Public Export Credit Insurance
as an Export Determinant:
Panel Data Analysis for The Netherlands
Max ten Have
July, 2018
University of Amsterdam
Amsterdam School of Economics
MSc Economics
International Economics & Globalisation
Student number:
10261605
Supervisor:
Dr. P. (Péter) Foldvari
i
Statement of Originality
This document is written by Maximiliaan Franciscus Johannes ten Have 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.
ii
Preface
You find before you my master thesis on the effect of Public Export Credit Insurance in the
Netherlands on Dutch exports, which was written to conclude the master program in Economics at the University of Amsterdam.
I would like to thank my supervisor, Dr. Péter Foldvari, for his availability and helpful academic feedback in conducting empirical analysis as well as my second reader, Dr. Dirk Veestraeten, for his inspiring guidance throughout this master program.
Furthermore, I would like to thank my former colleagues at the Ministry of Foreign Affairs, Jaap and Saskia in particular, for getting me acquainted to the practices of international trade policy. This internship experience sparked my personal interest for the subject of this thesis among many other things.
In addition, I would like to express my appreciation towards a second group of former colleagues, Jaco and Sander, who challenged me in my ability to analyze critically and formulate my
observations with care.
Finally, and on a more personal note, I would like to thank my friends and family, my parents in particular, for their support. This applies not only during this thesis but throughout the whole course of my education. Thereto, a special word of gratitude is addressed to my girlfriend Marta, who supported and encouraged me in uncountable ways.
I hope you all enjoy reading this thesis.
Max ten Have
iii
Table of contents
Abstract List of abbreviations1. Introduction
2.Theoretical Framework
2.1 Historical background2.2 Public Export Credit Insurance 2.3 The Dutch perspective
2.4 Gravity model of Trade 2.5 Related Research 3.
Empirical Framework
3.1. Empirical strategy 3.2. Data 3.3 Model specification 3.4. Results3.4.1. Fixed Effects Model for Dutch Exports
3.4.2. Delayed Effects 3.4.3. Comparison to literature 4.
Discussion
5.Conclusion
Bibliography Appendixiv
Abstract
This thesis addresses the question to what extent Dutch merchandise exports are stimulated by the public export credit insurance (PECI). In doing so, a Gravity analysis on panel data, covering the years from 2005 to 2016, is conducted by using a basic Fixed Effects model. Whereas estimators do not provide strong statistical evidence, the results do imply that PECI can be identified as an export enhancing determinant of merchandise exports in the year following the grant. This claim is supported by theoretical foundation, characteristics of the Dutch PECI-scheme and a comparison with related literature. The main findings could pave the way for more elaborate research concerning the Dutch PECI-scheme, as this subject provides extensive room for distinguishing different effects within different country groups, industries and time-varying market conditions.
v
List of abbreviations:
ADSB: Atradius Dutch State Business
CIRR: Commercial Interest Reference Rate (agreed on by OECD)
CBS: Centraal Bureau voor de Statistiek (Dutch National Statistics Office) CPB: Centraal Planbureau (Dutch Bureau for Economic Policy Analysis) CPI: Consumer Price Index
ECA: Export Credit Agency ECB: European Central Bank
ECI: Export credit insurance (interchangeably used with PECI) EMU: Economic and Monetary Union of the European Union EU: European Union
FE: Fixed Effects
GDP: Gross Domestic Product
GFC: Global Financial Crisis (2008 and aftermath) GMM: Generalized Method of Moments
IMF- IFS: International Monetary Fund - International Financial Statistics NER: Nominal Exchange Rate
OECD: Organization for Economic Co-operation and Development PECI: Public export credit insurance (interchangeably used with ECI) RE: Random Effects
RER: Real Exchange Rate
RLE: Rule of Law Estimator (Political Risk variable provided by WGI) SMEs: Small and Medium Enterprises
UNSD: United Nations trade Statistics Division WBG: World Bank Group
WDI: World Development Indicators WGI: World Governance Indicators
6
1. Introduction
“Apart from purely economic variables it is likely that political or semi-political factors play a part in determining the volume of trade between countries.”
Jan Tinbergen, 19621
These words were expressed by Tinbergen as he introduced his Gravity model of Trade, which provides a framework to identify key drivers of bilateral trade flows as well as variables limiting it. The quote, expressed over half a century ago, remains relevant and could in fact refer to the concept of contemporary export finance.
Whereas the WTO more recently labeled trade finance as ‘the oil of commerce’ and ‘vital for trade’ (2013, p. 251), a certain form of market failure exists in covering export credit risk. This mechanism is more extensively described in section 2.2, but in short, the political risk component of export credit contributes to the private sector being reluctant to cover the export transactions regarding capital goods specifically. At this point public export credit insurance (PECI) could provide sufficient risk coverage in order to facilitate the export transactions concerned. Hence, (semi-)political factors are indeed assumed to play their part in both limiting as well as facilitating certain export transactions, and thus determining the volume of trade between countries.
For a small and open economy such as the Netherlands, international trade contributes for a large amount to the national welfare. In fact, the Dutch national statistics office (CBS) estimated the contribution of exports to GDP to be around 31 per cent over the last 20 years (2015, p.45). To foster exports, the Dutch government established a PECI-scheme through the facilities of Atradius Dutch State Business (ADSB), which was recently acknowledged to be of significant contribution to the real economy by Van den Berg, Lemmers, Span, Van Beveren and Walker (2017). This raises the hypothesis that Dutch PECI has an enhancing effect on merchandise exports.
The gravity model has been extensively applied to condutct empirical research on analyzing trade patterns and export determinants. For example, political risk has been numerously identiefied as a significant export barrier (Moser, Nestmann and Wedow, 2008). However, research on export stimulating policy, such as public export credit insurance on a macro level, is more scarce and yielding wide-ranging results with respect to significance and effectiveness. Moreover, export promotion policy exists in different forms, varying per country, magnitude of the program and Export Credit Agency (ECA) assigned with the execution of the policy.
1 From: Tinbergen, J. (1962), Shaping the World Economy: Suggestions for an International Economic Policy (New
7 Whereas analyses on export promotion for Germany and Austria are available, research on Dutch PECI is still in its infancy as the only reputable publication in this respect concerns the aforementioned Input-Output analysis by Van den Berg et al. (2017). This thesis employs the gravity model on the Dutch perspective, thereby complementing existing literature with new results using a dataset that is facilitated with non-publicly available data2, containing the most recent years and
covering a relatively long timespan.
Whereas the study by CBS implied a significant export enhancing effect of PECI, the main aim of this thesis is to confirm and quantify PECI’s contribution to exports. Hence, the main research question states: To what extent does public export credit insurance in the Netherlands stimulate Dutch
merchandise exports?
In answering this question, the next chapter will set out an theoretical framework to place this research in context. It presents an overview of international trade developments and elaborates on theoretical foundation of PECI-facilities, followed by a section describing the Dutch economy and Dutch PECI-scheme more specifically. Subsequently, Tinbergen’s Gravity model of trade is elucidated more in detail, to conclude with an overview of literature related to this thesis’ topic applying such a model. Chapter three contains the empirical analysis, in which the presentation of results is precided with sections describing the empirical strategy, data collection and choice of variables and the resulting model specification. The presentation of results itself is divided into analyzing the model estimation, interpreting the results of direct and lagged effects and a comparison of implications to the literature related. To finalize the empirical framework, the observations are discussed in chapter four, accompanied by a validity assessment of results with respect to data used, model estimation technique and theoretical foundation. Finally, the fifth chapter briefly summarizes the research and concludes by answering the main research question along with suggestions for further research in this topic.
8
2.
Theoretical framework
In order to fully grasp the potential contribution of PECI for economies, one ought to take into account the framework of international trade as well as the academic tools to conduct analyze in this field of study. This chapter aims to do so by briefly outlining the historical background in international trade and its relation with international finance in the first section. The second section explains how PECI enables export promotion and places it in an international context. The third section provides a brief overview of the Dutch economy and introduces the Dutch PECI-scheme. Thereafter, the fourth section presents the Gravity model of Trade, identified as being the main workhorse of empirical analysis in this field of study. Finally, section five presents existing literature containing a Gravity-based analyses the effect of PECI on exports for other countries.
Prior to outlining the context of this research, certain clarification on terminology is required as existing literature is not fully consistent in abbreviations nor definitions.
First of all, Export credit insurance (ECI) is to large extent provided by the private sector. However, this thesis as well as the literature included focus on the export credit insurance provided by the public sector (PECI) as a policy instrument. Following the example of existing literature on this topic, the abbreviations ‘ECI’ and ‘PECI’ are interchangeably used in referring to public export credit insurance, unless perspicuously indicated otherwise.
Second, export credit insurance agencies is also labelled public export credit guarantee (Egger and Url, 2006, p.400). Whereas the Dutch state in 2009 actually introduced a so-called ‘export credit guarantee scheme’ aside from the longstanding ‘export credit insurance scheme’3, only the latter is
being assessed in this research. In order to prevent confusion with respect to related literature, the terms ‘insurance’ and ‘guarantee’ will refer to one and the same instrument: public export credit insurance aiming to cover a certain percentage of export credit.
Third, by referring to export credit agencies (ECAs), institutions undertaking the official provision of export credit insurance on behalf of governments or with state guarantee are intended. Finally, exports can be distinguished into export of goods, also merchandise exports, and export of services. This thesis focusses explicitly on the effect of PECI on merchandise exports.
3 For details, see:
ondernemen/ondernemen-in-het-buitenland/exportkredietverzekering and https://www.rijksoverheid.nl/onderwerpen/internationaal-ondernemen/ondernemen-in-het-buitenland/exportkredietverzekering/exportkredietgarantie
9
2.1 Historical background
Historical trends in international trade are extensively described by the World Trade Organization (WTO, 2013, pp.44-108), which bundles a large amount of economic research. It identifies the industrial revolution, entailing technological innovation in transportation and communication, as the starting point of the modern world trading system. Following the decline in transport and communication costs a ‘first wave’ of globalization is observed, resulting in the unbundling of factories from consumers. However, the First World War, Great depression and Second World War proved to destabilize the foundation of international economic system. In response, political and economic cooperation intensified in an attempt to decrease trade barriers and endure peace (p.51-52). This resulted in new multilateral economic institutions such as the World Bank Group (WBG), International Monetary Fund (IMF) and the (predecessor of) the Organization for Economic Co-operation and
Development, OEEC/OECD). From this point onwards a second wave of globalization unfolded. This
time declining trade barriers (such as tariffs, non-tariff political barriers, transportation cost) contributed to the geographical division of manufacturing stages (WTO, 2013, p.55). This process, also referred to as the formation of ‘global supply chains’, is more extensively presented by Baldwin (2013). Fueled by the increase in trade of intermediate goods, international trade increased dramatically over the last decades (WTO, 2013, pp. 55).
Against the backdrop of historical trends and drivers of international trade, the WTO also discusses its challenges, among which the necessity of sufficient volume of trade finance at an affordable price (2013, pp. 251-256). The link between international trade and international finance becomes more apparent during financial crises. For example, Amiti and Weinstein confirm this link as they observed exports declining much faster than domestic output during financial crises. In addition, they find producers are more reliant on credit for their exports than for domestic sales and that SME exporters and industries with long transportation time suffer more from credit constraint (2011, p. 1874).
2.2 Public Export Credit Insurance
The policy document (Tweede Kamer der Staten Generaal, 2016, pp. 10-14) explains this mechanism behind export finance. It points out that export transactions regarding capital goods or contractor activities typically rely on trade finance as the cost of production and delivery of export goods are not immediately counterbalanced by the client’s payment. In this case, a bank can serve as third party intermediary by issuing credit to the foreign client and ensuring payment to the exporting firm upon delivery of agreed goods. However, the risk of default on such credit is exacerbated by incomplete information regarding the creditworthiness of the foreign buyers, political or financial instability, or
10 hazards involving the enforcement of international contracts. Due to these causes, the buyer might not, or not fully, repay its debt to the intermediary bank. As a result, banks can be reluctant to finance the transaction.
To mitigate such credit risk, commercial banks and insurance companies step in as a fourth party in the transaction by offering various policies to indemnify the risk-taking party against incomplete payment. Thus, in case the buyer does not fulfil the payment, the insurer will reimburse the exporter and/or the intermediary bank and attempt to retrieve the damage at the foreign client. However, even in the absence of financial crises, the private export credit insurance is found reluctant to cover large transactions (limiting adequate risk diversification), transactions with a long-term maturity and transactions involving substantial amount of political risk (Baltensperger & Herger, 2008, pp. 545-546).
To provide further indication on political credit risk, Jensen (2008, p. 1043) mentions three broad categories. The first corresponds to risks associated with (civil) war, uprisings or terrorist attacks, which could turn operations unprofitable. The second group relates to government intervention such as direct nationalization and expropriation of assets along with breach of contracts between the firm and government. The third category covers the risk of capital flow restrictions harming business.
In acknowledging this form of market failure, an increasing number of emerging, countries have installed official export credit agencies (ECAs) operating on behalf of the government or backed by state guarantee (Moser et al., 2008, p. 782). Whereas public insurers might not be able to bear the commercial risk concerned with large and long-term transactions, governments usually can compensate for incidental losses. Moreover, official government support also increases feasibility of recovering damage in case of political risk, especially if the foreign buyer is another government. Official ECAs provide government-backed insurance to the country’s exporting firms with the aim to promote national exports. The rationale for such policy could be found in the concept of export-led growth, which states that export is a substantial determinant of economic growth. In support of this theory, Feder (1983) points out that exports do not only contribute directly to GDP in terms of aggregate output, but also facilitate the exploitation of economies of scale. In addition, Feder remarks that international trade stimulates technological innovation and the optimal allocation of resources in the economy. Hence, it would be to the benefit of countries to ensure that export opportunities are not limited by the absence of trade credit.
According to Egger and Url (2006, p.415), the export enhancing effect of PECI manifests both in the short and in the long term. In the short term, direct sales could be realized which would not
11 have been facilitated by the private sector. In addition, a short-lagged effect might arise due to the practical issue that the time in which the insurance is granted does not correspond one to one with the shipment of the good and thus the actual recording of the export in foreign statistics. Furthermore, Egger and Url argue that PECI can be particularly valuable in facilitating exporters first access to a foreign market. Consequently, exporting firms can repeat sales to credible customers and abandon less credible buyers according to progressive insights.
However, the state-backed component originally threatened to conflict with WTO’s guidelines on export subsidies. As described by the OECD (1998, p. 27), there was increasing competition in export financing in the beginning of the 1970s. In an attempt to increase exports amidst world-wide economic recession and oil crises, countries made substantial efforts to offer better export credit terms than their competitors and the role of publicly supported ECAs grew more important. Following the debate on illegal state aid, an agreement4 with respect to public export credit insurance was made
by the OECD-members in 1999. It ensures an international level-playing field by establishing minimum risk-based premia and required transparent reporting on collected premiums, disbursed claims and net operating cashflows. From these conditions follows the rationale that PECI covers ‘unmarketable’ export credit risks, which exempts such PECI-schemes from condemnation by the WTO (Moser et al., 2008, p. 967).
2.3 The Dutch perspective
As we can see in Figure 1, Dutch merchandise exports amounted to about 460 billion euros in 2017. It displays an upward trend, only interrupted by a decrease in 2008 and a stagnation in 2014. In observing this trend, the CBS states that the Netherlands thereby occupy a high position in the ranking of both absolute merchandise exports as well as relative to a country’s GDP (2018, p.11-20). However, it also identifies a substantial increase in the share of re-exporting and transit-trade within Dutch exports as a consequence of the emergence of global supply chains described in section 2.1. Despite the relatively low contribution to GDP per euro, re-exports are found to provide a considerable contribution to the economy as their volume is substantial (p. 49). Furthermore an increase in the share of service exports within total exports is observed, which can also attributed to the emergence of global value chains as it caused Dutch firms increasingly specialised in services prior to production (CBS, 2015, pp.44-51).
Overall the CBS stresses the contribution of exports to the Dutch economy in terms of
4 Refers to the OECD’s ‘’Arrangement on Officially Supported Exports Credits’’ , also known as the ‘Knaepen
12 time jobs provided and GDP, which is estimated to have been more or less constant around 31 per cent over the last 20 years (2015, p.45). Considering the historical contribution of trade for Dutch economy, it is no surprise that the Dutch government facilitates a PECI-scheme to foster exports.
In the Netherlands, Atradius Dutch State Business (ADSB) is the official ECA. It operates on behalf of and for the account and risk of the Dutch State but is in fact a full subsidiary of Atradius Group, a private insurer. ADSB is exclusively assigned with the task of executing the Dutch PECI-scheme, for which it receives a yearly fee (Atradius DSB, 2018 & Tweede Kamer der Staten Generaal, 2016).
Figure 2 displays the total amount of newly granted PECI per year, which, in combination with the total Dutch merchandise export expressed in Figure 1, is used to derive the coverage ratio of ECI over Exports for the period between 2005 and 2016 (Figure 4). It is interesting to see the decline in coverage ratio following the global financial crisis of 2008, as the PECI instrument is supposed to lower the trade barrier of insufficient trade credit by the private sector. In response to this discrepancy, the Dutch ECA was directed to engage more actively into facilitating export credit demand, which resulted in the issuance of products specifically aimed at exporting small and medium enterprises. However, Atradius DSB remains to report a predominantly positive net operating cashflow, implying that operations are refrained from hidden export subsidies (Tweede Kamer der Staten Generaal, 2016, p.81).
Furthermore, ADSD (2018) states to offer ECI insurance mainly, but not exclusively, for the delivery of capital goods with credit or completion periods longer than twelve months and applies a certain commitment ceiling per country in order to keep its portfolio diversified. Figure 3 presents the distribution of new ECI over geographical regions from 2005 to 2016, which indeed represents a diversified portfolio.
As referred to in the introduction, the Van den Berg et al. (2017) recently assessed the Dutch public export credit insurance facility (PECI) and its contribution to Dutch GDP and employment. Their analysis consists of a relatively simple input-output (IO) approach, using data from 2010 to 2014. They find that the economic activity (total value of 11.2 billion euro) insured by the public ECI contributed to GDP by about 7.8 billion euro, 0.24% on yearly average. On the employment side, they estimate the contribution of the ECIF to be 95,000 FTE jobs over these five years.
The IO-framework used is however subject to simplifying assumptions with respect to firm heterogeneity, for which these results could be interpreted an upper boundary of the true contribution. Also, the analysis relies on the additionality of the PECI-instrument to be 100 percent and thus assumes each transaction could not have been realized without it. On the other hand, PECI
13 might have facilitated economies of scale, for which the full contribution to output growth might even be underestimated.
The methodology followed does allow for interesting sector-specific findings. For example, 79 percent of the publicly insured export during 2010-2014 was related to civil engineering (mainly offshore activities and dredging) and manufacture of transport equipment (mostly ship construction) (2017, pp. 16-18).
2.4 Gravity model of Trade
Analyses of bilateral export relations and the effectiveness of export promotion on the macro-level have previously been conducted using the standard gravity model of trade. The Gravity model and its development over time is extensively described by Shepherd (2013) as well as by the WTO (Bacchetta, et al., 2012). They recognize the model as a key instrument in applied international trade literature and credit Jan Tinbergen for being the first to introduce this concept in describing trade patterns in 1962.
The model is named after the concept of gravity in the field of physics, since its key variables include the ‘economic mass’, usually measured in GDP, and a proxy for transportation costs, usually measured in absolute distance in between the two countries. In reference to Newton’s law of gravity, the basic Gravity model of trade states that comparably to planets being mutually attracted in proportion to their size and proximity, larger countries are expected to trade more. Vice versa, it implies that countries further apart from each other are likely to experience less bilateral trade, since transportation costs increased. Extensive empirical research confirms these claims (Shepherd, 2013, p 10).
This basic equation can then be elaborated upon by adding various explanatory and control variables. Also, the form and estimation method of gravity models has been further developed to facilitate the use of panel data and to ensure consistent and robust results in the case of zero-trade observations. These calibrations were illustrated by Westerlund and Wilhelmsson (2011).
2.5 Related research
Empirical research on the effectiveness of PECI-schemes on a macro-level is rather scarce. Yet, as far as literature on this topic is explored, the analyses are conducted within the framework of the gravity model of trade.
For instance, Baltensperger and Herger (2009) analyze the link between exports and public export insurance schemes in OECD countries from 1999 to 2006. Using the OECD’s country risk index
14 and the OECD’s insurance data, they find public ECIF programs to be modestly trade extending with respect to high- and middle-income countries. Yet, they find no significant effect on exports towards low-income countries.
However, other research indicates stronger effectiveness of the instrument. In narrowing down to the Austrian perspective, Egger and Url (2006) analyze the effects on Austrian exports from 1996 to 2002 using a standard RE-model as well as a Mundlak-RE, which allows for estimating long-term dynamics. They address the distinction between trade creation and trade diversion and distinguish the short and long-term effects of newly granted public insurances. With Austrian PECI covering about 5 to 10 % of Austrian exports during the simple period, they find the short-term impact of Austrian public ECIF to be negligible compared to a substantial long-term trade multiplier. Furthermore, they find exports towards higher-risk markets stimulated significantly (pp.413-415). Whereas this research could provide a relevant comparison with the Dutch case since both concern a small open economy with a well-developed ECIF, the dynamics of PECI in this timespan differ fundamentally as a substantial part of the sample dates from before the harmonization of PECI-schemes (section 2.2).
Moser et al. (2008) apply an identical model on the German case, using data ranging from 1991 to 2003. They use the ICRG-index as their political risk variable and do find strong evidence that political risk constitutes an important friction to international trade activity. Using both a static and a dynamic panel data analysis, they find the German public ECIF to be fostering exports. However, the time span analyzed again encompasses a fundamentally different institutional framework.
Finally, Felbermayr and Yalcin (2013) analyze the German perspective with more recent data, covering yearly observations from 2000 to 2009. Using sectoral data provided by the German ECA, they control for different forms of unobserved heterogeneity. As a result of this timespan, some findings are specifically with respect to financial liquidity issues arising from the global financial crisis in 2008. In general they find political risk, taken from the OECD country risk index, to be only weakly related to the effects of the public ECIF. Furthermore, they observe no significant export enhancing effect towards low-income countries and note show that the effectiveness of public insurance varies strongly between sectors depending on financial vulnerability and credit constraints.
Taking into account the timespan covered, similarity between the PECI-instrument analyzed and the modelling technique applied, this last research by Felbermayr and Yalcin (2013) is considered the best suited benchmark in discussing the results obtained in this thesis. However, all papers analyzing from the perspective of one exporting country make use of sectoral data. As elaborated upon in the following chapter, this entails a fundamental side note for placing obtained results in the context of existing literature.
15
3.
Empirical framework
Given the theoretical insights obtained in the previous chapter, this chapter starts by elucidating the strategy for conducting empirical research, thereby taking feasibility for into account.
From this follows a description of the data collected, in which while theoretical justification and estimator expectation as well as drawbacks for the selected variables are being discussed. Combining the first two sections, section three presents the model specification. Finally, the fourth section displays the obtained results and provides interpretation of the model and, first based on its direct coefficients, then regarding the lagged ECI-variables and third in comparison with literature. Finally, section five places the results in context of existing literature and discusses the validity of the observations.
3.1 Empirical Strategy
Analyses of bilateral export relations and the effectiveness of export promotion on the macro-level have previously been conducted using the gravity model of trade. The same model provides the foundation to answer the main research question of this thesis, for which a panel dataset can be constructed. It includes all bilateral export partners for as far as data availability allows.
As described in the literature review, a gravity analysis aiming to dissect the Dutch export flows altogether has recently been conducted by the CBS (2018) and is therefore not the main goal of this thesis. This research is narrowed down to determine whether the Public Export Credit Insurance in the Netherlands can be identified as a significant determinant of Dutch exports. Hence, the model has to focus on accurately estimating the coefficient of a time-invariant variable while using panel data with respect to one exporting party only.
Taking all of the above into account, a Fixed Effects (FE) regression model will be used. As extensively outlined by Stock and Watson (2012, pp. 396-400), the FE-Model controls for omitted variable bias in panel data when the omitted variables vary across countries but remain constant over time. Using this characteristic, time invariant determinants of Dutch exports, such as Distance, shared
border and cultural similarities, drop out of the equation. Still, accurately estimating the determinants
of export flows has to be taken with caution, since omitted time-variant variables would lead to estimation bias.
Whereas (Anderson & van Wincoop, 2003) developed the concept of Multilateral Resistance Terms (MRT) to capture unobserved bilateral trade costs, implementation of this extension was hindered due to aggregate data. Moreover, Felbermayr and Yalcin, who did manage to account for MRT as they increased observations through adding an industry-market dimension, proved the
16 overestimation of the estimates in absence of MRT to be ‘very minor’ (2013, p. 979. Hence, the basic FE-model is preferred as it is theoretically suitable to provide insights while keeping the econometric scope feasible. Based on existing literature, the most relevant time-variant variables will be included. In doing so, the usual assumptions regarding linearity in parameters, absence of perfect multicollinearity and strict exogeneity of the data are applied (Stock and Watson, 2012, p.405).
3.2 Data
In order to conduct this research, a unique panel dataset was constructed using data from multiple sources. It contains 2448 observations over a 12-year time-span (from 2005 to 2016) and covers all Dutch trading partners that data availability reasonably allowed for. In this respect, the main rationale for dropping certain countries from the data panel concern: no observations for the independent variable whereas zero-trade flows are denied by other sources5 (e.g. Liechtenstein, Monaco, Kosovo),
severe lack of independent variable availability (e.g. Turks and Caicos Islands) and territorial discrepancies causing data to be dispersed over multiple sovereign statuses in between data sources (e.g. Sudan, Dutch Antilles, Puerto Rico). This brings the number of countries down to 200 unique partner countries. A detailed description of the dataset is attached in the appendix (Table 1).
Concerning the dependent variable, the yearly bilateral exports between the Netherlands and its partner countries was obtained from the UNSD Comtrade database via the WITS-software of the World Bank Group (WBG). This dataset is publicly available online and contains a large number of partner countries compared to the online database from the Dutch National Statistics Office (CBS). The data is published under the name ‘Trade Value’ and contains total merchandise export (thus excluding services) expressed in thousands of US Dollars at constant prices. It was manually transformed to single US Dollar for corresponding to the independent variables in scale.
The main variable of interest, newly granted public export credit insurance on behalf of the
Dutch state, was kindly provided by ADSB upon explicit request. It contains all new commitments
granted since 2005 per country per year, expressed in US Dollars, and represent the maximum obligations the Dutch State has committed insurance coverage to. The total amount is assigned to the country in which the transaction-partner inducing the credit risk is actually located and the year reported is the year in which ADSB entered into the ECI obligation. As described by Felbermayr and Yalcin (2013, pp. 974 & 980) and Van den Berg et al. (2017, p. 16) the PECI instrument is dominantly applied in sectors with long time-to-build lags, high upfront financing costs and long shipment time to
5 CBS – the Dutch National Statistics Office – does show exports from The Netherlands to these countries,
17 destination (e.g. dredging machinery and ship construction). In addition, such foreign sales might hang over several years. Therefore, the year in which the ECI is reported does not necessarily correspond to the exact time of its potential export creation. Considering the implications of this practical note in combination with the theoretical assumptions that PECI helps exporters gaining first access into initially unserved markets, the effect of ECI might very well be lagged to future exports. As the PECI-instrument practically implies a reduction in trade barriers, the public ECI PECI-instrument is expected to positively affect exports. In fact, this expectation coincides with the main hypothesis of this thesis. To control for price level variations, the variables Exportsi,t and ECIi,t were deflated using the
CPI-deflator. This index overall inflation rates for the common consumer goods, which varies from export goods by nature. Thus, we would ideally deflate variables using custom-made deflators, such as an export price index. However, these were not available. Hence, to example of Moser et al. (2008), CPI is used as it was openly available through the World Development Indicators (WDI). This deflator corresponds to the yearly CPI-averages and is indexed with respect to base year 2010. In order to deflate, the variables are divided by the decimal equivalent of the CPI-level for each specific year. For as much as the classical Gravity variables are concerned, the GDP and GDP per capita data were also retrieved from the WDI. These variables are included for proxying ‘economic mass’. As stated by Shepherd (2013, p. 9) we expect larger country pairs to trade more, whereas the per capita variant is used to express for a country’s economic mass relative to its population. For each of these variables (GDPNL,t , GDP_pcNL,t , GDPi,t and GDP_pci.t), a positive coefficient is expected. The data is
denominated in US dollars and reported in constant prices, thus corresponding to real (per capita) GDP.
Regarding the exchange rate indicator, Auboin and Ruta (2013) outlined a strong relationship between exchange rates, relative price levels and bilateral trade flows. By using real exchange rates (RER), this relationship can be expressed not only between currencies but also within currency unions such as the European Monetary Union (EMU). As the RER variable was not readily available for the full sample, it was manually composed. The nominal exchange rates (NER) reflects the specific country’s local currency units (LCU) per Euro at a specific year and was obtained at the IMF’s financial statistics. The variable RERi,t is expressed as
( 𝑁𝐸𝑅𝑖,𝑡 𝑥 𝐶𝑃𝐼𝑖,𝑡 )
𝐶𝑃𝐼𝑁𝐿,𝑡 and thus reflects the LCU’s purchasing power over
the Dutch consumption basket in a specific year. In other words, a decrease in RER implies that the LCU price of Dutch exports decreased, which could boost exports. Thus, we expect a negative coefficient for RERi,t. The CPI-levels used are the same as the earlier applied deflators, which leads to
a RER indexed towards base year 2010. In order to avoid sudden jumps in the variable due to intermediate adoption of the Euro, the exchange rates for countries concerned are expressed in the (former) local currency throughout the full sample timespan. The exchange rate applied ex-post
18 accession date corresponds to the official irrevocably fixed exchange reported by the ECB.
As shown in all related literature, a risk measurement variable related to political (in)stability, institutional strength or country specific credit risk qualifies as a significant determinant for bilateral exports. Whereas Moser et al. (2008) use the International Country Risk Guide (ICRG), Jensen (2008), Baltensperger and Herger (2009) and Felbermayr and Yalcin (2013), base their risk variable on the OECD’s Country Risk indicators. However, using the OECD’s risk parameters seems questionable as it excepts high-income OECD and high-income Euro Area countries from any appropriate risk indicators (OECD, 2018, p. 18 & p. 21), whilst the ICRG is subject to a paid license. In this analysis, the variable of choice is the Rule of Law estimator, composed by the World Bank Group as an element of the World Governance Indicators. This index is in fact partly based on the ICRG, among 31 different sources, (Kaufman, Kraay & Mastruzzi, 2011, p. 225). It assigns a value in between -2.5 and 2.5 to each country ‘‘capturing perceptions of the extent to which agents have confidence in and abide by the rules of
society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence’’ (2011, p. 223). As the quality of contract
enforcement is particularly mentioned, this index should pre-eminently be able to provide a suitable risk variable concerning exports. Using this risk indicator, a positive coefficient is expected with respect to exports.
Finally, binary dummy variables for signaling membership of the European Union (EUi,t) as well
as the European Monetary Union (Euroi,t) are added. The value of these dummies takes the value ‘1’
in case of membership, and ‘0’ in the absence of it. The dummies were manually constructed using accession information retrieved online6. Since a respectable number of countries acceded the EU or
Eurozone during the sample time-span, the effects are not fully absorbed by the Fixed Effects model. In both cases, membership is expected to be positively affecting Dutch exports.
Whereas Egger & Url (2006), Moser et al. (2008) and Felbermayr and Yalcin (2013) introduced
Manufacturer Imports and Capital Formation to the analysis, these variables are not included here.
The main rationale for including those variables stems from their interpretation of characteristics considering the German economy and the fact that they conduct industry specific regressions. However, these variables are not identified as key determinants for Dutch aggregate export flows by CBS (2018, Figure 2.5.2, pp. 62-63). This implies that the risk of a severe omitted variable bias due to not including these variables in this research is surmountable and results remain comparable with the existing literature.
5 Referring to official online communication platforms:
https://europa.eu/european-union/about-eu/countries_en#tab-0-1, and, https://www.ecb.europa.eu/euro/changeover/html/index.en.html. Accesed on: June 24th, 2018.
19 Dissimilar to Moser et al. (2008), Felbermayr and Yalcin express exports, guarantees, GDP and exchange rate variable in nominal terms. This preference is probably related to incorporating MRT later on in their analysis, since MRT effectively serves as special price indices (Shepherd, 2013, p.12).
3.3 Model Specification
Following the empirical strategy in accordance with the collected data, the preferred estimation equation is given by:
ln (𝐸𝑥𝑝𝑜𝑟𝑡𝑠𝑖,𝑡) = 𝛽0 + 𝛽1𝑙𝑛(𝐸𝐶𝐼𝑖,𝑡) + 𝛽2𝑙𝑛(𝐺𝐷𝑃𝑁𝐿,𝑡) + 𝛽3𝑙𝑛 (𝐺𝐷𝑃𝑝𝑐𝑁𝐿,𝑡) + 𝛽4𝑙𝑛(𝐺𝐷𝑃𝑖,𝑡) + 𝛽5𝑙𝑛 (𝐺𝐷𝑃𝑝𝑐𝑖,𝑡) + 𝛽6𝑙𝑛(𝑅𝐸𝑅𝑖,𝑡) + 𝛽7𝑅𝐿𝐸𝑖,𝑡+ 𝛽8 𝐸𝑢𝑟𝑜𝑖,𝑡 + 𝛽9𝐸𝑈𝑖,𝑡+ 𝜂𝑖+ ɛ𝑖,𝑡 in which the ‘ln’ signals the usage of a natural logarithm over the variable in brackets. The subscript ‘i’ refers to the specific entity (partner country) and the subscript ‘t’ indicates the specific year in the panel data analysis. The element ‘𝜂𝑖’ reflects the unobserved country specific effects that are naturally accounted for in the Fixed Effects model. The error term ‘ɛ𝑖,𝑡’ is assumed to have a zero-mean and a constant variance. Using this model, we can test whether public export insurance in the Netherlands had a significant effect on exports from 2005 to 2016.
For the sake of interpretation, the ECI-variable can be included in the model in different forms as Felbermayr and Yalcin exhibited in analyzing the German case (2013, pp. 975-978). In order to provide ground for comparison, three variants of the ECI-parameter were tested in separate regression models: a binary dummy variable, a logarithm variable and a Coverage-ratio variable.
The binary dummy (dECIi,t) variable takes the value ‘1’ if the ECI instrument was active in both the
specified year and the specified market (country) and ‘0’ otherwise. Therefore, the variable indicates whether the downright presence of Dutch PECI for a market affected the exports.
The logarithmic variable (lnECIi,t) is computed by taking the natural logarithm of the base
variable (‘ECIi,t’). Its estimated coefficient can be interpreted as the elasticity of exports with respect
to the quantity of public export credit insurance. Since the logarithm of value ‘0’ is undefined, Stata generates missing values following the transformation. This significantly reduces the number of observations in the sample (from 2091 in the non-lag regression to 325), thereby undermining the explanatory power of the model and raising concerns about sample selection bias (Shepherd, 2008, p 55). Following the example of Felbermayr and Yalcin (2013, p. 978), we correct for this by setting the missing values of lECIi,t equal to zero. This ad hoc transformation is theoretically justified by the fact
that the missing values were not caused by missing observations in the first place, but that Atradius DSB actually reported zero ECI-flows to the combination of these specific markets and specific years.
20 Coverage ratio (Covi,t) relates to base variable ECIi,t in the sense that it corresponds to the
share of exports covered by ECI. Its estimate measures the instrument’s effectiveness with regard to additional share of exports covered. For interpreting the point estimate, the unity value plays a pivoting role. As a coefficient equal to unity indicates that the per cent increase in exports due to the ECI-instrument equals the share of exports covered, an estimate lower than unity indicates that part of the increase in exports would have taken place without ECI coverage. Therefore, this variable could provide an indication concerning the additionality of the ECI-instrument. (2013, pp. 975-978).
Following the characteristics of the PECI instrument, these variants will also be analyzed for their delayed effects.
For all regressions conducted, the Hausman test suggests using Fixed Effects over the Random Effects model and robust standard errors are used to control for heteroskedasticity.
As pointed out by Shepherd (2013, p.54 & 55), the presence of zero-export flows potentially leads to inconsistent estimates. However, in this sample the number of zero-export flows is limited to eight observations, which reduces the need for cumbersome adjustments to the model. The limited number of zero-export flows is likely due to using aggregate exports data as well as the double check on validity of zero-export flows as specified in the data section.
3.4. Results
In this section, the results for export-stimulating effect of Dutch Public Export Credit Insurance are presented. The model will first be evaluated on a broad level, after which focus will shift to interpreting the direct ECI-coefficients and lagged effects. Next, the results will be compared to the model employed by Felbermayr and Yalcin in order to provide a benchmark for both the size of points estimates as well as model estimation technique (2013).
In analyzing statistical significance of coefficients, the p-values assigned provide a foothold for interpreting the results. Although selection of certain significance levels is fairly arbitrary, Stock and Watson (2012, p. 119-120) acknowledge that the level of choice should reflect the context of the research. Whereas scientific guidelines have conventionally set the suggested significance level at 5-per cent, related literature on this topic applied a degree of claims concerning significance levels of 10-percent (‘weak’), 5-percent (‘sufficient’) and 1-per cent (‘strong evidence’) (Moser et al., 2008; Felbermayr & Yalcin, 2013). To provide a fair comparison and prevent stretching subject specific jargon, those significance levels are adopted.
All interpretations regarding point estimates are subject to the ceteris paribus condition, the assumption that all other relevant elements remain unchanged.
21
3.4.1. A Fixed Effects Model for Dutch Exports
Table 2 presents the results for all three variants of ECI-variables as well as the control variables. The first observation to be made is that the relatively low within-group R-squared (0.15) implies that a large extent of the variance in the dependent variable could not be explained by variance in the independent variables. This is not surprising, regarding the gravity analysis by CBS (2018) indicating that the lion share of Dutch export can be assigned to geographically oriented fixed effects, such as Distance, which are filtered out by the FE-model.
Secondly, neither of the direct ECI-variable coefficients is statistically significant. Furthermore, over all three variant models we observe GDP of the partner country to be the most significant export determinant. This is in line with existing literature and research from CBS (2018, p. 62) as a bigger market tends create increased demand for Dutch exports. This variable is followed by the Euro Currency Area dummy and Rule of Law estimator, which indicates that a common currency as well as political stability play a role in attracting Dutch exports.
Despite most variables lacking significance, the control variable coefficients mainly display the expected plus- and minus signs, thereby refraining from challenging the theoretical foundation of the model. Although refrained from statistical evidence, we can at least continue to assume that a depreciation of the Dutch real exchange rate (increase in RER) tends to decrease exports towards the partner country concerned and that a common customs union tends to increase it.
For GDP per capita in both The Netherlands and the partner country, mainly negative coefficients and relatively large standard deviations are observed. Taking a glance at the summary statistics, a possible explanation for this inconsistency with literature could relate to the fact that these variables seem to be relatively fixed over time. As mentioned earlier, Fixed Effects models are ill-equipped for estimating the coefficients of such variables.
3.4.2. Delayed effects
Leaving observations concerning the general model aside, we return to the core of this thesis by continuing analysis on the ECI-variables. As direct effects of ECI fell short of providing significant results, the instrument is tested for potential lagged effects. As elaborated upon in section 3.2 and the theoretical framework, the export enhancing effect of PECI might manifest itself with a minor delay, following from practical implications, and in long term effects following from theory. By testing for lagged effects, it seems that the first lag effect of all ECI-variants dominates and is backed by an indication of statistically significant evidence, albeit ‘sufficient to weak’ (see Table 3). While the point
22 estimates of the control variables remain roughly unchanged, neither of the second and third lags reach beyond the boundary of statistical significance. Subsequently, the model estimations are being repeated including both the base and the first lag of the ECI-variants. In testing the combined effects of the two, the logarithm of ECI continues to indicate a significant effect.
Supported by results regarding the first lag coefficients, we can identify some implications based on the interpretations applied by Felbermayr and Yalcin (2013, pp. 976-978).
The first column contains the binary ECI-variable. The point estimate of the first lag, 0.0384, suggests that exports towards a certain country, in which the PECI instrument was active, were increased by 3.84 per cent the year after the PECI grant compared to the situation in which the instrument was not available. It thus moderately supports the market access facilitation hypothesis of the instrument as it relates increased exports to the mere availability of PECI in a certain country in the preceding year. To better grasp the magnitude of this implication, a 3.84 per cent increase in the average merchandise export per year, corresponding to approximately 400 billion euro, (See Figure 1) would correspond to an additional export of 15.36 billion euro7 in the year following the presence of
the ECI-instrument. Certainly, this simplified back-on-the-envelope calculation is applied on total merchandise exports per year and does not control for the average country in which this export expansion could reasonably be established. Nevertheless, it provides a frame of reference for valuing the coefficient appropriately.
The second column exhibits the coefficients for the logarithmic variant of the ECI-variable. Since both the first lag as well as the combined effect of the direct variable and the first lag together yields a moderately significant point estimate, this column allows for interpreting the sum of both estimates, which accounts to 0.00419. The elasticity suggests that a 1 per cent increase in PECI stimulates exports by approximately 0.0042 per cent in the year of issuance and the following year combined. Following through on the reasoning of Felbermayr and Yalcin, we can subsequently derive the instrument’s effectiveness ratio. Taking into account that the share of total merchandise exports covered by PECI in the Netherlands roughly corresponds to 0.006 per cent over the full time-span (see figure 4), the effectiveness ratio (Euros of exports spread out over the two years concerned per Euro provided as PECI) assigned to Dutch PECI corresponds to 0.6988. In doing so, this seemingly minor
point estimate seems to indicate a sound effectiveness of the instrument.
Finally, the third column employs the ratio of ECI over exports as ECI-variable. In this case, the
7 Follows from (400bln x 0.0384)
8 Calculations to the example of Felbermayr and Yalcin (2013, p. 978, footnote 16): Let exports be 100 and the
level of ECI coverage 0.6 (according to the average coverage ratio). Now a 10 per cent increase in ECI (level change of 0.06) leads to a 0.0419 per cent increase in exports (level increase of 0.0419). Subsequently, the effectiveness ratio can be derived by (0.0419 / 0.06 = 0.698).
23 coefficient of 0.058 suggests that an increase in the share of exports covered by PECI in a certain market by one point results in an increase of exports by 0.058 per cent in the following year towards this specific market. As this estimate falls substantially short from unity, the increase in exports is smaller than the additional share in exports underwritten by PECI. To illustrate, a one-point increase in coverage of total merchandise exports would account to an insurance grant of 4 billion euros, which would in turn boost exports by only 232 million euros9 the following year. This implies that a vast part
of exports would have taken place in absence of PECI, which places the instrument’s additionality principle into discussion in case it would be expanded to a larger share of the export transactions. Whereas the effectiveness implications following from the different point estimates might seem contradictory at first glance, the results are in fact in accordance with the theory concerning PECI. First of all, one has to bear in mind that the overall coverage ratio of PECI over aggregate exports in the Netherlands is fairly low. Now in short, the point estimate from column one indicates that sheer availability of PECI for exports towards a certain country had an export enhancing effect of 3.84 per cent in the following year. The second column indicates that every euro used in PECI resulted a certain degree of additional exports distributed over two years implies. In addition, it implies a relatively large effectiveness ratio of euros allocated to ECI, which, as we can see in the footnote calculation, is too a large extent due to the meagre denominator (aggregate coverage ratio). The third column does not relate to availability nor elasticity, but tests for effectiveness of the instrument related to the share of exports covered by PECI. The relatively low point estimate in this case implies the additional exports induced by extending PECI coverage to a larger share of export transaction to be fairly low in proportion to the number of euros that would be additionally allocated to the PECI scheme for realizing this increase. The latter confirms the views expressed by Van den Berg et al. (2017), that the PECI instrument is particularly valuable to a very specific type of industry and less so on the aggregate level.
3.4.3. Comparison to literature
As statistical evidence for these claims remains ‘sufficient to weak’, placing the obtained results in a broader perspective would be welcomed in order to verify the implications. We can establish such comparison in literature by taking aside results of the basic FE-model on German exports presented by Felbermayr and Yalcin (2013, Table 1, p.977), which are included in the Appendix (Table 4). A comparable gravity model is displayed, aiming to determine the export enhancing effect of so-called ‘Hermes guarantees’, the German counterpart of Dutch PECI.
As outlined in the related literature overview, this model employs the industry dimension of
24 exports, thereby substantially increasing the number of observations in the data.
Apart from slight deviations concerning the variables included (see data section) and difference in timespan covered, the point estimates provide certain ground for comparison as the model specification is similar.
In doing so, we observe the general outlook of the models coinciding to a large extent with respect to significance and interaction signs of the control variables’ estimators. Felbermayr and Yalcin find the export enhancing effect of PECI is dominated by the direct effect, while they find no significant estimators for any delayed effects (2013, p.980). However, they did obtain both remarkably larger point estimates and more robust significance levels for the direct effect of ECI-variables. The latter is rationally facilitated by the increased number of observations due disaggregation of exports to industry levels. In any case, these findings support the implication that PECI could be identified as a significant driver of exports using a basic FE-model. However, the differences in extent of effectiveness and the discrepancy between dominance of delayed and direct effects require further discussion. At this point, Felbermayr and Yalcin continue their research controlling for MRT and see opportunity to reproduce the RE-model and the RE-Mundlak model as employed by Moser et al. (2008) on their aggregate data (2013, pp. 978 & 996). While these models do allow for dynamic modelling, they conclude that working with the basic FE-model does not lead to a strong bias in results for estimating short-term trade multipliers. These findings indirectly add to the validity of this thesis research, as the estimation bias due to the model specification seems to be surmountable.
Whereas the exploitation of the sectoral dimension allow Felbermayr and Yalcin to differentiate the effect of PECI across countries (geographical regions, income level classifications and financial market maturity), the use of aggregate exports and PECI unfortunately does not. In an attempt to conduct further research on the heterogeneity of the panel entities, lack of explanatory power in the model proved to hinder further statistically supported insights.
25
4.
Discussion
In discussing the above observations, this chapter continues by addressing the seemingly discrepancies between obtained results and literature related. This is followed by an assessment of the validity of results with respect to data used, model estimation technique and theoretical foundation.
In establishing the comparison with Felbermayr and Yalcin’s findings, one ought to keep in mind the difference in timespan covered in the analysis as well as deviations regarding the variables included in the regression. However, as the size and significance of coefficients do prompt further discussion, a certain degree of discrepancy between the results obtained by Felbermayr and Yalcin and this thesis could be attributed to potential differences in between the Dutch and German PECI-scheme.
Concerning the latter, Felbermayr and Yalcin illustrate that the average coverage ratio of the German scheme amounts to approximately 3 per cent of total German exports (2013, p. 972). This implies that the size of the German scheme relative to total merchandise exports is about five times larger than the relative size of the Dutch scheme. Therefore, as the size of the German instrument is a multiplication of the Dutch counterpart, the discrepancy in size of initial estimation coefficients is not surprising. Secondly, the disparity of dominance in between direct effects and first lagged ECI-variants gives rise to the question of whether there exists a fundamental difference in data reporting, institutional settings of book-keeping or modus operandi of the agency executing PECI-scheme in between countries.
In analyzing the size of ECI-coefficients as well as the effectiveness ratio obtained through the basic FE-model, Moser et al. provide valuable insights through their dynamic Mundlak-model specification. They remark that the short-run trade multiplier is typically substantially below unity (2008, p. 794), which relates to the fact that PECI tends to be granted for a period exceeding one year. Furthermore, the legitimacy of results ought to be critically assessed regarding use of data, model estimation and theoretical validity.
Concerning data, the usage of aggregate data on a yearly base indubitably limits the model in accurately identifying significant estimators as Felbermayr and Yalcin observed ECI to be significantly sector-biased (p. 971). Although constructing a well-balanced industry specific database did not prove to be feasible in this research, it could have produced a number of advantages for the analysis. Firstly, disaggregation of data to industry levels leads to a considerable increase in number of observations (2091 versus 42669), which would allow for inclusion of numerous dummy-variables
26 capturing MRT. Whereas, the overestimation due to omitted MRT were previously identified as ‘very minor’, the results may suffer from aggregation bias.
Moreover, Garrett (2003, p. 64) points out that statistical significance of a coefficient is positively related to coefficient size and negatively related to its standard error. He continues by illustrating the possibility of coefficients from an aggregated regression to be statistically significant, whereas identical coefficients from less aggregated regressions turn out statistically insignificant, and vice versa. Following these findings, the comparison with related literature would be more accurate if the level of data aggregation is harmonized.
In addition, Jensen (2000, p.86) remarks that using disaggregated data allows for uncovering trends within a heterogeneous group of countries and facilitates more in-depth analysis on the export enhancing effects of PECI. As a consequence, disaggregated data analysis provides more refined results for policy analysis.
With respect to model specification, the limited amount of observations increase vulnerability for ‘overfitting’ the model due to the relative number of explanatory variables compared to regression entities. To mitigate this risk, only variables with a strong theoretical argumentation were included.
However, one might question whether a basic FE-model is the best instrument for measuring the effects of PECI. For instance, GMM-estimation would allow for autoregressive dynamics as it facilitates the inclusion of lags for the dependent variable (Exports) thereby accounting for possible endogeneity. As shown by Moser et al. (2008), this would allow for estimating the long-term trade multiplier of ECI. However, accurate results from dynamic modelling would again be subject to a disaggregated dataset and is therefore beyond the scope of this thesis.
On a more general note, the validity of conclusions is subject to reflections on theoretical aspects of the analysis.
Apart from point estimates containing only moderate statistical power, caution is advised in interpreting the effects as mere trade creation. Whereas the availability of PECI-instruments to facilitate exports towards a certain market could potentially restrict manufacturing capacity for exports towards markets that do not require PECI to complete the transaction. Consequently, one might wonder to what extent PECI leads to trade creation as opposed to trade diversion. On the other hand, Dutch PECI is explicitly aimed at being additional to the private sector (Tweede Kamer der Staten Generaal, 2016) and EU regulation limits ECAs activities to ‘non-marketable risks’ (Moser et al., 2008, p.785). Both elements strengthen the case of trade creation due to PECI.
27 be the case that a shock on exports causes ADSB to grant or withdraw public export credit insurance to a specific market in a specific year. However, this elucidation is again not in accordance with the policy objective communicated by the Dutch ministry of Finance and moreover mitigated by the usage of country coverage limits applied by Atradius DSB (Tweede Kamer der Staten Generaal, 2016, p.71). Taking the caveat into account, the result analysis nevertheless presumes a dominant one-way causal effect of PECI on Exports.
28
5.
Conclusion
This thesis took off with an antiquated quote by Jan Tinbergen, insinuating bilateral trade to partly determined by political factors. In analyzing the relation between international trade and international finance, the export enhancing policies conducted by governments and the dynamics around export credit, this quote remains credible in modern times. On the one hand, specific export transactions seem to be jeopardized by the existence of political risk, whereas public intervention in the export credit insurance market indicates a trade-enhancing impact of politicized schemes.
Considering importance of trade to the Dutch economy, this observation led to questioning to what the extent public export credit insurance in the Netherlands stimulate Dutch merchandise exports. In order to answer this main research question, a Fixed-Effects model was constructed based on theoretical foundation of the Tinbergen’s own Gravity model of trade.
Although the model does not provide strongly significant coefficients, the results do imply that Dutch PECI can be identified as an export enhancing determinant of Dutch merchandise exports in the year following the grant. More specifically, the mere availability of PECI in a specific market and year is estimated to increase exports to this market by 3.84 per cent in the following year. Furthermore, the instrument is regarded to having an effectiveness ratio of 0.698 over two years. This implies a short-term trade-enhancing effect of roughly 70 cents per euro reserved in the scheme. However, a third estimate variant does not imply that an extension of the scheme would lead to substantial increase in total merchandise exports.
These estimators are lacking behind in both size and significance levels compared to related research by Felbermayr and Yalcin (2013) analyzing the German case. This discrepancy could be explained partly explained by the fact that the relative size of the German scheme with regard to exports is about five times larger than the Dutch counterpart.
On the other hand, the results are in accordance with earlier established views that the PECI instrument is particularly valuable to a very specific type of industry and less so on the aggregate level. Also, the fact that PECI-coefficients are dominantly observed in the first lag relates well to the characteristics of the Dutch PECI scheme, as it mainly targets transactions with credit or completion periods longer than twelve months.
Although the analysis conducted did not allow for estimating long-term effects , the prospect of conducting more thorough analysis on the export enhancing effect of Dutch PECI has at least become more promising in light of the listed observations. As a result of progressive insights, the use of industry-level data and value-added merchandise exports would be suggested to allow for deeper understanding of country-group and industry specific effects. In addition, the feasibility for a more sophisticated modelling technique would be recommended. In this respect, a RE-Mundlak model and
29 GMM-estimation might be of use to capture both short- and long-term effects as well as the autoregressive effect of exports.
