-691. [34] Roberts, M.J. and J.R. Tybout, 1997. The decision to export in Colombia: an empirical model of entry with sunk costs. American Economic Review 87(4), 545-564. [35] Ronci, M., 2004. Trade …nance and trade ‡ows: panel data evidence from 10 crises, in J. Wang and M. Ronci (Eds.), Access to trade …nance in times of crisis, Washington, D.C.: International Monetary Fund. [36] Rose, A.K., 2004. Do we really know that the WTO increases trade? American Economic Review 94(1), 98-114. 25

[37] Rose, A.K. and M.M. Spiegel, 2011. The olympic e¤ect. Economic Journal 121(553), 652-677. [38] Santos Silva, J.M.C. and S. Tenreyro, 2006. The log of gravity. The Review of Economics and Statistics 88(4), 641-658. [39] Schmidt-Eisenlohr, T., 2011. Towards a theory of trade …nance. University of Oxford Discussion Paper Series No. 583. [40] Schwartz, R. A., 1974. An economic model of trade credit. Journal of Financial and Quantitative Analysis 9(4), 643-657. [41] Stock, J.H. and M. Yogo, 2005. Testing for weak instruments in linear IV regression, in D.W. Andrews and J.H. Stock (Eds.), Identi…cation and inference for econometric models: essays in honor of Thomas Rothenberg, New York, Cambridge University Press. [42] Swiss Re, 2006. Credit insurance and surety: solodifying comments. Sigma 6/2006, Zurich. [43] Van der Veer, K.J.M., 2011. Private trade credit insurers during the crisis: the invisible banks, in J. Chau¤our and M. Malouche (Eds.), Trade Finance during the Great Trade Collapse, Washington DC: The World Bank, 199-212. [44] Winter, R.A, 1994. The dynamics of competitive insurance markets. Journal of Financial Intermediation 3(4), 379-415. [45] Wooldridge, J.M., 1995. Selection corrections for panel data models under conditional mean independence assumptions. Journal of Econometrics 68(1), 115-132. [46] Wooldridge, J.M., 2002. Econometric analysis of cross section and panel data. Cambridge, Mass.: MIT Press.

26

Appendix A. Data Sources, Country List, and Correlation Matrix

Table A1: Data Sources

FOB exports in US dollars are taken from IFS Direction of Trade CD-ROM. The …gures are converted to euros at the average annual exchange rate. Pre-1999 exchange rates were calculated as the weighted bilateral dollar exchange rate of the 11 countries participating at the start of the euro in 1999 (Source: FT/Reuters). All …gures are de‡ated by the Harmonised Index of Consumer Prices (HICP), overall index, taken from Eurostat, 2000=1. Population and real GDP per capita (rgdpl) taken from PWT Mark 6.2. If PWT data are unavailable, I use World Development Indicators. The …gures are converted to euros at the average annual exchange rate. Country-speci…c data (on location, area, island-nation status, contiguity, language, colonizer, and independence) taken from CIA World Factbook website. Currency-union data taken from Glick and Rose (2002). Regional trade agreements taken from WTO website http//www.wto.org/english/tratop_e/region_e/eif_e.xls The credit insurance data comes from one of the "Big Three" internationally active private credit insurers; company details are con…dential.

27

Table A2: Country List Afghanistan Albania Algeria Angola Antigua & Barbuda Argentina Armenia Australia Austria Azerbaijan Bahamas Bahrain Bangladesh Barbados Belarus Belgium Belize Benin Bhutan Bolivia Bosnia & Herzegovina Botswana Brazil Brunei Bulgaria Burkina Faso Burundi Cambodia Cameroon Canada Cape Verde Central African Republic Chad Chile China, P.R.: Mainland China, P.R.: Macao Colombia Comoros Congo, Dem. Rep. Congo, Republic of Costa Rica Cote D’Ivoire Croatia Cuba Cyprus Czech Republic

Denmark Djibouti Dominica Dominican Republic Ecuador Egypt El Salvador Equatorial Guinea Eritrea Estonia Ethiopia Fiji Finland France Gabon Gambia Georgia Germany Ghana Greece Grenada Guatemala Guinea Guinea-Bissau Guyana Haiti Honduras Hong Kong Hungary Iceland India Indonesia Iran Iraq Ireland Israel Italy Jamaica Japan Jordan Kazakhstan Kenya Kiribati Korea, Rep Kuwait Kyrgyzstan

Laos Latvia Lebanon Lesotho Liberia Libya Lithuania Luxembourg Macedonia Madagascar Malawi Malaysia Maldives Mali Malta Mauritania Mauritius Mexico Moldova Mongolia Morocco Mozambique Myanmar Namibia Nepal Netherlands Netherlands Antilles New Zealand Nicaragua Niger Nigeria Norway Oman Pakistan Palau Panama Papua New Guinea Paraguay Peru Philippines Poland Portugal Qatar Romania Russian Federation Rwanda

28

Samoa Sao Tome & Principe Saudi Arabia Senegal Seychelles Sierra Leone Singapore Slovakia Slovenia Solomon Islands Somalia South Africa Spain Sri Lanka St. Kitts & Nevis St. Lucia St. Vincent & Grens Sudan Suriname Swaziland Sweden Switzerland Syria Tajikistan Tanzania Thailand TogoBahamas Tonga Trinidad & Tobago Tunisia Turkey Turkmenistan Uganda Ukraine United Arab Emirates United Kingdom United States of America Uruguay Uzbekistan Vanuatu Venezuela Vietnam Yugoslavia Zambia Zimbabwe

Table A3: Correlation Matrix

Expijt InsExpijt Disij P op1it P op2jt GDP pc1it GDP pc2jt CU ijt Lang ij RT Aijt Borderij Islij Areaij CColij Colijt EColij SameC ijt RT Aijt Borderij Islij Areaij CColij Colijt EColij SameC ijt

Expijt 1:00 :66 :45 :19 :49 :03 :48 :27 :04 :39 :31 :28 :27 :03 :01 :01 :00 RT Aijt 1:00 :21 :26 :08 :02 :01 :13 :03

InsExpijt Disij

P op1it

P op2jt

GDP pc1it GDP pc2jt CU ijt

1:00 :37 :02 :27 :03 :42 :22 :04 :35 :24 :14 :06 :02 :04 :10 :02 Borderij

1:00 :19 :04 :04 :28 :35 :14 :64 :43 :37 :26 :05 :02 :07 :02 Islij

1:00 :05 :36 :17 :10 :12 :18 :01 :05 :37 :02 :00 :17 :01 Areaij

1:00 :03 :13 :04 :14 :12 :06 :33 :62 :00 :05 :12 :05 CColij

1:00 :05 :07 :07 :06 :04 :05 :15 :03 :01 :05 :00 Colijt

1:00 :22 :12 :30 :16 :06 :19 :01 :02 :14 :01 EColij

1:00 :07 :27 :24 :12 :08 :01 :01 :01 :01 SameC ijt

1:00 :11 :01 :09 :01 :02 :01

1:00 :10 :01 :02 :04 :01

1:00 :02 :07 :07 :06

1:00 :00 :01 :00

1:00 :11 :83

1:00 :09

1:00

29

Lang ij

1:00 :16 :06 :19 :00 :00 :07 :48 :06

Appendix B. Alternative estimators to deal with "multilateral resistance" to trade In this appendix, I test the sensitivity of the results to two di¤erent speci…cations of the gravity model. Both speci…cations deal with the possibility of misspeci…cation in the benchmark model because of "monadic" problems. These refer to omitted factors that are speci…c to a single country but may vary over time, such a those associated with "multilateral resistance" to trade [e.g. Anderson and Van Wincoop (2003)]. Baier and Bergstrand (2009) propose a simple method for approximating trade-costs e¤ects in the presence of multilateral resistance. They call their approach "Bonus vetus OLS" since they suggest a linear appromixation of the multilateral resistance terms, motivating a reduced-form gravity equation that can be estimated using OLS. In practice, this involves estimation of each country’s multilateral resistance to trade with other countries based on the simple or –to eliminate approximation errors – GDP-weighted averages of the indicators of trade barriers with all countries (such as distance, borders etc.). Baier and Bergstrand (2009) provide more details. The results are presented in the …rst six columns of Table B1 and con…rm the positive e¤ect of insured exports on trade. All point estimates are statistically signi…cant, but the model based on simple averages gives a much smaller range of .01 to .06, compared to the benchmark results. Also, the size of the estimate for the subsample with insured exports above 10 percent is lower than the estimate for the subsample with insured exports above 5 percent, which is implausible. The estimates of the GDP-weighted transformation seem to suggest that approximation errors are indeed relevant in this sample; accounting for such errors results in a range of .03 to .34, much more in line and even bigger than the benchmark estimates. Another way to deal with the presence of multilateral resistance is the "method of tetrads", advocated by Head, Mayer and Ries (2010).24 Under this method, consistent estimates can be attained in the presence of multilateral resistance by comparing export observations to exports for a pair of base countries for the same year. See Head, Mayer and Ries (2010) for more details. The method presents two special issues. First, one needs to select a base exporter and importer to do the tetradic calculations. To check the sensitivity I use two di¤erent pairs of countries: a) United Kingdom and The Netherlands; and b) Australia and France. Second, the observations are likely to be dependent as the error terms in the tetrads appear repeatedly across observations. I therefore use

30

multi-way clustering to correct the standard errors, as proposed by Head, Mayer and Ries (2010). Since the tetradic calculations consume observations, I could only estimate the system for the full sample and the subsample with insured to total exports above 5 percent. The results presented in Table B1, Columns 7 to 10, again show a positive and statistically signi…cant e¤ect of private export credit insurance on exports, regardless of the base exporter and importer taken.

31

Table B1: Bonus Vetus OLS and Tetradic Estimates of Private Export Credit Insurance E¤ect on Exports Fixed E¤ects: Country-Pair and Year Transformation of bilateral trade costs variables à la Baier and Bergstrand (2009) Using Simple Averages Using GDP Weights Base Exporter Base Importer Insured to Total Exports InsExp

( Expijtijt ) in Percent

All (1)

> 5% (2)

> 10%

All

(3)

(4)

> 5% (5)

> 10% (6)

Log Insured Exportsijt

:01

:06

:05

:03

:24

:34

Currency Unionijt

:15

:28

:15

:17

:34

:35

RTAijt Currently Colonyijt R2 RMSE Observations

(:00)

(:06)

:06

(:07)

:21

(:01)

(:08)

:12

(:09)

:23

(:02)

(:16)

:02

(:00)

(:04)

:15

(:04)

(:13)

:03

(:40)

(:19)

(:11)

:98 :35 14256

:99 :31 4724

:99 :31 2815

:04 (:03)

:98 :34 14256

(:02)

(:05)

:17

(:05)

(:03)

(:11)

:18

(:07)

Tetrads à la Head, Mayer and Ries (2010) UK NLD

UK NLD

AUS FRA

AUS FRA

All

> 5%

All

> 5%

(7)

:01

(:01)

:26 (:06)

(:03)

:11

(:05)

:00

:19

:38

(:03)

(:03)

(:21)

:99 :27 2815

9863

(9)

:02

(:01)

:02 (:09)

:21

(:11)

:14 :99 :29 4724

(8)

:11

(:05)

(10)

:21

(:04)

:97

(:12)

:01

(:16)

:08 (:12)

611

8582

1051

Robust standard errors (clustered by country-pairs in the Bonus Vetus OLS estimation and (multiway) clustered by dyad, exporter and importer for the Tetrad estimation) in parentheses. Signi…cance: ***1%, **5%, *10%. Regressors included but not recorded: Log Distance; Log Exporter Population; Log Importer Population; Log Exporter Real GDP p/c; Log Importer Real GDP p/c; Common Language dummy; Common Border dummy; # Islands; Log Product Area; Common Colonizer dummy; Ever Colony dummy; and Common Country dummy.

32

Notes 1

Relatedly, others have found that exports of …rms that are more dependent on external …nance are the most a¤ected during a …nancial crisis (Bricogne et al. 2012; Chor and Manova 2012; Iacovone and Zavacka, 2009), and that the availability of short-term credit a¤ect a country’s exports (Ronci, 2004; Berman and Martin, 2012). 2 In Van der Veer (2011), I discuss the possible role of private trade credit insurers during the world trade collapse in 2008-2009, and review policy measures taken within the European Union to support the market for short-term export credit insurance in that period. In that paper I also extrapolate the private export credit insurance e¤ect that I …nd in this paper to the period of the world trade collapse. 3 The focus in this paper is on private export credit insurance, but private trade credit insurance is also used to cover domestic trade. 4 One reason why an importer could refuse to approach its bank to request a payment guarantee or letter of credit is that it is charged against the importer’s overall credit limit set by the bank (Jones, 2010). As a result, banking products to cover nonpayment risk reduce the importer’s borrowing capacity. Private export credit insurance, in turn, facilitates international trade based on supplier credit without reducing the importer’s access to bank credit. In fact, private export credit insurance is generally purchased by the exporter even without the importer knowing it. Another di¤erence is that the banking products generally cover a single transaction for a single importer, whereas private export credit insurance policies are usually "whole turnover", covering all of an exporter’s trade receivables with the exception of intercompany sales, exports to governments or (risky) companies the insurer is not willing to cover. Due to these di¤erences, the banking products are normally more expensive than export credit insurance (Jones, 2010). 5 The data set does not include information at the level of the …rm. 6 Egger and Url (2006) …nd an average long run trade multiplier of Austrian public guarantees of 2.8, implying that every euro spend on public guarantees creates 2.8 euro worth of exports. Moser, Nestmann and Wedow (2008) account for possible endogeneity issues and trade dynamics, and …nd a somewhat lower long run trade multiplier of public guarantees in Germany of 1.7. Moser, Nestmann and Wedow (2008, p.794) note that the short-run trade multipliers found are typically smaller than 1. Felbermayr and Yalcin (2011) examine whether the German export credit insurance scheme has alleviated …nancial frictions during the recent international …nancial crisis, and …nd public export credit guarantees to have increased sectoral exports. They also report an "e¤ectiveness ratio" (euros of exports per euros used as guarantees) of 0.47; i.e. a short-run trade multiplier below 1. 7 Company details are con…dential. 8 This is called the "proximity to the risk" principle. 9 Other conditions attached to cover policies are an important way by which insurers compete. For example, customers have a self-retention, which can di¤er in terms of cover percentage, claim deductibles, aggregate …rst losses. 10 The subsample of observations with insured to total exports of respectively 3, 4, 6, 7, 8, and 9 percent revealed a trade multiplier of 1.4, 1.4, 1.3, 1.2, 1.2, and 1.3. 11 Focusing on public guarantees, Funatsu (1986) shows that by using a credit guarantee, an exporter can reduce its pro…t uncertainty in the foreign market thereby increasing its optimal output level. Abraham and Dewit (2000) further demonstrate that government guarantees can stimulate …rms to export even without subsidization by charging a fair premium. 12 See Ahn (2011) and Schmidt-Eisenlohr (2011) for two other recently developed models of trade …nance. Both models assume that …rms are risk neutral and do not demand export credit insurance. 13 Likewise, Baier and Bergstrand (2007) test for "strict exogeneity" of free trade agreements. 14 The various groups of countries are as classi…ed by the IMF’s International Financial Statistics’country codes. 15 Since the sample only includes 25 exporters, the number of RTAs included is limited compared to other studies. 16 Notice that a fully balanced data set would include 68250 observations (15 years x 25 exporting countries x 182 importing countries). However, I lose 2578 observations because Belgium and Luxembourg are not included before 1997 (trade statistics do not report these countries separately before 1997), and the (formerly) centrally managed economies are missing in 1992. I lose another 920 observations by excluding observations with insured exports below EUR1000. The benchmark gravity model regressions lose 1742 observations due to missing data on the gravity controls. 17 See also Egger and Nelson (2011). Cross-section procedures as in Helpman, Melitz and Rubinstein (2008) are not applicable in this case, as pointed out by Wooldridge (1995). Also, the poisson quasi-maximum likelihood model suggested by Santos and Tenreyro (2006) is not suitable to handle sample selection due to the observations with zero insured exports. In a Poisson model, the right-hand-side of the speci…cation is logarithmically transformed but not the dependent variable. Hence, the zero insured trade ‡ows (including the 5082 observations that also correspond to zero exports) are still dropped by log-transforming the model. 18 Although not reported, the estimates for the inverse Mills ratio indicate little signi…cant selection into the sample. 19 These variables are calculated as in Helpman, Melitz and Rubinstein (2008, see p. 480). Data are from the CIA’s World Factbook and "The Global Social Change Research Project" available at http://gsociology.icaap.org. 20 In Winter’s capacity constraint model, insurers must hold equity to guarantee that they will be able to pay all claims, and external capital is assumed to be more costly than internal capital.

33

21 For example, the ICC Global Survey report (2010) reports that "Total claims paid to insured customers by all Berne Union members more than doubled from 2008 to 2009 and reached USD2.4 billion. As the total premium stayed roughly the same at an estimated USD2.8 billion, the loss ratio jumped from 40 to 87 percent. The Berne Union is the leading international organisation of public and private sector providers of export credit and investment insurance. 22 For instance, even if a supplier has an insurance contract covering its accounts receivable, the insurer can e¤ectively reduce this cover to zero over night. This ability to set and manage exposures distinguishes trade credit insurance from other kinds of insurance and many other credit instruments (Swiss Re, 2006). 23 Note that I estimate the system using the contemporaneous claims ratio as instrument for insured exports. This way, I maximize the number of observations and the F-statistic for the excluded instruments, while being conservative on the size of the estimated private export credit insurance e¤ect (compare Columns 2 to 5 of Table 5). 24 See for another application Rose and Spiegel (2011).

34