The effect of the financial crisis of 2007-2008 on business cycle synchronicity in the EMU

(1)

1

The effect of the financial crisis of 2007-2008 on business cycle

synchronicity in the EMU

Master’s Thesis International Economics and Globalization Marieke Berendsen

10104798

mberendsen1@hotmail.com Supervisor: dr. D.J.M. Veestraeten Second reader: dr. M. Micevska Scharf

(2)

2 Abstract

When countries become a member of the European Monetary Union, they lose their independent monetary policy tool. Instead, the ECB needs to use a one-size-fits-all approach. This could be suboptimal when countries are hit by an asymmetric shock. It is therefore important that member countries have a high degree of business cycle synchronicity. This paper studies the effect of the recent global financial crisis of 2007-2008 on business cycle synchronicity in the EMU. Four types of analyses are used to assess this effect. First, the mean synchronicity levels for each country are calculated for the pre-crisis and the post-crisis period. The results show that in most countries the mean decreased. However, this decrease is not significant. In addition, the synchronicity measure is regressed on a time trend variable in both periods, which showed a positive coefficient suggesting convergence in the pre-crisis period and a negative coefficient suggesting divergence in the post-crisis period. Furthermore, recursive and rolling regression analysis is used to assess the stability of the parameters of the regression models. The results of the recursive regression show that overall; there is no clear change after the crisis. In contrast, the rolling regression analysis shows that the time trend coefficient increased in 2008 and decreased in 2010 for most countries. This indicates that only the European sovereign debt crisis had a negative effect on business cycle synchronicity.

(3)

3

Statement of originality

This document is written by Marieke Berendsen who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of

(4)

4

1. Introduction

In 1999, a monetary union in Europe was formed. Countries face advantages and disadvantages when joining a currency union. Potential benefits are the elimination of exchange rate risk, a decrease in transaction costs, more transparency in prices and more political cooperation between European countries. The main costs of entering a monetary union include the loss of national monetary and exchange rate policy (Minford, 2002).

From a theoretical point of view, potential entrants of a monetary union and thus also of the Economic and Monetary Union (EMU) should satisfy certain criteria, the Optimal Currency Area (OCA) conditions. The more these conditions are fulfilled, the more suitable countries are and the lower are the costs mentioned earlier (Mongelli, 2002). One of these conditions, the convergence of countries' business cycles, is the main focus of this paper. Business cycle synchronicity is important for the following reason. When member countries of a currency union are hit by an asymmetric shock, they cannot use independent monetary policy to counteract the effects of this shock. Indeed, the ECB needs to use a one-size-fit all approach when designing the Eurozone’s monetary policy. For example, when the business cycle of country A is negatively correlated with the business cycle of other Eurozone countries, then it is possible that country A experiences an economic slowdown and at the same time other Eurozone countries experience an economic expansion. Then, with respect to monetary policy, country A needs a lower interest rate than the other countries. The ECB thus needs to choose the best policy option available by weighing the costs and benefits of the effects of each policy choice on the member countries. This eventually could make country A worse off in comparison with the situation when country A would pursue its own monetary policy (Frankel, 2004). When the ECB sets a policy rate that is not optimal for a country in question, it could be the case that a boom or a crisis will be intensified in that country. Therefore, a high correlation of countries' business cycles makes the loss of independent monetary policy less costly, because in this case ‘one size fits all’ is desirable (Frankel, 2004). Although a significant amount of literature about business cycle convergence after joining a currency union is available, not much research is available about the consequences of the global financial crisis of 2007-2008 on business cycle synchronicity in the EMU. This is an important topic for policy considerations, since the sustainability of the Euro-area is threatened when differences between member countries have, for instance, increased after the crisis. The recent economic crisis had a negative effect on all Eurozone countries. However, some countries were affected differently. For example, the Periphery countries, including

(6)

6 Greece, Ireland, Italy, Portugal and Spain (GIIPS) were especially hit hard by the crisis, because of, inter alia, debt, deficit and structural problems. The question thus arises whether the business cycles of the member countries behaved differently after the crisis, and what the effect of this potential divergence could be on the sustainability of the Eurozone. My research question thus is: how did the financial crisis of 2007-2008 affect business cycle synchronicity in the Eurozone? This question is answered using a regression analysis with data from Eurostat.

The second section of this paper outlines the existing literature. It reviews the theoretical reasons according to which business cycle convergence or divergence will strengthen after joining a currency union. It also includes an analysis of the potential effects of the recent crisis on business cycle synchronization. After that, existing empirical research about business cycle synchronicity is presented in a table and evaluated. In the third section, the methodology used in this paper for the measurement of business cycles and synchronicity is examined. In addition, it is outlined how the effect of the crisis will be assessed. Then, a description of the data is given. This is followed by the results of the empirical exercise. Finally, a conclusion will be made based on the findings in the literature and the empirical results.

2. Business cycles in a monetary union

This chapter is divided into four parts. First, the relevance of business cycle synchronicity in a monetary union is discussed. This is followed by a review of the channels through which business cycle synchronicity could increase or decrease after joining a currency union. Then, the theoretical reasons behind the effect of the financial crisis of 2007-2008 are discussed. Finally, the empirical research done about business cycle synchronicity in the EMU is presented in a table and discussed.

2.1 Relevance of business cycle synchronicity in a monetary union

Business cycle synchronicity in a monetary union is important because of the following reason. Countries with a Gross Domestic Product (GDP) level above its long term trend experience a boom and require a contractionary monetary policy. In contrast, countries with a GDP level below its long term trend experience an economic slowdown and require an expansionary monetary policy. Therefore, when these countries form a monetary union, the loss of independent monetary policy could be non-optimal for some member countries (Mink

(7)

7 et al., 2007). When a country is experiencing a recession and contractionary policy is used, then the policy interest rate will be too high for the latter country. As a result, the recession will be deepened. In contrast, when a country is experiencing a boom and expansionary policy is used, then the policy interest rate will be too low for the former country. As a consequence, the boom will be fuelled further. Therefore, booms and crises could be intensified by a one-size-fits-all approach by the unique Central Bank, which results in more divergent business cycles.

In addition, when entering a monetary union, countries lose their independent exchange rate policy. This makes it also important for these member countries to have similar business cycles. This can be illustrated via the following example. Suppose that country A is not a member of the EMU, and country B is. Assume that in both countries the fear of default increases due to an external shock. Country A and B are affected differently by this fear of default. When the fear of default increases in country A, investors would sell their holdings of government bonds of country A, and invest the proceeds elsewhere. This will result in a depreciation of the currency until other investors would be willing to buy the currency of country A again. Therefore, because of the flexible exchange rate, the money stock and the liquidity in country A’s bond market will not change by much. The government of country A could therefore still obtain funds. The situation is different in country B. When the fear of default increases in country B, investors would sell their holdings of government bonds of country B and invest the receipts in less risky government bonds from for example other EMU member countries. This will not lead to a depreciation of the currency of country B, since country B does not have a flexible exchange rate. Therefore, the money supply and liquidity in country B decrease. As a result, it will be more difficult for the government of country B to obtain funds. This increases the probability of default of country B, which in turn increases the fear of default and eventually could lead to a liquidity crisis (de Grauwe, 2011). Therefore, country B needs to use other policy tools to counteract the crisis. A possibility could be monetary policy. However, when other member countries do not experience a recession, the ECB could design a monetary policy that is non-optimal for country B. This thus shows that it is important for countries to have similar business cycles.

(8)

8

2.2 Reasons behind a change in business cycle alignment within a monetary

union

There are several theoretical reasons why there could be convergence or divergence of business cycles in the Eurozone. These theoretical explanations are outlined in this section.

2.2.1. Reasons for convergence in business cycles

In the early 1960s theory began to emerge about the pros and cons of flexible and fixed exchange rates. This and the economic integration of European countries, led to the introduction of the traditional OCA theory. The traditional OCA theory describes the conditions that countries need to fulfil when entering or forming a currency union. The more these conditions are satisfied, the more suitable countries are for joining a currency union. One of these criteria is the so called business cycle synchronicity (Mongelli, 2002).

A theory that emerged in the 1990s is the endogenous OCA hypothesis. The endogenous OCA hypothesis, proposed by Frankel and Rose (1998), suggests that the formation of a currency union will automatically lead to fulfilment of the OCA criteria, including business cycle convergence. According to this hypothesis, countries do not need to achieve the conditions ex ante, because countries will satisfy the conditions ex post, i.e. after entering a currency union. This thus suggests a self-fulfilling, i.e. endogenous mechanism in creating an optimal currency area. The question arises what the theoretical reasons behind this prediction are.

First, the formation of a currency union could lead to an increase in trade for several reasons. For example, due to the creation of the Euro, bilateral exchange rate volatility is eliminated, transaction costs are reduced and market transparency is increased (Micco et al., 2003). The increase in trade through the formation of a currency union is considered to be the most influential transmission mechanism affecting business cycle synchronicity. For example, when a foreign country is affected by a negative demand shock, this will lead to a decrease in its imports, which decreases output in the domestic country. Therefore, demand shocks in one country, could affect the GDP of other countries, which leads to a higher correlation in business cycles (Otto et al., 2001).

In addition, financial integration is increased due to the formation of the EMU, because of the elimination of exchange rate volatility and the coordination of policies in the financial markets (Kalemli-Ozcan et al, 2010). As a result, investors with different nationalities will invest part of their income in the asset markets of other member countries.

(9)

9 Therefore, when the asset prices in one country increase, this could have a positive welfare effect in other countries, because of their increased holdings of foreign assets. This positive welfare effect could lead to an increase in consumption in both countries (Kose et al., 2003), thus increasing business cycle synchronicity. Indeed, Lane and Milesi-Ferretti (2007) concluded that being a member of the EMU resulted in more cross-country equity holdings between Eurozone countries.

Moreover, financial intermediaries are becoming more interconnected when a currency union is formed. Problems with one financial institution might signal to the public that the whole financial market is in trouble. This contagion effect will reduce the confidence in the financial system, which eventually could lead to a withdrawal of capital from domestic and foreign financial institutions, inducing a recession (Kalemli-Ozcan et al., 2013). Therefore, a negative shock in the financial market can have a negative effect on countries’ economies when they are financially integrated, which makes the business cycles more alike. In the economic crisis of 2007-2008, this contagion effect occurred.

In addition, business cycle alignment can be explained by the supply side. Due to an increase in trade, it could be the case that more domestic inputs are used in foreign final products. An increase in the production of these foreign goods, through for example a positive productivity shock or a demand boom, can lead to positive spill over effects in the production of domestic inputs. As a result the output in both countries is increased, which thus suggests more business cycle synchronicity (Kose et al., 2003).

2.2.2. Reasons for divergence in business cycles

According to the specialization hypothesis, more trade can also lead to divergence in business cycles through specialization. It states that countries will specialize in goods and services in which they have a comparative advantage when trade linkages increase (Liu, 2012). Specialization can also be explained by the New Trade Theory (NTT). It suggests that countries could benefit from trade when they specialize in goods with increasing returns to scale. In addition, the economic geography theory, points out that trade will increase the clustering of business activities at certain locations, which could lead to regional and/or national specialization. Therefore, when countries are hit by shocks, countries may be affected differently, leading to divergence in business cycles.

Moreover, Artis et al. (2008) mentioned labor market rigidities as a cause of specialization. Labor market rigidities lead to more specialization, because differences in the institutional structure of labor markets can give countries a comparative advantage. For

(10)

10 example, if autarky is assumed and two industries experience an asymmetric shock, then countries with higher labor market rigidities find it more difficult to reallocate labor across industries. This results in lower productivity in these countries than in countries with a more flexible labor market. As a consequence the relative prices across countries could change, which could lead to specialization. Countries with an inflexible labor market will specialize in goods from low volatility sectors, because in these sectors there is no need for the reallocation labor. In contrast, countries with a flexible labor market will specialize in goods from high volatility sectors, because they can easily shift labor from one product to the other when market conditions change. Therefore, differences in labor market rigidities are a reason for the fact that trade liberalization could lead to specialization (Cuñat & Melitz, 2005). As a consequence, business cycle synchronicity will decrease. The labor market structures of countries in the Eurozone have remained heterogeneous since 1999 (Merkl & Schmitz, 2010). This thus suggests that the formation of the EMU could have induced more specialization, and thus business cycle divergence.

In addition, in a model developed by Obstfeld (1994), the increase in financial linkages enables countries to hold a globally diversified portfolio of risky assets. As a result, in his model, investors will choose projects that are relatively risky with high profits, instead of safe investments with lower profits. Specialization contains high risk, which thus implies that an increase in financial integration is positively related with specialization, inducing business cycle divergence (Obstfeld, 1994).

2.3 The effect of the financial crisis of 2007-2008 on business cycle

synchronicity in the EMU

Eurozone countries were affected differently by the financial crisis of 2008. Especially the GIIPS countries were hit hard. It is therefore possible that the business cycles of the GIIPS countries and the other Eurozone countries started to diverge. However, it could also be possible that the required reforms in the GIIPS countries have led to business cycle convergence. This section analyses the channels through which the crisis could have had an effect on business cycle synchronicity in the crisis period and the post-crisis period.

In the previous section, it is stated that financial integration can have an effect on business cycle synchronicity in the EMU. This channel was also present during the crisis and could be explained by several events. First, the Greek debt restructuring case in March-April 2012 consisted of a debt exchange of approximately 200 billion euro and a buyback of Greek

(11)

11 government bonds. The restructuring led to a transfer from private investors to Greece of about 100 billion Euros. Since part of Greek government bonds were held by foreign private investors from the Eurozone (Zettelmeyer et al., 2013), the countries of residence of these investors could have been negatively affected. Therefore, problems in one country can affect other countries through cross-country asset holdings, making business cycles more similar during the crisis. It should be noted, however, that the reverse can be true as well which is explained by the following example. The start of the European crisis in 2008 was characterized by large sovereign bond yield spreads between the GIIPS countries and the core, like Germany. The sovereign debt problems in the GIIPS countries increased the risk of default in these countries’ sovereign bonds, resulting in a flight to quality, which increased the demand for German government bonds (Acharya & Steffen, 2015). This in turn increased the prices of German bonds resulting in more welfare for investors holding German bonds. However, due to the flight to quality, the demand for GIIPS government bonds decreased. This led to a decrease in the prices of GIIPS government bonds resulting in lower welfare for investors holding GIIPS bonds. Thus, depending on the composition of holders of GIIPS and German bonds, countries could be affected differently resulting in business cycle divergence. Therefore, the effect of financial linkages on business cycle synchronicity during the crisis is ambiguous.

Moreover, the sovereign debt problems in the GIIPS countries increase the risk that these governments will default. As a result, investors will sell their holdings of government bonds with a high risk of default, and buy government bonds from other countries with lower risk, the so called flight-to-quality explained earlier. This would make it more difficult for high-risk countries to obtain funds. In a situation in which a high-risk country is not a member of a monetary union, the exchange rate would depreciate to offset the flight-to-quality, as it makes its government bonds less expensive. However, since member countries cannot pursue an independent exchange rate policy, the exchange rate will not depreciate to offset the flight to quality. This makes it harder for the government to acquire funds, which could lead to or intensify a sovereign debt crisis (de Grauwe, 2011), leading to lower business cycle synchronicity.

Instead of a change in exchange rates, the GIIPS countries needed to use internal devaluation to restore competitiveness. Due to the wage cuts unit labor costs relative to the Eurozone decreased by 19.4% in Greece, 11.3% in Spain, 12.3% in Ireland and 8.2% in Portugal from 2009 to 2014. This led to an improvement of competitiveness and the current account in the GIIPS countries. However, it did not lead to an increase in productivity and it

(12)

12 induced higher unemployment and a deflationary spiral (Coeuré, 2014), which resulted in lower economic growth. In contrast to the GIIPS countries, other Eurozone economies improved after the crisis (Hallet & Richter, 2014). Therefore, it could be possible that the business cycles of the GIIPS countries and EMU diverged.

Before the crisis, the Eurozone could be divided into two areas, the core countries and the periphery countries. This division is based on the divergence of certain macroeconomic variables between these areas, such as production costs, price level, the balance of payments situation, GDP growth, government debt, bank debt and foreign debt. For example, in the years before the crisis, Germany’s unit labor costs remained quite stable. However, the unit labor costs of Spain increased by 40 percent. In addition, the periphery countries had a higher inflation rate than the target rate of two percent set by the ECB. In contrast, Germany’s inflation rate was lower than the target rate (Landmann, 2011). Since May 2010 four periphery countries obtained financial assistance from the European Union (EU) and the International Monetary Fund (IMF), namely Greece, Ireland, Portugal and Cyprus. The financial assistance is complemented with the requirement that these countries need to implement certain reforms proposed by the Troika (Sapir et al., 2014). In addition, other periphery countries like Spain and Italy have implemented reforms as well. These reforms could lead to more convergence in the characteristics of the Eurozone countries’ economies, which in the end could lead to an increase in business cycle synchronicity.

In addition, almost all the member countries were hit by the crisis at the same time. The timing and strength of domestic policy measures in the recovery phase, however, were different for several countries. The reason for this is that the debt levels of Eurozone countries differed substantially before the crisis. Therefore, countries with a low Debt-to-GDP-ratio could use expansionary fiscal policy. In contrast, countries with a high ratio, such as in the GIIPS countries, needed to use austerity measures to continue to meet the conditions of the Stability and Growth Pact (SGP), which intensified the crisis in those countries. In addition, the weak structural foundations of those countries, such as tax evasion and corruption, increased the severity of the crisis even more (Gächter et al., 2012). This thus suggests that the business cycles of countries with an initial low debt level diverged from those of countries with a high debt level.

Thus, based on the theoretical reasons it is not clear if the business cycles of the Eurozone countries have converged or diverged after the recent economic crisis.

(13)

13

2.4 Empirical research on business cycle synchronicity

A significant amount of empirical research has been performed about business cycle synchronicity in the EMU. Therefore, in this section a table is presented in which previous empirical research is summarized in chronological order. The articles presented in the table focus on two issues, namely, whether the business cycles in the Eurozone converged or diverged in a certain time period and which variables had an effect on this convergence or divergence. The dependent variable in the table is a measure of business cycle synchronicity. When more equations or models were used in a certain paper, then the equation or model with the best fit, highest significance or robustness is presented in the table.

Table 1: Literature overview

Authors Data source

& time Period

Synchronicity measure

Main findings Regression method if applicable Independent Variables Relationship Artis & Zhang (1997) Monthly Industrial Production (IP) data from the OECD, 1961-1993 Cross-correlation with Germany and US1,5,6

Results are not dependent on filter method Formation of the ERM led to convergence - - - Koo & Wynne (2000) Annual GDP data from Penn World Tables, 1963-1992 Pairwise correlation3 Higher correlation between the US states than between EU members - - - Inklaar & de Haan (2001) Monthly IP data from the OECD, 1960-1997 Cross-correlation with Germany1 No relationship between exchange rate volatility and business cycle synchronicity Formation of the ERM did not led to convergence - - - Fidrmuc (2004) Quarterly GDP and IP data from the International Pairwise correlation2 Business cycle synchronicity depends on the composition of 2SLS Bilateral trade indicator Intra-Insignificant

(14)

14 Financial Statistics database, 1991-2001 trade

Plans about the formation of the EMU led to business cycle convergence Central Eastern European Countries (CEECs) converged to the EU member countries Industrial Trade indicator (ITT) Positive Darvas et al. (2005) Monthly and annually real GDP and unemployment from the OECD, 1964-2003 Pairwise correlation1,3,7 Fiscal convergence leads to business cycle synchronicity Results are not dependent on filter method

OLS & IV Measure of fiscal divergence Negative Mink et al. (2007) The International Financial Statistics database, 1970-2005 Measure of synchronicity4 Measure of amplitude4 Overall synchronicity is high in the Eurozone, but does not increase. Synchronicity between countries and the Eurozone cycle varies over time and is different for several countries

-

(15)

15 Artis et al. (2008) Quarterly real GDP from the International Financial Statistics database, 1970-2004 Contempo-raneous correlation1,2 The government finance criteria of the Maastricht Treaty led to convergence Some channels lead to convergence and some lead to divergence. Thus, the enlargement of the EMU may have an ambiguous effect on synchronicity IV Bilateral Trade Indicator Intra-Industrial Industry Trade Indicator FDI intensity indicator Labor market rigidities Measure of fiscal divergence EMU-dummy Positive Positive Positive Negative Negative Positive Gonçalves et al. (2009) World Economic Outlook database, 1980-2007 Pairwise correlation1 The formation of the EMU led to convergence in business cycles Differences- in-differences EMU dummy Bilateral trade indicator Positive Negative Degiannakis et al. (2011) Quarterly GDP (dataset unknown), 1980-2009 Time-varying correlation index1,3 Until 2007: convergence After 2007: Belgium, the Netherlands and Ireland diverged - - -

(16)

16 Gächter et al. (2012) Quarterly real GDP and IP data from Eurostat, 1995Q1-2011Q3 and 2000-2012 respectively Correlation coefficient1 Standard deviation1 Results show divergence since the crisis for both measures Parallels are observed between the crisis of 2004 and the recent crisis of 2007-2008 - - - Crespo-Cuaresma & Fernández-Amador (2013) Quarterly real GDP from the OECD and Eurostat, 1960-2008 Cross-country weighted standard deviation8 Seventies and early eighties: convergence Late eighties: divergence Nineties: convergence Start of the EMU: divergence Since 2004: convergence Policy coordination is needed for a currency union to be sustainable - - - Gächter & Riedl (2014) Annual real GDP from Eurostat, 1993-2011 Time-varying correlation index1,3 The formation of the EMU led to convergence in business cycles Strong relationship between trade and business cycle GMM EMU dummy Bilateral trade indicator Industrial specialization Measure of fiscal Positive Positive Insignificant Negative

(17)

17 synchronicity divergence EU dummy Not significant Hallet & Richter (2014) Quarterly real GDP from the OECD, 1970Q1-2012Q3 Coherence8 No common GDP growth pattern for the GIIPS countries Italy, Spain, Ireland and Greece eventually converged to the Eurozone after the crisis

OLS Lagged dependent variables Result differs per frequency and country

¹ Using the Hodrick-Prescott filter ² Using the Fourth-differences of logs ³ Using the Baxter-King filter

4

Using the Christiano-Fitzgerald filter

5

Using the Phase-Average-Trend (PAT) method

6

Using linear trending

7

Using first-differences

8

Using the Kalman Filter

It can be seen from the table that the results of previous empirical research are mixed. For example, Artis and Zhang (1997) found that the introduction of the Exchange Rate Mechanism in 1979 led to business cycle convergence between member countries. This result is contradicted by Inklaar and de Haan (2001). In addition, a few papers such as Gonçalves et al. (2009) and Gächter and Riedl (2014) found that the introduction of the Euro led to business cycle convergence in EMU member countries. However, Artis et al. (2008) shows that there are some factors that affect business cycle synchronicity in a positive way and some in a negative way. They therefore suggest that becoming a member of the EMU could lead to not only convergence, but also divergence of business cycles, which is in line with the theoretical section of this paper.

Also, theory suggests that bilateral trade should be an important factor driving business cycle synchronicity. However, empirical research shows mixed results. Artis et al. (2008) and Gächter et al. (2004) found a positive relationship between bilateral trade and business cycle synchronicity, which suggests that the endogenous OCA hypothesis holds. In

(18)

18 contrast, Gonçalves et al. (2009) found a negative relationship and Fidrmuc (2004) no significant relationship.

In addition, the studies of Fidrmuc (2004) and Artis et al. (2008) include an indicator of intra-industrial trade (IIT). This is a continuous measure of specialization, in which a value of zero means full specialization between countries. On the other hand, a value of one means that there is only intra-industrial trade, which means that no specialization is present. In addition, a value between zero and one indicates a combination of intra-industrial trade and inter-industrial trade. Both studies found a positive relationship between IIT and business cycle correlation. This suggests a negative relationship between specialization and business cycle correlation, which is also expected by theory. In addition, Artis et al. (2008) added a measure of labor market rigidity to the regression. It can be seen from section 2.2.2 that this is also a potential source of specialization. As can be seen from the table, Artis et al. (2008) found a positive relationship. Thus, labor market rigidities result in more specialization, and this in turn will lead to less business cycle synchronicity. These findings are contradicted by Gächter and Riedl (2014), who studied the effect of the creation of the Euro on business cycle alignment of EMU member countries, with specialization as a control variable. The results showed that no significant relationship exists between business cycle patterns and specialization.

Moreover, Darvas et al. (2005), Artis et al. (2008) and Gächter and Riedl (2014) studied the effect of fiscal divergence on business cycle synchronicity. They all found a negative relationship, suggesting that differing fiscal policies could lead to divergence in business cycles between countries. Since the fiscal stance of countries in the EMU differed at the start of the crisis of 2008, this could thus have led to divergence in business cycles during or after the crisis.

Since the topic about the effect of the crisis on business cycle synchronicity is quite new, not much empirical research about this topic is available. Hallet & Richter (2014) showed the effect of the crisis on the coherence of business cycles of the GIIPS and Eurozone countries using a time-varying spectra analysis. They defined coherence as the link between two business cycles at a certain frequency. When looking at the coherence between GIIPS and the Eurozone, they found that the Eurozone and Italy, Spain, Ireland and Greece converged after the recent crisis. However, the coherence of Portugal and the Eurozone did not change. In contrast, Gächter et al. (2012), studied the cross-country business cycle similarities of seventeen Eurozone countries, using correlation coefficients and the standard deviations of

(19)

19 GDP and industrial production as measures. Their results show that the crisis led to divergence in Portugal, Spain, Ireland and Greece.

In addition, Degiannakis et al. (2011), estimated the correlation of the business cycle from 14 EU countries with the combined EMU12 business cycle for the period 1980-2009. Their results show that for most countries, the convergence that was present before the crisis, stabilized during 2008 and 2009. However, the correlation coefficient of Belgium, the Netherlands and Ireland decreased in 2009. This could be explained by the study of Gayer (2007), who estimated the cross-country correlation of GDP and industrial production of Eurozone countries. The author found that the early recovery phase of a crisis is often characterized by a sharp decline in business cycle similarity. After this phase, the synchronization increases again. It could be possible that the results of Belgium, the Netherlands and Ireland found by Degiannakis et al. (2011) are subjected to this short drop in synchronicity. Their data includes only two years of the crisis period. Therefore, inclusion of more years is needed to see if synchronization is increased after the early recovery phase or if countries experience a prolonged detachment.

3. Empirical application

This chapter is divided into four parts. First, the methodology section discusses the way in which the research question of this paper will be answered. Then, a description of the data is given, which is followed by the overview of the obtained results. After that, the results will be discussed.

3.1 Methodology

This section discusses which method is used to estimate the effect of the crisis on business cycle synchronicity in the EMU. The exposition of the methodology of this paper is divided into six parts. First, it is evaluated which filter method is desirable to decompose the cyclical component of GDP from the trend component. Then, for the measurement of business cycle synchronicity the approach of Bordon et al. (2013) is analysed. After that, the method for measuring the Eurozone’s reference cycle is discussed. Then, the regression model used to estimate the effect of the crisis on business cycle synchronicity is presented. Finally, recursive and rolling regression analysis is used to test how the parameters of the model change around the crisis.

(20)

20 3.1.1 Filter method

To develop a measure of business cycle synchronicity it is first necessary to give a definition to the notion of business cycle. Two definitions are commonly used in the literature, namely the ‘classical business cycle’ and the ‘deviation business cycle’. The identification of a classical business cycle is based on the determination of turning points. A peak is identified as the point that is followed by a decrease in GDP in absolute terms and a trough by the point that is followed by an increase. In contrast, the deviation cycle is based on extracting the cyclical component of GDP from the trend component. Since classical cycles are quite rare in the European growth economies, in this paper the deviation cycle is studied (Artis, 2003).

From table 1 it can be seen that there are several filter methods available to estimate the deviation cycle. In the literature there is some disagreement about the effect of the use of different filters on the outcomes. Some authors like Artis and Zhang (1997) and Darvas et al. (2005) found that their outcomes were not sensitive to the filter method used. Bordon et al. (2013) use four filters in their study, namely the Hodrick-Prescott (HP) filter, the Baxter-King (BK) filter, the Christiano-Fitzgerald (CF) filter and the Butterworth (BW) filter. They also conclude that their results do not depend on which filter method is used. In contrast, Massmann and Mitchell (2003) concluded that their results were dependent on the filter method. Comparing the results of several filters is beyond the scope of this research. Therefore, one of the filters used by Bordon et al. will be used in this paper.

The most commonly used filter in the literature is the HP-filter. It decomposes the growth (or trend) component, 𝑔𝑡, from the cyclical component, 𝑐𝑡, by minimizing the variance

of the sum of the squared deviation from the growth component plus a penalty for the variation in the sum of squared second differences in the growth component (Mise et al., 2002). The main drawback of the HP-filter for the purpose of this paper is the end point estimation problem. This means that the HP-filter can only be useful for a sample with an infinite time interval or for approximations of the middle part of a sample with a long time interval (Mise et al., 2002). Indeed, Baxter and King (1999) recommend that when interpreting the results of the HP-filter with quarterly data, 12 quarters should be dropped at the beginning and end of the time interval since they do not produce useful results. In this paper, the pre-crisis period (1996-2007) is compared with the crisis period and the post-crisis period (2008-2015). Using the HP-filter thus means that for the crisis and post-crisis period only data can be used for the time period 2008-2011. This will mean a substantial loss of data. Therefore, the HP-filter is not seen as optimal for the goal of this paper.

(21)

21 The same is true for the band-pass filter developed by Baxter & King (1999). This filter is a so-called symmetric moving average filter, which means that the trend component is extracted from GDP using observations from the time period before and after the observation for which the trend is calculated. At the endpoints this is not possible, leading to non-optimal results at the tails (Nilsson & Gyomai, 2007).

A modification of the BK-filter is offered in Christiano and Fitzgerald (2003). Instead of using a symmetric filter, they use an asymmetric filter and for decomposing the trend component from the cyclical component the complete time series is used. This reduces the end-point estimation problem. In addition, for practical purposes it outperforms the filter of Baxter and King (Nilsson & Gyomai, 2007). In addition, Issever-Grochova & Rozmahel (2015) found that the CF-filter is the most similar to an ideal filter in comparison with the HP-filter, the BK-filter and the BW-filter. Therefore, in this paper the CF-filter will be used. Burns and Mitchell (1946) stated that the duration of business cycles last between 1.5 and 8 years or equivalently 6 and 32 quarters respectively. Therefore, these values are used as the minimum and maximum of the cycle’s duration.

3.1.2 The business cycle synchronicity measure

The table in section 2.4 showed that the majority of studies used the correlation coefficient as a measure of business cycle synchronicity. Therefore, using this measure makes comparison with other studies easier. However, certain drawbacks exist. First, correlation coefficients cannot be estimated on a per-observation basis. Instead, they need to be estimated over a period of time (Wälti, 2009). Therefore, when quarterly data of GDP is available, the correlation coefficient could be calculated for example for a period of five years (Flood & Rose, 2009). However, in this study the effect of the crisis on business cycle synchronicity is measured. The start of the crisis was in 2008. Therefore, using a five year period to calculate the correlation coefficient means that relatively few data for the synchronicity measure is available. This would make it difficult to draw a conclusion. In addition, selecting a time interval for the calculation of the correlation coefficients is arbitrary and the outcome could be dependent on the particular choice of the time interval (Wälti, 2009).

Moreover, for a synchronicity measure to be sufficient the measure needs to take into account two properties of synchronicity, namely the sign and direction of the relationship between the output gaps of two countries and differences in amplitude between the output gaps of two countries. The correlation coefficient does not take proper account of these two features. For example, it could be the case that the correlation of the output gaps of two

(22)

22 countries is equal to unity while the amplitude differs substantially. This situation is sketched in figure 1. In this case, the correlation coefficient would suggest that a one-size-fits-all approach by the ECB is suitable for these two countries. However, the figure shows that this is not the case. The business cycle of country 1 fluctuates by a larger extent than the business cycle of country 2. As a consequence, country 1 needs a higher policy rate in booms and a lower policy rate in economic slowdowns when compared with the ideal policy rate for country 2. This example thus shows that the correlation coefficient needs to be interpreted carefully (Mink et al., 2012).

Figure 1: The output gaps of two countries. Adapted from Mink et al., 2012.

To overcome the shortcomings of the correlation coefficient, Mink et al. (2007), proposed two measures that estimate the sign of the relationship and amplitude separately. These measures can be estimated for a group of countries or for an individual country. The measures to compare individual countries with a reference cycle are calculated by the following equations

(1) Sign:

𝜑

_𝑖,𝑡

=

𝑔𝑖,𝑡𝑔𝑟,𝑡 |𝑔_𝑖,𝑡𝑔_𝑟,𝑡| (2) Amplitude:

𝛾

_𝑖,𝑡

= −

|𝑔𝑖,𝑡−𝑔𝑟,𝑡|

|𝑔_𝑖,𝑡|

where 𝑔_𝑖,𝑡 stands for the output gap of country i at time t and 𝑔_𝑟,𝑡 stands for the output gap of the Eurozone reference at time t. The first equation measures whether the output gap of country i and the reference output gap have the same sign at time t. It has a value of 1 when the output gaps of country i and the reference cycle have the same sign at time t, a value of 0 when one of the output gaps is equal to zero at time t and a value of -1 when the output gaps

(23)

23 of country i and the reference cycle have opposite signs at time t. The second equation measures the extent to which the amplitudes of the output gaps are related to each other. It has a value of 0 when the amplitudes are equal and turns negative when the amplitudes differ from each other.

Although the correlation coefficient is easier to understand, the approach of Mink et al. (2007) is more informative than the correlation coefficient. However, the disadvantage of this approach is that it is necessary to focus on two measures instead of one. In addition, according to Bordon et al. (2013), an encompassing synchronicity measure needs to satisfy two conditions.

First, the measure needs to be bounded, or in other words, it should not take on extreme values. Extreme values can have a significant impact on the mean and standard deviation of the synchronicity measure. For example, suppose that there are five countries in a monetary union, four with a small economy and one with a large economy. When synchronicity is calculated using aggregate GDP of the monetary union as a reference, than it is reasonable to think that the country with a large economy has a high synchronicity level and the countries with a small economy have a low synchronicity level. The mean could be distorted because the average synchronicity could be high, although most countries have a low synchronicity level. A distorted mean and standard deviation can lead to distorted results of regression models. In addition, an unbounded measure has no limits. A specific number does not provide much information, because the range of the measure is unknown. Therefore, extreme values should be avoided.

Second, an optimal measure needs to range between the following values: a value of 1 when two cycles are perfectly synchronal, a value of zero when there is no relationship between two cycles and a value of -1 when two cycles are the opposite of each other. These two conditions make it easier to interpret, compare and evaluate the results. The amplitude measure, γ, of Mink et al. (2007) can take on values that are lower than -1. Therefore, Mink et al. (2007) used standardization to attain bounded variables. However, as already stated, extreme values can have a significant impact on the mean and standard deviation of a measure. The mean and standard deviation is used in the process of standardization. Therefore, standardization of an unbounded measure to the [-1,1] interval can have a distortionary effect on the results and reduces the usefulness of equation (2) (Bordon et al., 2013).

A measure that overcomes these problems is proposed by Bordon et al. (2013) and is presented by the following equation.

(24)

24

(3)

𝜗

_𝑖,𝑡

= 1 −

(𝑔𝑖,𝑡−𝑔𝑟,𝑡)²

𝑔_𝑖,𝑡2 +𝑔_𝑟,𝑡2

where 𝜗𝑖,𝑡 stands for a measure of business cycle synchronicity of country i and the Eurozone

reference, 𝑔_𝑖,𝑡 for the output gap of country t at time t and 𝑔_𝑟,𝑡 for the output gap of the Eurozone reference at time t. A high and positive value of 𝜗_𝑖,𝑡, means that a high synchronicity level between the business cycles of country i and the Eurozone reference exist. A low and negative value, however, means a low level of synchronicity.

This measure takes into account not only the sign of the business cycle synchronicity between a country and the reference business cycle, but also the difference in amplitude between the output gaps of a country with the reference business cycle. In addition, instead of using two measures, it combines the sign and amplitude measure into one measure. It also satisfies the two conditions proposed by Bordon et al. (2013). Therefore, the measure of Bordon et al. (2013) will be used in this paper.

3.1.3 The Eurozone reference cycle

There are several authors who study whether a European business cycle exists. For example, Artis (2003) found that it is difficult to assume one single European business cycle, since some countries follow the same business cycle and others do not. In addition, Camacho et al. (2006) did not find a European business cycle either. However, they point out that the results of the estimation of a European business cycle should be interpreted carefully when an European business cycle is estimated using the weighted average of the business cycles of the European countries, the business cycle of the largest European economy or the cycle of a mutual factor. Another way to estimate the Eurozone business cycle is to choose the median output gap in every period t of the output gaps of the Eurozone countries. One half of the data is higher than the median, and one half of the data is lower than the median. Therefore, in every period the Eurozone output gap, or equivalently the median output gap, lays as close as possible to the output gaps of each country (Mink et al., 2007). This is the method that is used in this paper.

3.1.4 Regression analysis

The methodology that is used in this study consists of estimating and comparing a regression model for two periods. Since quarterly data for most Eurozone countries is available from 1996, this year will be the start of the first period. The financial crisis in Europe started at the

(25)

25 beginning of 2008 (European Commission, 2009). Therefore, 2007Q4 will be the end of the pre-crisis period. Thus, the first period, 1996Q1- 2007Q4, is the period from 1996 until the crisis in the Eurozone has started. The second period, 2008Q1-2015Q1, will be the crisis and post-crisis period. For notational brevity, in the remainder of the paper this period will be called the post-crisis period. Kappler & Sachs (2012) proposed to include a linear time trend t into the regression model to estimate whether the correlation coefficient increased or not. This approach is followed in this paper, however, with the synchronicity measure of Bordon et al. (2013), which then gives the following equation

(6)

𝜗

_𝑖,𝑡

= 𝛼 + 𝛽𝑡 + 𝜀

_𝑖,𝑡

where t is a linear time trend and 𝜀𝑖,𝑡 is the error term. A positive value of beta means that the

business cycles of the Eurozone converged. In contrast, a negative value indicates that divergence has occurred. In addition, the size of the coefficient measures the extent of divergence or convergence (Kappler & Sachs, 2012).

It can be expected that the synchronicity in year t could depend on the synchronicity in year t-1. For example, when business cycle synchronicity increased between the member countries of the Eurozone in t-1, then it is reasonable to think that, for instance, the governments of these countries will implement fiscal policies that are more alike since their economic situation is more similar. This could in turn lead to higher business cycle synchronicity in period t, implying a positive relationship between synchronicity in period t-1 and period t. Therefore, this paper also includes a lagged variable to account for this dynamic effect. The following regression model will therefore be estimated.

(7)

𝜗

_𝑖,𝑡

= 𝛼 + 𝛽𝑡 + 𝛾𝜗

_{𝑖,𝑡−1}

+ 𝜀

_𝑖,𝑡

Where again t is the linear time trend, 𝜗𝑖,𝑡−1 is the lagged dependent variable and 𝜀𝑖,𝑡 is the

error term.

3.1.5 Recursive analysis of time series

An assumption that is made when estimating a model over time is that the estimated coefficients are constant over the considered time period (Zivot & Wang, 2003). For the purpose of this paper it is also interesting to know whether the parameters changed significantly in subsequent years. Therefore, recursive regression analysis is used to test for the stability of the parameters of the model and to see whether the business cycles in the Eurozone converged or diverged at different points in time. In this paper, the starting period is fixed at 1996Q1 and ends at 2001Q1, to see what happens in the periods before and after the

(26)

26 crisis. After that, the regression model is estimated for the time periods with the starting point at 1996Q1 and a growing window size of one quarter. For example, the first sample period is 1996Q1-2001Q1, the second sample period is 1996Q1-2001Q2, the third sample period is 1996Q1-2001Q3 and this goes on until the end period of 2015Q1 is reached. This method can show how the crisis affected business cycle synchronicity in different periods and how the parameters change when more periods are included in the regression.

3.1.6 Rolling regression of time series

The problem with recursive regression analysis is that the starting period is held constant. Therefore, adding a quarter to a regression model has more effect with a small window size compared to adding a quarter to a regression model with a large window size. For example, a in the regression model a higher weight is attached to an additional quarter when the window size is 1996q1-2001q1, in comparison with a window size of 1996q1-2015q1. Rolling regression analysis is therefore included in this paper as well. With rolling regression the window size is held constant. Thus, adding a quarter in one window has the same weight as adding a quarter in another window. This could make it easier to analyse the effect of specific years. The window size in this paper is 5 years or equivalently 20 quarters.

3.2 Data description

Twelve Eurozone countries are included in the sample, namely Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain. Slovenia, Malta, Lithuania, Latvia, Estonia, Cyprus and Slovakia are not included, because they became a member of the Eurozone just before or during the crisis period. Therefore, for these countries it is not possible to calculate the effect of the crisis on business cycle synchronicity separately from the effect of entering the EMU. For all the countries we use quarterly real GDP data from Eurostat for the period 1996Q1-2015Q1 to estimate the business cycle, with Luxembourg and Ireland as an exception, because quarterly GDP data for Luxembourg is only available from 2000Q1 and for Ireland only from 1997q1. This results in a strongly balanced dataset.

(27)

27

3.3 Results

3.3.1 Descriptive analysis

The following figure shows how the business cycle synchronicity measure changed over time for each country. The synchronicity measure fluctuates quite heavily for some countries. For other countries, such as France, Finland, Germany and the Netherlands, the synchronicity measure fluctuates less extensively. The graphs of most countries show a drop in synchronicity around 2002-2003 and 2006-2007. It could be the case that the drops in synchronicity in 2002-2003 and 2007 are related to the dotcom-crisis and the current global financial crisis, respectively. In addition, from 2010 onwards in most countries, synchronicity seems to fluctuate more. This could be due to the European sovereign debt crisis of 2010.

Figure 2: Business cycle synchronicity with the Reference output gaps per country.

The following table makes the comparison between the pre-crisis and post-crisis periods of the financial crisis clearer. The average level of the synchronicity measure is calculated for the entire sample period, the pre-crisis period and the post-crisis period. In addition, a

mean--1 -.50 .5 1 -1 -.5 0 .5 1 -1 -.50 .51 -1 -.50 .51 -1 -.5 0 .51 -1 -.5 0 .51 19 9 5 20 0 0 19 9 8 20 0 3 20 0 5 20 0 8 20 1 0 20 1 3 20 1 5 19 9 5 20 0 0 19 9 8 20 0 3 20 0 5 20 0 8 20 1 0 20 1 3 20 1 5 Austria Belgium Finland France Germany Greece Ireland Italy Luxembourg Netherlands Portugal Spain syn ch ro n ici ty Year Graphs by Country

(28)

28 comparison test is performed to evaluate whether the difference between the means of the pre-crisis and post-pre-crisis periods is significantly different from zero. As can be seen from the table, especially Greece, Luxembourg and Portugal have low average levels of synchronicity during the entire period. A reason for this could be that Greece and Portugal have the least developed economies (or lowest per capita income) of the Eurozone-12 countries and Luxembourg the most developed economy (or highest per capita income). Their business cycles could therefore differ from most countries, leading to low synchronicity levels. In addition, in most countries the average synchronicity decreased when comparing pre-crisis period to the post-crisis period. This could suggest that the reasons for divergence explained in section 2.3 outweigh the reasons for convergence. However, the difference is not significant for each country.

Countries 1996Q1-2015Q1 average Pre-crisis period average (1) Post-crisis period average (2) Difference: (2)-(1) Significance Of the difference (p-value) Austria 0.670 0.696 0.628 -0.068 0.576 Belgium 0.579 0.478 0.746 0.269 0.058 Finland 0.515 0.518 0.511 -0.007 0.957 France 0.584 0.600 0.557 -0.043 0.752 Germany 0.718 0.720 0.715 -0.004 0.964 Greece -0.054 -0.080 -0.011 0.067 0.639 Ireland 0.519 0.446 0.630 0.184 0.108 Italy 0.670 0.610 0.769 0.159 0.206 Luxembourg 0.419 0.500 0.330 -0.170 0.220 Netherlands 0.670 0.635 0.727 0.092 0.450 Portugal 0.416 0.430 0.393 -0.036 0.803 Spain 0.692 0.718 0.651 -0.067 0.545

(29)

29 3.3.2 Regression analysis

Table 3 demonstrates the results of the regression models of the pre-crisis and the post-crisis period using an Ordinary Least Squares (OLS) model and a Fixed Effects (FE) model. Theoretically, the FE-model is reasonable to use for the purpose of this paper, because each country has individual characteristics that in principle lead to omitted variable bias when excluded from the regression. The standard errors are presented within parentheses and are robust in the case of the OLS model. The significance levels are indicated by asterisks. The table shows that for both models, each variable included in the regression is statistically significant.

The time trend variable shows a positive coefficient in the pre-crisis period, which indicates convergence of business cycles in this period. In the post-crisis period, however, the coefficient has a negative sign, which indicates divergence of business cycles in this period. Using a chi-squared test, a statistically significant difference is found between the coefficients of the time trend variable for the pre-crisis and the post-crisis period at the 1 percent significance level. This thus implies that the crisis had a negative effect on business cycle synchronicity in the Eurozone. The difference (0.01243=0.00623+0.00620 and

0.01488=0.00728+0.00760 for OLS and FE respectively), however, is quite small. The

convergence in the pre-crisis period could mean that the channels behind the endogenous OCA-theory outweigh the channels behind the specialization hypothesis. In addition, in the post-crisis period, the channels through which the crisis affects the synchronicity levels negatively could outweigh the channels that affect the synchronicity levels positively.

In addition, as expected the first lag of the synchronicity measure has a positive sign. In contrast, a regression analysis is also performed including a second lag of the synchronicity measure. The results show a negative sign for the second lag. This could indicate that the synchronicity measure follows a cyclical pattern, and that an increase in synchronicity in period t-2 is followed by a decrease in synchronicity in period t. However, the coefficient of the second lag is not significant in the OLS regression and is therefore not included in the table.

(30)

30 Variable OLS pre-crisis period OLS post-crisis period FE pre-crisis period FE post-crisis period t 0.00623*** (0.00174) -0.00620** (0.00267) 0.00728*** (0.0016) -.00760*** (0.00284) 𝝑𝒊,𝒕−𝟏 0.51554*** (0.04539) 0.62721*** (0.06037) 0.43200*** (0.03877) 0.55324*** (0.04708) constant 0.09551* (0.05437) 0.58428*** (0.18331) 0.11158** (0.04497) 0.71459*** (0.19067) n 544 348 544 348

Table 3: Regression model using Ordinary Least Squares, Fixed Effects and Random Effects

* means significant at the 10% level ** means significant at the 5% level *** means significant at the 1% level

3.3.3 Recursive analysis of time series

Applying recursive regression on the data gives the recursive regression coefficients presented in figure 3. The quarters on the x-axis represent the end of the window size for each time interval. For each window the graphs display the recursive coefficients of the time trend for each country.

The graphs show that the coefficient is quite stable over time. However, the patterns differ per country. Most countries are characterized by a declining coefficient of the time trend. For most countries this means a decrease in the pace of convergence, since the coefficient is positive. The coefficient of the time trend switches from positive to negative in Greece when the years 2010 and 2011 are included in the regression and in Portugal when the year 2010 is included in the regression. This could mean that the sovereign debt crisis of 2010 had a negative effect on the synchronicity level of Greece and Portugal. After that, the coefficient turns positive again, indicating convergence. Reforms implemented in Greece and Portugal could be a reason for this. In addition, Luxembourg has a negative coefficient in most years, indicating divergence. From these graphs it is difficult to draw a conclusion about the exact cause of divergence. It seems like a decrease in the pace of convergence or divergence was present after the crisis in most countries. However, this could also be the result of factors other than the crisis.

(31)

31

Figure 3: Recursive coefficient of the time trend per country

3.3.4 Rolling regression of time series

The results of the rolling regression are presented in figure 4. The quarters on the x-axis represent the end quarter of the rolling window. For each window the graphs display the rolling coefficients of the time trend. A red reference line is drawn where the first year of the post-crisis period is included in the regression model.

As can be seen from the graphs, the time trend coefficient is not stable over time for each country. Austria, Germany, Luxembourg, Finland, Ireland and Portugal are characterized with a trough in the year 2007, indicating a decrease in the pace of convergence for some

(32)

32 countries and divergence for other countries at the start of the crisis. For these countries, the time trend coefficient seems to decrease just before 2008q1. After 2008, the time trend coefficient increases for most countries, suggesting that the crisis of 2008 in Europe had a positive effect on synchronicity.

In contrast, including the quarters of 2010 in the regression model has a negative effect on the time trend coefficient of Austria, Finland, Germany, Greece, Ireland, Italy, Luxembourg and Portugal. This could suggest that the European sovereign debt crisis of 2010 had a negative effect on business cycle synchronicity. In most countries synchronicity started to increase at the end of the post-crisis period. This could suggest that the reforms implemented in the periphery countries brought the business cycles of the Eurozone countries closer together.

(33)

33

Figure 4: Rolling coefficient of the time trend per country

3.4 Discussion of the empirical results

The results obtained in this paper should be carefully interpreted. For example, it should be noted that the results could depend on which filter method is used. There is no filter method that extracts the cyclical component of the business cycle perfectly, because these methods are based on certain assumptions. It is therefore possible that the results obtained in this paper are subjected to the choice of filter. The CF-filter is compared to the ideal filter for each country, which can be found in the appendix (A1). Although the CF-filter and the ideal filter

(34)

34 follow the same path, they are not completely equal to each other. Therefore, the measurement of synchronicity could be subjected to an estimation error.

In addition, the results could depend on which synchronicity measure is used. It could be the case that the correlation coefficients and measures of Mink et al. (2007) lead to different results. However, it is beyond the scope of this research to test whether the results change when other synchronicity measures are used.

Furthermore, the Eurozone reference cycle could be calculated in multiple ways. Therefore, to check the robustness of the results, the OLS and FE regression models are also estimated when aggregate Eurozone-12 data from Eurostat is used to calculate the reference output gap. The results can be found in the appendix (A2). The pre-crisis period is again characterized by convergence in business cycles (positive time trend coefficient). In addition, the post-crisis period is characterized by divergence in business cycles (negative time trend coefficient). The chi-squared test shows that in both models a significant difference exists between the time trend coefficients of the pre- and post-crisis period at a significance level of 1 percent. This thus suggests that the results are independent of whether the median output gap is used or the Eurozone output gap.

Moreover, it could be possible that the factors leading to divergence mainly have an effect on synchronicity in the short run and that the factors leading to convergence have an effect on synchronicity in the long run. For example, it is reasonable to think that it takes a significant amount of time before the reforms implemented in the GIIPS countries have an effect on business cycle synchronicity. In contrast, the flight to quality which results in divergence could have a more immediate effect. To study this difference, more years need to be included in the sample. In addition, Gayer (2007) found that the early recovery phase of a crisis is often characterized by a sharp decline in business cycle synchronicity. After this phase, synchronicity increases again. The question arises how long this phase persists. These two topics could be interesting for further research.

The effect of the financial crisis of 2007-2008 on business cycle synchronicity in the EMU