Common Currency Effect in European Union
Bachelor Thesis University of Amsterdam Faculty of Economics and Business
BSc in Economics and Business Specialization: Economics
Name: Kaspars Burka Student Number: 10085432 Email: kasparsburka@gmail.com
Supervisor: Lukáš Tóth
ABSTRACT
In this paper I estimate the effect of a membership in a currency union on the trade of the participating countries. The most recent data is employed and the results are compared with previous works. The case of European Monetary Union due its possible extension in the very nearest future is of particular interest. Previous research suggests that trade for the participating countries is expected to rise, however, by employing the most recent unilateral trade data of the 27 European Union countries of the last 15 years this paper does not confirm the existence of the effect in neither of four different Gravity model specifications. The four specifications refer to the replication of historical results of Rose (2000) with a few minor changes with the addition three other specifications that take into account country specific effects, time specific effects and both.
TABLE OF CONTENTS
Abstract
ii
Table of contents
iii
1. Introduction
1
2. Literature Review
3
3. Methodology
2.1 Model
7
2.2 Data
10
4. Findings
3.1 Results
12
3.2 Discussion
15
5. Conclusion
18
Bibliography
19
1. INTRODUCTION
More than 500 million Europeans were subject to the one of the greatest monetary experiments of our time, when the common currency euro started circulating in eleven European countries in 1999. Even though during the first three years the currency was only digital, the actual notes and coins started circulating in 2002, thus entirely taking over the national currencies of the first twelve member countries. However, it was not only the design of notes or price tags in grocery stores that changed. By agreeing to permanently fix the exchange rates, the involved countries gave up all individual monetary tools and power and were prepared to bear the cost but also reap the benefits that come with it.
In order to make an economically sound decision about joining a currency union the potential member has to asses benefits against cost. The price to pay is the loss of monetary independence and thus the ability to use specific monetary tools in order to absorb idiosyncratic business cycle shocks. Frankel and Rose (1998) argue that the less integrated the potential member is with the existing currency union before actually committing to the new currency, the more severe this cost is going to be due to asymmetric business cycles. On the other hand the main benefit is associated with decreased exchange rate volatility and the gains in efficiency that come with it. Hedging the exchange rate against possible shocks is pricey, thus causing a deadweight loss that can be avoided.
The lack of academic consensus about the exact size of both the positive and negative effects of a membership in a currency union seemed to have ceased the research in the area until Rose (2000) included a common currency union dummy variable in the Gravity model. His estimates show that a mere membership of a currency union is increasing trade of more than 200% for the involved countries. Furthermore, the effect is much more substantial than reduction of the exchange rate volatility to zero. The study of Rose (2000) was motivated by the upcoming introduction of euro; however, the sample he employed consisted of mainly very small and very poor nations, thus diminishing the external validity as far as application to the European countries is concerned. Nevertheless, the effect seemed so unrealistically high, that the study has been replicated by many; different samples have been used and theoretical mistakes of Rose (2000) have been corrected. However, the efforts seemed to have ceased with Rose’s Meta-Analysis (2005) and the summary by Baldwin (2006), with the general conclusion being that a common currency area does indeed enhance trade and that the initial estimate of 235% is not applicable to the European Monetary Union. The currency union effect works more like a ‘’magic wand’’, meaning that it is relatively persistent in different samples and different settings but the controversy of the causal factors and their importance remains.
Several authors have argued that the case of the European Union and European Monetary Union in particular is special due to the fact that convergence of policies leading to multilateral integration have stemmed over decades (Mengelli, Dorrucci and Agur, 2005). That, however, is counterintuitive to the previously obtained significant estimates of the common currency union effect for the European Union, which connect the particular year a country becomes a member of the European Monetary Union with the increase in trade.
Since the year 2006 when Baldwin wrote the summary of the Gravity model application in international trade literature in order to measure the common currency effect, data on 5 more years have become available with additional countries joining the euro zone. The new data are employed in this paper in order to replicate the earlier findings and to answer the research question: ‘’Does the common currency union effect exist between European Union countries?’’ The main critiques of the initial results obtained by Rose (2000) are taken into account to avoid biases and have more accurate results. Furthermore, to separate the effect of introducing a common currency from gradual integration between European Union countries the chosen data period of 15 years allows only for two reference years, 1997 and 1998, before the euro started circulating. As already explained the meta-analysis of Rose (2005) shows that common currency effect is persistent in different model settings, thus also the prediction of this paper is to observe an effect significantly different from zero.
2. LITERATURE REVIEW
For a long time economists have tried to quantify the effects of a decreased exchange rate volatility on trade leading to the general academic consensus that there should indeed be small mutual gains. The research in this area seemed to have dried up until the mere addition of a common currency union dummy variable to the Gravity model made the work by Rose (2000) one of the most influential thus also quite provocative international economics paper of the previous decade. The work of Rose (2000) was so hallmark that it allows to review the previous research in three sections: firstly, the Rose (2000) itself, secondly the academic rebuttal of the Rose (2000), and thirdly the most recent research that already employs the available data on the European Monetary Union after the introduction of the common currency- euro.
In the simplest form the Gravity model explains the flow of international trade between a pair of countries to be proportional to their national income and inversely proportional to the distance between them. Rose notes that the previous literature on the subject (e.g., Frankel and Rose, 1998) presumes that common currency is equivalent to reducing the exchange rate volatility to zero. However, by using an augmented Gravity model and applying it to a sample of 33903 trade observations (no data on the European Currency union was available yet) over the years 1970-1990 Rose finds that effect of entering a currency union on trade is far more magnificent than reducing the exchange rate volatility to zero. Moreover, the effect of currency union on trade is statistically significant and economically large: countries sharing the same currency trade on average three times as much as countries with different currencies. Even though Rose could only speculate about the reasons of this substantial effect, naming serious government commitment to long term integration and the possible difficulties to hedge the exchange risk among others, the result shock the academic world and thus is called the ‘’Rose effect’’ ever since. The results of this paper challenge the historical results of Rose (2000). The replication of the historical Gravity model specification, when applied to the sample of 27 European Union countries, fails to show a significant and positive common currency union dummy variable coefficient.
Much of the response to the work of Rose has been critical mostly questioning the applicability and the robustness of the results. The sample used by Rose (2000) is widely criticized in order to shrink the external validity of the paper. Rose was investigating the effects of a currency union, however, actual currency unions accounted for less than one percent of the total bilateral country pairs thus making standard regression techniques inappropriate (Nitch, 2001). As far as the country size is concerned: countries from the sample that have participated in currency unions are mostly very small with very weak
economies thus also very poor. European countries on the other hand have bigger economies thus making the results inconclusive for the European Monetary Union (Frankel, 2008). Also Baldwin (2006) notes that the country pairs sharing the common currency are far from average thus questioning the external validity. Furthermore probably the most important concern refers to causality problems. Frankel (2008) explains that it refers to the endogenuity of a country’s choice to join a currency union. Optimum currency area theory suggest that countries should peg if they are small and open and should peg to partners they trade a lot. If that is indeed the case, then the Rose (2000) results become spurious. He extends the discussion noting that similar logic might apply to European countries that have decided to join the Euro. Countries that have pushed the introduction of the Euro, e.g. Germany, the Netherlands and Luxemburg are the ones that have been the most thoroughly integrated long time before. Persson (2001) investigates that the countries in the sample of Rose (2000) are not randomly selected in addition to trade costs being not linear. Matching principle technique is suggested instead and the obtained estimates are more modest.
The result of the critique is the paper by Rose and van Wincoop (2001). In order to solve the endogenuity of a country’s choice to join a currency union and thus the omitted variable bias, the model with the multilateral trade resistance term is introduced. Multilateral resistance refers to the average trade barrier a country faces, thus also accounting for differences in trade a particular country faces, once any of its trading partners change their trade towards a third country. The most obvious example is Australia and New Zealand. Even though both countries are far apart, they trade a lot simply because other developed countries are even further away (Baldwin, 2006). The practical solution is to introduce country-fixed effects dummy variables (Anderson, van Wincoop, 2001). Taking into account the multilateral resistance shrinks the initial common currency union effect to 58% instead of the initial 300% (Rose, van Wincoop, 2001). In order to obtain unbiased results this paper takes into account the critiques of the Rose (2000) thus accounting for both country and time fixed effects in addition to the replication of the initial specification used by Rose (2000).
The discussed approach so far has been focused on cross-section data analysis, thereby answering the question ‘’by how much is trade between members of a currency union larger than between countries that have their own currency and thus do not participate in any currency union’’. For policy purposes, however, the question ‘’what happens to trade of the involved country once a currency union is created or dissolved’’ is more relevant. The problem is that in order to note the variation the required sample is large. Rose and Glick (2002) addressed the issue by using large panel data set covering 217 countries from 1948 to 1997. It is emphasized that during the time period many countries left the currency unions, thus avoiding the critique experienced by Rose (2000), namely, that the sample cover insufficient number
of countries with the parameter of interest. The conclusion, however, is similar to that of Rose (2000), namely, joining or leaving a currency union doubles or halves the bilateral trade respectively.
To further review the existing literature it is necessary to shift the focus to post euro introduction research, as once the early Euro data became available the research has been almost entirely focused on estimating the ‘’Rose effect’’ for the countries that joined the euro as well as spillover effects on other European Union countries. Before the provocative findings of Rose (2000) relating common currency to increased trade, academics threated the Euro project skeptically. Although, Frankel and Rose (1998) argue in favor of business cycle synchronization across the members of a currency union, thereby lowering the cost of loss of national monetary policy, the very low labor mobility in the early stages combined with sticky prices and lack of budgetary transfers to stabilize asymmetric shocks emphasizes the huge cost of losing monetary independency, which has been considered as the primary concern of joining a currency union (Flandreau and Maurel, 2001).
The huge estimates of the ‘’Rose effect’’ in the earlier research challenged the academics to estimate the effect that exists between the European countries in particular. It was clear that the results obtained by Rose (2000) could not be extended to Europe, if only for the reason of completely different samples: none of the countries in Europe was as small and poor in the same time, as most of the countries researched by Rose. The estimated effects did indeed differ, depending on the assumptions and methods different authors employed. The first thorough and well received research was conducted by Micco, Stein and Ordonez (2003). The difference was drawn between two samples and two methods, namely, only the 15 EU nations or 22 developed countries and with or without country-pair specific dummy variables. The obtained results range from 4% to 16%, which is not even close to the results reported by Rose (2000) but still statistically significant. In order to avoid the possible endogeneity bias, country pair specific dummies are introduced. Pair dummies absorb the time invariant characteristics of countries, thus if two countries have historically traded a lot, and only therefore decided to join a union, it is absorbed in the dummy, thus reducing the bias. An important finding also is that there is lack or trade diversion towards non-EU countries, the only observable phenomena is that relative trade towards non-euro countries decreased and the relative trade towards non-EU countries decreased even more, which means that countries that have joined Euro increased their trade overall but more so towards other euro countries. The results of this paper challenge also the modest estimates of Micco, Stein and Ordonez (2003), since even after accounting for country and time fixed effects no positive and statistically significant coefficient of the common currency dummy variable for the European Monetary Union is observed.
By employing the adequate technique it is possible to estimate the country specific effects, Spain and the Netherlands have the largest effect, however, Greece has a statistically significant negative effect on trade (Micco, Stein, Ordonez, 2003). Additionally, removing the former countries of the Deutsche-Mark Bloc makes the estimated currency union effect statistically insignificant, thus clearly showing that the effect of the currency union has been the greatest between the countries that have the most closely been integrated beforehand. Furthermore, the overall effect, even though smaller than what Rose (2000) estimated still seems too large, especially once compared with the effect that the countries experience after joining the European Union. However, by replicating the study and including additional year of data Berger and Nitch (2005) estimate the Rose effect that is higher than the Micco Stein and Ordonez (2003), thus also showing that the results vary within time.
Furthermore, according to estimates the removal of all trade barriers and a creation of single common market have only slightly higher effect than the introduction of common currency in a situation when the exchange rate volatility was already tiny. However, taking into account the gradual trade integration over the last decades and adding a time trend variable makes the coefficient of common currency union dummy variable insignificant (Berger, Nitch, 2003).
Moreover, it is possible to estimate the currency union effect by splitting trade in exports and imports instead of taking the sum of bilateral exports. The results show an increase in exports within countries, but also the increase in exports to zone countries, however, no increase of exports of euro-zone countries to non-euro countries is observed. In sum, despite the increase in bilateral trade the results indicate that the euro-zone countries become better importers than before (Flam, Nordstrom, 2003). The results of this paper are also based on unilateral trade data.
Lastly, it should be examined what industries and what types of firms benefited the most from the common currency. Common market lowers trade costs that previously have been associated with the exchange rate volatility, thus Baldwin and de Nino (2006) suggest that trade is increased due to what is called extensive margin of trade. Due to lower costs new firms that previously were above the cost threshold that would allow them to export, now with the resulting drop in costs are able to start exporting. The results, however, are only supportive but not conclusive. Furthermore the literature suggests that trade is increased in sectors that involve differentiated, high value-added products. More precisely products such as beverages, tobacco, chemical products, e.g. pharmaceuticals and manufacturing products experienced the biggest increase. It can be argued that reason behind is the vertical product specialization across borders due to lower trade costs and high investments in marketing and distribution.
3. METHODOLOGY
3.1 Model
Dating back to 1960s the Gravity model has been the pillar upon which to base the international trade research. Its success lies in the easy empirical applicability and consistency throughout (Rose, 2000). In the simplest form the Gravity model predicts that trade is directly proportional to the economic mass of the countries involved and inversely proportional to the distance between them (one can draw similarities to the way Newton was describing the force of gravity). It can be written as follows:
logXij = c+b1logGDPi+b2logGDPj+b3logD+eij
Where X refers to the exports of the country i to country j, GDPs refers to the respective gross domestic products of both countries, D refers to the trade costs estimated by distance between both countries and e is the error term. In sum, the bigger and the more closely located both trading partners are, the bigger the trade.
The revolution in the international trade literature occurred when Rose (2000) added a common currency union dummy variable to an augmented gravity model. Rose (2000) did augment the simple Gravity model with the following variables:
• GDP per capita
• Common language dummy variable
• Common border between both countries dummy variable
• Dummy variable representing whether both countries are members of the European Union
• Dummy variable representing whether both countries are members of the European Monetary Union
• Exchange rate volatility
Despite the further critic of the Rose (2000) findings, the analysis of this paper will begin by replicating the model with one difference: exclusion of the exchange rate volatility variable. Author believes that the cumbersome calculations of the exchange rate volatility combined with lack of neat data do not provide the necessary insights thus are not worth the effort.
The reason to believe so is that previous research indicates that even though skipping the exchange rate volatility biases the common currency union dummy, the effect is minor. Rose (2000) finds that the magnitude of the effect of entering a currency union on trade is far bigger than reducing the exchange rate
volatility from one standard deviation to zero. In fact the estimated coefficient on the common currency dummy is 1.21, when the exchange rate volatility coefficient is only -0.017. Berger and Nitch (2005) confirm this by investigating what effects of the inclusion of the exchange rate volatility has on the common currency dummy and arrive at the conclusion that even though the exchange rate volatility takes on the expected negative sign and is significant, by including it in the model, the value of the common currency dummy fall only very slightly. Authors conclude that exchange rate volatility does not indicate long-term trend in the European Monetary Union trade intensity.
The replication of the initial Rose (2000) for the 27 European Union countries will yield only intuitive results. The purpose is not to use the estimates for policy purposes, merely observe whether the initial paradigm-shifting work by Rose could be applied to the European Union and the euro.
The gold medal mistake of the initial Rose equation, as called out by Baldwin (2006), refers to the problem of endgeneity and omitted variable bias thus making the initial results and also the intuitive reproduction of this paper biased. The most practical way to address it within the framework of the Gravity model is to introduce country fixed effects.
By assuming ‘’love of variety’’ preferences for the consumers, and single product with increasing returns of scale for each producer Anderson and van Wincoop (2001) came up with what essentially was a demand function. After aggregating across industries and applying simple derivations authors derived a model that has been called ‘’Gravity with Gravitas’’ model ever since and certainly received less critic by other academics then the initial Rose (2000).
The most notable feature is the inclusion of the outward multilateral resistance as well as the inward multilateral resistance terms in the Gravity equation. Outward multilateral resistance refers to the fact that exports from one country to another depend on trade costs across all possible suppliers. Inward multilateral resistance refers to the fact that imports into a country depend on trade costs across all possible suppliers. Together both terms help to account for the fact that a change in cost of one bilateral trade route, due to relative price effects, affects all other trade routes as well. Thus together both terms correct most of the omitted variable bias and engoneneity flaws in Rose (2000).
There is no accurate data referring to multilateral resistance thus permitting inclusion of it as a data point in the Gravity model. The most common strategy (also applied by e.g. Micco, Stein, Ordonez (2003)) refers to the fixed effects estimation. Accounting for fixed effects means creating a dummy variable that is equal to unity each time a particular exporter is present in the data set and assuming that the key assumptions are met, the ordinary least squares estimate stays constant, unbiased and efficient estimator.
According to Shepherd (2012): ‘’country specific dummy variables can be understood as mean for accounting for all sources of unobserved heterogeneity that are consistent for a given exporter across all importers, and constant for a given importer across all exporters’’. Furthermore, Baldwin (2006) emphasizes that inclusion of country fixed effects only partially solves the endonegeity problem and arising biases. It is noted that after the inclusion of the country fixed effects, there is still a time varying residual in the error term left, which means that the results are still biased. Furthermore, the size of the bias will depend upon the volatility of the trade cost over time.
Introduction of the fixed effects to the model comes with several restrictions, namely, the model doesn’t allow for separate inclusion of variables that vary in the same dimension as the fixed effect due to perfect colinearity. In other words, if the fixed effect for Germany is included in the model, the time invariant variables specific to Germany such as distance to a specific country or common language cannot be included in the model. The aim of the model is to predict the effect of a common currency union on trade and due to the fact that common currency dummy variable varies bilaterally and is not country specific, it is possible to estimate the effect.
In sum, the methodology part has provided the reader with reasoning behind different choices regarding methods that the paper will employ to estimate the common currency union effect within the 27 European Union countries. First of all, a regression similar to the one estimated by Rose (2000) is conducted for merely intuitive results. Furthermore, country specific fixed effects are dealt with. The third equation substitutes the country specific fixed effects with time specific fixed effects. Lastly, both country specific and time specific fixed effects are employed to yield the cleanest insights.
3.2 Data
In order to estimate the model it is necessary to gather the specific data and create a data base of panel data. The model will focus on the 27 European Union countries: Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxemburg, Malta, the Netherlands, Poland, Portugal, Romania, Slovakia, Spain, Sweden, and United Kingdom. The time period when the countries joined the European Union varies, thus also allowing for the estimate of the effect of the free trade area and a comparison between the effect of a free trade area and a monetary union. The chosen time period is 15 years, more precisely, 1997-2011, thereby taking advantage of the most recent data as well accounting for few years before the common currency- euro, started circulating but also separating gradual multilateral integration of the European Union countries from the pure common currency effect. Moreover, the panel database consists of 10530 observations, that is in terms of size and number of observations sufficiently comprehensive.
Despite the fact that Rose and also other authors do employ the average of bilateral trade date (exports+ imports) in the model as the dependent variable, this paper will use unilateral trade, e.g. Germany’s exports to Austria and Austria’s exports to Germany will be treated as two separate equations to be estimated. This method does come with one benefit and one flaw. Already the basic theory explains that the Gravity model relationship holds for bilateral exports, and even though there is nothing wrong with using the average of bilateral trade instead, the mistake is committed when it is presumed that log of the average is the same as average of the logs. As long as the involved countries experience a balanced trade, the method works fine, however, the more unbalanced trade of the involved countries, the more biased the results will be. The same logic applies to differences in the common currency union effects that both countries experience. In order to avoid it, exports of every country will be treated as a separate variable (Baldwin, 2006). The flaw is that the equation of each country pair will have the same independent variables, only the exports vary, thus making the estimates less precise than the model employed by Rose. However, as the existing literature shows, different European countries have experienced different common currency union effects, also the trade is not balanced, thus the benefits do outweigh the imprecision.
Furthermore, it is important to distinguish between real and nominal data. Even though some authors have employed real data, this paper will focus on nominal data. According to Shepherd (2012) the inclusion of the multilateral resistance terms that also serve as special price indices does effectively deflate the exports. Thus this paper employs nominal data in US dollars. There are however, 10 missing data points in the sample. These are estimated using the method of averages, meaning that the missing value is
estimated using the average of two years before and two years after the particular period. Furthermore, by employing a different deflator, e.g. GDP deflator or CPI, would not fully account the multilateral resistance terms, thereby biasing the results. Shepherd (2012) also emphasizes that the aggregate GDP, no the GDP per capita should be used.
Data on nominal GDP and bilateral exports is taken from the United Nations Conference on Trade and Development (UNCAD STAT) database; distance between countries is taken from www.distancefromto.net and is calculated as line segment between the geographical centers of the involved countries, lastly, data on common language is taken from www.eucountrylist.com .
4. Findings
5.1 Results
Before carrying out the regressions to test the impact of the common currency union dummy variable on trade, it is useful to graph the most basic and intuitive relationships, namely the relationship between the trade and the GDP, and the relationship between the trade and distance between both countries. The purpose is to familiarize the reader with the historically empirically successful relationships that the model entails and provide with merely intuitive grasp of the concepts of basic international trade.
The first scatterplot depicts the relationship between trade (US dollars) and prospective nominal GDPs (US dollars). The bilateral trade, as already explained in the methodology part, in the specification of this model is proxied by bilateral exports of both countries as two different equations.
As can be seen by observing the trend line, the relationship is clearly positive. Furthermore, also the other intuitive relationship, namely the negative relationship between trade (US dollars) and distance (measured in kilometers and refers to the line segment between the geographical centers of the countries) holds, as depicted by the red trend line with negative slope in the graph below.
0 40,000,000 80,000,000 120,000,000 160,000,000 200,000,000
0.0E+00 4.0E+24 8.0E+24 1.2E+25
GDPPRODUCT T R A D E 0 40,000,000 80,000,000 120,000,000 160,000,000 200,000,000 0 1,000 2,000 3,000 4,000 Dist. T R A D E
Furthermore, the results of the regression in the pre-specified order are presented:
• Column 1 shows results of the replication of Rose (2000) and does not account for any fixed effect
• Column 2 shows results accounting for country (cross-sectional) specific fixed effects • Column 3 shows results accounting for time specific effects
• Column 4 shows results accounting for both country and time specific effects
EMU -0.08*** (0.02) -0.07*** (0.02) -0.11*** (0.02) 0.01 (0.02) EU -0.22*** (0.02) 0.15*** (0.02) 0.07*** (0.03) 0.12*** (0.02) Nominal GDP1 1.02*** (0.01) 0.75*** (0.02) 1.02*** (0.01) 1.22*** (0.03) Nominal GDP2 0.85*** (0.01) 0.58*** (0.02) 0.844*** (0.01) 1.06*** (0.03) Distance -1.45*** (0.02) - -1.45*** (0.02) - Common Language 0.24*** (0.06) - 0.18*** (0.06) - Border 0.14*** (0.04) - 0.13*** (0.04) - Country fixed effects - + - +
Time fixed effects - - + +
R2 0.88 0.98 0.88 0.98
The results obtained and represented in column 1 are an intended reproduction of the original Rose (2000). As can be observed the variables explain the trade very well as R-squared is 0.88. Furthermore, all coefficients but EU and EMU have the pre-supposed signs and are statistically significant. According to recent debates the GDPs coefficients should be around 1 and that is indeed the case. Lastly, the results do not confirm the hypothesis that the common currency union effect, also known as the Rose effect, occurs in the European Union.
Second column presents the results, once country specific fixed effects are assumed. This means that the time-invariant variables cannot be included in the regression but gives the benefit of isolating country specific variation in trade from the common currency effect. R-squared value is 0.98 (adjusted R-squared 0.98) which indicates that almost all variation in trade can be explained by the independent variables. The respective coefficients of GDPs are plausible as they are positive, and higher than 0.5, also the coefficient of EU takes a positive value, that would have been expected. However, the hypothesis of the common currency effect in European Monetary Union is rejected as the coefficient of EMU is below 0 and statistically significant.
Third tables shows the results, once time specific fixed effects are accounted for. It means that the time specific effects in trade, like the economic crisis are isolated from the effect of introducing euro. R-squared is still relatively high: 0.88 percent of the variation in trade can be explained by independent variables, once time specific effects are accounted for. Coefficients of GDPs are in line with theory, namely positive and quite close to 1, also the EU has a positive coefficient. Again the coefficient of the European Monetary union effect is negative, thus denying the common currency effect in Europe.
Fourth table employs both country and time specific fixed effects. It is expected that it increases the explanatory power of the regression, thus the relatively high value of R-squared (0.98) is no surprise. Furthermore, coefficients of the respective GDPs are positive and close to 1, EU coefficient is statistically significant and positive. Finally, the obtained common currency union coefficient is positive, however, not statistically significant at 10 percent significance level.
The relatively very high R-squared values, once the fixed effects are introduced, might seem unrealistic, however, the previous research suggests that high values of R-squared is a common occurrence in Gravity model applications. For example by also employing unilateral trade data and accounting for fixed effects Flam and Nordstrom (2003) indicate a R-squared values of 0.99.
5.2 Discussion
It is striking that the results are not only out of line with most of the earlier findings, but also that the common currency dummy variable does not take a positive and statistically significant value in any of the specifications. Furthermore, even the replication of the most basic form of Gravity equation, to match that used by Rose (2000), does not yield the expected results, namely that the coefficient of common currency union dummy variable would take a significant positive value. Also the more modest and more recent findings of Micco, Stein and Ordonez (2003) or Flam and Nordstrom (2003) are not met, once the country specific fixed effects or time specific fixed effects or both are employed. In order to find the proper reasons for these differences, the discussion will be split into three parts. First of all, the sample and the European Union as such will be discussed. Secondly, the model specification will be discussed in order to look for possible biases and lastly recommendations for a further research will be given.
With a combined population of over 500 million and a magnificent economic power the European Union was subject to probably the biggest monetary experiment to date, once the common currency euro began circulating. Even though the common currency, albeit not in a physical form at first, was introduced in 1999, the shift was not overnight, but merely the final stepping stone of integration efforts of the last century. On the other hand Rose (2000) findings were based on a sample of countries that were mostly very small and also very poor: nothing like the sample of European Union countries.
The results of this paper are based on sample of the 27 European Union countries over the time period of 15 years: 1997-2011. The time period was chosen to account for both: the most recent data available and also observe the official introduction of euro in 1999. Few additional years added were based on findings by Micco, Stein and Ordonez (2003) that in fact 1998 was already the pivotal year and 1997 would serve as a reference. However, different authors suggest that the relevant time period for the case of European Union might extend much longer. Mongelli, Dorrucci and Agur (2005) investigate the link between economic integration and the overall institutional process of regional integration for the case of European Union. The results are not striking, merely, confirming that the shift to monetary integration did not occur overnight. By proving a positive relationship between the institutional integration and monetary integration, thus also trade, authors find that the relevant time horizon for the case of the European Union extends to over 50 years, starting perhaps with Treaty of Rome back in 1957. It took time to achieve a common market and even more time to achieve a common currency but both of these stepping stones were a result of continuous coordination and harmonization of individual policies regarding tariffs, regulations and economic policies, thus stretching the Rose effect over a long period of time, mostly
beyond the sample of this paper. The same would hold for the new members of the Union, namely, the effort to harmonize their policies, fix exchange rates to the common currency began long before the official admittance date, thus also stretching out of the sample of this paper. Additionally, Berger and Nitch (2005) show that by taking into account the gradual development of the trade relationships of the European Union countries and adding a time trend variable to the classical Gravity model estimation, the coefficient of common currency variable becomes insignificant. In other words, this means, that once the gradual integration is taken into account, the “magical wand” also known as the Rose effect loses its magic. Noteworthy, that the papers that serve as the reference for the expected values of the coefficient of the common currency union dummy variable in this paper, e.g. Flam and Nordstrom (2006) and Micco, Stein and Ordonez (2003) do employ longer time horizon before the official year the common currency began circulating, that serve as the reference and increases the value of the coefficient, if the gradual trade integration is not taken into account, which is indeed the case. Also as Micco, Stein and Ordonez (2005) show, that by including additional developed countries in the sample in addition to the European Union countries, which is not the case in this paper, the effect of common currency union is elevated.
Furthermore, it is necessary to examine the specification of the model and the variables that were used in order to discuss the implication it might have had on the results. In essence, there are several major differences between the employed model and the one used by Rose (2000), each of whom might serve as the reason for not observing the positive common currency union dummy coefficient.
Firstly, Rose (2000) employed the bilateral trade as the dependent variable, this paper, however, takes unilateral trade as the dependent variable, thus two separate equations are estimated, that have the same independent variables but different dependent variables (assuming that exports of the countries in question differ). As already explained in the methodology part, this method makes the results less accurate, however, accounts for imbalanced trade between country pairs. Apart from the added inaccuracy, the problem could be that the sample in this paper does not include countries outside the European Union for reference purposes, thus in case the common currency spillover effect is substantial not only between the members of the European Union, but also increases trade towards or from non-European Union countries, the sample fails to capture the effect, thus underestimates the actual effect. The essence is captured by Flam and Nordstrom (2006) that despite finding a positive and significant common currency effect coefficient, note that the common currency makes the member countries better importers not exporters. It is grounded on findings that exports of Eurozone countries towards non-Eurozone countries did not increase, but the exports of non-non-Eurozone countries towards the non-Eurozone did increase, once the common currency was introduced. In addition, authors also note that the common currency effect seems larger, the more reference countries are taken into the sample. In sum, the lack of
reference non-European Union countries would potentially eliminate the effect, as is the case in this paper, if trade relationships after the introduction arise as depicted by Flam and Nordstrom (2006).
Secondly, this paper does not include a variable to account for exchange rate fluctuations. Even though the reasoning is explained in the methodology part, some authors argue that the bias might be substantial, especially once the unilateral trade not bilateral trade is taken into account (Baldwin, 2006). In case bilateral trade is taken as the dependent variable, then the exchange rate fluctuations might balance out, as once the exports of one country become cheaper and increase, the exports of the other country become more expensive and shrink. In case, the fluctuations of the euro, especially after its initial launch were big enough to substantially change the trade patterns, albeit temporarily, the employed model in this paper, fails to capture the effect, thus biasing the results.
The models in this paper accounts for fixed country and time effects, however, the positive statistically significant, coefficient of common currency union dummy variable is not observed. These models fit data extremely well, as the R-squared values are high but the predicted effect is missing. The purpose to introduce fixed effects is to avoid the omitted variable bias. In practice there are many more variables than the model can encompass that influence trade but are also correlated with the countries choice to participate in a common currency union, thus causing bias, e.g. multilateral resistance was explained in the methodology part. In order to explain these phenomena, the same logic as at the beginning of this discussion applies. Namely, the results are not conclusive that the common currency effect does not exist, merely indicate that in order to observe the possible effect, the sample should entail longer period of time as well as include more developed countries for reference.
Lastly, in order to add to the existing literature, further research should set the common currency union effect of each sample in as much historical context as possible. While it is useful for intuition to provide results that seem to be caused by a ‘’magic wand’’, historical perspective might solve the mechanism of this wand. Also, there are more countries about to join the euro zone in the upcoming years, e.g. Latvia in 2014, Lithuania in 2015, thus studies examining the trade effects of countries that share similar properties and have joined the Euro in last years might yield insights for policy purposes of the potential new members. Additionally, it would be useful to examine, whether the gains in trade occur only after the day that the euro begins circulating in those economies or on the other hand, the lengthy preparation process and gradual integration already has increased the trade.
5. CONCLUSION
In order to replicate the earlier findings of Rose (2000) a panel database was constructed to account for the most recent data available and observe, whether the common currency effect varies over time. Four different model specifications were used to estimate the common currency effect on trade; however, none of the models captures a positive and significant coefficient of the currency union dummy variable thus providing a negative answer of the central research question of this paper. According to the results a significant common currency effect does not exist in the European Union.
One reason for not observing the expected results might be the model specification. Even though the particular specification was grounded in the work of several authors, not including the exchange rate volatility variable and accounting for unilateral trade instead of bilateral could bias the results and make them less accurate.
The results, however, should not be misinterpreted to state that there has been no change in trade patterns once a common currency started circulating between members of the European Monetary Union. The discussion suggests that in order to observe the effect more reference countries and also longer time span should be taken into account, since the increase in trade might be a result of gradual integration and harmonization of policies between the members of the European countries instead of overnight magic.
BIBLIOGRAPHY
Anderson, J. E., & van Wincoop, E. (2003). Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review, 93, 170-192.
Berger, H., & V. Nitch (2005). Zooming Out: The Trade effect of the Euro in Historical Perspective. CESifo Working Paper Series, No. 143.
Baldwin, R. (2006). The Euro's Trade Effects. European Central Bank Working Paper Series, No. 594. Baldwin, E., & Di Nino, V. (2006). Euros and Zeros: The Common Currency Effect on Trade in New
Goods, NBER Working Paper No. 1267.
De Nardis, S., & Vicarelli, C. (2003). Currency Unions and Trade: The Special Case of EMU. Review of World Economics, 139, 626-649.
Frankel, J. (2008). The Estimated Effects of the Euro on Trade: Why Are They Below Historical Effects of Monetary Unions Among Smaller Countries? NBER Working Paper No. 14542
Frankel, J., & Rose, A. (1998). The Endogenity of Optimum Currency Area Criteria. The Economic Journal, 108, 1009-1025.
Flam, H., & Nordstrom H. (2003). Trade Volume Effects of the Euro: Aggregate and Sector Estimates. Stockholm University Seminar Paper
Flandreau M., & Maurel M. (2001). Monetary Union, Trade, Integration, and Business Cycles in 19th Century Europe: Just Do It. Centre for Economic Policy Research Discussion Paper No. 3087 Frankel, J., & Rose, A. (2002). An Estimate of the Effect of Common Currencies On Trade and Income.
The Quarterly Journal of Economics, 117, 437-466.
Glick, R., & Rose, A. K. (2002). Does a currency union affect trade? The time-series evidence. European Economic Review, 46, 1125-1151.
Mongelli, F. P. & Dorrucci E.. & Agur I. (2005). What Does European Institutional Integration Tell Us About Trade Integration. European Central Bank Occasional Paper Series No. 40
Micco, A., Stein, E., & Ordonez, G. (2003). The currency union effect on trade: early evidence. Economic Policy, 18, 315-356.
Nitch,V. (2001). Honey, I Just Shrank The Currency Union Effect. The World Economy, 25, 457-474 Persson, T. (2001). Currency Unions and Trade: How Large Is the Treatment Effect? Economic Policy,
16, 433-448.
Rose, A. K. (2000). One Money, One Market: The Effect of Common Currencies on Trade. Economic Policy, 15, 7-45.
Rose, A. K. (2005). A Meta - Analysis of the Effect of Common Currencies on International Trade. Journal of Economic Surveys, 19, 347-365.
Rose, A. K., & Van Wincoop, E. (2001). National Money as a Barrier to International Trade: The Real Case of Currency Union. The American Economic Review, 91, 386-390.
