1
A. van der Veen Supervisor: S. Chan
Cooperation or Confinement? Evidence on EMU Membership Trade Effect for Iceland
Megan Haasbroek 10385274
ABSTRACT
In the last decade, the currency union effect on trade has been the subject of an increasing body of research. In this paper, a specific contribution will be made in the form of estimating the trade impact of EU and EMU membership for Iceland. It is found that the EMU effect on trade is economically and statistically significant at 27 percent, and that acceding to the EMU will induce an estimated increase in Iceland’s long-run real GDP per capita equal to minimum 1.5 percent.
1 Introduction
On 17 July 2009, Iceland formally submitted its application for EU membership (European Commission, 2010, p. 2). Before the elections in 2009, such an advance towards European integration would have been surprising at least, as historically, the Icelandic government has adopted a “wait and see” policy towards integration of any kind (Thorhallsson, 2000, p. 262). However, as the significant economic and social consequences of the Icelandic crisis engendered political turmoil, the established right-wing party was repudiated (EC, 2010, p. 3). The struggle to mitigate the consequences of the Icelandic crisis induced the newly appointed left-wing government to further European integration: the then prime minister Sigurdardottir commented after the 2009 elections that if Iceland were to apply for EU membership, “[it] will be able to adopt the euro within four years” (In: Totaro, 2009). The EU and EMU were considered to be a credible anchor, which could assist Iceland in their economic recovery. In February 2010, the European Commission published a report, recommending that negotiations for Iceland’s accession to the European Union should be started (p. 8). In June 2010, the negotiations were officially opened, and because Iceland is already highly integrated with the EU framework through its membership in the European Economic Area (EEA) 1, the ‘screening exercise’ processed satisfactorily (EC, 2012, p. 74).
1 Signed in 1994, this agreement between all EU members and Iceland, Norway and Lichtenstein guarantees free flow
Statement of Originality
This document is written by Megan Haasbroek 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.
Nevertheless, the integration process stalled relatively soon after it was launched. The sovereign debt crisis and the subsequent instability in Europe, combined with the legal conflict about Icesave, generated increasing dissatisfaction among the populace with the course towards European integration (Council of the European Union, 2012, p. 2, 19). Consequently, at the 2013 elections, the right-wing Independence Party, which is opposed to EU membership, was reappointed into parliament. Ultimately, the negotiations were officially suspended by the Icelandic government in May 2013 (Council of the European Union, 2014, p. 3). Even though the negotiations have not been resumed since then, Iceland did not yet formally retract their application. Therefore, the Council of the European Union has committed themselves to proceed the negotiations at Iceland’s request, as they believe further integration will be mutually beneficial (2014, p. 3).
In this paper, it is endeavoured to quantify one of the benefits of integration for Iceland. More specifically, the trade effect of EMU membership will be assessed through quantitative analysis, and this result will be extrapolated to the situation in Iceland. The next section will provide a theoretical background on monetary unions in general and the European Monetary Union in particular, as well as the political characteristics of Iceland insofar they are relevant for the topic of integration. In the third section, a review of past academic research on currency union effects on trade will be conducted, after which the methodology adopted in this study will be described. Then, the results will be discussed and interpreted for Iceland. Finally, a comprehensive conclusion to the research will be provided.
2 Theoretical framework
In this section, a comprehensive overview is provided of the European Monetary Union (EMU), the benefits and costs of acceding to a monetary union and the political situation in Iceland.
2.1 From continent to cooperation
The decade after the Second World War was marked by a general consensus that the only way to maintain peaceful coexistence was increased cooperation (Feldstein, 2011, p. 2). To that end, in 1952 six countries – Belgium, Germany, France, Italy, Luxembourg and the Netherlands – established the European Coal and Steel Community, a common market in aforementioned commodities (ECB, 2011, p. 4). In an effort to strengthen their communion evermore, the same six nations founded the European Atomic Energy Community (EURATOM) and the European Economic Community (EEC) six years later. This collective became known as the European Communities, and was eventually restructured and augmented into the European Union with the adoption of the Maastricht Treaty in 1993. The European Union not only provided a large free trade area but also freedom of capital and labour mobility (Feldstein, 2011, p. 3). Besides the ever-deepening relationships between the member countries, the union also experienced a broadening membership base: Denmark, Ireland and the United Kingdom entered in 1973, and since then eighteen more countries have acceded to the European Union (ECB, 2011, p. 4).
However, the European Union was but the penultimate stage of European integration. Already after the Second World War, an economic and monetary union in Europe was the ideal aspired by several policymakers (ECB, 2011, p. 4). According to Feldstein, such a vision was motivated by the idea that a common currency would create a greater feeling of social cohesion among Europeans (2011, p. 1). The first official attempt towards a monetary union was the Werner Report in 1970, which described three stages to attain a monetary union by 1980. However, due to the international currency turmoil after the Bretton Woods system breakdown and the international recession anticipating the first oil crisis, the project was never realized (ECB, 2011, p. 5). Despite, or rather because of the international unrest, the nine member states of the EEC at that time introduced the European Monetary System in 1979, with installed fixed exchange rates among their currencies. This proved to be a proper foundation for further steps towards integration, such as the Single European Act in 1986. This act allowed for a single market; nevertheless, it was recognized that a single currency would allow for more trade benefits (Delors, 1989, p. 29). To that end, the Delors Report (1989) was published, detailing a three-stage strategy towards a European and Monetary Union. It was advocated that in the first stage (1990-1993), the European single market should be realized. The second stage (1994-1998) entailed technical preparations for a single currency and improvements in convergence; this period is marked by the formation of the European Monetary Institute. The last stage started in 1999 with irrevocably fixed exchange rates, delegating monetary policy to the ECB and the introduction of the euro.
An EU member state can enter the Eurozone conditional on the convergence criteria as described in the Treaty of Maastricht (1992). These convergence criteria are the following: (1) a high degree of price stability; (2) a sustainable government financial position; (3) a stable exchange rate; (4) stable long-term interest rates (Treaty of Maastricht Art. 109 j, 1992, p. 24). Furthermore, the central bank of the prospective member country needs to be independent to ensure proper delegation of monetary policy to the European Central Bank (ECB, 2011, p. 6). Currently, nine countries – Bulgaria, Croatia, Czech Republic, Denmark, Hungary, Poland, Romania, Sweden and the United Kingdom – are members of the European Union but have not adopted the euro. Of these countries, seven2 have a derogation, meaning that they do not yet conform to the convergence criteria. The protocol dictates that if a country succeeds in accomplishing all five criteria, it is committed to acceding to the Eurozone. The only notable exceptions thus far are Denmark and the UK: as “Members with a Special Status”, they have negotiated an opt-out clause. In effect, regardless of whether they conform to the convergence criteria, Eurozone membership is only realized at the initiative of the country (ECB, 2011, p. 6). In the case of Denmark, the country was allowed to defer participation in the third stage towards EMU. In a referendum, the citizens ultimately voted against entering the Eurozone, so that the exemption came into effect (Treaty of Maastricht Protocols, 1992, p. 35). For the United Kingdom, the membership decision was delegated to the government: in 1997, the government under Tony Blair expressed the intention of
joining the Eurozone provisional on five economic tests. Thus, the UK only wishes to adopt the euro if there is evidence that this will induce higher growth, stability and employment.
As mentioned before, every EU member nation with a derogation will eventually be expected to adopt the euro if all criteria are met. This introduces an additional consideration in the decision on EU membership: most new EU member states want to adopt the euro as soon as possible, as they expect more trade, increased macroeconomic stability and reduced interest rates. However, as De Grauwe and Schnabl assessed in their research on EMU entry strategies, it will be more difficult for “late joiners” to comply with the convergence criteria (2005, p. 537). The “economic catch-up process” towards EMU membership is likely to induce increased inflation rates, effectively violating the first Maastricht criterion. This Maastricht dilemma, or the trade-off between exchange rate stability and price stability, can be avoided if the member states would employ a gradual compliance to the Maastricht criteria (De Grauwe & Schnabl, 2005, p. 553). Therefore, a hard peg to the euro may not be most desirable in circumventing the Maastricht dilemma; despite this argument, several Eastern European countries have adopted this strategy. Nevertheless, De Grauwe and Schnabl argued that once a commitment to a fixed exchange rate has been made, it would be detrimental to a country’s credibility to adopt a different strategy (2005, p. 553). Eventually, the nations that adopted such a hard peg (Estonia, Lithuania, Latvia and Slovenia), managed to enter the Eurozone within ten years after their EU entry. Finally, De Grauwe and Schnabl (2005, p. 538) expressed that the optimal strategy for a prospective member state depends on its budgetary position as well, and whether fiscal consolidation is necessary. To ensure greater European stability ex post, it is important that the entry strategy chosen is most ideal considering the country’s circumstances.
2.2 Monetary unions: a good idea?
New EU member states expressed themselves to be most willing to enter the Eurozone in anticipation of welfare gains and increased stability. These gains stem from different consequences of adopting a common currency. The main benefit of a currency union, and indeed one of the official motivations behind the establishment of the EMU, is the increase in trade (Rose, 2000, p. 10). The extent of the trade benefit has been examined in numerous researches, and has been estimated to be positive and between five and forty percent (Bun & Klaassen, 2007, p. 473). A common currency, superseding the national monies of the member states, will reduce the transaction costs of trade among these member nations (Rose, 2000, p. 10). Furthermore, a common currency will facilitate price transparency, which fosters international competition. A final reason for increased trade is the elimination of exchange rate risk among member countries and thus the need to hedge such risk (Micco, Stein & Ordoñez, 2003, p. 322). In their meta-analysis, Anderson and Van Wincoop (2004) endeavoured to quantify these separate effects. They estimated total trade costs to amount to 170 percent on average, as expressed in terms of their ad-valorem tax equivalent – in other words, trade costs are as much as 170 percent of the value of
the good traded (2004, p. 692). Of this 170 percent, 74 percent3 entails international trade costs; the other 55 percent represents retail and wholesale distribution costs4 (2004, p. 693). To see how a currency union can increase trade among its members, it is insightful to split up the international trade costs. Eight percent of the 74 percent refers to policy costs, which represent tariffs and non-tariff barriers to trade (NTBs); these are completely eliminated in a currency union as it implies a free trade area (Anderson & Van Wincoop, 2004, p. 719). According to Eaton and Kortum (2002), seven percent can be attributed to language barriers (in: Anderson & Van Wincoop, 2004, p. 719). Furthermore, Rose and Van Wincoop (2001) ascribed fourteen percent to currency barriers: this includes both the costs involved in exchanging monies and the influence of exchange rate volatility (In: Anderson & Van Wincoop, 2004, p. 719). More specifically, according to Emerson et al. (1992), these costs may be as high as half a percent of GDP for the European Union as a whole (in: Micco et al, 2003, p. 322). As the exchange rates are irrevocably fixed in a monetary union, both currency risk and exchange costs are eliminated. The information cost barrier amounts to six percent of the 74 percent, and it has been shown the cost is lower if the good in question has a reference price (Anderson & Van Wincoop, 2004, p. 720). Since a common currency increases price transparency across countries, the information barrier may be reduced substantially. Finally, insurance and contracting costs are estimated to be around three percent. However, this is not an important trade barrier in developed countries, as it is caused by corruption and enforcement problems (Anderson & Marcouiller, 2002; In: Anderson & Van Wincoop, 2004, p. 722). In conclusion, a currency union can potentially decrease trade costs substantially.
Another benefit of joining a currency union is increased price stability. Alesina, Barro and Tenreyro argued that if a country with high inflation shares a currency with a credible anchor, its new inflation rate will be equal to that of the anchor plus the change in its price level relative to the anchor (2002, pp. 308-309). In the case of the EMU, the role of credible anchor has been fulfilled by the European Central Bank, maintaining the reputation of the German Bundesbank as an inflation fighter. Therefore, the prospect of low and stable inflation rates can also be a stimulus to join a monetary union.
However, adopting a common currency is not necessarily a guarantee for a successful union. Mundell, in his theory of optimal currency areas, advocated that the domain of a currency area is limited (1961, p. 657). He opined that the optimal currency area is a region defined by internal factor mobility and external factor immobility (1961, p. 661). In determining whether a collective of countries would be part of a potential OCA, one should consider three other interrelations besides the degree of factor mobility: the extent of trade linkages; the degree of business cycle synchronization; and the fiscal transfer system (Frankel & Rose, 1998, p. 1011). The greater the degree of fulfilling these criterions, the more appropriate is a common currency. First, the more two nations trade ex ante, the greater will the trade benefit be after adopting a currency union (Barr, Breedon & Miles, 2003, p. 577). The other three
3 All values are based on an elasticity of substitution of 𝜎 = 8, as the actual value for 𝜎 is to be in the range of five to
ten (Anderson & Van Wincoop. 2004, p. 716).
characteristics are related to the concept of automatic stabilizers5. Members of a currency union depend on a supranational monetary authority as they necessarily give up monetary policy as one of the national stabilization tools (Micco et al., 2003, p. 317). However, especially in the case of idiosyncratic business cycles among member states, one central policy tool may be too blunt to appropriately respond to asymmetric shocks on a national level (Frankel & Rose, 1998, p. 1009). Nevertheless, the absence of monetary policy autonomy is not necessarily unfavourable if automatic stabilizers are active in the currency union. These stabilizers are best illustrated with an example as formulated by Mundell (1961, p. 658). Consider two countries, A and B, which are initially in equilibrium. If a positive demand shock hits country A, this introduces inflationary pressures in country A and unemployment in country B. If the countries share a common currency and thus a common monetary authority, this disequilibrium cannot be appropriately resolved with monetary tools: if the supranational authority were to increase the money supply to alleviate the unemployment predicament in country B, country A would be subject to even higher inflation. However, with automatic stabilizers, equilibrium may be attained without imposing negative consequences on one of the countries. If automatic stabilizers are implemented on a supranational basis, by which the negatively impacted country automatically receives wealth transfers from the country that is better off. Furthermore, if labour mobility is high, the unemployed workers in country B will relocate to country A, thus restoring equilibrium at minimal cost.
In conclusion, a currency union is more suitable if the countries have synchronized business cycles or, lacking that, employ automatic stabilizers to serve as a substitute for autonomous monetary policy. Then, one can wonder to what extent the EMU is an optimal currency area. This is an empirical problem of considerable interest to policymakers and academics alike (Mundell, 1961, p. 660). Alesina, Barro and Tenreyro (2002) endeavoured to answer this question by comparing the co-movement of output, price and trade relations of countries with the then EMU, the United States and Japan, in order to determine the compatibility of these countries with the euro, dollar or yen. They found that a currency union in Europe – including both the EMU countries and the non-members such as Sweden – could be appropriate based on aforementioned characteristics (2002, p. 331). Also, using a probit regression, they found that both comovement of prices and comovement of outputs had a positive and significant effect on the propensity of joining a currency union (2002, p. 341).
On the other hand, De Grauwe and Schnabl (2005, p. 354) raised concerns about the new Eastern European member states: even though the prospective high inflation is a symptom of equilibrium transition, the dangers of “highly heterogeneous real interest rates” will prevail in the enlarged EMU as a consequence of the loss of monetary policy autonomy. Furthermore, several countries that are either willing or wanted to join the EMU have idiosyncratic business cycles as compared to the current EMU
5 The concept of automatic stabilizers relates to how fiscal instruments mitigates extreme swings in the economy as tax
revenues and demand for social security payments varies with economic growth. In recessions, for example, incomes are lower and unemployment is higher. This means, ceteris paribus, that the government receives less tax revenue and has to pay more in unemployment benefits. The increase in government spending limits the decrease in aggregate demand.
countries; especially Iceland is an example of this phenomenon (Breedon & Pétursson, 2006, p. 724). Given this evidence, one could be inclined to argue that the European Monetary Union is not an optimal currency area. However, Frankel and Rose (1998, p. 1013) advocated that the OCA criteria are jointly endogenous, noting that the second criterion, the correlation of business cycles among the members of the monetary union, is in itself dependent on trade integration. They found that historically, a greater degree of integration has resulted in increasingly synchronized business cycles (1998, p. 1031). In that respect, currency unions may be more optimal ex post than ex ante (Frankel & Rose, 1998, p. 1031).
2.3 Iceland and the EMU
Whether or not Iceland should accede to the EMU is a matter of considerable interest for several reasons. First, as argued by Breedon and Pétursson (2006, p. 723), Iceland is a strong candidate for an exchange rate peg in any form as it is a small country with a floating currency. Furthermore, they stated that Iceland’s trade with EMU countries has augmented in the past years, suggesting an increasing importance of European trade (2006, p. 726). This finding suggests that further European unity can be favourable for Iceland. They also noted that Iceland’s trade share of GDP of 30 to 40 percent is relatively low compared with the average of 50 percent for small countries. One explanation for this phenomenon may be that Iceland’s economy is mainly based on exporting natural resources such as fish and sustainable energy and importing final goods; this in contrast to many other small European countries that import intermediate goods and export final goods, implying a large trade share of GDP (Breedon & Pétursson, 2006, p. 725). This discrepancy is related to industrialization, which started relatively late in Iceland compared to other European countries (Thorhallsson, 2001, p. 265). Moreover, the Icelandic trade share may be relatively low due to its remote location. Regardless of trading volume, as Iceland’s greatest trade partner is the EU, accounting for two-thirds of its trade, European integration may indeed be highly beneficial.
Nevertheless, Iceland did not follow its fellow Nordic states in adopting policies regarding to European integration (Thorhallsson, 2001, p. 258). Historically, the Nordic states have been opposed to European unity; it was only after the collapse of the Soviet Union that most countries started a process of integration (Thorhallsson, 2001, p. 258). Still, up until 2008, Iceland has been characterized as adopting a “wait and see” policy regarding European integration; and it seems that after the right-wing government was voted into parliament in 2013, that stance has been resumed. Several considerations can be provided for this persistent political reluctance.
According to Ingebritsen, the main argument against further European integration is the European Union’s Common Fisheries Policy, which is said to be jeopardizing Iceland’s fisheries (2000, p. 129). The policy was implemented to ensure long-term sustainability in the fishing sector and entails several regulations and restrictions all member countries need to comply with, including Total Allowable Catches and fair competition for all EU members (EU, 2013, p. 22). Politicians and interest organizations are afraid that the latter arrangement will threaten the small fisheries as they cannot keep
up with the increased competition (Ingebritsen, 2000, p. 39). On the other hand, the larger and more export-oriented fisheries will likely profit from the increase in market size. As mentioned before, the Icelandic economy is founded on primary sectors such as farming and fishing. Therefore, fish is regarded as a vital natural resource and the Icelandic government is set on preserving it as the primary sector dominates Iceland’s economy (Ingebritsen, 2000, p. 126).
Furthermore, Thorhallsson (2001, p. 264) argued that Iceland’s reluctance to accede to European integration has several other, more political, causes. First of all, interest groups have a great influence on Icelandic politicians: according to Ingebritsen (2000, p. 168), political decisions are formed by a partnership between economic interest groups and politicians, with little room for the public. Of these interest groups, the one representing farmers and fisheries is most prominent, and they campaign against European unity (Thorhallsson, 2001, p. 264). This position has manifested in 1993, when Iceland considered EEA membership. The EEA agreement was promoted as a pact that would produce an unlimited and almost tariff-free export market for fish. Still, several interest groups voted against the EEA, spurring a large debate, so that the president considered refusing to the sign the agreement (Thorhallsson, 2001, p. 265). While there is currently general consensus on the EEA agreement, politicians are reluctant to enter a similar turmoil over EU membership (Thorhallsson, 2001, p. 265). Especially with the current right-wing government, the situation is not likely to change. Moreover, this conflict between pro- and anti-European parties is not only prevalent in politics; labour organizations experience a similar animosity (Thorhallsson, 2001, p. 269). For example, the Icelandic Federation of Labour has become pro-European, recognizing the benefits of EEA membership, but the organization for farmers and fisheries is still opposed to it.
Another reason for the politicians’ unwillingness to proceed with European integration is caused by the voting system. According to Thorhallsson (2001, p. 270), the rural areas have an unfair number of seats compared to the cities and their surroundings. As the farmers and fisheries are situated in the pastoral localities, the share of the public opposing European integration is larger there than in the urban areas. Therefore, the chosen politicians are less likely to challenge the “wait and see” policy and act in support of EU membership. Thorhallsson argued that this situation is exacerbated by Iceland’s policymaking: rather than considering the opposition in the decision-making process, as is the norm several European countries, they do not have a tradition of coalitions (2001, p. 270). Thus, only the demands of the leading party are considered; for the last sixty years this has been the right Independence Party which has connections with fisheries’ and farmers’ associations (Thorhallsson, 2001, p. 271).
A final cause for hesitation in attaining European unity is the state of the governmental administration. First of all, the administration is not independent: it only compiles information at the request of the ministers (Thorhallsson, 2001, p. 273). This means that the staff cannot investigate the issue of European integration or consider novel ideas, as they are not expected to do so. Besides the fact that the administration has no autonomy over its research, it is also small in terms of both staff and available resources (Thorhallsson, 2001, p. 273). These two characteristics restrict the government
administration in its ability to influence politics. While in the other Nordic states, the administration has become more influential, especially in the light of the copious amounts of data that are currently available, this trend has not continued in Iceland (Thorhallsson, 2001, p. 273).
In conclusion, while the Icelandic right-wing government’s main concern seems to be the threat to the fisheries and farmers, there are several structural problems underlying Iceland’s decision-making, including the primary sector being more represented and the smallness and dependency of the government administration. It would go beyond the scope of this paper to address all these issues in more depth, but it is important to be aware of the backgrounds to Iceland’s political reluctance in European integration.
3 Literature review
Introduced in 1962, the gravity model is used to describe trade flows between two countries by assuming them to be proportional to income and inversely related to the distance between the countries (Tinbergen, 1962. In: Rose, 2000, p. 13):
𝑇𝑖𝑗= 𝐴 × 𝑌𝑖× 𝑌𝑗/𝐷𝑖𝑗
where 𝑇𝑖𝑗 is the bilateral trade (average of exports and imports) between two countries 𝑖 and𝑗, 𝐴 is a
constant term, 𝑌𝑖 and 𝑌𝑗 are the Gross Domestic Products of country 𝑖 and 𝑗 respectively and 𝐷𝑖𝑗 is the
distance between the two countries. In practice, the logarithmic model is often estimated.
Besides intuitively appealing in its theoretical underpinnings, the gravity model has provided consistent and robust results (Leamer & Levinsohn, 1995, p. 1384. In: Rose, 2000, p. 13). According to Rose (2000, p. 13), the gravity model has been used extensively: e.g. to prove that trade induces growth (as in Frankel & Wei, 1993) and to investigate anomalies with respect to the Law of One Price (Engel & Rogers, 1996).
Rose (2000) was the first researcher to estimate the effect of a currency union on trade, in the light of the newly formed Eurozone. Using an augmented gravity model6, and a sample of 186 countries over five different years, he found that countries with a common currency will trade three times as much compared to countries with different currencies (2000, p. 17). However, this controversial result has been subject to criticism and efforts to “shrink” the currency union effect. The main critique is based on the type of analysis. In his paper, Rose used cross-sectional analysis to investigate whether countries with the same currency trade more. However, it has been argued this is not very relevant from a policy perspective and instead, panel data should be used to study the effect on trade of joining a currency union (Micco et al., 2003, p. 323-324). Also, Rose’s results do not exhibit external validity to the Eurozone, as he acknowledged himself (2000, p. 15). In his sample, less than one percent of the countries was in a monetary union, and those, as small developing countries, were not representative for the EMU member states (Breedon & Pétursson, 2006, p. 727). A third criticism refers to the endogeneity involved in the monetary union dummy variable: as the selection of countries into monetary unions is not random,
6 The Rose specification regresses log of bilateral trade on output per capita, a contiguity and language dummy, dummies
all factors that influence such a decision need to be accounted for (Persson, 2001. In: Micco et al., 2003, p. 324). Thus, if countries expect an increase in trade after adopting a common currency, the monetary union coefficient cannot be interpreted as the monetary union effect (Breedon & Pétursson, 2006, p. 727). One approach to resolve this, adopted by Rose (2000), is to add a cornucopia of variables that may influence trade links (Breedon & Pétursson, 2006, p. 727). Furthermore, to isolate the effect of a common currency from other variables that influence trade – a close cultural or historical relationship for example – country-pair fixed effects can be included in the regression model. Using country-pair fixed effects entails including a dummy variable for every country pair in the sample. Then, the currency union coefficient is only significant if there is change in trade post-EMU rather than higher average trade (Breedon & Pétursson, 2006, p. 727). Despite controlling for omitted variables and fixed effects, there may still be a potential endogeneity problem. According to Tenreyro (2001), an omitted variable that influences the propensity to join a monetary union and predicts higher trade ex post may lead to positive bias in the OLS estimates (In: Micco et al., 2003, p. 324). To resolve this problem, an instrumental variable estimation may be in order.
Barr, Breedon and Miles (2003) adopted the latter approach and used output co-movements as an instrument, since it has a strong relationship with EMU entry (p. 582). Their sample included all EU and EFTA members except Luxembourg and Lichtenstein7 over a timespan from 1978Q1 to 2002Q1. Using a similar regression model as Rose (2000), they found an EMU effect on trade of 29 percent with an OLS regression; of 27 percent with the fixed effects regression and the instrumental variable estimation yielded an effect equal to 23 percent (2003, pp. 581-582). However, even though it is suggested in OCA literature that output co-movements and monetary unions are closely related, using output co-movements as an instrument is not ideal: according to Frankel and Rose, cycle correlations are closely associated with trade intensity as well, thus impairing the independence of the instrument (1998, p. 1013). Also, output co-movements are time invariant, so that even if they can predict currency union membership, they cannot predict when the accession will occur (Barr et al., 2003, p. 583).
Based on this argument, Micco et al. (2003) used fixed-effects regressions to estimate the effect of the EMU on trade. Another justification for not using IV regression is their relatively short sample period from 1999 to 2002, as it mitigates the probability that the decision to join is instigated by an event (Micco et al., 2003, p. 330). Micco et al. use two samples: one including 22 developed countries and the other comprising the EU 15. While the first sample provides the advantage of a larger sample size, the latter sample is more homogeneous so that it is less likely that the results will be impaired by omitted variables (Micco et al., 2003, p. 327). Their gravity model is similar to the Rose specification, although they replace exchange rate volatility with the real exchange rate of both countries in each country pair and they added an EU Trend variable to account for increasing adoption of EU laws in national
7 Luxembourg and Lichtenstein were excluded due to limited data availability. The sample included Austria, Belgium,
Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and the UK.
constitutions (2003, p. 348). They found that using an OLS regression, trade increased by approximately 25 percent when joining a monetary union; in the case of a fixed-effects regression this effect decreased to 7.3 percent. It thus seems that endogeneity is impacting the OLS results, and that adopting a fixed-effects approach mitigates this concern (Micco et al., 2003, p. 330).
A later study, which draws on Barr et al. (2003), is conducted to estimate the trade effect for Iceland. Breedon and Pétursson (2006) used the results reported by Barr et al. to analyse the implications for Iceland if it were to accede to the Eurozone. They find that the total increase in trade with EMU members for Iceland is 59 percent, composed of an EU effect of 28 percent, an EMU effect of 29 percent and an exchange rate volatility effect of two percent (2006, p. 733). As in 2002 the trade share of Iceland with the EMU-12 members was 40 percent with a trade-to-GDP ratio of 50 percent, this implies an increase in the trade-to-GDP ratio of twelve percentage points (Breedon & Pétursson, 2006, p. 733). If, as according to the HM Treasury (2003), one percentage point increase in the trade-to-GDP ratio increases GDP by one-third percentage point, EMU membership could augment Iceland’s long-term GDP per capita by four percentage points (Breedon & Pétursson, 2006, p. 735).
In general, most panel data studies on the currency union effect on trade yield roughly similar estimates, ranging from five to forty percent depending on the estimation method (Bun & Klaassen, 2007, p. 473). However, there are still limitations to be addressed. First, as mentioned before, while an instrumental variable estimation is desirable, no proper instrument has been found yet (Micco et al., 2003, p. 330). Second, and perhaps most important, is the fact that most studies were conducted when the EMU was recently established, thus allowing for the possibility that any long-term effects are not captured in the currency union effect. In this paper, this is the main limitation that will be addressed.
4 Methodology
In this section, the sample will be defined and the process of data collection will be described. Also, a comprehensive overview of the data will be provided, as well as a detailed description of the model.
4.1 Data collection and sample statistics
The sample includes the eighteen countries that were EMU members as of 20148 plus Iceland, and ranges from 1999 to 2014. It is important to note that this study cannot assess the EMU effect on trade for the initiators, as there is no pre-EMU data. Instead, in this paper it was endeavoured to estimate the trade effect for the late joiners, as it was presumed that these countries are more representative for Iceland in both characteristics and situation. For example, the majority of the late joiners is Eastern European, and only after the Iron Curtain was dismantled and the ties with the Soviet Union were abolished, the countries focused on their economic development. As mentioned before, Iceland underwent the process of industrialization in the 1960s and is thus a recently developed economy as well.
8 Specifically, these countries are Austria, Belgium, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy,
In the study, the dependent variable is the log of bilateral trade, which is the average of imports and exports between two countries. Data on bilateral trade flows in current US dollars was obtained from the IMF Direction of Trade Statistics (DOTS) database for all countries in the sample. To ensure the data is least impaired by reporting errors, the average of four measures was taken for each country pair: imports and exports from country 𝑖 to country 𝑗 and imports and exports from country 𝑗 to country 𝑖. Then, the trade amounts were deflated using the 2005 US Consumer Price Index retrieved from the World Bank’s World Development Indicators. The independent variables included real GDP and real GDP per capita as a proxy for economic size. These values (in 2005 US dollars) were obtained from the World Bank’s World Development Indicators until 2013. As there was no data available for 2014, the real GDP growth for that year was used to compute the missing values. Real GDP growth was obtained from the IMF World Economic Outlook 2015 database. To compute real GDP per capita in 2014, the countries’ real GDP was divided by the respective population statistics obtained from the EU website. Besides these traditional gravity variables, several proxies of difficulties to trade were formulated. The distance variable was computed as the Great Circle Distance between the central points in the respective countries as listed in the CIA World Factbook. The CIA World Factbook was also used to code the contiguity dummy that takes on the value of 1 if the country pair shares a common border. The language dummy, that takes on the value of 1 if the country pair shares a common or similar9 official language, was coded using Lindsay’s research on mutual intelligibility in Slavic, Germanic and Roman languages. In defining the exchange rate volatility, the monthly bilateral exchange rates for all sample countries were derived from OANDA and the standard deviation for the returns was computed. Data on the Free Trade Agreement, EU and EMU membership for the respective dummies was obtained from the European Commission’s overview of regional trade agreements.
A first look at the data shows a possible positive currency union effect on trade. In figure 1, the amounts of bilateral trade with the EMU-18 are presented for the late joiners Cyprus, Estonia, Greece, Latvia, Malta, Slovak Republic and Slovenia. It can be inferred that most, if not all, countries enjoyed an increase in trade in the process of entering the EMU. As expected, the trade amounts for the smaller economies (Cyprus, Estonia, Latvia and Malta) are lower than that for the larger economies Greece, Slovak Republic and Slovenia. Furthermore, except for a decrease in trade coinciding with the sovereign debt crisis, most late joiners have experienced persistent trade growth after they entered the EMU, suggesting that the trade impact may hold in the long run. Moreover, the data on Iceland corroborates Breedon and Pétursson’s (2006, p. 726) argument that the EMU is increasingly important for Iceland in terms of trade. As shown in figure 2, except for the Icelandic and sovereign debt crisis, Icelandic bilateral trade with the EMU-18 has been increasing over the past decades. This suggests that entering the EMU may be appropriate to cultivate this interdependence.
9 Here, similarity is defined as mutual intelligibility, where speakers of the respective languages can understand each
Figure 1. Bilateral trade with EMU-18
Source: Author’s calculations with IMF Direction of Trade Statistics.
Figure 2. Icelandic bilateral trade with EMU-18
Source: Author’s calculations with IMF Direction of Trade Statistics. 0 5.000 10.000 15.000 20.000 25.000 30.000 35.000 Am o un t in m illi o n 2 0 0 5 $ Year Cyprus Estonia Greece Latvia Malta Slovak Rep. Slovenia 0 500 1000 1500 2000 2500 1999200020012002200320042005200620072008200920102011201220132014 Am o u n t in m il li o n 2 0 0 5 $ Year
4.2 Definition of the model
In this paper, an adapted version of the Rose specification, derived from Barr et al. (2003), will be used to assess the effect on trade, using both an Ordinary Least Squares and a fixed effects estimation. To account for the fact that Iceland is in the European Economic Area, the model was extended by including a Free Trade Agreement dummy. The model used in the OLS estimation is formulated as follows:
ln 𝑇𝑖𝑗𝑡 = 𝛼𝑖𝑗+ 𝛽1ln Υ𝑖𝑡Υ𝑗𝑡+ 𝛽2ln ( Υ𝑖𝑡 𝑃𝑖𝑡 Υ𝑗𝑡 𝑃𝑗𝑡 ) + 𝛽3ln 𝐷𝐼𝑆𝑇𝑖𝑗+ 𝛽4𝐸𝑋𝑉𝑂𝐿𝑖𝑗𝑡+ 𝛽5𝐶𝑂𝑁𝑖𝑗+ 𝛽6𝐿𝐴𝑁𝑖𝑗 + 𝛽7𝐹𝑇𝐴𝑖𝑗𝑡+ 𝛽8𝐸𝑈𝑖𝑗𝑡+ 𝛽9𝐸𝑀𝑈𝑖𝑗𝑡+ 𝜖𝑖𝑗𝑡 Where:
ln 𝑇𝑖𝑗𝑡 is the log of bilateral trade between countries 𝑖 and 𝑗.
ln Υ𝑖𝑡Υ𝑗𝑡 is the log of the product of the countries’ respective real GDP.
ln ( Υ𝑖𝑡
𝑃𝑖𝑡
Υ𝑗𝑡
𝑃𝑗𝑡) is the log of the product of the countries’ real GDP per capita. ln 𝐷𝐼𝑆𝑇𝑖𝑗 is the log of the distance between the two countries.
𝐸𝑋𝑉𝑂𝐿𝑖𝑗𝑡 is a measure of exchange rate volatility, defined as the standard deviation of returns of the
bilateral exchange rate for the two countries.
𝐶𝑂𝑁𝑖𝑗 is a contiguity dummy that takes on the value of 1 if the country pair shares a common border
and 0 if otherwise.
𝐿𝐴𝑁𝑖𝑗 is a language dummy that takes on the value of 1 if the country pair shares a common or similar
official language. Countries that have multiple official languages are coded with fractional values10. The language dummy is equal to zero in other cases.
𝐹𝑇𝐴𝑖𝑗𝑡 is a dummy that takes on the value of 1 if both countries in the pair are part of the same Free
Trade Agreement. In this sample, the only relevant FTA is the European Economic Area (EEA) and the values are coded as such.
𝐸𝑈𝑖𝑗𝑡 is a dummy that is coded as 1 if both countries are members of the European Union.
𝐸𝑀𝑈𝑖𝑗𝑡 is a dummy that takes on the value of 1 if both countries in the pair are members of the EMU.
𝜖𝑖𝑗𝑡 is an error term.
The time fixed effects model extends this regression with time dummies (𝜏𝑖𝑗𝑡), which will
account for year-specific effects. It can be inferred that introducing time dummies will improve the model as for example the sovereign debt crisis will be accounted for. The final specification is the time and entity fixed effects model with both time and country pair dummies (𝜂𝑖𝑗𝑡). As mentioned before,
the entity fixed effect will comprise of all characteristics that influence trade between two countries, both observable and unobservable, while time dummies. Therefore, in this fixed-effects regression, the listed “difficulty-of-trading proxies” (in this model: distance, contiguity and language) will be excluded
10 E.g. for Belgium, which lists both Dutch and French as official languages, the language dummy takes on a value
of 0.5. While German is also an official language in Belgium, it is natively spoken by less than one percent of the population, which is why it coded as a 0.
from the model since they have no time series variation (Barr et al., 2003, p. 610). As this model limits the possibility of omitted variable bias the most, this specification will be the preferred estimation.
The expected sign of the variables entering the regression has been well-documented in the literature, for example in Micco et al. (2003). The variables accounting for economic size and wealth and living standards, real GDP and real GDP per capita respectively, are expected to have a positive effect on trade as they are related to both production and consumption. Exchange rate volatility will affect trade negatively, as it introduces the need for exchange rate hedging and thus is a barrier to international trade. Propinquity, in both the cultural and geographical sense, is reflected in the distance variable and the contiguity and language dummies. Sharing a border or language increases trade, while geographical distance is detrimental to trade. Finally, the dummies for FTA, EU and EMU membership are expected to positively impact trade, as such agreements eliminate tariff and non-tariff barriers.
5 Results 5.1 Overview
In table 1, the results of three variants of the basic equation are provided. The first column shows the OLS estimation of the model, the second column shows the fixed effects regression with time dummies only and the last column describes the result for an estimation accounting for both time and entity fixed effects.
First of all, the R-squared, provided chi-squared and F-statistics indicate that the three models are a good fit with the data, as the null hypothesis that all coefficients are zero is rejected. All variables, except contiguity, enter the model with the appropriate sign: as expected, economic size and propinquity in terms of language or community have a positive influence on trade, while exchange rate volatility and distance affect trade negatively. Contiguity is expected to improve trade, but here, the coefficient is below zero. However, it should be noted that this number is small and not significantly different from zero in both the OLS and time fixed effects estimation. Another effect that is documented differently in the previous literature is the influence of exchange rate volatility: for this sample, in all specifications, the effect – albeit not significant – was relatively low. For example, Barr et al. (2003) estimated the coefficient to be equal to -0.12. One explanation for this discrepancy is the sample employed in the research. For example, in Rose’s research, which spanned 168 geographical units, the sample mean exchange rate volatility was five percent and its standard deviation amounted to seven percent (2000, p. 17). In contrast, the exchange rate volatility in this sample exhibits a mean of 0.5 percent and a standard deviation of one percent. This is consistent with the fact that more than half of the countries in this sample are in a monetary union and the other countries had adopted a fixed exchange rate pegged to the euro. Furthermore, it can be observed that the effect of joining a free trade agreement is relatively low and not significant: in the preferred fixed effects model, the trade impact amounts to 5.13 (exp (0.05) − 1) percent. Interestingly, the size of the EU coefficient varies greatly across the different model
specifications, ranging from 27 percent in the OLS specification to 7.25 percent in the fixed effects model.
Table 1. Trade Effect EMU membership
(1) (2) (3)
OLS Time FE Time and Entity FE
Real GDP 0.83*** 0.82*** 0.06
(0.04) (0.03) (0.43)
Real GDP per capita 0.34*** 0.18 1.27***
(0.10) (0.11) (0.41)
Exchange Rate Volatility -0.01 -0.03* -0.02
(0.02) (0.02) (0.02) Distance -1.21*** -1.23*** (0.16) (0.15) Contiguity -0.12 -0.08 (0.35) (0.32) Common Language 0.66 0.66 (0.50) (0.47)
Free Trade Agreement 0.07 0.10 0.05
(0.36) (0.36) (0.41) EU Membership 0.24 0.11 0.07 (0.35) (0.35) (0.40) EMU Membership 0.12** 0.21*** 0.24*** (0.05) (0.06) (0.06) Observations 2736 2736 2736 R-squared 0.9072 0.9167 0.9837
Test statistic Wald 𝜒2 (9) Wald 𝜒2 (24) F(21, 170)
2552.29*** 3476.56*** 45.77***
Time Fixed Effects No Yes Yes
Country-pair Fixed Effects No No Yes
Notes: Clustered standard errors in parentheses: * p < 0.10, ** p < 0.05, *** p < 0.01
Constant and time dummies not reported.
Curiously, the EMU effect, as tabulated in table 2, increases across the different estimations; in previous literature the effect either slightly or significantly decreased when a fixed effects model was used. There, it was hypothesized that reverse causality may have been the cause of this discrepancy, or the idea that high trade flows induce currency union membership rather than vice versa. However, as the sample statistics showed, trade flows improved for the late joiners after their accession to the EMU,
so that reverse causality is seemingly not present here. One reason may be that the founding members of the EMU already had high levels of trade among each other before the common currency unlike the newer member countries, especially since most late joiners entered the EMU relatively soon after joining the EU. Nevertheless, as the country-pair fixed effects model is the preferred estimation11, an EMU effect of 27 percent will be considered in the remainder of this paper.
Table 2. The euro’s effect on trade Implied proportional impact on trade (%)
(1) (2) (3)
OLS TIME FE CP FE
EMU membership 12.75 23.37 27.12
Note: These values are calculated as exp(𝛽9) − 1, where 𝛽9 is the coefficient for the EMU membership variable
in all three model specifications.
5.2 Implications for Iceland
Now the coefficients have been estimated, the total effect of acceding to the EMU can be calculated for Iceland. The computations can be found in table 3. As in Breedon & Pétursson (2006), the total increase in trade comprises three effects: the EU effect, the EMU effect and the effect of eliminating exchange rate volatility. As mentioned before, the exchange rate volatility coefficient is much larger than in previous literature, which suggests that the estimated trade effect will exhibit upwards bias.
Table 3. Trade effects of entering EU and EMU
EU effect Pure EMU effect Exchange rate volatility effect
Total effect
(%) (%) (%) (%)
Iceland 7.25 27.12 1.23 35.6
Note: The trade effects described above are derived from the country-pair fixed effects model in table 1. The
exchange rate volatility is found by computing the predicted increase in trade if exchange rate volatility had been naught since 2014 (see appendix).
Thus, the trade cost of not acceding to the EMU amounts to 35.6 percent, which is an economically substantial figure. However, this number is only representative for intra-EMU trade and does not account for the possibility of trade diversion with non-EMU members. Nevertheless, while Micco et al. stated that adopting a common currency may result in preferential trade liberalization, they found that entering a monetary union induced more openness in general, increasing trade with all countries (2003, p. 334). Thus, as the effect of trade diversion is negligible, the given estimate can be used to compute the effect on Iceland’s GDP. As Iceland’s current trade-per-GDP ratio is 31 percent, and about 41 percent is attributable to EMU countries, the implied increase in trade as a share of GDP is 4.512 percentage points.
11 Tests for the appropriateness of the fixed effects can be found in the appendix. 12 31% × 41% × 35.6% = 0.045
This figure implies a subsequent increase in real income, although this is difficult to estimate. Breedon and Pétursson (2006) used two different estimates to compute the ultimate effect on long-term real income. A conservative estimate is provided by HM Treasury (2003), stating that a percentage point increase in trade as a share of GDP raises real GDP per capita by “at least 1/3 percent in the long run” (Breedon & Pétursson, 2006, p. 735). This implies an increase in long-run real GDP per head of 1.5 percent. Another measure was presented by Frankel and Romer (1999), who suggested that there is a one-to-one relationship between the trade-to-GDP ratio and long-run GDP per capita (Breedon & Pétursson, 2006, p. 735). Correspondingly, Icelandic long-run GDP per capita could increase by 4.5 percent if Iceland were to enter the EMU.
6 Conclusion
In entering a monetary union, there is a trade-off to be made between the benefits of increased trade and price stability and the cost of losing monetary policy autonomy. Besides, social and political considerations should also be recognized in the process towards acceding to a currency union. The latter proved to be highly relevant in Iceland, as the fishermen and farmers are opposed to European integration. These groups opine that further unity will harm their vocation, and since they are influential in both the political and social environment, they induce the Icelandic government to oppose EU membership. Indeed, save a brief period between 2009 and 2013, in which Iceland obtained official candidate status, the ruling party has consistently defied European unity.
In order to provide another, more economic perspective on EU and EMU membership for Iceland, the trade benefit for Iceland was quantified. By means of an extended gravity model, the impact on trade of EU and EMU membership were estimated and applied to the Icelandic economy. It was found that while the EU effect was not significant, the trade effect of entering the EMU was economically and statistically significant at 27 percent. It was estimated that the total potential increase in intra-EMU trade, if Iceland were to adopt the euro, would be 35.6 percent. This would induce a raise in the trade-per-GDP ratio of 4.5 percentage points, augmenting Iceland’s long-run GDP per capita by 1.5 percent.
Still, there are several limitations to this research. First, as the possible other benefits and the costs of adopting the euro were not quantified in this study, it would be premature to infer conclusions from this study only about Iceland’s European integration. Other consequences in the economic sphere, as well as the social and political environment, need to be addressed to develop a balanced opinion on the future of Iceland. Second, while there is general consensus that an instrumental variable regression is appropriate, a proper instrument has not been documented yet. An interesting future research perspective for this topic would be the formulation of an instrumental variable, as to develop a more defined estimate of the currency union effect on trade.
Bibliography
Alesina, A., Barro, R. J., Tenreyro, S. (2002). Optimal Currency Areas. In: Gertler, M. and Rogoff, K. (2003). NBER Macroeconomics Annual 2002 Vol. 17. Cambridge: MIT Press, 301–356. Anderson, J., and Van Wincoop, E. (2004). Trade Costs. Journal of Economic Literature 42(3), 691–
751.
Barr, D., Breedon, F. and Miles, D. (2003). Life on the Outside: Economic Conditions and Prospects outside Euroland. Economic Policy 18(37), 575–613.
Breedon, F. and Pétursson, T. G. (2006) Out in the Cold? Iceland's Trade Performance outside the European Union and European Monetary Union. Cambridge Journal of Economics 30, 723– 736.
Bun, M. J. G. and Klaassen, F. J. G. M. (2007), The Euro Effect on Trade is not as Large as Commonly Thought. Oxford Bulletin of Economics and Statistics 69, 473–496.
Council of the European Union (2012), Council Conclusions on a Homogeneous Extended Single Market and EU Relations with Non-EU Western European Countries. Retrieved from the Council website: http://www.consilium.europa.eu/uedocs/cms_data/docs/pressdata/en/ er/146315.pdf
De Grauwe, P. and Schnabl, G. (2005). Nominal Versus Real Convergence – EMU Entry Scenarios for the New Member States. Kyklos 58(4), 537–555.
Delors, J. (1989). Report on Economic and Monetary Union in the European Community. ISBN 92-826-0655-4. Retrieved from: https://infoeuropa.eurocid.pt/files/database/000000001-000005000/000003936_2.pdf
European Central Bank (2011). The European Central Bank, the Euro system and the European System of Central Banks. Luxembourg: Imprimerie Centrale.
European Commission (2010). Communication from the Commission to the European Parliament and the Council: Commission Opinion on Iceland’s application for membership of the European Union. Retrieved from the EC website: http://ec.europa.eu/enlargement/pdf/ key_documents/2010/is_opinion_en.pdf
European Commission (2012). Communication from the Commission to the European Parliament and the Council: Enlargement Strategy and Main Challenges 2012-2013. Retrieved from the EC website: http://ec.europa.eu/enlargement/pdf/key_documents/2012/package/strategy_paper_ 2012_en.pdf
European Union (1992). Treaty of Maastricht. Retrieved from: http://www.eurotreaties.com/ maastrichtec.pdf
European Union (2013). Regulation No 1380/2013 on the Common Fisheries Policy. Official Journal of the European Union 354. Retrieved from: http://eur-lex.europa.eu/LexUriServ/ LexUriServ.do?uri=OJ:L:2013:354:0022:0061:EN:PDF
Frankel, J. A. and Rose, A. K. (1998). The Endogeneity of the Optimum Currency Area Criteria. The Economic Journal 108(449), 1009–1025.
Ingebritsen, C. (2000). The Nordic States and European Unity. Ithaca: Cornell University Press. Micco, A., Stein, E. and Ordoñez, G. (2003). The Currency Union Effect on Trade: Early Evidence
from EMU. Economic Policy 18(37), 317–356.
Mundell, R. A. (1961). A Theory of Optimum Currency Areas. American Economic Review 51(4), 657–665.
Rose, A. K. (2000). One Money, One Market: the Effect of Common Currencies on Trade. Economic Policy 15(30), 7–46.
Stock, J. H., and Watson, M. M. (2012). Introduction to Econometrics. Harlow: Pearson Education Limited.
Thorhallsson, B. (2001) The Distinctive Domestic Characteristics of Iceland and the Rejection of Membership of the European Union. Journal of European Integration 23(3), 257–280. Totaro, P. (2009). Iceland may join EU after left-wing victory. Retrieved from The Age website:
http://www.theage.com.au/world/iceland-may-join-eu-after-leftwing-victory-20090426-ajbg.html
Appendix
Calculating the exchange rate volatility effect in table 3
In calculating the exchange rate volatility effect on trade, it was assumed that there would be no exchange rate volatility from 2014 onwards. As this is a log-linear estimation, the percentage change in the dependent variable for a change in the independent variable is equal to the 100 × 𝛽% × ∆𝑋 . Accordingly, the exchange rate volatility trade impact was computed by multiplying the coefficient with Icelandic krona’s exchange rate volatility regarding the euro in 2014:
0.006948 × −0.02 × 100% = −1.23%
As this negative trade impact will be eliminated if Iceland had adopted the euro, the impact on trade for zero exchange rate volatility amounts to 1.23 percent.
Assumptions fixed effects regression
Stock and Watson (2012) list the following assumptions necessary for fixed effects regressions: (1) the error term has a conditional mean of zero; (2) the variables are independent and identically distributed across entities; (3) large outliers are unlikely; and (4) there is no perfect multicollinearity (p. 405). Also, it needs to be established whether heteroscedasticity or autocorrelation is present in the data. In the remainder of this section, these properties and assumptions will be elaborated on.
The first assumption that needs to be addressed is the zero conditional mean criterion for the error term. This means that on average, the observed value coincides with the fitted value so that the expected value of the error term is zero for any value of the independent variables (Stock & Watson, 2012, p. 238). Figure A1 shows a scatter plot of the observed value of the log of bilateral trade versus the fitted values as predicted by the time and entity fixed effects model.
Figure A1. Observed versus fitted values plot
10 15 20 25 L n (B ila te ra l tra d e ) 10 15 20 25 Fitted values observed = fitted
The plot is highly suggestive of a linear relationship, implying a nonlinear model is not appropriate. Furthermore, in almost all cases, the fitted value is very close to the observed value as the fitted values are clustered around the 45-degree line. Moreover, the mean standardized residual was 0.00003, and was not significantly different from zero with 𝑡(2735) = 0.0016. In conclusion, the first assumption seems to hold.
The second assumption is that all variables are independently and identically distributed (i.i.d.). The problem is that time series or panel data is not i.i.d. by definition, as observations across time are often correlated with each other (Stock & Watson, 2012, p. 167). To see whether this property holds for the dataset in this study, the Woolridge test for autocorrelation in panel data was used. The null hypothesis for this test is that no first-order autocorrelation is present. The F-statistic was calculated to be 𝐹(1,170) = 83.130 with 𝑃𝑟𝑜𝑏 > 𝐹 = 0.000, so that there is enough statistical evidence to reject the null hypothesis. In other words, the data exhibits autocorrelation. Another property that needs to be tested is heteroscedasticity. To do so, a modified Wald test for group wise heteroscedasticity in a fixed effect model was used. The null hypothesis dictates that the variance is constant across all cases: 𝐻0: 𝜎𝑖2= 𝜎2 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖. The chi-squared test statistic was calculated to be 𝜒2(171) = 3.3𝑒 + 05 and
was significantly different from zero at the 1% level. Therefore, the null hypothesis was rejected, meaning that heteroscedasticity is present in the dataset. To account for these properties of the dataset, clustered standard errors were used in the estimations. This allows for the existence of autocorrelation and heteroscedasticity in a manner consistent with the second assumption listed above (Stock & Watson, 2012, p. 406). In other words, the fact that autocorrelation exists is not detrimental to the second assumption of i.i.d. if clustered standard errors are used, so that the assumption is fulfilled.
The third assumption dictates large outliers are unlikely. In mathematical theory, large outliers are unlikely if the kurtosis for all variables in the regression is nonzero and finite (Stock & Watson, 2012, p. 169). These results are presented in table A1:
Table A1. Kurtosis values for the regression variables
Variable Kurtosis
Bilateral trade 2.487694
Real GDP 2.398252
Real GDP per capita 2.936433
Distance 4.068335
Exchange Rate Volatility 11.95309
Contiguity 8.169213
Language 19.74941
FTA Membership 4.088627
EU Membership 2.143398
EMU Membership 1.032068
Regarding the actual outliers, first the most extreme values for the log of bilateral trade, together with the values of its main determinants, were tabulated in table A2.
Table A2. Extreme values for the dependent variable Obs. Country pair Year Log of Bilateral
Trade Log of Real GDP Log of Real GDP Per capita Log of Distance 1921 Iceland-Luxembourg 2014 10.80134 45.69459 20.3152 8.267426 1922 Iceland-Malta 2000 11.62944 45.80641 20.40807 8.267426 643 Cyprus-Iceland 2000 11.7813 46.81268 20.79151 8.461527 2309 Latvia-Malta 2002 11.95335 45.79536 18.25652 7.815541 2306 Latvia-Malta 1999 12.065214 45.57236 18.04349 7.815541 1290 France-Germany 2007 25.38715 57.22995 21.0262 6.631702 1614 Germany-Netherlands 2011 25.39665 56.09204 21.25499 5.633537 1615 Germany-Netherlands 2012 25.40398 56.08582 21.24301 5.633537 1610 Germany-Netherlands 2007 25.45974 56.09332 21.25418 5.633537 1613 Germany-Netherlands 2010 25.466 56.10427 21.254 5.633537
The table above is not indicative of outliers: it can be observed that extreme values are either associated with similarly high or low values for the other variables. For example, while the high trade between Germany and the Netherlands is associated with relatively high values for Real GDP and Real GDP per capita and a low distance, it is also probable that there exists some country-pair effect that needs to be taken into account. Therefore, it is more illuminating to consider the leverage-versus-residual-squared plot. Figure A2 plots the leverage of a particular observation, which depends on the respective distance from the means of the independent variables, against the normalized residuals squared of that observation. This means that the plot accounts for all possible effects on trade included in the model specification. Observations above and to the right of the red lines have both higher-than-average leverage and normalized residual squared values. In this sample, there are but a few outliers: the only outlier that may affect the results is the one in the top right corner, as it is extreme in both leverage and normalized residual squared values. Nevertheless, the outlier cannot be removed as its extremity is not due to a mistake in the data.
Figure A2. Leverage-versus-residual-squared plot
.0 7 .0 8 .0 9 .1 L e ve ra g e 0 .01 .02 .03
Regarding the fourth assumption, the only variables that exhibited perfect multicollinearity were Distance, Contiguity and Language in the time and entity fixed effects model. This is because those variables are time-invariant and are thus absorbed already in the country-pair fixed effect. Therefore, these three variables were excluded from the time and entity fixed effects specification, so that in the end, no perfect multicollinearity was present in any of the three specifications.
Comparing the three specifications: is a fixed effects model appropriate?
In past literature, a fixed effects model is recommended over a random effects model due to the possibility of endogeneity. To assess which model is the most appropriate in this sample, several tests were undertaken. First, a joint F-test was employed to see if time dummies are jointly significant from zero, thus asserting whether or not time variance had an influence on bilateral trade. The null hypothesis was rejected, as the F-test yielded an F-statistic of 𝐹(15, 2544) = 5.38 which was significant at the 1% level. This provides an indication that the OLS regression suffers from omitted variable bias. Second, a Hausman test was applied. The Hausman test assumes that the random effects model adequately estimates all individual coefficients, and can thus be used to compare the random effects model with the time and entity fixed effects model. Again, the null hypothesis was rejected at the 1% level, with chi-squared test statistic equal to 𝜒2= 124.83. Both results imply that the time and entity fixed effects specification is the best fit with the data.
Robustness tests
To assess the validity of the model employed in this study, two robustness tests were considered. First, a variable exclusion test was performed. Table A3 shows the impact on the EMU membership coefficient of omitting individual variables for all specifications of the model. It can be seen that the EMU coefficient remains statistically significant at the 1% level across all estimations but the OLS specification for the distance, contiguity and language variables.
Table A3. Variable exclusion test for EMU effect
Variable excluded (1) (2) (3)
OLS Time FE Time and Entity FE
Real GDP 0.18*** 0.25*** 0.24***
Real GDP per capita 0.13*** 0.21*** 0.21***
Distance 0.12** 0.21*** 0.24***
Exchange Rate Volatility 0.12*** 0.22*** 0.25***
Contiguity 0.12** 0.21*** 0.24***
Language 0.12** 0.21*** 0.24***
FTA Membership 0.12** 0.21*** 0.24***
EU Membership 0.12*** 0.21*** 0.24***
The second robustness check assesses the effect of excluding a country from the sample on the EMU trade impact. In table A4, the results across all three model specifications are tabulated. Again, the EMU coefficient remains significant at the 1% level for both fixed effects regressions for every exclusion. For the OLS estimation, the coefficient is less robust with varying samples, in terms of both significance and figures. The only country exclusion that has a relatively large impact on the EMU coefficient in all model estimations is Iceland. This may be because it is the only country in the sample that it is not part of the monetary union in the full sample period; the other countries are either initiators or late joiners. Still, in general the fixed effects models seem to be robust to adjustments of both the precise model specification and the countries contained in the sample. The OLS model fares less well in that respect.
Table A4. Country exclusion test for EMU effect
Country excluded (1) (2) (3)
OLS Time FE Time and Entity FE
Austria 0.13*** 0.21*** 0.25*** Belgium 0.12** 0.22*** 0.26*** Cyprus 0.12*** 0.22*** 0.23*** Estonia 0.07 0.17*** 0.21*** Finland 0.15*** 0.22*** 0.26*** France 0.14*** 0.23*** 0.26*** Germany 0.13*** 0.22*** 0.25*** Greece 0.09* 0.20*** 0.24*** Iceland 0.10** 0.14*** 0.18*** Ireland 0.13*** 0.21*** 0.23*** Italy 0.12** 0.22*** 0.25*** Latvia 0.14*** 0.29*** 0.28*** Luxembourg 0.14*** 0.19*** 0.23*** Malta 0.10** 0.20*** 0.22*** The Netherlands 0.11** 0.22*** 0.26*** Portugal 0.11** 0.20*** 0.23*** Slovak Republic 0.10* 0.19*** 0.23*** Slovenia 0.10** 0.21*** 0.24*** Spain 0.11** 0.21*** 0.24*** * p < 0.10, ** p < 0.05, *** p < 0.01
