University of Amsterdam
Faculty of Business and Economics
Private Trade Credit Insurance Promoting
International Trade:
More so when credit markets are constrained?
Master Thesis July 2017
Author: Marta Radinovic Lukic Student no.: 11400536
Statement of Originality
This document is written by Marta Radinovic Lukic who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is 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.
Abstract
With this study, I examine whether private trade credit insurance promotes international trade, especially in the presence of credit market constraints in the importing and exporting economies. The underlying idea here is that through reducing trade-related uncertainties and through mitigating credit constraints, insurance can help promote international trade. I tested for these ideas through applying a new unique dataset on privately insured trade credit covering the period of 2000-2015, provided by one of the three market-leading private insurers, on a theoretical gravity model of trade.
The results found in the study imply that trade credit insurance has positive and significant effects on trade, and that each euro spent on trade credit insurance increases international trade by more than the euro invested. I find these positive effects to be larger in cases when insurance is targeted towards countries with tighter credit market conditions. Thus, these results suggest that private trade credit insurers play a significant role in promoting trade through minimizing trade-related uncertainty, potentially more so in times of financial distress when credit constraints are tighter.
1 Introduction ...2
1.1 Motivation and Research Question ...3
1.2 Organization of the Paper ...4
2 Background and Theory ...5
2.1 Trade Finance...5
2.2 Private Trade Credit Insurance ...6
2.2.1 Past and Current Trends in Private Trade Credit Insurance ...7
2.3 Theoretical Underpinnings of Trade Credit Insurance ...10
3 Literature Review ...13
4 Data and Methodology ...16
4.1 Empirical Model ...16
4.1.1 Specification of the Benchmark Model ...17
4.1.2 Extensions to the Benchmark Model ...18
4.2 Econometric Considerations ...19
4.2.1 Endogeneity ...19
4.2.2 Zero (insured) Trade Flows ...20
4.2.3 Robustness Checks...20
4.3 Sample and Data ...22
4.3.1 Sample...22
4.3.2 Data ...22
5 Results and Discussion ...25
5.1 Benchmark Model Results ...25
5.1.1 Strict Exogeneity ...28
5.1.2 Sensitivity Analysis ...29
5.1.3 Other Estimation Approaches ...31
5.2 Extenstions to the Benchmark Model ...33
5.2.1 Financial Market Maturity ...35
6 Conclusion ...39
6.3.1 Further Research ...40
References ...41
1 Introduction
International trade is an important component of the global economy, and it significantly contributes to global and domestic economic growth. Encouraging international trade should thus, at least in theory, be of great interest for individual countries and the world as whole.
Encouraging international trade can be done through multiple channels, for instance through policies which aim at lowering border tariffs, introducing free trade agreements, or improving the quality of contract enforcement. Through reducing the costs of trade, these policies increase the probability that domestic firms engage in export-related activities. Another factor which proves rather important in terms of trade, is the exporting firms’ access and management of (external) trade finance, and it is in fact estimated that 80% of all exporters rely on some sort of trade related finance (CGFS, 2014).
An important feature related to trade and trade financing regards the fact that exports often are made on credit, allowing for the shipment of the goods to be completed before the payment is to be made. However, when exporters offer to extend credit, they face a substantial risk of non-payment by their customers, due to commercial- or political reasons. Further, because of the time lag between the production of the good, delivery, and actual payment, the exporting firms’ liquidity is affected negatively (Auboin & Eggerman, 2012; CGFS, 2009). This added risk exposure and need for liquidity has necessitated the purchase of trade credit insurance for many exporters, and today 11% of all international transactions are covered by trade credit insurance (ICC, 2016). Through purchasing trade credit insurance exporters reduce the costs that follow from various trade-related uncertainties, such as for instance losses stemming from unexpected customer insolvencies, which in turn can prove stimulating for the exporting company’s business activity. Several studies confirm this effect, implying not only that trade credit insurance has a positive effect on trade, but also that trade credit insurance has a multiplying effect on trade through, among other channels, allowing for transactions that otherwise may not have occurred due to being too risky (Chor & Manova, 2012, Morel, 2010).
Although empirical results seem affirming, the data availability on trade credit insurance is rather limited, hindering an overall approach to the subject matter. This point is made particularly valid for private trade credit insurance1, and to my knowledge, only the study of Van der Veer (2015) directly addresses the effects of private trade credit insurance on trade. Seen that private insurers are important actors in the field of trade credit insurance, covering approximately 2,300 billion EUR worth of exports in 2015 (ICISA, 2017a), it becomes important to further investigate this link.
1.1 Motivation and Research Questions
With background in above, in this study I aim to account for the size of the trade multiplier caused by private trade credit insurance, with the main purpose being to test how the multiplier is affected by (changes in) credit market conditions. By doing so, I aim to contribute to a small but significant expansion to the literature on private trade credit insurance, and provide useful insights into why, and through what mechanisms, private trade credit insurance proves valuable for international trade.
To accomplish the above, I first apply a theory-consistent gravity model on trade by using a unique dataset on private trade credit insurance, as supplied by one of the market-leading private insurers. In the model, the main independent variable is that of private trade credit insurance, which is then regressed on the levels of bilateral trade. This approach has by large been guided by the study of Van der Veer (2015), and allows me both to consistently estimate insurance’ effects on trade, and to account for the robustness of Van der Veer’s (2015) results.
I then go on by extending the model in two ways, to account for the main purpose of the paper which is to test for the changes in the multiplier in relation to (changes in) global credit markets. First, because my dataset covers the period of 2000-2015, I estimate the changes in the multiplier during the years of the recent financial crisis. Because the crisis had large disruptive effects on global trade and on the global credit markets, the effects of insurance on trade can here be expected to vary. Second, I extend the model through including variables that serve as a proxy for the condition of the financial and credit markets in the destination country, and test for whether the multiplier is affected by these. With background in above, I have in my study been guided by the following questions:
a) Does private trade credit insurance have a trade-multiplying effect on international trade?
b) In what way is the relationship between private trade credit insurance and trade affected by (changes in) the credit market conditions?
For question a), my hypothesis is that insurance does have a multiplying effect on international trade, as suggested by previous literature. For question b), I expect that the multiplier varies with the credit market conditions, because trade is largely dependent on the availability and cost of trade financing. Further, due to the mechanisms which will be explained below, I expect the multiplier to be larger when exports are targeted towards market that are relatively more credit constrained.
1.2 Organization of the paper
This paper is organized as follows. In the next section, I first provide a background on private export credit insurance, together with relevant theoretical underpinnings. Here, I also account for the potential mechanisms in which insurance is believed to influence trade. In the third section, I present some of the previous literature on the topic, followed by a section on the applied empirical methodology, including any econometric considerations related to the study. Subsequently, I provide the empirical findings together with a thorough discussion of the results, and finally, I present my conclusions.
2 Background and Theory
In this section, I briefly introduce the background on trade finance in general, and trade credit insurance in particular. I also shortly explain the difference between public and private trade credit insurance. Finally, I present some of the mechanisms in which trade credit insurance is expected to influence international trade, which serves to support the analysis of the empirical findings of the study.
2.1 Trade Finance
Trade finance, much like it sounds, regards the financing processes connected to commerce and international trade. Every firm that considers expanding abroad needs to decide what type of financing to use in their transactions. Either the firm can choose to support the transaction through bank-intermediated trade finance, in which a bank guarantees the payment of the transaction on the behalf of the purchasing entity (importer) to the supplying entity (exporter), or it can choose to manage its’ finances through inter-bank credit, either on open account terms (shipment before payment) or on cash-in-advance terms (payment before shipment) (CGFS, 2014).
What by large determines the choice of trade financing is the underlying risk that a buyer defaults on the seller, or that the seller does not deliver goods as originally agreed in the contract (Antras & Foley, 2014). Also, due to the time required to ship goods globally, another important determinant of the choice on trade finance is the bilateral distance between the trading partners. Antras and Foley (2014) show that in countries that are further away from each other, and in which contract enforcement laws are weak, thus trade-related risk is higher, transactions are more often made on pre-payment terms. In other, less risky situations, exporters often choose to extend credit to the importers. However, because the underlying risk is never fully eliminated, firms often choose to protect their accounts receivable through purchasing trade credit insurance. As such, trade credit insurance represents a valuable trade finance product, which by large depends on the choice of trade financing (Jones, 2010).
Through helping companies manage their international payments, reduce the risks associated with trade, and provide the needed working capital, trade finance plays a key role in promoting international trade (CGFS, 2014). The lack thereof however, can have significant impeding effects on trade. In fact, many studies confirm the negative consequences the lack of trade finance left on international trade in the aftermath of the financial crisis (Chor & Manova, 2011, Antras & Foley, 2014). Because trade credit financing had to compete with other parts of the credit markets for the same amount of
reduced liquidity in the crisis period, short-term trade financing experienced a drop by 11.5% (Korinek et al., 2010; Auboin & Eggerman, 2012). This gap in trade financing, which was estimated to 25-500 billion USD, proved to be the second biggest contributor behind the great trade collapse in 2008, following the drop in the global demand (Auboin & Eggerman, 2012; CGFS, 2014). These large disruptions in trade finance and global credit markets successively left large impacts on the market for private trade credit insurance, and as Van der Veer (2011) estimates, the lack of insurance potentially accounted for 5-9% of the drop in world exports during the crisis period.
2.2 Private Trade Credit Insurance
Trade credit insurance first emerged in the end of the 19th century, and is today a multi-billion line of business. As explained above, the need for insurance stems from the fact that exports are often made on credit, in which the exporting firm (seller) extends credit to the importer (buyer). While this action is beneficial to the importing company, among others because it is now allowed to resell the purchased goods before a payment is to be made to the exporter, the exporting company instead faces a significant amount of credit risk, due to commercial and/or political reasons (Jones, 2010). The graph below represents a simplified version of the mechanism behind trade credit insurance:
As can be seen above, an exporting company buys insurance for the price of a premium, charged based on its turnover and credit risk. In return, the now insured company receives protection on any losses against late- or non-payments by its’ customers, normally covering 75-95% of the transaction value (Jones, 2010, ICISA, 2017b). This type of insurance mainly covers commercial and political risk, where commercial risk refers to the situation in which a trading partner is unwilling, or unable to pay for the underlying transaction. Political risk on the other hand, covers the risk of a loss resulting from political
Graph 1. Trade Credit Insurance
* The right of subrogation provides a trade credit insurer that has paid a claim the contractual right to collect directly from the buyer who has failed to pay.
actions of the importing country, and is a risk which is beyond control of the involved entities. Such risks include those of expropriation (seizure or nationalization of a business, government regulations), currency inconvertibility (inability to convert money from local currency to foreign exchange), and political violence (civil wars, domestic unrest, revolutions) ((ICISA, 2017b, Waters, 2015).
Besides offering a peace of mind to the exporter, trade credit insurance also leaves a positive influence on economic stability, as risk is now shared with private insurers who are better able to absorb the losses stemming from trading partners insolvencies (Jones, 2010). Risk is further mitigated as traded goods can be used as collateral, and because the transactions are easy to track (Korinek et al., 2010). Numbers also show an exceptionally low loss rate for this field of business (calculated as claims over total exposures), amounting to 0.17% for the period of 2005-2012, pointing at sound underwriting practices (CGFS, 2014).
Finally, it is important to mention that trade credit insurance is offered either publicly, through the export credit agencies (ECA’s), or through private insuring companies. While both private and public trade credit insurance function in a comparable manner, a few significant differences need to be lifted between the two. In the early periods, up to the beginning of 1980s, trade credit insurance was mainly provided publicly either because the private actors were not willing to underwrite this risk, or because premium levels charged by ECA’s were much lower than those of the private insurers (Egger & Url, 2006). However, through setting subsidized premium rates, the public actors were potentially distorting international trade, and were facing substantial losses because the premium incomes were insufficient to finance the losses they faced. As a response, the OECD, the EU and Berne Union2 together tried to correct for this failure through posing restrictions on the public providers of insurance, allowing them to operate only in cases were risk was deemed non-marketable (EU Council Directive, 98/29/EC). Following these restrictions, public insurance was limited to long credit maturity transactions (2-5 years), and to high-risk countries, whereas private insurance is instead concerned with relatively less high-riskier countries, and with transactions with credit maturity of less than a year (although most often 30-90 days)3. The regulatory restrictions, together with the emergence of local banking sectors and the establishment of global inter-bank links, has since given the private actors more ground in the field of trade credit insurance (Auboin & Eggerman, 2012).
2.2.1 Past and Current Trends in Private Trade Credit Insurance
Trade credit insurance has been a relatively steady field of business during the last decades, except for the major challenges the industry faced during the period of the recent global crisis. When the crisis hit the economies worldwide in 2008, international trade immediately suffered, and trade levels fell by approximately 20% during the third quarter
2 Berne Union is the leading association for the global export credit and investment insurance industry (Berne Union, 2017a).
3 Since 1998, short-term public export credit insurance to borrowers from the OECD region are limited to commercial conditions (EU Council Directive, 98/29/EC).
of 2008. At the same time, trade finance was becoming relatively more expensive and prices of letters of credit were doubling or tripling for certain countries, which is why the exporting companies resorted to other forms of trade financing (Malouche, 2009). In this unstable environment, firms were facing much higher risks of insolvency and started delaying payments (Antras & Foley, 2011, Auboin & Eggerman, 2012). As a response, exporting firms sought to protect themselves from possible defaults which spurred the demand for trade credit insurance. On the other hand, trade credit insurance companies sought to reduce their risk exposures and avoid potential default themselves, which resulted in the shortage of supply of trade credit insurance, together with an increase in pricing.4 Due to these developments on the global markets, insurers were now facing significant losses and increase in claims as well, as shown by the graph below (Morel, 2010):
Despite the challenging environment however, the private insurers proved resilient to the crisis mainly due to two reasons. First, because they maintain the right to reduce or cancel credit limits in the case the buyer’s financial situation or overall political situation deteriorates5, they were quickly able to reduce their overall exposure and thus limit their losses (Van der Veer, 2015, Jones, 2010). Second, the rapid policy responses following the crisis, from both national and international actors, worked in great favour for the insuring companies6. One of the main policy responses, brought forward by the London G-20 trade finance initiative in April 2009, included an increase in credit insurance and risk mitigation capacity by export credit agencies, which stepped in with programmes for short-term lending (Auboin & Eggerman, 2012). For these reasons, private insurers could continue supporting international trade during the challenging period. In fact, private insurers have since the start of the financial crisis paid out claims amounting to 56 billion euros, thus showing ample support for international trade (Berne Union & ICISA, 2017).
4 As shown by the IMF survey, in this period pricing of export credit insurance went up almost double, while 37% of exporters or banks requested more export credit insurance (Asmundson et al., 2011).
5 In this case the changes solely apply to future businesses and previously accepted risks remain covered. (Jones, 2010). 6 A framework containing best-practices regarding risk-sharing between multilateral development banks, export credit agencies and private insurers was suggested by the IMF already in 2003 (Auboin & Eggerman, 2012).
Graph 2. Trends in Private Trade Credit Insurance
Source: ICISA, 2017a.
0,00% 20,00% 40,00% 60,00% 80,00% 100,00% 0 1000 2000 3000 4000 5000 6000 7000 M ill io n E UR
In the years following the crisis, the private trade credit insurance has been on an upward trend, and in 2015 private credit insurance was approximately covering 2,300 billion euros worth of exports (see below) (ICISA7, 2017a). However, the current global developments, which are marked by high uncertainty in the commodities sector and in the global politics, leave a negative influences both on international trade and on trade credit insurance. This setting, where trade is facing a downward trend, suggests both that private trade credit insurance will become relatively more important in the future, but also that private trade credit insurers will undergo some challenges. Already now, it can be noticed that insurance is gaining relatively more importance, demonstrated by the fact that the total insurance coverage (insured over total exports) went up from 10% in 2015, to 11% in 2016. However, insurers have also seen a decline in the value of total turnover covered, a drop by 7% from 2015 to 20168, accompanied by an increase in the claims ratio, up 38% between 2015 and 2016 (ICC, 2016). A further concern is the rising loss ratio, calculated as claims paid over premium income earned, which stood at 75% in 2015 as opposed to 54% in 2014. However, it is important to note here that this loss ratio is higher for public export credit insurers than that of private actors.
7 ICISA is “The International Association for Private Trade Credit Insurance”, whose members cover over 95% of the world’s private trade credit insurance (ICISA, 2017a).
8 This number regards the whole industry of short-term insurance coverage, i.e. both private and public insurers (ICC, 2016).
Graph 3. Insurance Coverage
Source: ICISA, 2017a.
0 500 1000 1500 2000 2500 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Bi lli o n E UR
2.3 Theoretical Underpinnings of Trade Credit Insurance
While no pure theory that looks solely on trade credit insurance has been developed, several findings from the literature support the hypothesis that insurance leaves a positive influence on trade. In this text, I present two of the main channels in which this may be the case, and building on these arguments I further argue why insurance could have played a particularly important role during the recent financial crisis.
The first, and arguably the main channel through which insurance leaves a positive effect on trade is through reducing the trade-related uncertainty. One of the first who looks at this channel is Funatsu (1986), who in his paper examined “the fundamental role of export
credit insurance and its impact on an exporter’s decision making within a theoretical framework” (p.680). Through applying a model in which firms act under uncertainty with
a utility maximizing objective, he finds that firms are expected to trade less than optimal due to the uncertainties that prevail in foreign markets. Building on these results, he finds that governments can enhance exports through setting a “more than favourable” insurance premium rate (Funatsu, 1986). Although that is not the case with private export credit insurance, Funatsu’s idea can be generalized to any risk-averse firm, as insurance in this case will reduce uncertainty and thus create business opportunities which would not have occurred due to being too risky. Several papers build on this idea, and among others, Abraham and Dewit (2000) show that guarantees indeed reduce uncertainty in profits for risk-averse exporters, and thus enhance exporting activities.
This reduction in uncertainty not only benefits risk-averse exporters, but it also significantly reduces costs associated with trade. These costs are potentially very large and can, according to Anderson and Wincoop (2004), amount to 170% of the free on board (hereafter f.o.b.) export price in rich countries (and more so in poor countries). Two of these costs are directly related to trade credit, namely cross-border information costs and costs of contracting and insecurity, which together amount for 9% of the f.o.b. price (Egger & Url, 2006). More concretely, some of these costs are assigned to the differences in language and jurisdiction, or to insufficient information about a buyer’s liquidity, financial status or payment behaviour (Egger & Url, 2006). These costs are however minimized with insurance coverage as private trade credit insurers have the right to recover any losses from the buyer, and can take legal action against any delinquent buyers (Jones, 2010). Further, the insurers hold valuable information on millions of companies worldwide, and are aware of any risks particular to the country of destination. This is why the costs associated with information are minimized when exporters purchase insurance (Jones, 2010).
So far, these mechanisms support the idea that insurance leaves positive effects on trade, but not necessarily that it leaves multiplying effects on trade as suggested by previous research. In this regard, a multiplying effect9 of insurance on trade would imply that an
9 In the empirical part I account for the size of the multiplying effect on trade, as caused by insurance, through looking at the size of the multiplier. The multiplier in this regard simply presents an effectiveness ratio which looks at how much a euro invested in insurance generates in euros of exports.
increase in insurance coverage would lead to a more than proportional increase in trade and for that to occur, it must be that insuring exports leads to an increase in uninsured exports. As Dixit (1989) shows, exporting companies’ current market participation is by large affected by prior experience. This idea is confirmed among others by Roberts and Tybout (1997), who show that prior export experience increases the probability of exporting by as much as 60 percentage points. Furthermore, Antras and Foley (2011) show that when trading partners establish a relationship with each other, the transactions are less likely to occur on cash in advance terms and instead other forms of payment will be preferred. Thus, when exporters learn about the creditworthiness of their trading partners, they are more likely to continue trading without the need for insurance coverage (Antras & Foley, 2011; Van der Veer, 2015). In summary, if insurance first helps companies expand their business through the above mechanisms, and later allows them to continue trading without the need of insurance, then a multiplier should naturally follow.
The second channel in which insurance leaves positive effects on insurance is through positively influencing the financial management of both the exporting and the importing companies. When an exporter offers to extend credit, this is beneficial for the importing company which now holds valuable access to supplier credit. Due to the time value of money, through having access to supplier credit, importers are now able to increase other purchases and extend their business to other companies (Van der Veer, 2015). Also, because supply credit reduces the need of precautionary money holdings and thus reduces the importers’ transaction costs, it potentially stimulates the demand for imports (Ferris, 1981). In this regard, access to supply credit can be deemed even more valuable when exports are directed towards markets with tighter credit market conditions. In these situations, the importing company might not be able to pay in advance or use bank-intermediated trade-finance products in order to guarantee their payment. Thus, insurance coverage can here act as an instrument which mitigates credit market restrictions in the importing economies and as such, the enhancing effect of insurance is expected to be largest in markets which are relatively more credit constrained (Felbermayr & Yalcin, 2011).
Insurance coverage is also valuable for the financial management of the exporting companies, as it can leave positive impacts on the firms’ liquidity (Becue, 2008, Jones, 2010). For instance, when covered by an insurance policy the exporter can now more easily borrow from a commercial bank against the now protected insured trade receivables which act as collateral. The borrowing is potentially made on better terms and conditions, as the trade credit insurer most likely has better credit risk ratings than the exporter itself. Thus, through the insurance, firms’ liquidity may be spurred relatively more in relation to the firms that are not insured (Jones, 2010). Also, firms with better access to credit tend to offer more trade credit, which can again stimulate the import demand presented in the channel above (Van der Veer, 2015).
Finally, it can be argued that insurance plays a more vital role in case of financial and credit market frictions. Because levels of uncertainty are relatively higher in these cases, as compared to the status quo, the uncertainty reduction following insurance coverage is also relatively larger in case of financial frictions. Further, because of the long duration of
the shipments present in international transactions, firms are in a large need of working capital which becomes relatively harder to access in case of financial frictions (Amiti & Weinstein, 2011). However, as explained above, insurance can help mitigate the liquidity constraints through unlocking bank financing (Morel, 2010), which is especially important in sectors which are financially vulnerable (Felbermayr et al., 2012). Lastly, insurance can allow the trading partners to continue their business as usual, even when other firms are forced to default. This way, through allowing for the trading relationship to continue during a crisis or a downturn, the involved companies can potentially profit from the strong export growth that normally accompanies post-crisis periods (Moser et al., 2008).
3 Literature Review
Due to the lack in availability of reliable data, the field of trade credit insurance is still relatively under-researched, a point especially made valid for private trade credit insurance. Because of not many articles have yet been published on private trade credit insurance, I present in this section some of the main academic findings on both public and private trade credit insurance, and argue how this paper contributes to these.
One of the first articles that studies the direct link of public trade credit insurance’ effects on trade is that of Egger and Url (2006). In their article, the authors explicitly test whether public export credit insurance creates a significant amount of additional exports, and whether insurance changes the regional or industry structure of international trade. In particular, their hypothesis is that “destination countries with higher informational
barriers and insecurity should benefit more from insurance coverage.” (p.400). They test
this hypothesis through adapting a gravity model of trade on Austrian goods exports data from 1996-2002, disaggregated at the two-digit industry level for each country and year. Their methodological approach includes separating between the short-term and long-term effects of insurance through a random model estimation. They further simulate the impact of insurance on the structure of Austrian exports, seen from both industry- and partner country levels. For the short-term effects, they find a significant but rather small effect with an elasticity of 0.04-0.05. In the long-term scenario, the effects prove larger and their preferred model reveals an elasticity of 0.44, suggesting a trade multiplier of 2.8. With this result, they conclude that insurance creates a multiplier through reducing trade-related uncertainties and through providing information about a trading partners’ credibility, thus allowing for repeated transactions without the use of insurance. They further assign the large difference between short and long-term effects to the difference in timing of the export credit guarantee and the actual shipment of the good, which in the case of public trade credit insurance can amount to 5 years. They further show that increased public export insurance coverage increases trading with higher-risk regions (Egger & Url, 2006).
Moser et al. (2008) instead look at export promotion and political risk in the case of Germany, testing the hypothesis that political risk represents an important hidden transaction cost for international trade. They also adapt a gravity model in order to test their hypothesis, in which they include a variable on political risk. Their dataset compromises data on export credit guarantees from the German export credit agency Euler Hermes, covering 130 countries over the period 1991-2003. Besides estimating a random effects model following Egger & Url (2006), they also apply a dynamic panel estimator which incorporates the effect of past exports on current exports. They find support for their hypothesis and show that political risk reduces exports, and that guarantees increase exports. When using a random effects model, Moser et al (2008) find a multiplier of approximately 6, claiming that this represents an upper bound of the multiplier. After
adjusting for the dynamic structure of exports, the multiplier instead drops to 1.7, concluding that a dynamic model is to be preferred when estimating the effects of insurance. They however note that the effects of guarantees are sensitive to the sample of countries included in the study (Moser et al. 2008).
Following these two studies, Felbermayr and Yalcin (2011) also test for the hypothesis of the export credit guarantees’ effects on exports. However, the authors extend their study by looking at whether public guarantees remedy the export-restricting effect of credit market imperfections. They do so by looking at sector- and country-specific export levels, and make use of an extensive dataset on German export credit guarantees dating from 2000 to 2009. They find that guarantees do increase exports by sector, and that an increase in level of guarantees by 1 percent increases exports by 0.012-0.017 percent, which in their study translates to a multiplier of 0.47. They further find that “export creation” is heterogeneous in nature, suggesting a relatively more profound multiplier in certain sectors, regions or income groups. They also find small support that the export enhancing effect was largest in 2008, the peak year of the financial crisis. Finally, they find that the quality of institutions in the destination country affects exports positively, and they find some evidence that guarantees helped reduce the drop in exports caused by the financial crisis, especially in sectors that were relatively more dependent on external finance (such as aviation, shipbuilding and transportation) (Felbermayr & Yalcin, 2011).
Following this study, Felbermayr and Yalcin now together with Heiland (2012), apply the same dataset in order to test for the (claimed) causal effect of guarantees on sales and employment of the exporting firms. Through applying a diff-in-diff model in which they create an appropriate control group of untreated firms (through a matching strategy), they find results suggesting that insurance increases firm-level sales and employment by on average 4.5 and 3.0 percentage points respectively. They also find that this effect is significantly larger during the financial crisis, again supporting the idea that guarantees mitigate financial constraints and help firms expand their scale of activity (Felbermayr et al., 2012).
The study most relative to mine, and the only one which directly addresses the role of private trade credit insurance on international trade, has been that of Van der Veer (2015). In his study, Van der Veer estimates the effects of private trade credit insurance on international trade through adapting a gravity model of trade. He retrieves the data on private trade credit insurance from one of the three main private insurers, with a sample ranging from 1992-2006. Van der Veer finds a significant trade multiplying effect of private trade credit insurance, suggesting that 1 euro of insured exports generates 1.3 euros of exports. Using claims ratios as an instrumental variable, Van der Veer reaffirms the results, implying that his findings are robust after accounting for possible endogeneity issues. Based on these findings, Van der Veer (2011) further finds that the drop in private trade credit insurance supply during the recent crisis can explain 5–9 percent of the collapse in world trade, and 10–20 percent of the drop in European exports during the crisis period.
Previous literature suggests consistent positive effects of trade credit insurance on trade. However, the effects have been found to be heterogeneous in nature, as they vary across different geographical regions and industry sectors. Further, previous academic findings also suggest that incorporating long-term effects results in a larger multiplier (Egger & Url, 2006; Moser et al., 2008). Incorporating the long-term effect of insurance is however less relevant in my study, seen that I deal with private trade credit insurance which is short-term in nature. Lastly, I note that insurance seems to have had a significant role in mitigating the negative effects of the financial constraints that followed the financial crisis. Through having access to a new unique dataset on private trade credit insurance covering the period of 2000-2015, I contribute to the relevant academic field in two ways. First, I fill the existent data-gap and test for the robustness of the previously found results. Second, I distinguish for the role of private trade credit insurance in relation to credit market frictions/conditions, both through looking at how the multiplier is behaving through the years of the crisis, and through interacting relevant financial market variables with the insurance variable, which has not been done before for the case of private trade credit insurance. The closest attempt has been that of Van der Veer (2014) who in his paper addresses the role of private trade credit insurance during the credit market frictions during the financial crisis. However, because the dataset he applies does not directly cover the years of the crisis, he is only able to stimulate and not to directly test for these effects.
4 Data and Methodology
In this part, I first present the theoretical model that has guided the whole study, followed by the specification of the benchmark model conducted to answer my research questions. Second, I give a brief description of the sample and data used in the study. Finally, I include a discussion on the econometric concerns related to the applied methods, and what I have done to address these.
4.1 Empirical Model
Following previous research in this field of study, the empirical method applied here is based on the gravity model of trade. This model was first introduced by Jan Tinbergen (1962) who, based on Newton’s gravity equation, suggested that levels of bilateral trade should be positively affected by the size of the trading economies, and negatively affected by the distance between these. In general, the gravity model of trade has represented a set of models that each comprise bilateral trade equations, and that can be expressed as follows (Head & Mayer, 2013):
𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡 = 𝐺𝑆𝑖𝑡𝑀𝑗𝑡𝜃𝑖𝑗𝑡 (I)
Here, 𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗 represent bilateral exports from country i to j at time t, the 𝑆𝑖𝑡 represents all factors common to an exporter that affect supply to all destination countries and 𝑀𝑗𝑡 represents all characteristics of the importing countries that promotes imports from all export sources. The 𝜃𝑖𝑗𝑡 term represents all factors influencing bilateral trade from country i to country j, and as such it captures a wide range of trade costs which are believed to affect bilateral trade, and lastly G represents a gravitational constant. As such, the gravity model of trade simply represents a bilateral trade cost function with aim to explain bilateral trade (Head & Mayer, 2013).
Because the model is originally presented in a multiplicative form (see above), a log transformation yields the following equation, which can simply be estimated through OLS estimation technique (Head & Mayer, 2013; Shepherd, 2013):
log𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖𝑗𝑡 = log 𝐺 + 𝑙𝑜𝑔𝑆𝑖𝑡+ 𝑙𝑜𝑔𝑀𝑗𝑡+ log𝜃𝑖𝑗𝑡 + 𝜀𝑖𝑗𝑡 (II) For long, this equation has been estimated through (naively) using importing and exporting countries’ GDP as proxies for 𝑆𝑖𝑡and 𝑀𝑡𝑗. The 𝜃𝑖𝑗𝑡 term, which is the primary interest of the model, was instead accounted for through including a set of bilateral trade variables that potentially explained bilateral trade flows. These variables include distance, regional
trade agreements, currency unions, being part of the same country, sharing the same border or language, or having a colonial past to mention a few (Head & Mayer, 2013).
However, although the gravity model has often been praised for its ability to predict international trade levels, it has since its introduction been subject to various critiques, mainly due to the lack of solid theoretical foundations. As a response to these critiques, many researchers developed an interest in providing theoretical support of the model, where two of the most influential researchers came to be Anderson and Wincoop (2003). In their famous paper, the authors lift the idea that bilateral trade is not solely dependent on bilateral trading costs and barriers as originally suggested, but also on the level of bilateral trade barriers relative to those barriers the trading partners face with the rest of the world. The inclusion of these multilateral trade barriers, more commonly referred to as multilateral resistance terms (hereafter MRTs), are a crucial element in any theoretically justified gravity model the authors claim10. More specifically, they claim that without accounting for these MRTs “… implies both that estimation suffers from omitted variable
bias and that comparative statistics analysis is unfounded” (Anderson & Wincoop, 2003).
Luckily Feenstra (2002)11 finds that, in studies based on cross-section data12, these MRTs are easily accounted for through the inclusion of importer- and exporter fixed effects. This methodology was later extended by Baldwin and Taglioni (2006), who show that through including time-varying importer- and exporter- fixed effects, together with country-pair fixed effects, the model is consistently estimated also in panel data settings. The main argument here is that without accounting for the time-varying fixed effects, the model would suffer from time-series bias (Baier & Bergstrand, 2007). However, including a wide range of fixed effects as proposed by theory comes at the cost of dropping some of the relevant explanatory variables due to multicollinearity. I explain further down in this paper why this is the case, what this implies for my study and I also show empirically that my benchmark model is consistent with theory.
4.1.1 Specification of the benchmark model
With basis in gravity model theory, I first employ a benchmark model which aims to test for the size of the trade multiplier and thus answer the first question posed in the introduction of the paper. The model is estimated using a fixed effects OLS approach, and is specified as follows:
ln(𝑒𝑥𝑝𝑜𝑟𝑡𝑠)𝑖𝑗𝑡= 𝛽0+ 𝛽1ln(𝑖𝑛𝑠𝑒𝑥𝑝𝑜𝑟𝑡𝑠)𝑖𝑗𝑡+ 𝛽2𝛾𝑖𝑗+ 𝛽3𝛿𝑖𝑡+ 𝛽4𝜃𝑗𝑡+ 𝜀𝑖𝑗𝑡 (III)
Here, the subscript i denotes the exporter, j denotes the importer and t denotes time; while ln represents the logarithm operator. Exports here represent the f.o.b. bilateral exports from country i to country j at time t, and ε represents the error term which catches all other
10 The theoretical justification behind the inclusion of the MRTs is that “the more resistant to trade with all others a
region is, the more it is pushed to trade with a given bilateral partner.” (Anderson & Wincoop, 2003).
11 The working paper of Anderson and Wincoop was published already in 2001. 12 The gravity model was in the beginning mostly applied to cross-section data.
omitted influences on bilateral exports. The coefficient of interest is 𝛽1, and it aims to
capture how insexports (the value of privately insured exports) affect bilateral exports, holding all other export determinants constant.
Most importantly, the benchmark model (III) accounts for three sets of fixed effects: country-pair (𝛾𝑖𝑗), time-varying importer (𝜃𝑗𝑡), and time-varying exporter (𝛿𝑖𝑡) fixed effects.
The inclusion of these fixed effects makes the model consistent with theory, as they simultaneously account for the multilateral resistance terms and for unobserved bilateral heterogeneity. More specifically, the inclusion of time-variant country-pair fixed effects correctly account for the bilateral effects on trade caused by for instance: bilateral distance, common language or border between the trading countries, or even country-pair specific regulation such as currency unions or free trade agreements, given that these effects are constant over the sample period. As such, these fixed effects account for all unobserved heterogeneity that potentially makes trade with a certain partner more, or less, likely. The
country-specific time-varying (importer and exporter) fixed effects on the other hand
account for all time-varying factors specific to a country. As such, they include various unobserved factors that potentially influence a company’s decisions of exporting to, or importing from, a certain destination. These factors for instance include the size of the respective countries’ economies (GDP) or population, or for the country-specific business cycle fluctuations, which are all believed to influence bilateral trade. Including these fixed effects also implies that the inclusion of the originally suggested variables in model (II) is not necessary, as they are now incorporated in the model (see discussion 4.2.1).
4.1.2 Extensions to the benchmark model
In order to answer the second question posed in my introduction, I extend the benchmark model in two ways. First, I include an interaction term and estimate the following model:
ln(𝑒𝑥𝑝𝑜𝑟𝑡𝑠)𝑖𝑗𝑡=
𝛽0+ 𝛽1ln(𝑖𝑛𝑠𝑒𝑥𝑝𝑜𝑟𝑡𝑠)𝑖𝑗𝑡+ 𝛽2𝑖𝑛𝑠 ∗ 𝑐𝑟𝑖𝑠𝑖𝑠𝑖𝑗𝑡+ 𝛽3𝛾𝑖𝑗+ 𝛽4𝛿𝑖𝑡+ 𝛽5𝜃𝑗𝑡+ 𝜀𝑖𝑗𝑡 (IV)
The model (IV) here is identical to the previous one, except for the inclusion of the interaction variable ins*crisis. This variable is constructed by multiplying a dummy variable, equal to one in the years of 2008 and 2009, to the variable of insured exports. Because the crisis can be considered as an exogenous shock to the credit markets (see Chor & Manova, 2012; Felbermayr & Yalcin, 2011), the dummy variable serves here as a proxy for the constraints in the credit markets that followed the crisis. As such, the interaction term is then believed to capture the effects of insurance in the case of negative shocks on the credit markets.
Second, I extend the benchmark model again, now through including variables on importer-specific financial market conditions, which are believed to affect the size of the multiplier. As such, I estimate the following model:
ln(𝑒𝑥𝑝𝑜𝑟𝑡𝑠)𝑖𝑗𝑡=
𝛽0+ 𝛽1ln(𝑖𝑛𝑠𝑒𝑥𝑝𝑜𝑟𝑡𝑠)𝑖𝑗𝑡+ 𝛽2𝐹𝑀𝑀𝑗𝑡+ 𝛽3𝑖𝑛𝑠 ∗ 𝐹𝑀𝑀𝑖𝑗𝑡+ 𝛽4𝛾𝑖𝑗+ 𝛽5𝛿𝑖𝑡+ 𝛽6𝜃𝑗𝑡+ 𝜀𝑖𝑗𝑡 (V)
In this model (V), FMM represents four different variables13: liquid liabilities, domestic credit to private sector, stock market capitalization, and stock market total value traded. These variables all serve as proxies for the financial market maturity in the destination country, as suggested by Beck et al. (2008) and applied to the case of trade credit insurance by Felbermayr and Yalcin (2011). The model also includes an interaction term between insurance and the FMM variables, ins*FMM, which is expected to catch the interplay between insurance and credit market conditions, and the consequences this interplay leaves on trade. Besides helping me in answering the second research question, the inclusion of this model is further important because, as Anderson and Marcoullier (2002) claim, omitting variables on the institutional and financial quality of importing economies potentially biases all the estimates in gravity models.
4.2 Econometric Considerations
Although the models presented here are in line with the underlying gravity model theory, it is important to note that the gravity model is under constant theoretical and empirical development. As Head and Mayer (2013) claim, the validity of the model often depends on the database applied, and the proportion of the variables included. For this reason, the authors further notice, different estimation techniques should always be included to ensure the robustness of the model. Also, it is important to correctly account for potential econometric issues, such as omitted variable bias or endogeneity, as failing to do so could bias the validity of the whole study (for a thorough discussion see Baier & Bergstrand, 2009; and Head & Mayer, 2013). In the following section, I thus present the main econometric concerns and what I have done to address them.
4.2.1 Endogeneity
One of the main econometric issues when applying the gravity model is that of endogeneity, which can stem from measurement errors, simultaneity, or omitted variables. Omitting variables which significantly influence both the independent and the dependent variables is, according to Baldwin and Taglioni (2006), the gold medal mistake of gravity models. The issue is here that if the omitted variables are positively correlated with the independent variables the model is trying to estimate, the variable coefficients are biased upwards and the effects become overestimated. In my respective models, the inclusion of the extensive set of fixed effects accounts for all invariant country-pair, and all time-varying country-specific omitted variables, which mitigates the severity of the omitted variable bias. I show in the empirical section that excluding the fixed effects would
significantly bias the obtained estimates upward, suggesting the presence of omitted variable bias (for further discussion see Baldwin and Taglioni, 2006). In a comparable manner, including fixed effects also deals with issues stemming from unobserved heterogeneity (see Arellano, 2003; Shepherd, 2013).
Endogeneity can also stem from reverse causality, which could be the case if exports would significantly influence the provision of insurance. In order to test for this issue, I adapt Wooldridge’s (2002) method of strict exogeneity, a method which has been readily applied to gravity models by Van der Veer (2015) and Baier and Bergstrand (2007). This method implies regressing the future value of insured trade (t+1), and the lagged value of insured trade (t-1) on current bilateral trade, to test for the correlation between the error terms and the explanatory variables. If the results suggest that future and lagged values do not have a significant effect on current trade, the assumption of strict exogeneity is satisfied (for further discussion, see Wooldrigde (2002)).
4.2.2 Zero (insured) Trade flows
Another critical issue in gravity models is that of zero (insured) trade flows. Because the dataset includes many missing values on bilateral trade flows, as not all countries trade with each other at all points in time, many observations are dropped out of the estimation. This issue becomes more severe in my regressions, because the main independent variable (insured exports), contains many missing values which causes even more observations to drop14. While no optimal solution to this issue has yet been found, an acknowledged method is the Poisson Pseudo Maximum Likelihood estimation method (hereafter PPML)15 by Santos Silva and Tenreyro’s (2006). Because this model allows the dependent variable to be estimated in absolute levels rather than logs, it naturally allows for the inclusion of observations in which bilateral trade is zero. However, the independent variables are still estimated in logs which is problematic in my study due to the large amount of missing values of insured exports. I solve for this issue through estimating the model through a dummy variable approach (a further explanation follows below).
4.2.3 Robustness checks
In order to account for the robustness of my results I have estimated the benchmark model (III) again, but instead changed the sample by dropping certain observations based on geographical region, income levels, or population size of the importing economies. This way, I check whether my results are robust to sample changes.
14 The observations drop here from 218 592 in total (66 exporters * 207 importers * 16 years) to 44,825 observations in the benchmark model (where insured exports > 0).
15 Another common method is to replace observations with zero observed trade (insurance) by 1, allowing these observations to be included in the regression. However, this method has proven to be largely dependent on units of measurement, which is why it should be (and has been) avoided (Head & Mayer, 2013).
Further robustness checks have been done by including other commonly applied control variables to the models, such as distance, population, GDP and more. However, because the majority of the control variables are country-specific (such as GDP or population), including the country-specific time-varying fixed effects would cause some of the control variables of interest to drop due to multicollinearity (Shepherd, 2013). In order to properly estimate for the size of the control variables, I am thus forced to exclude the time-variant country-specific fixed effects, which makes these estimations less consistent with theory. Nevertheless, estimating the model with control variables and without the time-varying country-specific fixed effects serves mainly to confirm, or reject, the results found by the benchmark model. In addition, this method allows me to compare the results with other common findings of the gravity models through looking at the behavior of the control variables.
Finally, following Felbermayr and Yalcin (2011), I also estimate the model through changing the insurance variable into a binary variable which is equal to 0 if no insurance is issued between a bilateral pair in a certain year, and equal to one if the value of insurance is positive. This way, I show whether the results are robust to different measures of the independent variable as proposed by Head and Mayer (2013). Another advantage of using the dummy variable approach is that all observations in which the variable of insured exports equals zero are now included in the study. Using the dummy approach, I also apply Santos Silva and Tenreyro’s (2006) PPML estimation method, which proves to be consistent in presence of heteroscedasticity and which can better account for zero trade levels as explained earlier.
4.3 Sample and Data
4.3.1 Sample
The sample used in the study covers the period of 2000-2015 and is by large dependent on the data availability on f.o.b. exports, as provided by the International Monetary Fund (IMF, 2017a), and data on insured trade, as provided by one of the market-leading private trade credit insurers (details are confidential). Thus, the number of exporting countries included in the sample is limited to 66, and includes those countries in which the private insurer is, or has been, active. The total number of importers is listed at 207, and covers independent countries and dependent territories16 that have appeared as a trading partner to any of the exporting countries in at least one point in time in the full sample. A full list of the included countries is presented in the Appendix 1.
4.3.2 Data
The database applied throughout this study includes variables retrieved from a wide range of sources and a full list is presented in Appendix 2. The main independent variable, value of privately insured bilateral exports (insexports), is retrieved from one of the three market leading private trade credit insurers and summary statistics are provided below.
There are several features regarding this dataset that need to be lifted. First, as already mentioned, the dataset does not cover those countries in which the insurer is not active, thus limiting the number of exporters to 66. Second, the data shows large variations, regarding both the share of insured to total exports, and regarding the amount of total observations per exporter (see Table 1 below). The variety in the share of insured exports will be accounted for in the estimation models through splitting the sample based on the different thresholds of insurance coverage (insured to total exports) The large differences in number of observations per exporter reflects the fact that different exporters enter the dataset in different point in time (once the insurer becomes active in that country), and the fact that different countries use insurance to a different extent (different number of destination countries). A way of accounting for these differences has been through applying sensitivity checks through excluding / including certain regions, which has been done further below.
Finally, the data may suffer from measurement issues, mainly due to the timing of the declaration of turnover, and due to the destination country to which the insurance is assigned. The dataset builds on information retrieved from the insurer’s customers, who declare their turnover at different frequencies, and so the insured exports may be assigned
16 For instance, countries like Andorra, Liechtenstein and Western Sahara which are not included in the IMF DOTS dataset are not included in the sample. Observations including independent areas, such as French Guiana or Réunion, are assigned to their respective dependency nations, in this case France. In cases where data covering independent territories from is available both in the IMF DOTS database, and in the private insurer’s database, the exports are assigned to their own territories (for instance, New Caledonia). (IMF, 2017b).
to the wrong period. For instance, the goods can physically be traded during 2014, but turnover only declared in 2015 and vice-versa. If this is the case, then we can expect the effect of insurance to be underestimated in 2014, and overestimated in 2015 (and vice versa). These types of measurement issues are however significantly reduced by the yearly frequency of data. The second measurement issue is that turnover is in certain cases assigned to the billing country rather than the destination country of exports. This potentially causes an over- and underestimation of the effects because certain insurance coverage is removed from the true receiving country, in which case the effect of insurance is underestimated, and assigned to another country, where now the effect of the insurance is overestimated.
Table 1. Summary Statistics over Insured Exports
Mean Observations SD Min. Max. Sharea
Total 64.34 46403 292.6686 7.34e-08 8582.668 5.97
Australia 27.00 2003 79.40091 6.46e-07 863.8652 9.57
Austria 61.26 1199 208.3096 .0000102 2823.172 7.06
Belgium 95.96 1874 364.1999 3.49e-06 3835.537 2.70
Bosnia and Herzegovina 8.22 97 13.13319 .0084859 77.26167 6.73
Brazil 3.28 51 8.831188 .0000123 60.95301 0.18 Bulgaria 3.16 394 7.015552 .0000307 44.07771 1.91 Canada 12.44 867 65.06267 9.97e-06 855.2424 0.87 Chile 53.10 27 77.02635 .011129 291.6443 5.21 Hong Kong 64.14 927 349.8281 .0000855 5800.148 2.43 China 1.49 132 3.677532 .0000229 29.20506 0.02 Colombia 0.19 10 0.223206 .0248584 .5491304 0.50 Croatia 1.29 42 2.252882 .006248 10.84062 0.43 Curacao 1.70 60 4.275438 9.45e-07 16.49746 6.89 Cyprus 2.37 44 4.766558 .0056576 23.86219 12.56 Czech Republic 30.29 634 128.2193 .0001005 2079.506 1.39 Denmark 79.48 2201 220.122 1.53e-07 1945.046 20.77 Dominican Republic 0.08 8 0.097349 .0117031 .2654727 0.13 Estonia 8.20 615 18.04683 .0001916 156.9312 7.63 Faroe Islands 5.16 294 9.584748 .0000274 60.46896 20.94 Finland 58.43 1439 164.0929 .0011418 3580.925 9.78 France 42.0 2264 140.4387 8.31e-06 1858.799 1.20 Germany 250.73 2176 764.5698 8.31e-06 8582.668 2.50 Gibraltar 1.67 49 5.425168 .001725 37.50575 24.11 Greece 8.64 248 21.89621 .0007931 123.5345 1.86 Hungary 16.07 590 35.57388 1.09e-06 275.4318 1.74 Iceland 7.52 366 19.83962 .0017015 151.2407 8.95 India 4.84 96 13.77513 7.34e-08 91.45538 0.20 Indonesia 3.34 18 6.586341 .001079 27.98115 0.90 Ireland 71.41 1322 307.6043 9.26e-07 4501.952 9.21 Israel 4.91 280 14.53663 .0000138 160.0639 0.95 Italy 39.19 1942 127.3544 1.07e-06 1507.56 1.39 Japan 21.59 545 125.9597 .0000113 2595.51 1.00 Korea, Republic of 0.38 31 0.547503 .0109095 2.059481 0.01 Latvia 9.40 420 22.48295 .0000399 196.6465 6.59 Lithuania 17.58 538 36.23416 .0018823 243.6391 8.91 Luxembourg 29.47 843 123.4879 .0002 2710.092 11.35 Macedonia, FYR 4.94 40 7.870645 .0000582 28.40808 8.50 Malaysia 2.71 200 6.448034 .0007018 57.46945 0.72 Malta 6.62 409 18.22649 .0006999 122.6376 24.40 Marshall Islands 0.08 18 0.1254 .0028938 .5435761 30.11 Mexico 24.15 1101 117.1466 .0000803 1525.548 2.89 Morocco 0.72 16 1.851902 .0019625 7.453032 0.12 Netherlands 149.91 2625 584.0751 1.94e-06 6998.138 7.77
New Zealand 6.57 983 20.12937 5.68e-07 213.2619 4.13
Norway 58.93 1285 141.4089 1.24e-07 1409.153 11.15 Panama 4.65 92 9.746429 .0043689 59.37922 18.80 Peru 30.68 61 68.79173 .0056375 347.6416 3.88 Poland 37.88 758 132.01 .0001275 1238.682 1.63 Portugal 6.61 623 20.29864 .0003317 320.1142 1.94 Romania 8.46 365 30.16737 .0010208 282.9601 1.56 San Marino 1.32 25 1.663999 .0021461 5.817631 7.08 Saudi Arabia 23.33 188 60.47377 .0020204 463.0182 2.56 Singapore 77.47 945 186.7281 .0000699 2433.427 8.70 Slovak Republic 31.63 571 109.3654 .000098 1323.811 2.25 Slovenia 5.94 303 23.21293 .0007422 232.9785 1.59
South Africa 4.34 301 12.77253 1.22e-06 100.4654 1.18
Spain 19.11 1587 60.25717 9.31e-07 697.8442 1.22 Sweden 83.62 1621 236.4814 .0000351 2419.754 7.60 Switzerland 88.40 1361 319.9515 1.02e-06 3817.802 5.92 Thailand 9.21 200 31.7746 .0000123 322.7435 0.48 Tunisia 2.44 6 2.520434 .0071099 6.035244 0.12 Turkey 4.07 463 13.07143 9.28e-06 149.7732 0.59
United Arab Emirates 25.58 672 68.20345 .0000321 690.4801 7.83
United Kingdom 123.89 2754 398.7861 2.26e-06 7342.748 7.39
United States 69.89 2152 180.8737 8.27e-07 2013.039 1.95
Vietnam 0.69 32 1.57648 5.63e-06 7.144106 0.05
Notes: a Insured over total exports over the total sample period. Sample covers the period of 2000-2015.
5 Results and Discussion
5.1 Benchmark Model Results
The results of the benchmark model (III) are presented below. Following Van der Veer (2015), the model is split based on different thresholds of insurance coverage (insured over total exports per observation). Further, due to measurement issues mentioned in the previous section, the observations in which insured to total exports exceed 100% have been dropped out throughout the study, as these data seem unreliable17.
Table 2. Benchmark model (III)
(a) (b) (c) (d) Insurance Coveragea >0% >1% >5% >10% Insured Exports 0.0108*** 0.108*** 0.216*** 0.297*** (0.00199) (0.00618) (0.0118) (0.0178) Multiplier 0,41 2,23 2,37 1,79 Observations 44,825 24,517 11,380 5,688 R-squared 0.983 0.989 0.992 0.994
Country-pair FE Yes Yes Yes Yes
Importer-year FE Yes Yes Yes Yes
Exporter-year FE Yes Yes Yes Yes
Notes: a Measured as insured over total exports. Robust standard errors in parentheses, clustered by
country-pair. All variables are log-transformed. Significance: *** p<0.01, ** p<0.05, * p<0.1.
As can be seen here, the results imply that insurance coverage has significant and positive impact on exports, and the results remain significant across all four sample estimations. In the full sample, column (a), the elasticity of insurance to trade amounts to 0.0108, implying that a 1% increase in insured exports increases bilateral exports by 0.0108%. Albeit this does not seem as a large effect at first, the results are better understood when related to their real economic value. This is easily done through applying the following formula: 𝛽1* meanexports / meaninsured 18, which now relates the estimated elasticity to the real economic values of insured and total exports. The number stemming from this equation is
17 This issue regards 909 observations, which are (inconsistently) present over various country-pairs. 18 The formula relates the elasticities as suggested by the 𝛽
1 coefficient, to the mean monetary values of insured
respective bilateral exports as measured in euros. As such, it relates the monetary value of a 1% increase in the mean of insured exports, to the monetary value of the % increase in the mean of bilateral exports, as suggested by the elasticity.
referred to as the multiplier19, and gives a measure on how much every euro invested in private trade credit insurance produces in euros of exports.
Applying the formula above, I get a multiplier of 0,41 for the full sample20. Contrary to the hypothesis, this estimation suggests that insurance does not have an export-enhancing effect. However, finding a multiplier below one over the full sample is not surprising and is in fact in line with some previous literature findings21. One reason behind finding this small multiplier could lie in the fact that for those observations where insurance coverage is below 1%, which constitute more than 40% of the full sample, the mean of insurance coverage amounts to only 0.29% against an average mean of 5.97%. As such, in these observations the effect of insurance is more likely to be concentrated around certain industries, and the observations might not be sufficient to capture the overall effects of insurance on the bilateral trade relationship.
Another reason behind finding a multiplier smaller than unity can potentially be found in the fact that private trade credit insurance is normally targeted to lower-risk countries. In these cases, the uncertainty reduction resulting from insurance coverage is smaller than in the case when trading with higher-risk countries. As such, the relative creation of new businesses due to the uncertainty reduction is smaller in low-risk countries, why the multiplier can be expected to be lower.
However, I find the small effect of the multiplier below unity not to be robust, and instead I find that the multiplier exceeds one when looking at the samples in which insurance coverage exceeds 1%. For the samples with insurance coverage above the thresholds of 1%, 5%, and 10%, the multiplier is positive and significant, amounting to 2.23, 2.37, and 1.79 respectively. The results of a positive multiplier are consistent with the findings of Van der Veer (2015), albeit my results suggest a somewhat higher multiplier than that of Van der Veer. However, my results do not point at an unusually high multiplier, as they go in line with the findings of Egger and Url (2006), who find a multiplier of 2.8, and Moser et al. (2008), who find a multiplier of 1.7 in their preferred models. However, because the multiplier shows variation across the chosen insurance coverage thresholds, it is difficult to assign a true value of the size of the multiplier. Nevertheless, with these results, I can confirm my hypothesis that insurance does leave a multiplying effect on trade. An interesting observation here is that the multiplier seems to decrease with the insurance coverage. After splitting the sample according to different insurance coverage thresholds (see Appendix 3a-3c), I found that observations where the insurance coverage exceeds 10% tend to be concentrated in what seems less-risky regions. As already referred to above,
19 The multiplier refers to the relative value creation of exports to insured exports. Note that if insured exports only stimulated those exports that they were already covering, the “multiplier” would simply be one.
20 In the full sample, the coefficient is equal to 0,0108; the mean of insured exports is 64 million; and the mean of exports is 2 452 million.
21 For instance, Van der Veer (2015) found a multiplier of 0,3 for the full sample, and Felbermayr and Yalcin (2011) found a multiplier of 0,47 in the full sample. Moser et al (2008) also find a multiplier smaller than unity depending on what country samples are included in the estimation, and show that for non-industrial countries the multiplier is 0,65. To note here is that in their sample, as in mine, the insurance coverage percentage varies widely depending on the sample that is chosen.
