Universiteit van Amsterdam August 2018
The Effect of Private Export Credit Insurance
on Bilateral Trade
By Thomas Clabbers 11946032 Hessel Oosterbeek
1
Abstract
Trade finance and trade risk mitigation have received much attention since the recent financial crisis of 2008. There have been several attempts to identify the effect of trade finance on trade flows, most of them have found positive effects. Although the drivers behind these effects are not always clear. This is mainly due to the fact that information is scarce. This paper investigates the role of export credit insurance on bilateral exports on a macro level, a tool to mitigate the risks of trade finance. A unique bilateral data set is used which spans over a time period of 15 years, nearly 2 cycles. The effect of export credit insurance is significantly positive throughout the entire period. Yet, the effect is less than proportionate. I.e. an increase of one euro of insured exports leads to less than one euro of exports. The paper examines the magnitude of the effect in different market conditions both in the country of origin and destination. The paper uses the workhorse model in international economics; the gravity model and an additional instrumentation strategy to address a potential endogeneity bias.
2 Statement of Originality
This document is written by Student: Thomas Bèrke Clabbers who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
1. Introduction
More than 200 years ago Adam Smith concluded, in ´The Wealth of Nations´1, that trading in the international sphere is riskier. Smith noted bankruptcy (default of payment) as the most common risk in international trade. New modes of communication, transportation and international agreements have reduced this difference (Köhn, 2017). Nonetheless, even today international trade faces longer times of transportation. Moreover, the working capital requirements for international trade are higher than local trades (in general), making international trade depending heavily on (trade) finance; between 80-90% of the transactions (Auboin, 2009). The exporter and importer allocate both risk and financing between them. When trading on an open account, which is the most popular contractual choice, all risks and financing are allocated to the exporter (IMF, 2009). The trade costs can be greater due to a different institutional context and judicial system (Antràs & Foley, 2015). The transaction can be perceived as too risky by the exporter, which makes him abstain from trade altogether. One way for exporters to mitigate the risk of non-payment is to buy an export credit insurance. Potentially, this would enable trade for risk-averse exporters, or may even create trade. This paper tries to estimate the effect of export credit insurance on bilateral exports on a macro level. The effect is translated into a trade multiplier, which shows the level effect. If the multiplier is higher than 1, it implies that every insured euro of export will lead to more than 1 euro of exports, i.e. creation of uninsured exports on top of the insured exports. A positive effect of export credit insurance on trade has been found in several papers, yet in slightly different contexts. Papers of Felbermayr & Yalcin (2013) and Schmidt-Eisenlohr (2013) focus specifically on public export credit insurance, which is provided by the state. Antrás & Foley (2015) look at private export credit insurance but from the perspective of only one exporter. Auboin & Engemann (2014) look at aggregate private export credit insurance on a bilateral
4 level with Germany as the only exporter. The only paper that investigates the effect of private export credit insurance on bilateral exports is van der Veer (2015). He discovers a trade multiplier of 1.3, i.e. every euro of insured trade creates 0.3 euros of uninsured trade. However, his database is limited both in number of exporters and in the years that are covered. Ergo, it is valuable to test the effect of export credit insurance on exports and its corresponding multiplier again in this paper.
The paper makes use of a unique data set, provided by one of the three major export credit insurers, which covers roughly two business cycles and includes the recent financial crisis of 2008. The topic of international trade finance and insurance has only been scarcely researched because information is not readily available in this sector. The database covers the period of 2000-2015 and includes bilateral data. Hitherto, no richer database has been used in similar research.
Furthermore, this paper will dig deeper into the possible drivers behind the effect of export credit insurance on exports, concentrating on different market conditions. More specifically, the quality of institutions, the development of the financial market and its efficiency. This is especially interesting for three reasons. Firstly, identifying which market conditions amplify or compress the effects of export credit insurance. This information might be useful for export credit insurers or governments that use export credit insurance as a tool to stimulate exports. Secondly, it is the first research that investigates the effect of market conditions and the institutions on a bilateral level, including both market conditions of the country of origin and destination. Consequently, these findings add to the existing base of literature. Thirdly, analysing the drivers adds robustness to the analysis.
To analyse the effect of export credit insurance on export the workhorse model in international economics is used, viz. the theoretically founded gravity model of international trade. Vast sets
5 of fixed effects are implemented to deal with possible endogeneity issues to get an accurate estimator. Furthermore, an additional instrumentation strategy is implemented to make sure that the results are consistent and not driven by any bias. The instrumentation strategy to analyse the effect of export credit (insurance) on trade has been implemented before by Auboin & Engemann (2014).
The research presented in this paper shows clear results, which the additional instrumentation showed to be consistent. The research finds a positive effect of export credit insurance on exports. A one percent increase in export credit insurance leads to a 0.011% increase in exports, translating this effect into a trade multiplier of 0.432. Other models, including the IV model 0.484, show a slightly higher multiplier. Yet, the trade multiplier never significantly exceeds 1. This suggests that the effect of export credit insurance is less than proportionate, i.e. one euro of additional insured trade leads to less than one euro of additional trade. One of the reasons might be that uninsured trade is substituted by insured trade, this could explain the positive effect that is less than proportionate.
An interesting result from the heterogeneous effects section is the importance of the market conditions of the origin country. A lower quality of institutions in the origin country increases the effect of export credit insurance on exports. Although this finding is supported by the contractual choice theory of Schmidt-Eisenlohr (2013) it had not been tested before. This research finds no evidence for the importance of the quality of institutions of the country of destination with regard to the effect of export credit insurance on exports. Looking at the financial market maturity and efficiency gives similar results. The origin market conditions seem to have a substantial effect on the magnitude of the effect of export credit insurance on exports. The exporter is the agent who buys the export credit insurance. Possible benefits for the importers might not be reflected in the corresponding bilateral exports, which does not imply that they are not there.
6 This paper concludes by performing a multifarious robustness analysis to show that the findings are consistent and not driven by a small subsample. Excluding small groups of exporters and importers, based on different criteria, does not change the results. The trade multiplier varies between 0.3-0.5, with an outlier of 1.5. Yet again, the trade multipliers are never significantly higher than 1. These findings add robustness to the analysis of the paper and its conclusions.
The paper is structured as follows. First, the theoretical background and logical model of export credit insurance and its effects on export are presented. Second, the data is shown and its limitations discussed. Third, the methodology of the gravity, IV and heterogeneous effects models are explained. Fourth, the results of the respective models are analysed and interpreted. Fifth, sensitivity tests are run to show that the results are not biased by a subsample. Sixth, a concise conclusion is added and future research will be deliberated.
2. Trade finance and insurance
Trade finance is a common practice in international trade that is necessary due to the time gap that exists between the moment the product arrives in the inventory of the importer, and the time that the exporter sends out the shipment. Trading on an open account implies that the seller (exporter) allows the buyer (importer) to pay after the product has arrived. Usually the credit is extended for a period of 30-180 days after placing the order. When credit is handed out by the seller to the buyer, all risk is allocated to the former. The importer can decide to default on the transaction. It would be, particularly in international trade, hard for the exporter to recover its product or the debt. Especially, when the distance between the countries is large and when institutional differences exists (Schmidt-Eisenlohr, 2013; Antràs & Foley, 2015). There are several reasons why an importer decides to default, inter alia insolvency, inconvertibility of currency, war and civil disturbances and undesirable government interventions (Jones, 2010). These risks could be the reason for exporters to demand payment ahead of the shipment, a
7 practice which is called ‘Cash in Advance’. This implies that the buyer finances the trade before it is shipped. In this case all risk is allocated to the importer because the exporter can decide not to fulfil his contractual obligations.
To conclude, due to the time gap trade finance is inevitable. The risks are considerably large and have to borne by the one who finances the transaction. Practice shows that the first few transactions between buyers and sellers are on a cash in advance, they change this slowly to an open account (Antrás and Foley, 2015).
Export credit insurance is a fairly common way to mitigate the seller’s risk of a defaulting buyer. When trading on an open account, the insurance protects the exporter against default of both commercial and political risks. The transaction between the seller and buyer does not change, the seller exports the goods or services on an open account. However, the seller pays a premium to a third party, viz. the credit insurer. If the buyer does not fulfil his contractual obligations the seller makes a claim to the export credit insurer. The insurer compensates the seller and has the right of subrogation, i.e. the right to recover the amount of the claim paid from the defaulting debtor (see figure 1).
The credit insurer is protected against adverse selection of exporters in two ways. First of all, exporters have to insure all their exports, i.e. it is impossible to insure just one transaction. The risk is pooled over all transactions. Secondly, the credit insurer has a database with the credit history of many companies that they insured. The credit insurer uses this information to set a credit limit. If an importer has a bad credit history or when little information is available, the credit limit will be lower to reduce the risks.
Export credit insurance is just one of the tools to mitigate the risk caused by trade finance. The IMF (2009) reports that open account is with 42% the most popular way of doing business, Export credit insurance operates in this area as well. 36% are bank intermediated transactions,
8 in which the bank guarantees payment of the importer. Cash in advance, which is used in 22% of the transactions, allocates the risk fully to the importer and is usually done with new clients. The height of the transaction and the location of the importer are important factors in determining which contractual form is used (Antràs & Foley, 2015). Schmidt-Eisenlohr (2013) was the first one to develop a formal economic theory which determines the contractual form that will be used by the exporter. The rule of law, interest rates and costs of litigations play an important role in this process and can explain this variability.
Figure 1: Trading on an open account with export credit insurance. The exporter pays a premium to the export credit insurer and is protected against the risk of non-payment by the importer. If the importer defaults the export credit insurer compensates the exporter and has the right of subrogation (recovery).
In general trade credit insurance covers a vast part of the international transactions. In 2016 around €2.3T of all exports were insured while international trade was roughly €14.5T; approximately 15 percent of the total exports (ICISA, 2018). During the financial crisis of 2008 the demand for credit insurance increased due to the growing numbers of insolvencies. The export credit insurers increased the price of the insurance and reduced the credit limits which resulted in a large drop of the coverage. This went hand in hand with the great trade collapse. The health of financial institutions, banks and export credit insurers have a great impact on trade flows (Amiti & Weinstein, 2011). When the export credit insurers incurred great losses
9 during the crisis in 2008, they became more cautious with insuring the trade credit. Auboin and Engemann (2014) find that around 11% of the great trade collapse can be explained by the reduction in supply of export credit insurance. Gradually the insurance coverage is increasing again, it exceeded pre-crisis levels in 2013 (See figure 2).
Figure 2: Annual export credit insurance coverage both in total and relative terms. Left axis depicts the total annual export credit insurance coverage in millions of euros. The right axis depicts the relative amount of export credit insurance to total world trade as a percentage. Export credit insurance seems to be increasing gradually, with the exception of the financial crisis in 2008. The share of Insured trade over total world trade seems to be increasing too and equals 15 %. This implies that export credit insurance coverage is growing faster than total world trade. (ICISA, 2018; WTO, 2017)
2.1 Logical model
Emerging economies are becoming more integrated in the world’s economy as both developing countries and emerging economies are trading more with emerging economies, moreover, the trade is deviating from the original North-South trade composition (Akin & Kose, 2008). Nevertheless, these fast-growing countries are more prone to geopolitical shocks because of the quality of their institutions (WJP, 2018). This uncertainty increases the perceived risk of default on credit and thus puts a hold on the growth of trade. The following sections presents an array of
0% 2% 4% 6% 8% 10% 12% 14% 16% 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 € -€ 500 € 1,000 € 1,500 € 2,000 € 2,500 In su red tr ad e o v er to tal w o rld tr ad e E x p o rt c red it in su ran ce in Millio n s
Annual export credit insurance coverage
10 channels through which export credit insurance can influence export, both positively or negatively.
2.1.1 Uncertainty and trade costs
Uncertainty has always had its effect on international trade, as Edward I, king of England noted in 1283 ‘many merchants [fearing uncertainty and lack of protection] are put off from coming to this land with their merchandise to the detriment of merchants and our whole kingdom’2. Büge (2010) also finds evidence for king Edward’s claim. Reducing uncertainty by ten percent increases exports by 2.4 percent, while adding controls for other institutional transaction costs. Improving institutional quality by 10 percent, increases trade by 3.8 percent.
Funatsu (1986) was the first one to research the effect of export credit insurance on the decision-making process of the exporter, whether to export or not. Funatsu set up a model which includes risk averse firms that are profit maximizing. He concludes that under uncertainty a risk-averse firm is more likely to abstain from trade. These companies are willing to pay for an insurance to reduce the risks. Having a trade credit insurance and thus mitigating default risks, induces the risk-averse firms to export. This would lead to more transactions by inclusion of risk averse firms in the market, thus more exports. Both on the intensive and the extensive margin of international trade. I.e. both an increase in the amount of export transactions and an increase in their value.
After tariffs and non-tariff barriers reduced during the second wave of globalization, policy research shifted more to the concept of trade costs. They include all costs made, except for the marginal costs of production. Weak institutions and higher uncertainty drive up these trade costs. This de facto uncertainty ad valorem tax is estimated to be about 170% of the freight on board (FOB) export price. In developing countries, with higher uncertainty due to weaker institutions
11 this ‘tax equivalent’ is even higher (Yotov, et al., 2016). Reducing this uncertainty indeed decreases the trade costs (Anderson & van Wincoop, 2004). Two components of these trade costs can be directly linked to export credit insurance, viz. cross-border information barriers (information about creditworthiness of importer) and cost of contracting and insecurity (different judicial systems and language). They account respectively for six and three percent of this ad valorem tax. Uncertainty increases with distance between the trade partners. Export credit insurance is a way to reduce these trade costs, and to guarantee payment. It could lead a significant effect on both the intensive and extensive margin of bilateral international trade (Egger & Url, 2006).
Jones (2010) states that the export insurance company minimizes the (impact of the) risks in two ways. When an importer does not pay back the credit to the exporter the credit insurer receives a claim from the latter. The credit insurer gets the right of subrogation, i.e. the right to collect the debt directly from the importer. With a team of legal experts this is, in most cases, more efficient than the recollection by the exporter himself. Furthermore, the credit insurer has access to the payment history of all their client’s importers (or national buyers), this gives them insights in the creditworthiness of the importer. Obviously, with rich information on importer’s payment history, exporters could abstain from export to importers with low creditworthiness ex-ante. Vice versa, If exporters have their doubts about the creditworthiness of a client, the credit insurer might enlighten them about the creditworthiness, which could generate trade. In short, the export credit insurer created its own extensive detailed database, which is impossible to replicate if you are an individual exporter. Taking insurance gives exporters access to that information, and its benefits.
2.1.2 Trade enhancing effect of export credit insurance
Next to allowing transactions to occur after insuring them, export credit insurance could also increase trade. The multiplier effect implies that for every euro of insured export credit,
12 uninsured trade increases as well. This leads to a trade multiplier larger than one. Van der Veer (2015) is the first one to investigate the effect of insured exports on export and he finds a multiplier of 1.3. Below some possible channels are described which could show a proportionate effect and a more than propitiate effect of export credit insurance on exports. Importers are granted capital from exporters, this means that they do not have to reduce their working capital before selling the product, or before the products even arrive (Jones, 2010). In general access to capital for importers could increase demand. This would lead to a more that proportionate effect if the importer buys from another seller in that country and the exporter does not insure the transaction. This effect is potentially larger when the importer has difficulties to access credit through other channels, this could lead to underinvestment or lower imports (Ferris, 1981). Felbermayr & Yalcin (2013) find evidence that this channel is more relevant in countries where credit markets are missing or underdeveloped. In practice this would maximally lead to a proportionate effect, since every euro of additional trade is also insured due to the whole turnover policy practice.
Another channel is the reduced need for insurance coverage in the future. If companies trade longer with each other they create a trustworthy relationship. In the medium run the insured transactions create transactions that are not insured, i.e. export creation. Although this effect is impeded by the custom of whole turnover policies and would only work if a company has a fixed set of importers and stops buying credit insurance all together. In the long run if companies build a relationship they start to trade with each other in a different manner. They switch from using cash in advance to open account with insurance to open account without insurance (Antràs & Foley, 2015; Jones 2010). the effect is only visible in the long-term.
The market signalling effect leads to the creation of uninsured exports. If a credit export insurer reduces its risk rating or starts selling coverage for a country. It implies that some exporters
13 might dare to start exporting, even without insurance. Hence, this would lead to a more than proportionate effect. This signalling effect also applies to a single importer. If a single importer receives a good credit rating from the export credit insurer, other exporters might start to export without an insurance (van der Veer, 2015).
The facilitation of access to capital for the exporters could also lead to an increase in trade. Receivables can make up a considerable part of their balance sheet. Financial institutions might consider uninsured receivables as a risky asset. which due to threat of non-payment is less likely to count as collateral for new loans. Exporters can be credit constrained if they will not insure the receivables. Insuring mitigates the risk on the receivables, and gives exporters access to credit, which in turn can lead to export creation. Due to the whole turnover policy structure this leads most likely to insured export creation but in theory it could lead to uninsured import creation (Becue 2008; Jones 2010).
Furthermore, export credit can be seen as a service from the exporter to the importer which could lead to a higher price, especially when competition on the international market increases. The export credit (insurance) in a way compensates for the drop in prices after supply on the world market increases (Demir & Javorick, 2018). Becue (2008) also states that receiving a credit limit as a buyer shows the creditworthiness of an importer to other financial institutions which in turn could increase or facilitate his access to credit. The export credit limit creates a signalling effect, which is beneficial for the buyer if he has a reputable credit record. If sellers communicate with each other, information about the creditworthiness of a buyer could induce uninsured trade. This occurs when sellers, who do not buy insurance, trade with a buyer due to its creditworthiness.
14 2.1.3 Less than proportionate effect on trade
There are also several drivers behind possible negative effects, that could lead to a multiplier below one. I.e. a one euro increase of insured trade leads to less than one euro growth of exports. A direct negative effect can be identified through the premiums, companies insure the transaction, which effectively reduces the profitability of the export by the amount of the premium. However, they are insured against negative shocks. Assuming that the export credit insurer makes a profit, this can be classified as direct negative effect. In extreme cases the costs of the premium reduce the profitability so much that the exporter abstains from trade.
The risk perception of a market might play an important role in the export decision making process. If many companies are insuring their transactions or if the premiums on a country go up. Companies might abstain from exporting at all to these countries. Which could lead to an increase in the trade insured and/or a decrease in the exports. Moreover, due to the risk perception exporters who do not insure their transactions might abstain from exporting all together. Export credit insurance and the costs of it, could induce a market signalling effect. Both would lead to a less than proportionate or negative effect.
The type of policies offered by the credit insurer to the clients are whole turnover policies. This implies that an exporter cannot just insure his most ‘risky transactions’. The exporter pays a premium over the total value of its exports. This means after taking export credit insurance, that probably some trade is created, yet some (uninsured) export is substituted by insured exports. This will most likely lead to a less than proportionate effect (van der Veer, 2015). This effect is the strongest and will probably drive the results downwards.
15
3. Data
This paper makes use of a unique bilateral database, which spans over a period of 15 years. The database belongs to one of the three mayor credit insurers. The data has its limitations, first of all since it excludes the other credit insurers and because the data still had to be cleaned.
3.1 Explanation of the data
This paper uses an extensive data base provided by one of the major export credit insurers. It is an annual bilateral trade database that covers the exports between roughly 41,000 country pairs in the period of 2000-2015. The sample can be considered an unbalanced panel data set, i.e. not all country pairs are present in all the years. This might be due to the fact that there was no insurance, or because the credit insurer did not cover the risks in that country. In addition to that, the database includes the value of the insured export credit and the value of the claims. This is self-reported data (by the clients of the insurer) and could therefore include measurement errors. Nevertheless, it is in the interest of the exporter to report the true value of the transactions otherwise the transactions are simply not insured. The database is completed by adding a whole set of covariates, inter alia common language, common border, distance. Afterwards country time variant and the country pair dummies are created to account for unobserved heterogeneity, I.e. unobservable characteristics that are correlated with both the independent or dependent variables and thus create a bias; such as culture of doing business, having a common diaspora, having a natural monopoly in a specific product.
16 Summary statistics Exporter N Amount of origin countries Average Share in (%) Average
Exports St.Dev Maximum
1 GBR 2440 184 6 131.25 414.19 7342.75 2 NED 2324 179 6.2 166.69 618.14 6998.14 3 FRA 2103 170 1.1 44.89 145.30 1858.80 4 DEU 1990 174 2.3 271.29 794.43 8582.67 5 USA 1982 172 1.7 73.78 186.70 2013.04 6 ITA 1773 164 1.1 42.53 132.60 1507.56 7 BEL 1716 156 2.4 104.63 379.43 3835.54 8 AUS 1687 168 5.4 29.56 84.15 863.87 9 DNK 1581 167 11 95.38 250.78 1945.05 10 ESP 1487 141 1 19.99 61.86 697.84 11 SWE 1438 136 5.5 87.85 247.53 2419.75 12 FIN 1307 130 7.3 53.63 132.35 1492.30 13 CHE 1237 130 4.5 91.80 331.99 3817.80 14 IRL 1157 130 4.8 50.88 197.81 2286.38 15 AUT 1082 119 4.4 65.00 217.21 2823.17 16 NOR 1079 118 6.8 63.26 149.75 1409.15 17 NZL 886 115 3 6.53 19.85 213.26 18 HKG 869 98 1.5 46.65 219.54 4028.43 19 MEX 856 66 2.4 29.07 131.55 1525.55 20 CAN 845 133 0.7 12.69 65.88 855.24 21 SGP 805 109 3.9 69.89 150.78 1437.07 22 POL 734 89 1.4 37.23 131.74 1238.68 23 LUX 726 90 5.9 20.93 68.34 697.49 24 CZE 612 70 1.6 6.60 20.43 320.11 25 PRT 611 89 1.2 28.01 100.69 1383.14 26 ARE 584 106 4.1 25.13 67.29 690.48 27 EST 584 67 5.8 7.98 17.27 156.93 28 HUN 573 86 1.3 16.36 36.02 275.43 29 SVK 551 71 1.7 27.49 89.01 895.68 30 JPN 516 87 0.6 21.53 129.13 2595.51 31 LTU 497 67 5.9 16.73 35.78 243.64 32 TUR 456 93 0.5 4.13 13.16 149.77 33 LVA 400 62 4.7 9.58 22.95 196.65 34 BGR 387 71 1.6 3.18 7.05 44.08 35 ROU 359 66 1.1 8.37 30.32 282.96 36 ISL 330 41 5.8 5.58 15.32 144.74 37 SVN 298 60 1.6 6.03 23.40 232.98 38 ZAF 279 71 1.2 4.63 13.22 100.47 39 ISR 271 64 0.9 4.90 14.72 160.06 40 MLT 271 54 7.6 3.45 9.39 83.27 41 GRC 247 48 1.7 8.27 21.15 123.53 42 THA 189 42 0.5 9.41 32.60 322.74 43 MYS 186 53 0.3 2.50 5.20 36.83 44 SAU 180 49 2.1 24.04 61.64 463.02 45 CHN 95 41 0.2 4.82 13.85 91.46 46 IND 91 22 4.2 7.53 12.92 77.26 47 BIH 58 19 3.5 32.21 70.23 347.64 48 PER 56 15 8.9 2.80 7.99 59.38 49 PAN 51 18 0.2 3.28 8.83 60.95 50 BRA 40 15 0.4 1.35 2.30 10.84 TOTAL 41100 196 3.7 67.08 300.00 8582.67
Note: Descriptive statistics, Only the 50 largest exporters are reported in this table. The share is the percentage value of mean Insured exports/exports. Average exports, the corresponding standard deviation and maximum are reported in millions of euros. The data on export credit insurance comes from the export credit insurer while the export data is based on the IMF Direction of Trade Statistics database (IMF,2018)
17 The database is similar in structure to the database used by van der Veer (2015). Both
databases operate in the bilateral dimension {ijt}3. Nevertheless, the database used in this paper is almost three times as large, both in number of observations as in the number of exporters {i}. Also, the database used in this research is more recent and includes the financial crisis of 2008.
The data has a clear advantage over data used in other papers that investigate the effect of trade finance (insurance) on trade flows. Auboin and Engemann (2014) use Berne Union4 data which is aggregated on the importer level {jt}. Felbermayr and Yalcin (2013) make use of a bilateral dataset while just including Germany as an exporter. Antràs and Foley (2015) use microdata on the company level, this includes only one exporter and his complete set of clients. Working in the {ijt} dimension gives richness to the analysis. The extra dimension, compared to {it} or {jt} generates more observations and possibilities to observe drivers behind the macro effect of export credit insurance on exports. Furthermore, it is a product that is bought in the {ijt} plane, both exporter and importer country conditions could matter for the magnitude of the effect.
Table 1 shows the descriptive statistics of the data used in this research. The data covers around 70 exporters, which are mainly located in the Western European region (60% of the observations). The number of observations per exporting country differs, it is based on the amount of destination countries that are covered by the credit insurer in a specific country of origin. E.g. the United Kingdom has export credit insurance on 186 countries. The distribution of destination countries is more equally divided over all regions, nevertheless there is a small dominance of Western European countries (roughly 25% of the observations). The average
3 The bilateral dimension can be interpreted as a country pair year. I.e. the situation of country i and j in year t. In this context exports that go from country i to country j in year t. {it} and {jt} can be interpreted as country year characteristics, while {i} and {j} are country characteristics.
4 The Berne Union is an international non-profit association for the export credit and investment insurance
18 share of insured exports over total exports also differs per country of origin. Denmark insures on average 11% of its trade whereas a country like Germany only insures 2% of its exports on average. This could be partly due to the presence of the credit export insurer or the culture of mitigating trade credit risks through export credit insurance. An array of robustness analyses will show how the results differ when leaving out several subsamples. Furthermore, a set of fixed effects is implemented to deal with country specific characteristics. In the appendix (table 9) a table is shown with all exporters and importers
The main limitation of the data is that is comes from only one of the three major credit insurers. This could lead to a bias since the market share of the credit insurer differs from country to country. The three major credit insurers represent almost 85 % of the market. Euler Hermes is the largest of the three credit insurers with a market share of 36%, followed by Atradius 25,4% and Coface 19,6% (AU group, 2016). Although each of them takes a fair share of the market, a selection bias might arise. Van der Veer (2015) argues that including country time variant fixed effects {𝛼𝑖𝑡}and {𝛼𝑗𝑡} will resolve this issue, which at the same is another argument for their inclusion in the model. The section ‘Endogeneity issues’ (4.1.1) will elaborate further upon this limitation.
Another issue is that the data comes straight out of the registration system of the credit insurer, hence it included some obvious measurement errors. The problem exhibited itself through the presence of extremely high shares. Around 3.5 % of the observations have a share (defined by insured exports over exports) that is higher than 0.3. Although not impossible, it is highly unlikely that more than 30 % of all trade is insured at only one of the three major credit insurers. The last section of this chapter will elaborate further upon this topic.
19 3.2 Additional data
Data on trade flows is taken from the IMF Direction of Trade Statistics database, which measures the FOB annual exports from country i to j in current USD (IMF, 2018). This data is converted to euros and adjusted with the HICP (harmonized index of consumer prices) (ECB, 2018). Many covariates, both country specific time variant and invariant, are taken from the World Bank development indicators database. This includes figures on income (GDP per capita) and population. A CEPII (Le Centre d’études prospectives et d’informations internationals)5 database contains data on distance between countries, proxied by the bilateral distance between the two capital cities (Mayer & Zignago, 2011), contiguity, if the countries speak the same language (Melitz & Toubal, 2014), if they had the same colonizer and if the countries where once one country (Head & Mayer, 2015). Data on currency union is taken from a paper on the effect of currency unions by de Sousa (2012). Data on trade agreements is extracted from the WTO Regional trade agreement information system. All data is updated on a regular basis and available throughout the whole period of the export credit insurance database; viz. 2000-2015.
To assess the different drivers behind the relationship of export credit insurance on export itself. This paper looks in depth at three factors: Quality of institutions, financial market maturity and financial market efficiency. The {ijt} dimension makes it possible to look at the importance of both the situation in the country of the importer and the exporter. Antràs and Foley (2015) look also at the influence of the quality of institutions. The data, in panel structure is taken from the World justice project; the rule of law. Financial market maturity can be approximated by using liquid liabilities, private credit, stock market capitalisation and value traded over capitalisation; this has been applied before in a paper of Felbermayr and Yalcin (2013) and Beck et al. (2009). Parameters such as importer and exporter net interest rate margin and overhead costs of banks
20 are used as a proxy for financial market efficiency, as done before in the papers of Schmidt-Eisenlohr (2013) and Beck et al. (2009).
3.3 Cleaning the database
In the database around 1.5% percent of the observations has a share that exceeds one. This implies that the amount of exports insured was higher than the amount of exports itself. Moreover, since only data of one export credit insurer is available, shares higher than 0.3 seem already highly unlikely, although not impossible. Determining a cut-off of high shares is ambiguous, therefore it is important to examine why the data shows these high shares. Hence, it is important to identify the source of the problem, to create a story behind these high shares, and consequently eliminate them if necessary.
𝑆ℎ𝑎𝑟𝑒𝑖𝑗𝑡 =
𝐼𝑛𝑠𝑢𝑟𝑒𝑑 𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡
𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡 (1)
The share essentially exists of two elements, the insured exports and the exports itself, all three are bilateral and calculated on an annual basis {ijt}. By looking at the transaction level, the high shares are identified and patterns observed. The specific problematique is elaborated upon in the appendix
Before using this database, it is necessary to undertake action. The vast majority of the above-mentioned issues that drive up the shares are located in the shares above 0.3 (see appendix). Therefore, the cut-off is placed at 0.3 and all observations with shares above 0.3 are excluded from the research. In the appendix the results are shown while using different thresholds. The exclusion of the high shares increases the coefficients of interests. This implies that high shares cause an upward bias. Exclusion is justified due to the faulty nature of these observations.
21
4. Methodology
This paper uses the standard procedures that are used similar journal articles in this field of research. The baseline model is an extensive theory founded gravity model with fixed effects to control for possible endogeneity biases. To address this fully an instrumentation strategy is proposed to deal both with reverse causality and omitted variable bias.
4.1 Baseline gravity model
The aim of this paper is to correctly estimate the effect of export credit insurance on bilateral exports flows on a macro level and to shed light on the possible drivers behind this process. This paper will make use of the workhorse model of international economics, the gravity model of international trade. A model that estimates trade flows by looking at the attraction between two economic ‘masses’. Tinbergen (1962) initiated and popularized this model. Although the model was successful in predicting trade flows, it was always subject to critique. Due to its origin as a physics’ model it lacked economic reasoning and microeconomic foundations. Anderson and van Wincoop (1979 & 2003) solved that issue by stating the importance of the multilateral resistance terms (MRTs). This implies that trade flows do not solely depend on characteristics of and between country A and B, but relative to all other countries as well. Every produced good has multiple possible destinations, depending on the relative attractiveness of an importer with respect to the other importers (Anderson, et al., 2003).
Building on the model of Head and Mayer (2015)6 this paper uses a gravity equation (2) that is based on micro-foundations. The general gravity equation exists of five terms. 𝑋𝑖𝑗𝑡 represents exports from country i to country j in year t. G is a general ‘Gravitational constant’, 𝜔𝑖𝑡 captures all the exporter’s characteristics, 𝜃𝑗𝑡 captures the characteristics of the importer’s market. 𝜑𝑖𝑗𝑡
6 Head and Mayer designed a model that is based on the micro foundations that are developed by Anderson and
22 is the bilateral accessibility, it combines the trade costs (proxied by distance) with their respective elasticity.
𝑋𝑖𝑗𝑡 = 𝐺𝜔𝑖𝑡𝜃𝑗𝑡𝜑𝑖𝑗𝑡 (2)
This model can be transformed to a structural gravity model which includes the MRTs.
𝑋𝑖𝑗𝑡 = 𝐺𝑌𝑖𝑡
𝛼𝑖𝑡
𝑌𝑗𝑡
𝛽𝑖𝑡𝜑𝑖𝑗𝑡
(3)
Export from country i to j depends positively on the gross production of country of origin (𝑌𝑖𝑡) and gross demand of the country of destination (𝑌𝑗𝑡) and the bilateral accessibility 𝜑𝑖𝑗𝑡. 𝛼𝑖𝑡 and 𝛽𝑖𝑡 can be interpreted as the MRTs (see equation 4). The MRTs thus show that the trade flow does not solely depend on characteristics of country i and j, but also on the multilateral accessibility of all other countries l, this justifies and states the importance the use of country time variant fixed effects in model 3.
𝛼𝑖𝑡 = ∑ 𝜑𝑖𝑙𝑡∗𝑌𝑙𝑡 𝛼𝑙𝑡 𝑙 𝑎𝑛𝑑 𝛽𝑗𝑡 = ∑ 𝜑𝑗𝑙𝑡∗𝑌𝑙𝑡 𝛽𝑙𝑡 𝑙 (4)
Both models are log transformed and estimated, to show the differences and the importance of the MRT’s. The log transformation of equation 2 results in:
𝐿𝑜𝑔(𝑋𝑖𝑗𝑡) = Log(𝐺) + 𝐿𝑜𝑔(𝜔𝑖𝑡) + 𝐿𝑜𝑔(𝜃𝑗𝑡) + 𝐿𝑜𝑔(𝜑𝑖𝑗𝑡) (5)
The log transformation is practical when interpreting the coefficient estimates. A log-log elasticity is more intuitive when interpreting changes in large numbers. Equation 5 can be estimated by including covariates that represent the exporter and importer characteristics and proxies of the bilateral accessibility. This is the classical gravity model without multilateral resistance terms.
23 𝐿𝑜𝑔(𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡) ≔ 𝛽1∗ 𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡) + 𝛽2∗ 𝐿𝑜𝑔(𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑖𝑗) + 𝛽3∗ 𝐿𝑜𝑔(𝐺𝐷𝑃 𝑒𝑥𝑝𝑜𝑟𝑡𝑒𝑟𝑖𝑡) + 𝛽4∗ 𝐿𝑜𝑔(𝐺𝐷𝑃 𝑖𝑚𝑝𝑜𝑟𝑡𝑒𝑟𝑗𝑡) + 𝛽5∗ 𝐿𝑜𝑔(Population 𝑒𝑥𝑝𝑜𝑟𝑡𝑒𝑟𝑖𝑡) + 𝛽6∗ 𝐿𝑜𝑔(Population 𝑖𝑚𝑝𝑜𝑟𝑡𝑒𝑟𝑗𝑡) + 𝛽7 (𝑅𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) + 𝛽8(𝐹𝑟𝑒𝑒 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) + 𝛽9(𝐶𝑜𝑛𝑡𝑖𝑔𝑢𝑜𝑢𝑠𝑖𝑗) + 𝛽10(𝐶𝑜𝑚𝑚𝑜𝑛 𝑙𝑎𝑛𝑔𝑢𝑎𝑔𝑒𝑖𝑗) + 𝛽11(𝐶𝑜𝑚𝑚𝑜𝑛 𝑐𝑢𝑟𝑟𝑒𝑛𝑐𝑦𝑖𝑗𝑡) + 𝛽12(𝐶𝑜𝑚𝑚𝑜𝑛 𝑐𝑜𝑙𝑜𝑛𝑖𝑧𝑒𝑟𝑖𝑗) + 𝛽13(𝑆𝑎𝑚𝑒 𝐶𝑜𝑢𝑛𝑡𝑟𝑦𝑖𝑗) + 𝛿𝑡+ 𝜀𝑖𝑗𝑡 (6)
The coefficient of interest is β1, insured exports explain part of the bilateral accessibility φ. Controls are added to account for 𝜔𝑖𝑡, 𝜃𝑗𝑡, 𝜑𝑖𝑗𝑡. A standard practice in the literature is to include observables for the trade costs (Yotov, et al., 2016). GDP per capita and population size of both the exporter and the importer are proxies for ω and θ, furthermore year fixed effects are added to the model. This standard model is enhanced by including country pair fixed effects (7). This accounts for the unobservable characteristics by adding dummy variables for each year and each possible exporter-importer pair. When including the country pair fixed effects, variables like distance and contiguity are left out of the model. They are captured by the fixed effects and inclusion would cause perfect multicollinearity. Country pair fixed effects are essential because they capture the effects of possible endogeneity caused by trade policy (Baier & Bergstrand, 2007) and all the unobservable time invariant bilateral barriers or trade costs (Egger & Nigai, 2015; Agnosteva, et al., 2014).
𝐿𝑜𝑔(𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡) = 𝛽1∗ 𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡) + 𝛽2∗ 𝐿𝑜𝑔(𝐺𝐷𝑃 𝑒𝑥𝑝𝑜𝑟𝑡𝑒𝑟𝑖𝑡) + 𝛽3∗ 𝐿𝑜𝑔(𝐺𝐷𝑃 𝑖𝑚𝑝𝑜𝑟𝑡𝑒𝑟𝑗𝑡) + 𝛽4∗
𝐿𝑜𝑔(Population 𝑒𝑥𝑝𝑜𝑟𝑡𝑒𝑟𝑖𝑡) + 𝛽5∗ 𝐿𝑜𝑔(Population 𝑖𝑚𝑝𝑜𝑟𝑡𝑒𝑟𝑗𝑡) + 𝛽6 (𝑅𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) +
𝛽7(𝐹𝑟𝑒𝑒 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) + 𝛽8(𝐶𝑜𝑚𝑚𝑜𝑛 𝑐𝑢𝑟𝑟𝑒𝑛𝑐𝑦𝑖𝑗𝑡) + 𝜕𝑖𝑗+ 𝛿𝑡+ 𝜀𝑖𝑗𝑡 (7)
In accordance with gravity theory, time varying country fixed effects should be included in the model (Feenstra, 2003). These fixed effects represent the multilateral resistance terms that are essential in the theoretical model provided by Anderson and van Wincoop (2003). Baldwin and
24 Taglioni (2007) argue that the omission of these MRTs (or importer and exporter variant fixed effects) is the gold medal mistake of the gravity equations. Due to perfect multicollinearity, adding fixed effects in the {it} and the {jt} dimension, eliminates the possibility of estimating heterogeneous effects or drivers in that field. Therefore, they have to be dropped when estimating the benchmark model (8). All three models (6-8), with year, country pair and country time-variant fixed effects are shown in the results section.
𝐿𝑜𝑔(𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡) = 𝛽1∗ 𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡) + 𝛽2 (𝑅𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) +
𝛽3(𝐹𝑟𝑒𝑒 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) + 𝛽4(𝐶𝑜𝑚𝑚𝑜𝑛 𝑐𝑢𝑟𝑟𝑒𝑛𝑐𝑦𝑖𝑗𝑡) + 𝜕𝑖𝑗+ 𝜔𝑖𝑡+ 𝜃𝑗𝑡+ 𝜀𝑖𝑗𝑡 (8)
4.1.1 Endogeneity issues
The relationship between export credit insurance and export is likely to work both ways. Export credit insurance could enable trade, however growth in trade could also influence the growth in export credit insurance. Simply, without trade there is simply no need for export credit insurance. In the appendix a test for strict exogeneity is shown. This test shows that the past and future values of export credit insurance are not correlated with current exports. This is mainly done to show that there are no feedback loops from past and future values. Strict exogeneity is satisfied.
Omitted variable bias might be an issue as well. This is a bias caused by a variable that is not included in the model and that has an effect on both the dependent and independent variables. Adding fixed effects to capture unobservable characteristics reduces the omitted variable bias. Nonetheless it impossible to know if it is eliminated completely.
As mentioned before in section 3.1, having information on just one of the three major credit insurers might be a problem. If the increase in insured trade is just driven by the fact that the company acquired a larger market share at the expense of the competition, it would be unlikely to have an effect on trade. This would probably lead to a downward bias. However, if the
25 increase in the insured trade is relatively lower than the change in total insured trade (of the market). This would create an upward bias. These biases cannot be completely eliminated; they can only be minimized by the inclusion of fixed effects, which capture the unobservable heterogeneity in the {it} {jt} and {ij} dimensions. Van der Veer (2015) argues that by inclusion of these fixed effects the benchmark model is ‘likely to adjust to changes in the supply of other insurers’, because supply and demand decisions by companies and trade credit insurers are made on the country level, and thus captured by {it} and {jt}.
A way to deal with reverse causality and a possible omitted variable bias is to use an instrumentation strategy, this is set out in the next section.
4.2 IV model
This paper uses an IV strategy to identify if endogeneity of the regressor biases the results of the gravity model. Instrumentation is desired to address the endogeneity issues of bilateral trade tools like trade finance and export credit insurance (Baier & Bergstrand, 2004; Egger & Larch, 2011).The instrumentation uses an exogenous variable that has a direct effect on the explanatory variable Log(insured), but does not influence Log(exports), this regression is the first stage. The second stage uses the predicted values of the first stage to explain the relationship between export credit insurance and exports. An important characteristic of an instrument is that it influences exports only through the export credit insurance, but does not have a direct effect nor an indirect effect through the error term on the dependent variable. The exogenous variation of the instruments on export credit insurance allows to isolate the effect of export credit insurance on exports by omitting the reverse causality channel (Murray, 2006).
The IV strategy used in this paper is similar to one used in an article by Auboin and Engemann (2014), they investigate the effect of trade credit on imports, while using export credit insurance as a proxy. I.e. they are estimating the effect of export credit insurance on imports. The first
26 instrument is the lag of net claims paid by the credit insurer divided by the total amount of exports that were insured (net claims per credit). The second instrument is the lag of the value of narrow money in the economy (‘M1’), which can be considered a measure for the liquidity of the economy. M1 includes all physical money, demand deposits and other liquid assets held by the central bank. The M1 data is subtracted from Macrobond, which in turn subtracts all the data from the countries’ national banks.
The variation of net claims per credit captures the supply side of export credit insurance, i.e. in the recent financial crisis of 2008 the net claims per credit increased. The export credit insurers reduced the supply of export credit insurance, simply because the market was too risky. The credit insurer has the right to always reduce or retreat current credit limits. The demand for export credit insurance is captured by M1, when there is much money circulating in the economy; credit becomes more readily available and cheaper. Hence, net claims per credit should have a negative effect on the amount of export credit insurance and the amount of narrow money should have a positive effect on export credit insurance.
4.2.1 First stage
The first stage (9) investigates the effect of the lagged values of net claims per credit { 𝑖𝑗𝑡 − 1} and M1 { 𝑖𝑡 − 1} on the amount of bilateral export credit insurance between country i and j in time t. The lagged values of claims per credit and M1 are used because one can assume that the market needs to adjust to changes in supply and demand. Also, the time of buying insurance does not correspond to the time-span of the insurance, i.e. insurance is bought in January for the period of January until December, in the case of a turnover policy. This implies that the decision is more likely to be based on last year’s situation.
𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡) = 𝛽0 + 𝛽1∗ 𝐿𝑜𝑔(𝐶𝑙𝑎𝑖𝑚𝑠 𝑝𝑒𝑟 𝑐𝑟𝑒𝑑𝑖𝑡𝑖𝑗𝑡−1) + 𝛽2∗ 𝐿𝑜𝑔(𝑀1𝑖𝑡−1) + 𝛽3∗
27
+𝛽5 (𝑅𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) + 𝛽6(𝐹𝑟𝑒𝑒 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) +
𝛽7(𝐶𝑜𝑚𝑚𝑜𝑛 𝑐𝑢𝑟𝑟𝑒𝑛𝑐𝑦𝑖𝑗𝑡) + 𝛼𝑖𝑗+ 𝛼𝑡+ 𝛼𝑗𝑡+ 𝜀𝑖𝑗𝑡 (9)
4.2.2 Second stage
The second stage (10) is similar to the equation used in the baseline model. Log of exports is regressed on the predicted values of Log (Insured). The control variables that are used are similar to those used in the OLS gravity model. The GDP per capita of the exporter is a proxy for the production in country i. Population captures in a similar way the magnitude of supply of exports. Trade agreements and common currency show the bilateral attractiveness from country i to j. Unobservable bilateral characteristics and the multilateral resistance terms are captured by the country pair fixed effects and importer year fixed effects. For more precision year fixed effects are added to the model. In comparison to the baseline model the TSLS model is unable to use the exporter time invariant fixed effects because the instrument ‘M1’ is in the {it} field.
𝐿𝑜𝑔(𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡) = 𝛿0 + 𝛿1𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑̂ 𝑖𝑗𝑡)+ 𝛿2𝐿𝑜𝑔(𝐺𝐷𝑃 𝑒𝑥𝑝𝑜𝑟𝑡𝑒𝑟𝑖𝑡) + 𝛿3∗
𝐿𝑜𝑔(𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑒𝑥𝑝𝑜𝑟𝑡𝑒𝑟𝑖𝑡) + 𝛿4 (𝑅𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) +
𝛿5(𝐹𝑟𝑒𝑒 𝑡𝑟𝑎𝑑𝑒 𝑎𝑔𝑟𝑒𝑒𝑚𝑒𝑛𝑡𝑖𝑗𝑡) + 𝛿6(𝐶𝑜𝑚𝑚𝑜𝑛 𝑐𝑢𝑟𝑟𝑒𝑛𝑐𝑦𝑖𝑗𝑡) + 𝛼𝑖𝑗+ 𝛼𝑡+ 𝛼𝑗𝑡+ 𝜀𝑖𝑗𝑡 (10)
4.2.3 Endogeneity issues
The results of the gravity model might be biased due to endogeneity issues. The endogeneity can arise because of an omitted variable bias or reverse causality. Instrumental variables deal with both of these issues at the same time. Omitted variable bias had already been minimized by inclusion of a large set of fixed effects, nevertheless it is impossible to know if it is eliminated completely. Furthermore, by including these fixed effects, the model sacrifices valuable degrees of freedom. Without the need of using fixed effects in the {it} plane, the model is more able to
28 estimate the drivers behind the effects, e.g. quality of institutions or development of credit markets.
The lagged values of net claims per credit are used as an instrument for insured trade in the first stage. The net claims per credit can be considered as the default risk. I.e. the risk that a transaction from country i to country j at time t will not be paid. The lagged value of the default risk will influence the supply of trade credit insurance but is not influenced by the value of imports (nor exports). Auboin and Engemann (2014) also raise the point that dividing net claims by the insured trade credit rather decreases the endogeneity concerns, since it truly represents the default risk (see equations 11-12).
𝐶𝑙𝑎𝑖𝑚𝑠𝑖𝑗𝑡= 𝑑𝑒𝑓𝑎𝑢𝑙𝑡 𝑟𝑖𝑠𝑘𝑖𝑗𝑡∗ 𝐼𝑛𝑠𝑢𝑟𝑒𝑑 𝑡𝑟𝑎𝑑𝑒 𝑐𝑟𝑒𝑑𝑖𝑡𝑖𝑗𝑡 (11)
𝐷𝑒𝑓𝑎𝑢𝑙𝑡 𝑟𝑖𝑠𝑘𝑖𝑗𝑡−1=
𝐶𝑙𝑎𝑖𝑚𝑠𝑖𝑗𝑡−1
𝐼𝑛𝑠𝑢𝑟𝑒𝑑 𝑡𝑟𝑎𝑑𝑒 𝑐𝑟𝑒𝑑𝑖𝑡𝑖𝑗𝑡−1 (12)
The instrument is relevant if it has a significant negative effect on the insured trade and if it does not influence exports directly. The default risk especially captures the supply of export credit insurance; therefore, another instrument is used to capture the demand side of export credit insurance. Liquidity (‘M1’) influences the take-up of export credit insurance, but it does not have a direct effect on exports. Only an indirect effect will take place through the financing of exports, hence export credit (insurance).
Next to the reverse causality the omitted variable bias is also resolved. One may still worry about some effects that influence both the default risk, liquidity and the bilateral exports. These effects are captured by the importer time variant fixed effects, country pair fixed effects and year fixed effects (Egger & Nigai, 2015; Agnosteva, et al., 2014). The model is estimated using a two stage least square regression and uses heteroskedastic-robust standard errors.
29 4.3 Heterogenous effects
To dig deeper into the possible drivers of the effect of export credit insurance on exports this section will add institutional and credit market conditions to the model. Excluding parameters about these market conditions can cause serious biases to the variable of interest (Anderson & Marcouiller, 2002). This paper builds further on papers that also tried to investigate heterogeneous effects, to identify how the effect varies in different markets. As discussed before, the heterogeneous effects can only be estimated with the use of country-pair fixed effects and year fixed effects {ij} and {t}. Dropping the country time variant fixed effects {it} and {jt}is necessary because of perfect multicollinearity with the institutional and financial market conditions. It is important to note that the heterogeneous effects should not be interpreted as a causal relationship. The effects of export credit insurance on exports change under different market conditions, but this should be considered as a correlation. This is interesting because the baseline model just looks at the average effect.
4.3.1 Rule of Law
The first addition to the baseline model is the quality of institutions. The rule of law is taken as a proxy for the quality of institutions. This is a clustered variable that looks, inter alia, at the protection of property rights, contractual enforcement and cost of litigation (World bank, 2016). The quality of institutions is especially important when looking at the importer. When trading on an open account, i.e. giving trade credit, the risk is allocated to the exporter. The quality of the institutions in the importer country determines how probable it is that an importer defaults and how costly it is to hold a defaulting importer accountable. The situation of the quality of institutions of the exporter also plays a role. Transactions on a cash in advance basis are less favourable if the exporter’s institutions are weak since the risk lies fully with the importer. In this case, it is more likely that the importer demands the transaction to be on an open account, which can be insured (Schmidt-Eisenlohr, 2013). Antrás and Foley (2015) find a significant
30 positive effect behind the importer’s quality of institution’s and the amount financed by the exporter, however if relationships are established, this effect declines. Therefore, both the rule of law of the exporter and the importer are included in the regression. The main variables of interest in this model are the interaction terms. They are used to estimate the heterogenous effects of export credit insurance in exporter and importer countries with varying degrees of the quality of their institutions. The rule of law estimator, is a measure that ranges from -2.5 to 2.5, and is available for almost all exporters and importers throughout the 2000-2015 period, except for 2001.
𝐿𝑜𝑔(𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡) = 𝛿0 + 𝛿1𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡) + ⋯ + 𝛿9(𝑅𝑂𝐿𝑖𝑡) + 𝛿10(𝑅𝑂𝐿𝑗𝑡)+𝛿11(𝑅𝑂𝐿𝑖𝑡∗
𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡)) + 𝛿12(𝑅𝑂𝐿𝑗𝑡∗ 𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡)) + 𝛼𝑖𝑗+ 𝛼𝑡+ 𝜀𝑖𝑗𝑡 (13)
4.3.2 Financial market maturity
Several proxies for financial market development are added to investigate the heterogenous effect of financial market maturity on the effect of export credit insurance on exports. Missing credit markets can have huge effects on importers and exporters, especially when considering long distance trade. Giving trade credit as opposed to letting an importer pay in advance gives the importer the freedom to not reduce its working capital, even before he has the product in inventory (Jones, 2010). Therefore, the effect of export credit (insurance) on importers, who face a missing or underdeveloped credit market, is probably larger than for importers who do not face these restrictions. Moreover, giving trade credit reduces transaction costs and the costs of lending (Ferris, 1981).
On the other hand, exporters do sap their working capital. Many financial institutions are only willing to give new credit to exporters if they are insuring their risky receivables (Jones, 2010). Especially if the receivables make up a large share of the assets. This seems to be more an issue if credit markets are underdeveloped (Becue, 2008). Also, trade credit seems a way to compensate for falling prices, this means that by giving trade credit prices remain higher, thus
31 higher exports, in comparison with cash in advance exports. Absence of proper credit markets could lead to export price reductions (Demir & Javorcik, 2018).
Schmidt-Eisenlohr (2013) also suggest that the interest rates play an important part in the contractual decision-making process. Handing out trade credit is more attractive if the local interest rates are low, and if the destination country’s interest rates are high. Felbermayr and Yalcin (2013) make use of four different variables to capture the ‘maturity’ of the financial markets and estimate the export-enhancing effect of export credit guarantees by looking at differences in the destination countries’ financial markets. They use liquid liabilities to GDP, private credit to GDP, Stock market capitalization to GDP, stock market total value traded to GDP. All are measured in percentage points, so a one-unit increase is in fact a percentage point increase. These measures are proposed as proxies for the financial market maturity by Beck et al. (2010). The difference with Felbermayr and Yalcin (2013) is that this paper includes proxies for financial market maturity in both the importer’s and exporter’s country.
The financial market maturity proxies are added separately from each other, The interaction variables are added to the model to estimate possible heterogenous effects, these are the variables of interest. The financial market maturity proxies are country time variant variables {it} and {jt}. Therefore, only country pair {ij} and year fixed effects {t} can be added to the model.
𝐿𝑜𝑔(𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡): 𝛿0 + 𝛿1𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡) + ⋯ + 𝛿9(𝐹𝑀𝑀𝑖𝑡) + 𝛿10(𝐹𝑀𝑀𝑗𝑡)+𝛿11(𝐹𝑀𝑀𝑖𝑡∗
𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡)) + 𝛿12(𝐹𝑀𝑀𝑗𝑡∗ 𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡)) + 𝛼𝑖𝑗+ 𝛼𝑡+ 𝜀𝑖𝑗𝑡 (14)
4.3.3 Financial market efficiency
Lastly, this paper incorporates financial market efficiency into the analysis. Financial market efficiency, proxied by the net interest rate margin; the average ex-post mark up of lending
32 activities of financial institutions. It differs from ex ante measures because it includes losses of non-performing loans. The second indicator is overhead costs, which is measured as the overhead costs as a share of its total assets (measured in %). Both measures capture the efficiency of the market. Lower efficiency could imply higher borrowing costs, so imperfect or absent credit markets (Schmidt-Eisenlohr, 2013). Both measures are suggested by Beck et al. (2009) as indicators for financial market efficiency. Again, the country time variant fixed effects have to be dropped, since the FME variables are country time variant. Country pair and year effects are in place. The variables of interest are the interaction terms, since they show how the effect of export credit insurance on export differs under different levels of efficiency of the financial market.
𝐿𝑜𝑔(𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡): 𝛿0 + 𝛿1𝐿𝑜𝑔(𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡) + ⋯ + 𝛿9(𝐹𝑀𝐸𝑖𝑡) + 𝛿10(𝐹𝑀𝐸𝑗𝑡)+𝛿11(𝐹𝑀𝐸𝑖𝑡∗
33
5. Findings
In this section the main findings of the research are presented. With both the gravity model and additionally the instrumentation strategy, this paper shows that there is a consistently positive but less than proportionate effect of export credit insurance on exports.
5.1 Baseline gravity model
The results of the gravity models are depicted in the table 2. The table includes in total three gravity models (1-3) corresponding equations (6-8). As theory suggest, there exists a potential bias in the regressors, the MRTs are captured by adding fixed effects. Since the regressors only capture bilateral resistance terms. Adding time invariant country pair and time variant country fixed effects would lead to unbiased estimators.
The main variable of interest is Log(insured) which captures the natural logarithm of the bilateral annual insured exports{ijt}. The dependent variable in all cases is Log(exports) which captures the natural logarithm of the annual bilateral exports {ijt}. Due to the use of natural logarithms, the coefficients should be interpreted as a log-log elasticity. The section below will show that model 1 and 2 are biased and that 3 is the preferred model. Nevertheless, when investigating the heterogeneous effects model 2 also shows its usefulness. Ergo will not be discarded.
Results and interpretation
Model 1
Model 1 is similar to the classical gravity model in its set up and outcomes. It basically portrays the gravity model with explanatory logical control variables, with the addition of year fixed effects, which intend to describe part of the variation in the bilateral annual exports. Looking at the 𝑅2(Around 0.81) one can conclude that these variables together do not perform badly and possess a fair amount of explanatory power. Almost all of the regressors come out as expected.
34 Table 2 Regression Results Dependent variable: Log(Exports) (1) (2) (3) Log(Insured) 0.177*** 0.028*** 0.011*** (0.005) (0.002) (0.002) Log(Distance) -0.680*** (0.023) Log(GDP/capita exporter) 0.703*** 1.146*** (0.014) (0.049) Log(GDP/capita importer) 0.487*** 0.592*** (0.026) (0.078) Log(population exporter) 0.699*** 1.135*** (0.009) (0.070) Log(population importer) 0.749*** 0.203 (0.012) (0.158)
Regional Trade Agreement 0.032 0.178*** 0.069
(0.053) (0.035) (0.055)
Free Trade Agreement -0.041 -0.229*** -0.030
(0.045) (0.025) (0.047) Contingency 0.438*** (0.082) Common Language 0.522*** (0.051) Common Currency -0.055 0.130*** 0.086* (0.050) (0.036) (0.047) Common Colonizer 1.551*** (0.131) Same Country 0.488*** (0.137) Trade Multiplier 6.756 1.073 0.432 (0.190) (0.082) (0.081)
Year FE Yes Yes No
Country-pair FE No Yes Yes
Country-Time-varying FE No No Yes Observations 41,100 41,100 41,100 R2 0.809 0.977 0.984 Adjusted R2 0.809 0.974 0.980 AIC BIC Residual Std. Error 1.022 (df = 41071) 0.375 (df = 36269) 0.327 (df = 32872)
Note: Three gravity models. Data set excludes all observations with a share (defined as insured exports/exports) higher or equal than 0.3. Robust standard errors clustered at the country pair are
35 Distance has a negative effect on bilateral exports, if distance increases by 1% exports decrease on average by 0.67%. The GDPs (per capita) of the importer and exporter have a positive effect. They represent the total production and the demand. Bilateral characteristics, such as contingency, having a common language, a common colonizer or if both countries were once one all have a positive effect on bilateral exports. The coefficient of interest, Log(insured), shows that a 1% increase in bilateral annual insured trade increases on average annual bilateral exports by 0.177%. The coefficient can be transformed in a trade multiplier to show how one unit increase in insured exports leads to an X unit increase in exports. The trade multiplier can be calculated in the following way: 𝛽1∗𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡
𝐼𝑛𝑠𝑢𝑟𝑒𝑑𝑖𝑗𝑡 . The coefficient of interest, is multiplied by
the average exports divided by the average insured exports (van der Veer, 2015). Across the sample, average annual bilateral exports are €2,573.5M, while average bilateral insured trade is €67.6M. This leads to a multiplier of 6.738 = 0.177 ∗2,573.5
67.6 . This multiplier implies that every euro of insured trade will increase exports by €6.74. However, this result could be subject to endogeneity issues, and should be interpreted as a correlation.
Model 2
Adding country pair fixed effects is important to control for all unobservable bilateral characteristics that could influence trade. Furthermore, they also capture part of the multilateral resistance terms. Excluding these could lead to substantial biases in the regressors. In total the sample counts 4808 unique country pairs, they are added as dummies to the model, together with year fixed effects. GDP per capita of the importer (as a proxy for demand) has a positive effect on the bilateral annual exports, on average a 1% increase of GDP per capita increases bilateral exports by 0.6%. The income level of the exporter (as a proxy for production) has even a larger effect, a 1% increase leads to 1.1% increase in bilateral trade. Furthermore, the common
