Asymmetric Shocks in the Economic and
Monetary Union:
An Empirical Analysis of Business Cycles
MSc Thesis
Pieter Speksnijder 6132936
University of Amsterdam
Department of International Economics Supervisor: dr. D.J.M. Veestraeten Second reader: prof. dr. F.J.G.M. Klaassen
Statement of originality
This document is written by Pieter Speksnijder 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.
Asymmetric Shocks in the Economic and Monetary Union:
An Empirical Analysis of Business Cycles
Pieter Speksnijder
Abstract
This thesis examines the vulnerability of the Economic and Monetary Union (EMU) to asymmetric shocks, by estimating the effect of the monetary integration process on the synchronization of business cycles in the EMU.
The results suggest that the business cycles of the euro area countries became more equal in sign, compared to the business cycles of the EMU countries that did not introduce the euro. However the business cycles of the euro area countries did not become more equal in terms of amplitude, compared to the non-euro countries in the EMU.
Furthermore it is shown that the Pearson correlation coefficient, which is used widely in the extant literature for this type of analysis, does not always properly reflect all relevant information about business cycles.
Contents
1 Introduction 5
2 Economic theory on monetary integration and business cycles 8
2.1 Specialization . . . 9
2.2 Endogeneity . . . 10
3 Extant empirical research on monetary integration and business cycles 12 3.1 Pre-EMU . . . 12
3.2 Pre-crisis . . . 13
3.3 Post-crisis . . . 14
3.4 Mixed results . . . 15
4 Research methodology 16 4.1 Calculating the output gap . . . 16
4.2 Synchronity and similarity . . . 17
4.3 Difference-in-difference model . . . 20
4.4 Data . . . 25
5 Results & analysis 26 5.1 Synchronity . . . 26
5.2 Similarity . . . 28
5.3 Analysis . . . 29
6 Conclusion 34 A Appendix 36 A.1 Cristiano-Fitzgerald filter . . . 36
A.2 Variable description . . . 37
A.3 Output gap per country . . . 38
A.4 Synchronity estimation results for different sample periods. . . 39
List of Figures
2.1 Geographical concentration. . . 10
4.1 Imperfect correlation despite equal sign of output gaps. . . 18
4.2 Perfect correlation despite differences in amplitude. . . 18
4.3 Mean similarity for treatment and control group. . . 22
4.4 Mean synchronity for treatment and control group. . . 23
5.1 Mean similarity and synchronity since 1999. . . 30
A.1 Output gap since 1999 . . . 38
List of Tables
4.1 Correlation in average similarity and synchronity. . . 235.1 OLS regression of the effect of the euro on the pairwise synchronity . . . 27
5.2 OLS regression of the effect of the euro on the pairwise similarity . . . 28
5.3 OLS regression of the effect of the euro on the output gap correlation . . . 32
A.1 Variable description. . . 37 A.2 OLS regression of the effect of the euro on the pairwise synchronity, 1969-2013 39 A.3 OLS regression of the effect of the euro on the pairwise synchronity, 1979-2013 40
Chapter 1
Introduction
Before the creation of the Economic and Monetary Union of the European Union (EMU) several economists questioned Europe’s potential to form a stable monetary union.1 In the years after its creation however, the economies of the EMU were growing steadily, inflation was low and interest rates were converging. Many economists then agreed that the creation of the EMU had been a succes, although acknowledging that the EMU did possess some flaws (Beetsma & Giuliodori, 2010). But then the sovereign debt crisis hit Europe in 2009 and painfully exposed the vulnerabilities of EMU, which rekindled discussions on its stability and questioned this perceived success.
An often neglected argument in these discussions concerns the diversification criterion in the optimum currency area (OCA) theory. This criterion states that all countries forming a monetary union should be highly diversified in production and consumption. When countries form a monetary union they lose the nominal exchange rate as a sovereign policy instrument because a single currency for all member countries is introduced. Losing the exchange rate instrument is however less costly for diversified countries because these countries are less vulnerable to asymmetric shocks and therefore have limited need of the nominal exchange rate as an adjustment mechanism (Kenen, 1969).
An asymmetric shock is an economic shock that affects different countries differently, i.e. asymmetrically. Asymmetric shocks can lead to imbalances in a monetary union. For instance, an increase in natural gas prices might positively affect the economy of the Netherlands being an exporter of natural gas, but negatively affect the Belgian economy which is a natural gas importer. If, for example, the Belgian and Dutch economies depend heavily on the import and export of natural gas, respectively, this gas price shock could lead to large imbalances between these countries. A depreciation of the Belgian franc vis-`a-vis the Dutch guilder then would, ceteris paribus, serve as an adjustment mechanism for these asymmetries by stimulating Belgian and dampen Dutch exports. In a monetary union however this adjustment is no longer possible as the countries share a single currency. Besides the loss of the nominal exchange rate, countries also lose the ability to indi-vidually set monetary policy when forming a monetary union. In the gas price example it would be optimal for monetary policy makers to ease policy in Belgium and tighten policy in the Netherlands, but in a monetary union this is impossible as policy can only be set
1
in one direction, for the union as a whole.
Diversification thus reduces the vulnerability to asymmetric shocks, which in turn limits the need of retaining sovereign policy instruments. It is therefore only optimal for highly diversified countries to form a monetary union. However, forming a monetary union, i.e. monetary integration, is likely to affect the extent of diversification in the countries forming the union, and through that affect the vulnerability to asymmetric shocks.
Krugman (1993) describes the case of Massachusetts, a region in the United States (US) that specialized in the production of high-tech goods. Massachusetts went through a significant economic slowdown in the 90s as demand shifted away from the high-tech products the region had specialized in. Krugman argues that the high level of integration in the US economy causes this type of regional specialization. Economic integration lowers transaction costs in the form of transportation expenses, tariffs or regulation disparities, which makes it more likely that any degree of economies of scale, either internal or external, lead to concentration of production (Krugman, 1993). He predicts that similar asymmetric shocks could occur in Europe as the creation of EMU would increase the level of integration to US levels and cause similar levels of regional specialization.
Frankel and Rose (1998) contradict Krugman’s hypothesis of increased vulnerability to asymmetric shocks after forming a monetary union. They use the Pearson correlation coefficient as a measure of the synchronization between the business cycles, and show that countries with closer trade links tend to have more correlated business cycles. This sug-gest that it is less likely that those countries will be hit by asymmetric shocks. Forming a monetary union is expected to increase trade among member countries, which will then lead to more correlated business cycles. This suggests that countries become less vulner-able to asymmetric shocks after they have entered a monetary union as this increases the correlation of their business cycles.
The synchronization of business cycles and the extent of diversification in an economy however are not two sides of the same coin. A higher correlated business cycle would suggest a lower probability to asymmetric shocks but it does not provide direct evidence of the economies being diversified. Similarly two highly diversified economies are less vulnerable to asymmetric shocks and are expected to, but need not necessarily, have a synchronized business cycle.2 Take for instance two countries, one that produces a wide variety of labour-intense cheap products and another one which produces a wide range of high-tech expensive products. Both these countries might be classified as having diversified economies but could have very different business cycles. It might be that the demand for high-tech expensive products is more volatile than for the cheap products due to larger price and income elasticities. This would imply that the business cycle of the high-tech country is also more volatile than the business cycle of the other country in this example. Nevertheless this thesis will focus on the synchronization of business cycles within the EMU, because it is a catch-all measure. If the synchronization of business cycles across
2
Another important note in this discussion is the difference between specialization and agglomeration. Specialization refers to an economy that produces only a small range of products or services. Agglomeration concerns geography, that is to which extent economic activity clusters in a certain area within an economy. It is however beyond the scope of this thesis to further analyse this distinction.
the EMU countries is getting lower for any given reason, specialization, agglomeration or something else, it implies that the EMU is getting more vulnerable to asymmetric shocks. Less synchronized business cycles can therefore be seen as a signal for underlying problems. On the other hand more synchronized business cycles suggest a lower vulnerability to asymmetric shocks, therefore the synchronization of business cycles is a useful tool to assess the vulnerability of the EMU for asymmetric shocks.
Economic theory however does not provide an unambiguous prediction of how the synchronization of business cycles between the euro area countries is likely to change as a result of the monetary integration. On one side there is Krugman who predicts a lower synchronization of business cycles, through increased specialization. On the other side there are Frankel and Rose who predict increased business cycles correlations through increases in trade. The aim of this thesis is therefore to empirically test the effect of monetary integration on the synchronization of business cycles, by trying to answer to following research question: Is the European monetary integration process affecting the synchronization of business cycles of the Eurozone countries?
Empirical research regarding this question has provided mixed results. This could be due to the fact that most empirical studies use the Pearson correlation coefficient as a measure for the synchronization of business cycles. Mink, Jacobs and de Haan (2007) argue that the Pearson correlation coefficient does not always properly reflect differences in the sign and amplitude of two business cycles. They propose to split the Pearson correlation coefficient into two new measures: a synchronity and a similarity measure.3 This thesis
will follow their methodology and use these new measures to assess the synchronization of business cycles.
A difference-in-difference technique is then used to measure the effect of the euro on the synchronity and similarity measures for the 11 countries that adopted the euro in 1999 against a control group consisting of Denmark, Norway, the United Kingdom, Sweden and Switzerland.4, 5
The remainder of this thesis is organized as follows. Chapter 2 will further elaborate on the two theoretical views on the effect of monetary integration on the synchronization of business cycles. Subsequently chapter 3 will provide an overview of related empirical research. Then the synchronity and similarity measures, as well as the difference-in-difference model will be discussed in chapter 4. The results of the difference-in-difference-in-difference-in-difference methodology will be analyzed in chapter 5, followed by a conclusion in chapter 6.
3A more detailed analysis of this methodology follows in chapter 4. 4
Greece is excluded from the treatment group because the euro was introduced in Greece in 2001, which gives difficulties with the difference-in-difference analysis.
5
These countries still possess their own currency and have broadly comparable economies to those of the Eurozone members.
Chapter 2
Economic theory on monetary
integration and business cycles
Since Mundell (1961) laid the foundations of the theory on optimum currency areas (OCA) a vast amount of literature has been written on the subject. In the field of OCA theory economists try to find criteria, which if fulfilled, would maximize the benefits and minimize the losses of forming a currency area or monetary union. One of these criteria is the diversification criterion that was already mentioned in the introduction. It states that it is optimal to form a monetary union for countries that are highly diversified in production and consumption (Kenen, 1969). It is reasoned that the loss of the exchange rate and independent monetary policy as policy tools is minimal for diversified countries because these countries are much less vulnerable to asymmetric shocks. Hence, they are in less need of monetary or exchange rate policy to dampen the effects of asymmetric shocks. The benefits of forming a monetary union e.g., less exchange rate uncertainty, increased price transparency and deeper and wider financial markets, then outweigh the costs associated with the loss of independent policy tools.
If countries fulfill the OCA criteria it is optimal for these countries to form a mone-tary union.1 It is however not unlikely, as mentioned in the introduction, that the forces unleashed by the formation of a monetary union might affect a country’s fulfillment of OCA criteria (Mongelli, 2002). Different theoretical views exist on the effects of forming a monetary union. On the one hand there is Krugman (1993) who predicts that monetary integration leads to increases in specialization and thus lower fulfillment of the diversifi-cation criterion. On the other hand there are Frankel and Rose (1998) who predict that monetary integration increases trade among the member countries, which will lead to more correlated business cycles. This implies that countries would become less vulnerable to asymmetric shocks after having made efforts of integration, which would mean more fulfillment of the diversification criterion.
This chapter will describe the two theoretical views on the effect of monetary
integra-1
In reality countries usually fulfill some of the extensive list of OCA criteria but others not. Making it very complicated to assess a country’s overall suitability for a monetary union because one then has to determine how much weight each criterion should receive. This is however an entirely different and complicated debate which is beyond the scope of this thesis.
tion in further detail. The arguments provided by Krugman (1993) and Frankel and Rose (1998) are referred to as the specialization and endogeneity hypothesis, respectively.
2.1
Specialization
The specialization hypothesis is based on international trade theory and especially on eco-nomic geography. This is a theoretical subfield studied in both ecoeco-nomics and geography and it concerns the location, distribution and spatial organization of economic activities across the world (Clark, Feldman, & Gertler, 2002). In economics the geographical dis-persion of where firms locate their production is often modelled as the outcome of two opposing forces, transaction costs and economies of scale. Consider the production of a single good, if transaction costs for this good are high, in the form of transportation ex-penses, tariffs or disparities in regulation, it is optimal that production is located close to the market in order to keep transaction costs low. If on the other hand the production of this good is associated with large economies of scale, either internal or external, it might be optimal to concentrate production in a single location in order to reap as much of these economies of scale as possible. Where production eventually will be located then depends on the relative strength of these two opposing forces.
There is a wide variety of models that explain the relationship between transaction costs and economies of scale. For the question at hand in this thesis it is however relevant to knwo how monetary integration affects these forces. Forming a monetary union is almost certain to decrease transaction costs as it completely abolishes exchange rate uncertainty and improves transparency. The effect of monetary integration on economies of scale is less straightforward as these are often very specific to type of good under consideration. However a reduction in transaction costs and increased transparency to both consumers and producers, increases the likelihood that any level of economies of scale will be sufficient to give rise to geographical concentration of production (Krugman, 1993).
To illustrate this argument consider figure 2.1 taken from Krugman (1993). Figure 2.1 describes the production of a good that is demanded in two locations and that can be produced in both locations as well. It is assumed that aggregate demand is fully price-inelastic so that the total demand for the good is fixed at OO∗. Demand in location 1 is then measured from the left by OQ and demand in location 2 is measured from the right by QO∗. Furthermore it is assumed that production in each location is associated with location-specific economies of scale represented by the downward sloping cost curves, CC for location 1 and C∗C∗ for location 2. At the current production level Q this leads to average costs c for production in location 1 and c∗ in location 2.
The situation illustrated in figure 2.1 shows that if however transaction costs are de-creased, for instance by monetary integration, the cost advantage could lead to increased production in location 1 and decreased production in location 2. This in turn would fur-ther decrease costs in location 1 and increase costs in location 2 leading to even more production in location 1 in the next period. This process continues until production is geographically concentrated in one location.
Figure 2.1: Geographical concentration.
Source: Krugman (1993)
2. The reduction of transaction costs therefore ordinarily leads to divergence between regions in terms of industrial structure and specialization (Krugman, 1993). Several more advanced models have been put forward in the literature but the conclusion drawn above remains the same, a decrease in transaction costs below a certain threshold can lead to geographical concentration of production.2
The specialization hypothesis thus predicts that the decrease in transaction costs caused by monetary integration eventually leads to more specialized regions in Europe, which increases countries’ vulnerability to asymmetric shocks.
2.2
Endogeneity
The endogeneity hypothesis states that countries will have more correlated business cy-cles after having formed a monetary union. The hypothesis is based on the empirical observation that countries with closer trade links tend to have more correlated business cycles (Frankel & Rose, 1998). The intuition behind the endogeneity hypothesis is that forming a monetary union is such a strong commitment between countries that it would stimulate trade beyond the elimination of costs related to exchange rate uncertainty alone. A common currency precludes future competitive devaluations, encourages foreign direct investment and eventually could lead to political integration. This leads to significant in-creases in trade which lead to increased business cycle synchronization among the member countries (Mongelli, 2002).
Besides bilateral trade the endogeneity hypothesis predicts that other channels caused by monetary integration are expected to affect the synchronization of business cycles of the member countries as well. The common monetary policy under a unified central bank
2See for instance Fujita, Krugman, and Venables (1999).
might unify demand shocks and through that increase synchronization of business cycles (Goncalves, Rodrigues, & Soares, 2009). The monetary policy is set in one direction i.e., contractionary or expansionary, for all countries, therefore affecting demand in all coun-tries of the union similarly. Other possible channels could be the integration of financial markets, common rules regarding fiscal discipline, and increased spill-overs in productivity shocks due to increases in trade (Mongelli, 2002).
Both the specialization and endogeneity hypotheses thus start from the notion that transaction costs will decrease as a result of forming a monetary union. Both hypotheses also predict that this decrease in transaction costs will lead to increases in bilateral trade.3 The hypotheses differ in their prediction of the effect of these increases in trade on the vul-nerability to asymmetric shocks. The endogeneity hypothesis predicts that the increases in trade lead to more synchronized business cycles and less vulnerability to asymmetric shocks, whereas the specialization hypothesis predicts the opposite i.e., less synchronized business cycles and more vulnerability to asymmetric shocks. Furthermore the endogene-ity hypothesis predicts that also other channels besides bilateral trade are expected to positively affect the synchronization of business cycles among countries that have formed a monetary union as well.
As described in this chapter, economic theory does not provide a decisive answer to the question whether monetary integration leads to, on average, more or less synchronized business cycles and vulnerability to asymmetric shocks. Therefore many researchers have turned to empirical research to find out what the effect of monetary integration on business cycles has been. The next chapter describes the empirical research which aimed to find out more about this effect.
3
It should be noted that increases in specialization also imply increases in trade, because production is now located in concentrated areas from which larger areas are supplied through trade.
Chapter 3
Extant empirical research on
monetary integration and business
cycles
Empirical evidence on the effect of monetary integration on the synchronization of business cycles can be categorized into three broad time periods, a pre-EMU period using data prior to 1999, a pre-crisis period from 1999 until 2008 and a post-crisis period from 2008 until today.
3.1
Pre-EMU
During the pre-EMU period the economic debate about EMU mainly focussed on the suitability of EMU as an optimum currency area. For the business cycle argument this meant that most analysis aimed to predict whether the creation of EMU would increase or decrease the correlation of business cycles amongst member states.
Bayoumi and Eichengreen (1993) compared the correlation of business cycle of 11 EU members with that of US regions and found that the EU member states were hit more often by asymmetric shocks than US regions. This result contradicts Krugman’s argument that far-going integration would lead to more asymmetric shocks as the US, especially at that time, was more integrated than the EU. Their result however, did not hold for Germany and its immediate neighbors, possibly explained by the close trade links or similarities in monetary policy in these countries. That the business cycles of US regions were found to be more correlated than those of EU members is not surprising, as ”within country” business cycles correlations are on average higher than ”across country” business cycles (Clark & Shin, 1998; Clark & Wincoop, 2001) That is because several adjustment mechanisms exist within countries opposed to across countries, such as fiscal redistribution mechanisms, common monetary policy and similar institutional setups, amongst others.
Other authors examined business cycles across EU countries and found that the cor-relations significantly increased for various countries since the 60s (Fatas, 1997; Angeloni & Dedola, 1998; Peiro, 2004; Furceri & Karras, 2006) implying that as these countries
opened up and trade increased, their business cycles became more correlated. Frankel and Rose (1998) confirmed this in their seminal publication on the endogeneity of OCA criteria which, as mentioned earlier, showed that countries with stronger trade links tend to have more strongly correlated business cycles. These result suggest that monetary integration would lead to a more closely related business cycles, as monetary integration is expected to foster trade which in turn would increase the correlation across business cycles in different nations.
During the pre-EMU period the empirical evidence seemed in favor of the endogeneity hypothesis as it was shown that for several European countries the pairwise correlation of business cycles increased during a time in which bilateral trade increased as well.
3.2
Pre-crisis
After EMU was created in 1999 it, for the first time, became possible to empirically test the effect of complete monetary integration on business cycles. Furceri and Karras (2008) calculated the correlation between the business cycle of each individual country with the EMU-wide business cycle and the correlation between the different components of country GDP with the components of EMU-wide GDP. They found that for all sample countries the business cycles were more synchronized in the period 1999-2004 than during 1993-1998. This result is less clear for the individual components of GDP where they found increases in import and export correlations for all countries but mixed results for gov-ernment spending. A possible explanation for the mixed results regarding the correlation of government spending is that, in absence of the exchange rate as policy instrument, government spending has been the only instrument available to respond to asymmetric shocks.
Similar results were found using different methodologies. Goncalves, Rodrigues, and Soares (2009) use a difference-in-difference model, similar to the one that will be used in this thesis, to compare bilateral business cycle correlations amongst EMU members and OECD countries during the period 1980-2007. They found that bilateral business cycle correlations increased by approximately 5% for all countries and roughly 20% for countries that adopted the euro. Furthermore they found a surprisingly negative effect of trade increases on bilateral business cycle correlations. This contradicts earlier research and implies that monetary integration increases business cycle correlations through other channels than bilateral trade, such as potentially common monetary policy, the rules regarding fiscal discipline or financial market integration.
Mink, Jacobs, and de Haan (2007, 2012) propose a different measure of business cycle synchronization, as stated in the introduction. They state the that standard correlation co-efficient does not accurately takes account of variations in business cycles and can therefore give misleading information. In order to compare business cycles one needs to know both the sign and amplitude of the business cycles under consideration. If these two variables are simultaneously represented by one measure, the correlation coefficient, it could well be
that some information regarding the sign and amplitude is lost.1 Therefore they use two
measures and the results of their analysis contradict earlier evidence from the pre-crisis period. They find that the business cycles of the Netherlands, Germany and France are similar to the EMU-wide business cycle, but that this is not the case for Finland, Greece and Italy. Using the same methodology they found, surprisingly, that the synchronization of business cycles in Europe has not changed much from the 70s until 2006, which marked the end of their data set. They did find however that business cycles in Europe have been more synchronized than business cycles in the US since the introduction of the euro.
Empirical evidence during the pre-crisis period seems mixed and does not clearly sup-port either the endogeneity or the specialization hypothesis.
3.3
Post-crisis
The financial crisis of 2007 and the subsequent sovereign debt crisis after 2009 steered the European economies into a deep recession and exposed the fragility of EMU. These crises affected the peripheral countries of EMU the most in terms of economic performance, and at the same time forced the governments of these countries to reduce spending, which both could have led to lower correlation in business cycles with the core EMU countries. It is therefore not surprising that empirical evidence since then points in the direction of decreases in business cycle correlation across the EMU members.
Christodoulopoulou (2014) uses a difference-in-difference estimation technique similar to Goncalves et al. (2009) but finds that the common currency negatively affects business cycle synchronization. This effect was found to be higher for the least developed countries, although less robust. Furthermore the results suggest that there are two diverging groups within EMU, the periphery with some core countries and the rest of the core countries (Christodoulopoulou, 2014). This could be a confirmation of Krugman’s specialization hypothesis but it could also be an effect of the crises and the measures taken to resolve these crisis. Further research is needed in order to find out what the underlying cause is of the findings of Christodoulopoulou (2014) but that is beyond the scope of this thesis.
A possible explanation for the contradicting results is that Goncalves et al. (2009) use a different set of countries as a control group. They have included non-European OECD countries in their control group. Another possible reason is that instead of using bilateral trade as an explanatory variable, Christodoulopoulou (2014) uses the variables that explain bilateral trade, as used in gravity models of trade, instead of bilateral trade itself. The reason for this alternative analysis is that using bilateral trade as an explanatory variable for business cycle correlations might give rise to simultaneity biases. It is not unlikely that a high correlation of business cycles, due to other reasons than close trade links, stimulates bilateral trade, which could give biased results.
Different from all previous analysis Fingleton, Garretson, and Martin (2015) use a spatial panel model on European regional data. Their findings suggest that a common contractionary shock in the Eurozone has the largest effect on the most geographically
1
A more detailed analysis regarding the sign and amplitudes of business cycles follows in chapter 4.
isolated regions, which, they state, are the least productive and suffer the most from the recent crisis. This suggests that a common contractionary shock has asymmetric effects in the Eurozone.
However not all evidence points published since the crisis points in the same direction. Antonakakis and Tondl (2014) estimate a simultaneous equation model and find that business cycles in the EU have become more synchronized, especially since the introduction of the common currency. Additionally they decompose the factors that drive business cycle synchronization and find that specialization does not lead to lower business cycle correlation, opposing Krugman’s argument.
3.4
Mixed results
During the pre-EMU period empirical evidence seemed to be in favor of the endogeneity hypothesis. After EMU was created however some different results were found supporting the specialization hypothesis, especially after the European sovereign debt crisis erupted. In summary the empirical evidence regarding the effect of monetary integration on the synchronization of business cycles can only be described as mixed. Amongst others that is because a wide variety of estimation techniques have been applied using different country samples, different time periods and different variables. Several studies for instance cal-culate bilateral business cycle correlations and average these to find a country’s average business cycle correlation. Whereas other studies calculate an EMU-wide business cycle and then pair that EMU-wide cycle to each country’s business cycle to find correlation coefficients. It is not unlikely that these different methodologies lead to different results.
Furthermore it might be that the Pearson correlation coefficient is not the most ap-propriate measure of the variation in business cycles, as described earlier. This thesis aims to add another argument to that discussion with an empirical analysis following the methodology of Mink et al. (2007) which will be described in detail in the next chapter.
Chapter 4
Research methodology
This chapter describes the methodology that is used to estimate the effect of monetary integration on the synchronization of business cycles. For this analysis a difference-in-difference technique is used to estimate the effect of receiving a treatment, in this case having introduced the euro as legal tender in 1999, on the pairwise synchronity and simi-larity of the European business cycles.
The treatment group consists of the 11 countries that adopted the euro in 1999; Aus-tria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain. The control group consists of European countries that have largely comparable economies but have not introduced the euro and is formed by Denmark, Nor-way, Sweden, Switzerland and the United Kingdom. Eastern European EU countries such as Poland, the Czech Republic or Hungary are excluded from the analysis because their economies, especially historically, are very different from the economies of the Western European countries.
To present this methodology in a clear manner first the statistical procedures used to calculate the output gap are described. Then the synchronity and similarity measures are discussed and finally the difference-in-difference technique is discussed.
4.1
Calculating the output gap
Before the synchronity and similarity measures can be can be calculated a notion of the output gap has to be established. It is common practice to use non-parametric filtering methods to decompose output data into a trend and a cyclical component (Mink, Jacobs, & de Haan, 2007):
yt= τt+ ct, (4.1)
where ytis output at time t, τtis the trend component and ctis the cyclical component.
The output gap of country i is then defined as the deviation of actual output from its trend value as a percentage of its trend value:
gapit=
yit− τit
τit
, (4.2)
where gapitis the output gap of country i at time t, yit is the observed value of output
of country i at time t and τit is the trend component of output of country i at time t.
Commonly used methods to determine the trend and cyclical components of a series are the Hodrick-Prescott (HP) filter, Baxter-King (BK) filter and Christiano-Fitzgerald (CF) filter (Mink et al., 2007). It should be noted that the estimation results often strongly depend on the chosen filter method (Christodoulopoulou, 2014). Unfortunately the choice of filter methods is rather restricted in this thesis because both the BK and the HP filter are not usable. The BK filter drops observations at the beginning and end of the sample period, which in this case implies that the latest year for which data is available would be 2001. This makes the BK filter inapplicable because the years after 2001 should be taken into account as well to properly estimate the effect of the euro on the business cycles of the euro area countries.
The HP filter is also not usable because it uses all observations to obtain the smoothest curve through these observations. For the analysis in this thesis, this implies that obser-vations after the introduction of the euro are included in the calculation of the pre-euro output trend, and vice versa. Since the purpose of this thesis is to estimate if the in-troduction of the euro represents a structural break in the output trends the HP filter is unusable. The CF filter however does allow for structural breaks in the data and does not drop observations at the beginning or end of the period, which makes it superior to both the HP and the BK filter for the analysis in this theses. Therefore the CF filter is used to calculate the output gap, following Mink et al. (2007).1
4.2
Synchronity and similarity
After calculation of the output gap a measure of output gap or business cycle synchro-nization can be obtained. It is common practice in the literature to the use the Pearson correlation coefficient for this purpose. Mink et al. (2007) however state that correlation coefficients do not always properly reflect differences in the sign and amplitude of business cycles. This statement is discussed first, before two alternative measures are presented.
4.2.1 Shortcomings of the correlation coefficient
To determine whether the output gaps of two countries are synchronous one needs to know both the sign and the amplitude of both output gaps. Two countries that have output gaps equal in amplitude but opposite in sign find themselves in completely different environments as this means that one country has a positive output gap whereas the gap of the other country is negative. Similarly two output gaps that are equal in sign but different in amplitude also represent different scenario’s. The output gap of one country might be slightly above the trend level whereas the other gap could be far above its trend value.
The Pearson correlation coefficient however does not always reflect these differences properly (Mink et al., 2007). The following figures illustrate this. Figure 4.1 shows the output gaps of two hypothetical countries. The output gaps are equal in sign everywhere
Figure 4.1: Imperfect correlation despite equal sign of output gaps.
Source: Mink et al. (2007)
in the figure i.e., simultaneously above and below the dashed line that represents an output gap of zero. Despite this similarity in signs the correlation coefficient of these output gaps is only 0.53, reflecting only a moderate correlation.
Figure 4.2: Perfect correlation despite differences in amplitude.
Source: Mink et al. (2007)
Figure 4.2 shows two other hypothetical paths that the output gaps of two countries can follow. In this case the correlation coefficient of the two series is 1, implying perfect
correlation. The two output gaps however differ substantially in amplitude and the notion of a one-on-one relationship between the two output gaps, as suggested by the correlation coefficient, seems incorrect.
Despite that the Pearson correlation is often used for empirical analysis of business cycles, figures 4.1 and 4.2 illustrate that it does not always properly reflects the relationship between two business cycles. It is therefore not unlikely that some of the contradictory empirical evidence is due to this shortcoming of the correlation coefficient. To overcome this problem two alternative measures proposed by Mink et al. (2007) are used in this thesis for the analysis of business cycles, which will be discussed in the next section.
4.2.2 Synchronity
The synchronity measure is designed to tackle the problem illustrated in figure 4.1. It indicates whether two output gaps are equal in sign or not and is defined as the product of the two output gaps divided by the absolute value of that same product:
φir(t) =
gi(t)gr(t)
|gi(t)gr(t)|
, (4.3)
where gi(t) is the output gap of country i at time t and gr(t) is the reference output
gap at time t, which could be the output gap of another country or a group of countries, depending on the type of analysis. The synchronity measure can take values from [−1, 1] where a value of 1 indicates that the sign of the output gap of country i is equal to the reference at time t. A value of -1 indicates that the signs of the two output gaps are unequal at time t. If one of the two output gaps is equal to zero the synchronity measure is set at zero as well, because it would otherwise imply a division by 0. For the output gaps illustrated in figure 4.1, φir = 1 for all t as the signs are equal everywhere in the
figure.
4.2.3 Similarity
Where the synchronity measure concerns the sign of two series, the similarity measure is designed to measure the extent of equality between the amplitude of two series. It is defined as 1 minus the absolute difference between the two output gaps as a share of the average absolute output gap of the sample in period t:
γir(t) = 1 −
|gi(t) − gr(t)|
Pn
i=1|gi(t)|/n
(4.4)
The similarity measure is defined on a [1 − n, 1] scale. A value of 1 indicates that both output gaps are equal in sign and in amplitude as in this case the fraction subtracted from 1 is equal to zero. If gi and gr are opposite in sign and all other output gaps are equal to
4.2.4 Reference output gap
Until now the reference output gap has remained undefined. If the problem at hand concerns the equality in business cycles of two countries the reference output gap should simply be the output gap of the other country. The purpose of this thesis however is to analyze if the monetary integration in the EMU is affecting the business cycles of the EMU member countries. The reference output gap could then be the European or the EMU output gap. But it should be noted that it is not accurate to ex-ante ”take it as given that the European cycle exists and that it coincides either with the cycle of a leading European economy, or the cycle of a weighted average of several European economies, or the cycle of a common factor” (Camacho, Perez-Quiros, & Saiz, 2006, p.1689).
Mink et al. (2007) therefore take a statistical approach and use the median output gap of all individual output gaps at time t as the reference, because this maximizes both the synchronity and similarity measures simultaneously. For the research question of this thesis it is however relevant whether the output gaps of the euro area countries have become more synchronized with each other as a result of the common currency than with the median. Therefore all countries in the sample are paired and for each country pair the pairwise output gap synchronity and similarity are calculated. The output gaps of all individual countries in the sample are thus used as a reference output gap.
In total the sample consists of 16 countries.2 This means that for each country the output gaps of the other 15 countries are individually used as a reference to calculate 15 pairwise output gap synchronities and similarities for each year in the sample. This gives a total of 16 ∗ 15 = 240 unique country pairs for which the similarity and synchronity of the output gap is calculated for every year in the sample. The data on GDP, which is used as the measure for output, covers the period from 1960 until 2013. Thus for each country pair there are 54 observations, implying a total of 54 ∗ 240 = 12.960 observations for both the synchronity and similarity measures.
4.3
Difference-in-difference model
To analyze the effect of the euro on the output gap synchronity and similarity a difference-in-difference estimation technique is used. Following Ashenfelter and Card (1985) the effect of a treatment can be extracted through the following regression:
yit= β0+ β1∗ Xit+ β2∗ Tit+ β3∗ Pit+ it, (4.5)
where Tit is the treatment dummy that is 1 for the treatment group and 0 for the control
group, Pit is a dummy that is 1 for the period after treatment and 0 for the period
before the treatment and Xit = Tit∗ Pit. Finally β0 and it are a constant and an error
term respectively. The OLS estimate of β1 is the variable of interest because it gives the
2
The 11 countries that adopted the euro in 1999: Austria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain and the 5 control countries: Denmark, Norway, the United Kingdom, Sweden and Switzerland.
treatment effect, which is calculated as follows:
ˆ
β1= (¯yt,2− ¯yt,1) − (¯yc,2− ¯yc,1) (4.6)
Which states that ˆβ1 is equal to the difference in the average effect in the treatment
group, denoted by t, between period 1 and period 2 minus the difference in the aver-age effect in the control group, denoted by c, between period 1 and 2. Hence the term difference-in-difference estimation.
The OLS estimate of β1 is unbiased under the assumption that the treatment group
would have followed a similar path as the control group had they not received treatment. This is necessary because it is impossible to establish a counterfactual for the treatment group i.e., what would have happened to the business cycles of the euro area countries had they not introduced the euro. Therefore it is assumed that the euro area countries would have followed the path of the control group if they had not introduced the euro. The choice of the control group is thus crucial for the difference-in-difference analysis (Christodoulopoulou, 2014). Therefore the next section will further discuss the validity of the chosen control group.
4.3.1 Control group
As described in the previous section the validity of the difference-in-difference analysis depends on the assumption that the treatment group would have followed the same path as the control group had it not received the treatment. The treatment in this case, having introduced the euro in 1999, is however not randomly assigned. This complicates the assumption of equality between the treatment and control group severely. It is not unlikely that the countries that introduced the euro in 1999 did this because they had largely synchronous business cycles and therefore did not expect much occurrence of asymmetric shocks. Or, even more likely, that one of the reasons why the countries of the control group did not introduce the euro was the fear of asymmetric shocks caused by divergencies in their economies with the rest of Europe.
The economies of the treatment and the control group however are all Western Eu-ropean and are roughly equal in terms of development, monetary and fiscal policy and consumption and production patterns. Or, at least roughly equal in these areas in com-parison to the rest of the world. Therefore Denmark, Norway, Sweden, Switzerland and the United Kingdom are chosen as the control group.
As described in section 4.2.4 all individual country observations are paired in order to form pairwise observations. This implies that for each country pair there are three possible situations regarding the treatment:
1. Both countries in the pair have the euro as legal tender.
2. One country in the pair has the euro as legal tender and the other country does not.
The goal of this thesis is to analyse the effect of the euro on the business cycles of the euro area countries. Therefore the treatment group consists only of country pairs for which both countries have the euro as legal tender. The control group in this case are all country pairs for which one or both countries do not have the euro as legal tender.
The control group includes both the country pairs of situation 2 and 3 i.e., pairs in which one or both countries do not have the euro, because if the euro affects bilateral business cycles, the business cycle synchronization of country pairs of situation 1 should be significantly different from those in situation 2 and 3. That is because the decreases in transaction costs and increases in transparency, created by the formation of the mon-etary union, should only apply for country pairs in which both countries have the euro. For example the introduction of the euro should increase price transparency between the Netherlands and Belgium, but not, at least not in the same magnitude, between the Netherlands and the UK.
Figure 4.3: Mean similarity for treatment and control group.
Source: Own graph based on data from IMF database.
One way to assess the validity of the control group is to plot the average synchronity and similarity during the sample period and assess whether they are similar over time. Figure 4.3 shows the average similarity for the treatment and the control group from 1960 until 2013. It seems that the average similarity in the treatment and the control group follow a roughly similar path. For the majority of the sample period the two groups seem to move in the same direction. Only at the end of ’70s and the beginning of the ’80s the two groups move in opposite direction.
At first sight the average synchronity in the treatment and control group, shown in figure 4.4, follow a less comparable path than the average similarity in both groups. Until the ’70s the two groups seem to follow a similar path but during the ’70s and especially
Figure 4.4: Mean synchronity for treatment and control group.
Source: Own graph based on data from IMF database.
the ’80s the paths diverge substantially. During the ’90s the treatment and control group seem to back on the same path again before diverging again around the end of the 20th century, after which another period of equality followed until 2010.
The assumption that the treatment group would have followed the same path as the control group had they not received treatment thus seems to be somewhat supported by figures 4.3 and 4.4. Furthermore the average output gap similarity and synchronity of the treatment and the control group are quite strongly correlated, as shown in table 4.1. This suggests that, on average, the treatment and the control group move in roughly the same direction for both the output gap synchronity and similarity.
Table 4.1: Correlation in average similarity and synchronity.
Measure Correlation treatment and control group Similarity 0.724
Synchronity 0.686
Source: Own table based on data from IMF database.
The output gaps of the control group however remain an approximation of what would have happened to the output gaps of the treatment group if they had not introduced the euro. Therefore any conclusions based on this analysis regarding causality should be interpreted with care as it will remain unknown how the output gaps of the treatment group would have behaved in the counterfactual case.
4.3.2 The model
To finally estimate the effect of having introduced the euro on the output gap similarity and synchronity two regression models similar to (4.5) are used. In which φit and γit are
the pairwise synchronity and similarity respectively:
φit= β0+ β1∗ Xit+ β2∗ Tit+ β3∗ Pit+ β4∗ tradeit+ β5∗ F Iit+ n X i=1 δt∗ αi+ it (4.7) γit= β0+ β1∗ Xit+ β2∗ Tit+ β3∗ Pit+ β4∗ tradeit+ β5∗ F Iit+ n X i=1 δt∗ αi+ it (4.8)
The variables X, T and P are again dummy variables, where T = 1 if both countries in the pair have the euro as legal tender and T = 0 if not, P = 1 for all observations from 1999 onward and P = 0 for all observations prior to 1999, and X = T ∗ P . Furthermore a measure of bilateral trade, trade, and financial integration, F I, are added as control variables. Earlier research (Frankel & Rose, 1998; Goncalves et al., 2009; Antonakakis & Tondl, 2014; Christodoulopoulou, 2014) has shown that these were significant variables in explaining pairwise output gap correlations, therefore it is to be expected that these variables also affect the output gap synchronity and similarity.
The variable trade is constructed by adding the exports and imports of the first country in the pair to the second country in the pair divided by the sum of the two country’s GDP:
trade = EXij+ IMij GDPi+ GDPj
, (4.9)
where EXij are the exports of country i to country j and IMij are the imports of country
i from country j. It should be noted that bilateral trade is expected to be an endogenous variable in explaining bilateral output gap synchronity and similarity, because it is shown to be endogenous in explaining output gap correlations as well (Frankel & Rose, 1998). Including it in the model may therefore give biased results which could be solved by using other estimation methods or variables. It is however beyond the scope of this thesis to further explore this direction.
The measure of financial integration, F I, is constructed similar to Christodoulopoulou (2014) and is as follows: F I = N F Ai GDPi −N F Aj GDPj , (4.10)
where N F Ai is the net financial position of the banking sector of country i calculated as
the difference between the claims and liabilities of the banking sector of country i abroad. This is an approximation of the financial integration between two countries because N F Ai
is the net financial position of country i with respect to the rest of the world and not to country j. Unfortunately no data was found on the actual bilateral claims and liabilities of the banking sectors, therefore this approximation is used. The measure is designed to capture the phenomenon that surplus countries are financing deficit countries in the euro area (Christodoulopoulou, 2014). Therefore the net position of country j is subtracted
from that of country i, so that F I takes high values for pairs with diverging net positions i.e., for pairs in which one country is a net creditor and the other a net debtor. A positive coefficient is expected for F I as it is expected that more financial interlinkages lead to higher correlated business cycles (Christodoulopoulou, 2014).
Finally n specific country pair dummies are added in order to capture other hetero-geneity between the country pairs that is constant over time.3
4.4
Data
The data used in this thesis comes from the online databases of the IMF, the OECD and the BIS.4 The output gaps have been calculated on the basis of annual GDP data from
1960 until 2013, taken from the IMF. For Germany and Ireland the data on GDP started in 1970 and for Switzerland in 1980. Data on bilateral trade unfortunately ranged a much shorter period, from 1995 until 2013, and was taken from the OECD database. The BIS data on the banking sector positions used to calculated the net financial positions of each country covered the period 1978-2013. Although for some country pairs data on the first few years of this period was missing.
3The dummy variable T then is dropped in the regression that contains the country pair dummies, in
order to avoid perfect multicollinearity.
Chapter 5
Results & analysis
This chapter describes the results of the difference-in-difference estimation of the effect of the euro on the synchronization of the European business cycles. It starts with a description of the estimation results for the output gap synchronity followed by the results for the output gap similarity. The subsequent section will then analyze these results.
5.1
Synchronity
Table 5.1 shows the estimation results of the difference-in-difference analysis for the output gap synchronity. For each variable the table reports the coefficient and the correspond-ing t -statistic, denoted in parentheses. The first column shows the baseline estimation that includes only the difference-in-difference dummy variables. In the second and third columns the control variables trade and F I are added respectively. Column (4) shows the full estimation of equation (4.7) with the country-pair fixed effects. To avoid perfect multicollinearity the dummy variable T is omitted in this last specification.
Table 5.1 shows that during the sample period the pairwise synchronity, on average, was positive for both the control and the treatment group. However the average pairwise synchronity was roughly 50% higher for the treatment group after the introduction of the euro. This is indicated by the coefficient for X which gives the treatment effect i.e., the effect of introducing the euro on the pairwise output gap synchronity. For all specifications that include one or multiple control variables the coefficient for this variable is positive and significant at 0.1% level. This implies that the pairwise synchronization on average was significantly higher for country pairs in which both countries adopted the euro after it was introduced in 1999.
These results suggest that the business cycles of the euro area countries became more equal in terms of synchronization than for the control group countries, after they in-troduced the euro. Although any conclusions regarding causality should be drawn with care, this analysis seems to suggest that the monetary integration in Europe increased the equality in signs of the business cycles among the euro area countries in the sample. These results thus seem to support the endogeneity hypothesis i.e., the hypothesis that the equality of business cycles would increase after the attempts of monetary integration. It should be noted that in the beginning of the sample period large divergencies
Table 5.1: OLS regression of the effect of the euro on the pairwise synchronity Pairwise synchronity (1) (2) (3) (4) X -0.00260 0.198∗∗∗ 0.211∗∗∗ 0.186∗∗∗ (-0.05) (4.07) (4.00) (3.38) T 0.236∗∗∗ 0.0278 -0.00937 (7.30) (0.66) (-0.20) P 0.161∗∗∗ 0.0218 0.0186 -0.00757 (4.16) (0.54) (0.43) (-0.17) trade 1.073 1.486∗∗ 5.546∗∗ (1.94) (2.61) (3.18) F I 1.144∗∗∗ 1.551∗∗∗ (4.23) (4.52) constant 0.310∗∗∗ 0.433∗∗∗ 0.399∗∗∗ 0.341∗∗∗ (15.08) (13.67) (11.51) (12.27) Observations 11820 5547 5320 5320
Column (4) includes country-pair fixed effects, therefore the dummy variable T is omitted t statistics in parentheses
∗
p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001
Source: Own calculations based on data from IMF, BIS and OECD databases
isted between the sample countries, which could have biased the results described above. Therefore the same regression is estimated for shorter time periods; 1969-2013 and 1979-2013.1 The estimation results however remain exactly equal for columns (2), (3) and (4) despite these different sample periods. That is because the variable trade is included in these estimation and the data for this variable only ranges the period 1995-2013. The statistical software that has been used only takes ”complete” observations into account when estimating a regression, which in this case implies that the regressions of columns (2), (3) and (4) are only estimated for the period 1995-2013.
Furthermore table 5.1 shows positive and significant coefficients for the control vari-ables trade and F I, indicating that countries with closer trade or financial links tend to have more equal business cycles in terms synchronity. This was expected as earlier empiri-cal research has shown that these variables were significant in explaining bilateral business cycle correlations (Frankel & Rose, 1998; Goncalves et al., 2009; Antonakakis & Tondl, 2014; Christodoulopoulou, 2014).2 However this analysis does not provide any evidence in the direction of the effect, it could be that it is the equality of business cycles which leads to closer trade or financial links and not the other way around.
Or, that another variable, which is excluded in this analysis, is causing the more equal
1
The results of these estimations are shown in appendix 4.
2Goncalves et. al (2009) however found a negative effect of bilateral trade on business cycle correlations.
business cycles and closer trade and financial links. The inclusion of the country-pair fixed effects nearly quadrupled the coefficient for trade, indicating that the coefficient for trade is probably biased due to omitted variables. Therefore further research is required to draw conclusions on the effect of bilateral trade or financial integration on the pairwise synchronity of business cycles. It is however beyond the scope of this thesis to further explore this direction.
5.2
Similarity
The estimation results for the effect of the euro on the output gap similarity are shown in table 5.2, which is similar in setup as table 5.1. So that the first column again represents the baseline estimation, which is extended by adding control variables as well as country-pair fixed effects in columns (2) to (4).3
Table 5.2: OLS regression of the effect of the euro on the pairwise similarity
Pairwise similarity (1) (2) (3) (4) X -0.0816 0.00342 0.0166 0.0159 (-1.58) (0.05) (0.23) (0.22) T 0.00217 -0.112 -0.145 (0.03) (-1.23) (-1.55) P 0.180∗∗∗ 0.123∗∗ 0.152∗∗ 0.156∗∗ (5.68) (2.60) (3.21) (3.27) trade 4.127∗∗∗ 4.087∗∗∗ 2.843∗ (4.66) (4.37) (2.34) F I 1.121∗∗∗ 1.318∗∗∗ (5.66) (4.76) constant -0.112∗∗ -0.116∗ -0.174∗∗∗ -0.232∗∗∗ (-2.75) (-2.29) (-3.29) (-8.95) Observations 11820 5547 5320 5320
Column (4) includes country-pair fixed effects, therefore the dummy variable T is omitted t statistics in parentheses
∗
p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001
Source: Own calculations based on data from IMF, BIS and OECD databases
The estimation results for the output gap similarity differ from those found for the output gap synchronity, described in the previous section. First, the introduction of the euro seems to have no significant differentiating effect on the pairwise similarity for the euro area countries in comparison to the control group countries, shown by the insignificant
3The regression for the output gap similarity is not estimated for alternative time periods, because the
results would be the same. As described in the previous section the variable trade is included in these regressions and therefore only observations from the period 1995-2013 are included in the estimation.
coefficients for X in all specifications. However it does seem that the creation of EMU did represent some structural break in the pairwise similarity for both the treatment group and the control group. The coefficient for the period dummy P is positive and significant for all specifications, implying that the average pairwise similarity increased for both the control group and the treatment group after 1999.
Table 5.2 however also shows some results that are similar to those shown in table 5.1. Again the variables trade and F I are highly significant indicating that it are variables that should not be omitted in this analysis. Furthermore the inclusion of the country-pair fixed effects changed the coefficient for trade quite somewhat, indicating that it was probably biased, although now the coefficient decreased.
Based on the results provided in table 5.2 it seems that the introduction of the euro did not have any effect on the equality of business cycles, at least not in terms of the similarity of business cycles as defined here. The average similarity in business cycles has increased after 1999, however for both groups equally. This implies that it was probably not the introduction of the euro that caused this increase in pairwise similarities. Based on the results of this section it therefore seems that neither the endogeneity nor the specialization hypothesis is supported.
5.3
Analysis
The results described in the previous two sections raise a number of questions. Of which the following are discussed in this section:
1. Why does the introduction of the euro have a significant effect on the output gap synchronity but not on the output gap similarity?
2. Why has the average similarity increased equally for both the treatment and the control group after the introduction of euro?
3. How are the results of the methodology used in this thesis different from the results of the traditional methodology that uses correlation coefficients?
5.3.1 Different effect of the euro on synchronity and similarity
The first question raised by the results of the previous two sections concerns the differ-ent effect of the euro on the output gap synchronity and the output gap similarity. The introduction of the euro seemed to have significantly increased the average pairwise syn-chronity whereas it did not have a significant effect on the pairwise similarity. The question that comes to mind is off course: Why this difference? How can it be that the pairwise synchronity increased for the countries that introduced the euro in comparison to the non-euro countries whereas the similarity remained rather unaffected by the introduction of the euro?
Figure 5.1 shows the average synchronity and similarity for the treatment and the control group from 1999 onwards and provides a little more insight in this difference. The figure shows that the average output gap synchronity increased substantially for the
(a) Synchronity (b) Similarity
Figure 5.1: Mean similarity and synchronity since 1999.
Source: Own graphs based on data from IMF databese.
treatment group right after the introduction of the euro. In 2002 however the average synchronity dropped back to a similar path as the control group from until 2010, after which another period followed where the average synchronity in the treatment group was nearly twice as high as in the control group.
For the output gap similarity it seems that the treatment and the control group have followed a roughly similar path and that during the majority of the period after 1999 the control group actually had a slightly higher average similarity than the treatment group. One explanation for this difference could be the simple fact that the two measures were designed to capture two different aspects of the business cycle. It could be that the introduction of the euro made the business cycles of the euro area countries more equal in terms of synchronity but not in similarity. The output gap synchronity measures if two business cycles have the same sign i.e., if both are in a boom or both in a recession. The output gap similarity however measures the extent of equality in the amplitude of two business cycles. It could thus be that the since the introduction of the euro the business cycles of the treatment group were more often in booms and recessions in the same period, but that these booms and recession were much less equal in terms of amplitudes. Phrased differently that for some countries the peaks and the throughs were much higher/lower than for others but that the highs and lows occurred during the same period.
Some evidence for this explanation of of inequality in amplitudes can be found in appendix 3, which shows the output gap for each country since 1999. The output gaps of the sample countries can be categorized in three broad groups. The first group consists of Austria, Belgium, France and Germany for which the output gap has remained rather close to 0 for both positive and negative gaps. For a larger group of countries consisting of, Denmark, Italy, Luxembourg, the Netherlands, Norway, Sweden, Switzerland and the UK, the output gap fluctuated within a band that stretched between -0.5 and 0.5%, which forms the intermediate group. Whereas for the third group consisting of Finland, Ireland, Portugal and Spain the output gaps exceeded this band and sometimes even reached values close or above 1%.
All countries of the control group are in the intermediate group, whereas the treat-ment group is distributed almost evenly across the three groups. It seems thus that the introduction of the euro has not made the business cycles of the euro area countries more equal in terms of amplitude. This could explain why the regression analysis showed no significant effects of the treatment of the output gap similarity.
It could also be that the period since the introduction simply has been to short to already find an effect of the monetary integration on the amplitude of the business cycles. If you take a business cycle to last between 6 and 8 years there could have been only 2 complete business cycles since the introduction of the euro. It could be that the similarity effects simply cannot be extracted for only 2 complete business cycles.
5.3.2 Increased similarity for both the treatment and the control group
Another reason why no significant effect of the treatment was found on the output gap similarity could be that another factor affected the similarity of both the treatment and the control group. Figure 5.1b shows that the output gap similarity increased steadily for both groups in the years before the great financial crisis and fell sharply after that.
In the years before the crisis all countries in the sample, just like the majority of the rest of world, experienced economic prosperity, represented by positive output gaps. It is therefore not surprising that the average similarity for both the treatment and the control group increased during that period. Similarly all countries in the sample and the rest of the world found themselves in recessions after the crisis, which is accompanied by a sharp drop in the average similarity for both groups. Note that the output gap similarity is not a measure of booms and/or recessions but that it is a measure of the extent to which the amplitudes of the sample country’s business cycles are equal.
It is however not unlikely that during good times the similarity increases but in crisis times decreases. In the years preceding the crisis all countries had positive output gaps, some a bit higher than others but they were all roughly equal. The crisis however probably had asymmetric effects i.e., for some countries the fall in output was much larger than for others. Especially the so called periphery countries like Italy, Spain, Portugal and Ireland saw much deeper drops in output than the core countries.
It could thus that the reason why the euro seemed to have no effect on the output gap similarity is the disturbance caused by the great financial crisis and/or sovereign debt crisis. The financial crisis occurred in 2007 followed by the sovereign debt crisis that started in 2009, but it seems that the European countries have started recovering from this crises only very recently. This means that nearly half of the years since introduction of the euro have been crisis years. Therefore the crises could have been a significant disturbance factor during the period after the introduction of the euro because they affected all countries in sample asymmetrically and not only the treatment group.
5.3.3 Differences with correlation analysis
The third question raised by the analysis with the synchronity and similarity measure is: Are the results obtained with this analysis different from the traditional analysis that used
the Pearson correlation coefficient? To answer this question the output gap correlation is calculated for each country pair in the period before and after the introduction of the euro. Then the same difference-in-difference model is estimated but now using the average correlation as the dependent variable. Table 5.3, which is again similar in setup as the previous tables with estimation results, shows the result of this traditional methodology.
Table 5.3: OLS regression of the effect of the euro on the output gap correlation
Correlation (1) (2) (3) (4) X -0.160∗∗∗ -0.130∗∗ -0.119∗ -0.118∗ (-3.88) (-2.87) (-2.50) (-2.49) T 0.281∗∗∗ 0.261∗∗∗ 0.247∗∗∗ (7.39) (6.40) (5.67) P 0.250∗∗∗ 0.230∗∗∗ 0.241∗∗∗ 0.242∗∗∗ (7.74) (6.48) (6.70) (6.71) trade -1.342∗∗ -1.531∗∗ -1.703∗∗ (-2.63) (-3.08) (-3.15) F I 0.217∗∗∗ 0.222∗∗∗ (4.05) (3.95) constant 0.372∗∗∗ 0.412∗∗∗ 0.398∗∗∗ 0.509∗∗∗ (13.87) (14.39) (13.14) (30.00) Observations 11820 5547 5320 5320
Column (4) includes country-pair fixed effects, therefore the dummy variable T is omitted t statistics in parentheses
∗
p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001
Source: Own calculations based on data from IMF, BIS and OECD databases
Table 5.3 shows that the output gap correlation was higher for all countries in the period after the euro was introduced than in the period before 1999, shown by the positive coefficient for P . Furthermore the positive coefficient for T shows that on average the output gap correlations of the treatment group were higher than those of the control group. However the output gap correlations of the treatment group increased less than those of control group, indicated by the negative coefficient for X. This implies that since the treatment group introduced the euro their output gap correlations increased less than those of the countries that did not introduce the euro.
This result is in line with Cristodoulopoulou (2014) who also finds a negative treatment effect using the difference-in-difference methodology. In contrast, Goncalves et al. (2009) find a positive treatment effect with the same methodology. However their control group also contains non-European countries such as Australia, Canada, Japan, New-Zealand and the US which decreases the validity of their analysis.
Although the negative effect for the treatment is in line with earlier comparable re-search it remains somewhat surprising that a negative effect was found. The treatment namely had a positive effect on the output gap synchronity and no significant effect on
