Financial Cycle Synchronization in the Eurozone
MSc Thesis
Name: Sebastiaan van der Weide
Student number: 10423249
Email address: sebas_weide@hotmail.com
Date: 13-‐07-‐2017
Name of supervisor: Péter Foldvari
Second reader: Dirk Veestraeten
Statement of Originality
This document is written by Sebastiaan van der Weide 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.
Abstract
This thesis performs a frequency based filter analysis to construct synthetic financial cycles for the core euro-‐area countries. The main goal of this thesis is to assess the degree of financial cycle synchronization within the Eurozone. Using concordance statistics, this thesis finds that there are several cases of significant synchronization. The outcome of the analysis also suggests that there is an increase in the degree of synchronization over time. Another result of this analysis is that financial cycles tend to converge during stress periods. The analysis in this thesis is complemented by a comparison with the business cycle, that finds a lower level of synchronization than expected.
Table of contents 1. Introduction ... 5 2. Literature review ... 7 2.1 Financial cycle ... 7 2.2 Synchronization ... 9
3. Methodology and data ... 10
3.1 Methodology ... 10
3.1.1 Constructing synthetic cycles ... 10
3.1.1 Cycle synchronization measures ... 12
3.2 Variable selection and data ... 14
4. Financial cycles in the core euro-‐area countries ... 15
4.1 Country-‐specific financial cycles ... 15
4.2 Financial cycle characteristics ... 18
5. Synchronization of financial cycles ... 23
5.1 Concordance statistics ... 23
5.2 Dispersion measure ... 26
6. Financial cycle vs. business cycle ... 28
6.1 Construction of the business cycle ... 28
6.2 Comparison and synchronization ... 29
7. Conclusion ... 33
References ... 36
Appendix A: Individual filtered time series ... 39
Appendix B: Financial cycles ... 42
Appendix C: Business cycles ... 45
Appendix D: Business cycles vs. financial cycles ... 48
Appendix E: Cycle comparison ... 51
1. Introduction
There is a pattern in the behaviour of certain financial variables that reflects the build-‐up of systemic risk in the financial system. This pattern is called the financial cycle (Claessens et al., 2011). Financial variables like credit, housing prices, and equity together form a cycle that is linked to the procyclicality of the financial system (Borio, 2012). The interactions between these financial variables can amplify economic fluctuations and lead to large economic downturns. Peaks of the financial cycle are associated with financial crises and the recessions that follow during the contraction phase of the cycle are known to be severe (Borio, 2012).
Although not a new concept, the financial cycle has only in recent years regained the interest of economists. The global financial crisis demonstrated the relevance of the financial cycle by showing that a build-‐up in systemic risk led to an unwinding of financial imbalances. Excessive risk taking and availability of cheap credit led to an accumulation of financial imbalances, up to an unsustainable point. It is crucial for the understanding of business fluctuations and their policy challenges to include the financial cycle. Economists have since tried hard to incorporate these factors into standard macroeconomic models (Borio, 2012). The macroeconomic costs of busts for the overall economy alone are validation enough for further research on the financial cycle. Understanding the financial cycle could enable policy makers to make predictions about its future course and maybe even anticipate future crises. The development of a generally accepted synthetic measure for the financial cycle could function as an early warning system for too much pressure in the financial system.
The global economy is highly integrated; it is thus important to see country-‐specific financial cycles not as isolated cycles. Financial cycles interact across countries, at times proceed in sync, and at other times proceed in different speeds across countries (Borio, 2012). This synchronization is driven by strong linkages between financial markets across countries. Financial variables such as credit and housing prices tend to co-‐vary across countries (Claessens et al., 2011). The main research question of this thesis will be: To what degree is there synchronization of financial cycles in the Eurozone? The choice for the Eurozone countries as the subject of analysis is both practical and logical. A relatively high degree of policy homogeneity (monetary union) and a high level of interconnectedness between the Eurozone economies, make the case of financial cycle synchronization in the Eurozone more likely. The practical aspect is that data on financial cycle variables are widely (and accurately) available for the Eurozone countries. To narrow down the scope, this thesis will analyse the
core euro-‐area countries (Austria, Belgium, France, Germany, Italy, Netherlands, Spain) as used by Breitung and Eickmeier (2005). Two main subquestions arise from the research question, namely how to measure the financial cycle and how to measure the degree of synchronization between financial cycles.
Little attention in the academic debate has been given to the synchronization of financial cycles across countries. Drehmann et al. (2012) briefly discuss it, but the authors look at cycles of financial variables in isolation, as opposed to looking at one aggregate financial cycle. It is the interaction between these financial variables that amplifies economic fluctuations and leads to economic distress (Borio, 2012). Conditions in credit markets influence asset prices and vice versa. The interactions between these variables can lead to credit and asset bubbles in the build-‐up to a peak of the financial cycle. Stremmel (2015) constructs synthetic country-‐specific financial cycles for eleven European countries and looks at the synchronicity of these cycles via a dispersion measure. In his article on the financial cycle, Stremmel (2015) uses two steps to produce synthetic financial cycles. First time series of individual financial variables are performed using a frequency-‐based filter. Then all financial indicators are standardized and the individual time series are aggregated to create synthetic financial cycles.
The novelty of the research in this thesis lies in the combination between two methods. On the one hand, this thesis uses a method of synthetic financial cycle construction similar to that of Stremmel (2015). On the other hand, this thesis uses concordance statistics to measure the degree of synchronization between financial cycles, like in the work of Drehmann et al. (2012). Combining a method that can construct an aggregate financial cycle with a cross-‐ country synchronization analysis of the core euro-‐area countries will be the main contribution to the academic literature by this thesis. In addition to the work of Stremmel (2015) and Drehmann et al. (2012), this thesis also analyses if there is a change in the degree of synchronization over time. This analysis combined with a comparison between the business cycle and the financial cycle, will provide insights to the dynamics of financial cycles in the Eurozone.
The main variables of analysis for the financial cycle in this thesis will be credit and housing prices. Both variables tend to co-‐vary rather closely and are a basic way to capture the core features of the financial cycle (Borio, 2012). The literature review in this thesis will further elaborate on the different variables used in the academic literature.
Academic literature on financial cycles mainly uses two methods to construct the financial cycle. The first method is widely used in business cycle literature and is called a turning point analysis. A turning point analysis uses an algorithm to identify local maxima and minima over a certain period and constructs peaks and troughs. The second method is a frequency based filter analysis based on the work of Comin and Gertler (2006). A frequency based filter analysis assesses how a variable fluctuates around its trend and then identifies a cycle as deviation from this trend (Claessens et al., 2011). This thesis will perform a frequency based filter analysis. The exact motivation for this choice will be provided in the methodology section.
To measure the degree of synchronization, two methods will be employed. First, concordance statistics based on the article of Harding and Pagan (2002) will be used to look at the degree of synchronization between the core euro-‐area countries. The second method will use a dispersion measure as used by Stremmel (2015) that evaluates whether cycles diverge or converge over time. Using this method will enable this thesis to assess whether financial cycles are less or more synchronized during stress periods.
The subsequent part of this thesis is structured as follows. Section 2 will provide the reader with a literature review on the concepts of the financial cycle and synchronization. Section 3 will consist of the methodology and data. Section 4 will provide an extensive analysis of the financial cycles of the Eurozone. Section 5 will analyse the degree of synchronization of financial cycles. Section 6 makes a comparison between the business cycle and financial cycle. Section 7 will conclude and make suggestions for further research.
2. Literature review
The literature review constitutes of two parts. The first part revolves around literature on the financial cycle, whereas the second part focusses on literature on the topic of synchronization.
2.1 Financial cycle
This section will focus on the concept of the financial cycle. The different variables used in the academic literature to describe and explain the financial cycle will be central in this section. Despite the recent boost of interest in the financial cycle, the amount of academic literature on the subject is still quite limited.
There is no clear consensus about the exact definition of a financial cycle. To understand the concept of the financial cycle, it helps to look at the booms and busts of the cycle. Peaks of the financial cycle are associated with financial crises (Borio, 2012). Financial imbalances build up, excessive risk taking and availability of cheap credit increases, eventually leading to an unwinding of financial imbalances and a subsequent recession. Vulnerabilities of the financial system are often based on the cyclical movements of financial variables (Stremmel, 2015). According to Borio (2012), an important characteristic of the financial cycle is that the booms of the financial cycle must cause the busts. Risk attitudes and perception of risk play a large role in building up the financial imbalances that lead up to the booms in the financial cycle. Financial developments reflect the fluctuations in expectations sentiment and degree of uncertainty, which are important indicators of aggregate economic activity (Borio et al., 2015).
Most authors seem to agree that the financial cycle has a lower frequency than the business cycle. According to Borio (2012), business cycles range between 1 and 8 years, whereas the average length of the financial cycle is around 16 years. Drehmann et al. (2012) find short-‐term cycles that range from 3 to 5 years and medium-‐term cycles that range from 8 to 18 years. Even though the business cycle and financial cycle have different frequencies, they are still closely related (Drehmann et al., 2012). Recessions following a downturn of the financial cycle are accompanied by large drops in GDP (GDP is often used as the main indicator for the business cycle).
One of the central questions of this thesis will revolve around the measurement of the financial cycle. The academic literature provides a wide array of indicators for the financial cycle. As discussed in the introduction, credit and housing prices are two of the main variables in the financial cycle. All prominent articles on the financial cycle use at least some measure of housing prices and credit in their analysis (Claessens et al., 2011; Drehmann et al., 2012; Stremmel, 2015). Borio (2012) describes the co-‐variation of credit and house prices as the most basic way to capture the essential features of the financial cycle. Credit can be linked to financing constraints and property prices are closely linked to the perceptions of value and risks. Drehmann et al. (2012) complement credit and housing prices with additional variables on equity prices and an aggregate price index. The usage of equity as an important indicator for the financial cycle is also used by Claessens et al. (2011). In their article on the financial cycle the authors focus on cycles in three different market segments, namely credit, housing,
and equity markets. It appears that credit and housing prices are the two prominent variables of the financial cycle in the academic literature. Housing prices and credit are often complemented with equity prices in the literature. However, not all articles use equity to construct the financial cycle. This thesis will therefore use credit and housing prices variables as the main indicators for the financial cycle.
There are also authors that use a more extensive list of variables to construct the financial cycle. Examples of other financial variables are leverage and liquidity, bank lending standards, the TED spread, and bank non-‐core liabilities (Ng, 2011). Ng (2011) acknowledges that the financial cycle is empirically not well defined and aims to cover the broadest range of financial variables. Although insightful, this is not feasible for the scope of this thesis.
2.2 Synchronization
There is little research on the synchronization of financial cycles. Therefore, this thesis will also look at business cycle literature to get a better understanding of the process of synchronization.
Claessens et al. (2011) look at cycles of credit, housing, and equity markets in isolation. The authors analyse the synchronization of these cycles within and across countries. Claessens et al. (2011) identify frequency, duration, amplitude, and slope as the main features of the financial cycle. To analyse the extent of synchronization across cycles, the authors use the concordance index developed by Harding and Pagan (2002). This index provides a measure for the amount of time two series are in the same phase of their cycles. A value of 1 means that series are perfectly procyclical, whereas a value of 0 means that series are perfectly countercyclical. Using this method, Claessens et al. (2011) find that there is a high degree of synchronization across countries of the different cycles. For credit and equity cycles, the authors find concordance statistics of respectively 0.75 and 0.70. Housing price cycles have a somewhat lower degree of synchronization, namely 0.59.
Another analysis of the synchronization of financial cycles is conducted by Stremmel (2015). In his article on the financial cycle in Europe, Stremmel (2015) constructs country-‐ specific aggregate financial cycles and analyses synchronicity by using a dispersion measure that evaluates whether financial cycles are diverging or converging over time. Analysis is performed on eleven European countries and leads to the conclusion that financial cycles are less synchronized during good times. This thesis will use the same dispersion measure to
assess whether the financial cycles of the core euro-‐area countries diverge or converge during stress periods.
Academic literature on business cycle synchronization is interconnected with literature on optimal currency areas (OCA). Frankel and Rose (1998) see the degree of business cycle synchronization as an important requirement for a successful OCA. Countries with a relatively high level of business cycle synchronization and close international trade ties are more likely to benefit from a common currency.
A method in business cycle literature to measure the degree of synchronization is to obtain the cyclical component of output and compute bilateral correlations of real activity (Calderon et al., 2007). A high level of correlation means a high degree of synchronization. An alternative measure for business cycle synchronization is developed by Eichengreen and Bayoumi (1996). Instead of looking at correlation, the authors develop a measure for the degree of asymmetry. The lower this value of asymmetry, the higher the degree of business cycle synchronization. Another alternative measure for business cycle synchronization is wavelet analysis. Wavelet analysis performs an estimation of spectral characteristics of time series, revealing different periodic components over time (Soares, 2011). Comparing wavelet spectra enables the author to check the contribution of cycles to the total variance and see if ups and downs of cycles occur simultaneously. It would be useful to asses if these methods can also be used for financial cycle synchronization. Using these different methods to measure financial cycle synchronization, could be a possibility for future research.
3. Methodology and data
3.1 Methodology
This section of the thesis will motivate the methodological choices and explain how the country-‐specific financial cycles are constructed. At the end of this section, two methods for measuring synchronization will be explained.
3.1.1 Constructing financial cycles
This thesis will use a methodology similar to that of Stremmel (2015), where two steps are followed to create synthetic country-‐specific financial cycles. It is important to keep in mind
that the financial cycle is a synthetic measure. There is no generally accepted method to construct this cycle and working with synthetic cycles means checking and verifying the appropriateness before coming to conclusions (Stremmel, 2015). On the other hand, artificial cycles allow us to analyse the joint behaviour of different financial indicators. Constructing country-‐specific financial cycles for the core euro-‐area countries enables us to assess the degree of synchronization between these cycles and check whether they diverge or converge during stress periods.
The first step in creating the country-‐specific financial cycles, is performing time series analysis on the selected financial variables. The choice of variables and the period of analysis will be discussed in the section on data and variable selection. This thesis will use four financial variables to construct the country-‐specific financial cycles (Credit-‐to-‐GDP ratio, House price index, Credit growth, House price growth). In the academic literature, there are two main methods for constructing financial cycles. A turning point analysis and a frequency based filter analysis. In line with the article of Stremmel (2015), this thesis uses a frequency based filter analysis. The main motivation for this choice is that individual filtered time series are additive (Drehmann et al., 2012). This characteristic of frequency based filter analysis allows us to construct synthetic financial cycles. Basically, what frequency based filter analysis does is isolate the cyclical patterns of individual time series. The frequency based filter analysis used in this thesis is based on the work by Comin and Gertler (2006) on business cycles. So as with the concordance statistic that will be used to check the degree of synchronization, this method is also borrowed from the business cycle literature.
There is a range of filters available when performing frequency based filter analysis. The two main filters used in the academic literature are the Hodrick-‐Prescott filter (HP) and the Christiano-‐Fitzgerald (CF) band pass filter. Other filters that are used in the literature are the Baxter-‐King filter and the Butterworth filter. The HP filter was created by Hodrick and Prescott (1981) and has become a standard method for removing trend fluctuations in business cycle literature. The CF band pass filter was created by Christiano and Fitzgerald (2003) and uses a two-‐sided moving average technique.
According to Hamilton (2016) there are several reasons why not to opt for the HP filter. One of the reasons why Hamilton discourages the use of the HP filter, is that the filter produces spurious dynamic relations. The HP filter yields different values for the middle and end of a sample, this is characteristic for spurious results. The CF band pass filter on the other
hand has some clear advantages. Nilsson and Gyomai (2011) state that the initial estimates of cyclical values of the CF band pass filter are the closest to the final long term value of the cycle. The CF band pass filter was designed to perform well for a larger class of time series and in the long run converge to the optimal filter. Like the Baxter-‐King filter, the CF band pass filter is frequency based. Frequency based filters are potentially an exact procedure in the case of infinitively long time series. The HP filter on the other hand works best for periods between 12 and 120 months, substantially shorter than the period of analysis in this thesis. A combination of the above arguments and the choice for the CF band pass filter in the relevant literature, make a clear case for the choice of the CF band pass filter in this thesis.
For each of the core euro-‐area countries, the CF band pass filter is applied to individual time series of the four financial variables. Here we make sure that all indicators are standardized to ensure comparability of the units (Stremmel, 2015). Once standardized, it is possible to create synthetic aggregate financial cycles for the individual core euro-‐area countries. The artificial cycles are created by aggregating the individual filtered cycles for each point in time. This is possible for two reasons. Firstly, because of the favourable characteristics of frequency based filter analysis. Secondly, because the components of the individual time series are comparable units of measurement. Aggregating the underlying time series yields individual financial cycles for the core euro-‐area countries. At this point it will be possible to do a graphical analysis of the individual cycles. A graphical investigation can help draw some preliminary conclusions before going over to statistical methods to assess the degree of synchronization.
3.1.2 Cycle synchronization measures
The previous paragraph explained how the individual financial cycles will be constructed. This paragraph will explain the methods used to assess the degree of synchronization. Two statistical methods, both used in business cycle literature, will be deployed to analyse synchronicity between the core euro-‐area countries.
The first statistical measure that will be used to assess the degree of synchronization is the concordance statistic. This concordance statistic was developed by Harding and Pagan (2002) and it assesses the amount of time two series are in the same phase of their cycles. To do this, the method looks at local maxima and minima in the sample path of the time series in quarterly data (Harding and Pagan, 2006). The concordance statistic yields a value between
0 and 1. Here, 0 being perfectly countercyclical and 1 being perfectly procyclical. A value of 0.5 means cycles are perfectly independent. The index is defined as:
𝐶𝐼#$ = 1 𝑇 [𝐶)# * )+, . 𝐶)$+ 1 − 𝐶)# . (1 − 𝐶 )$)] where:
𝐶)# = {0, if x is in recession phase at time t; 1, if x is in expansion phase at time t} 𝐶)$ = {0, if y is in recession phase at time t; 1, if y is in expansion phase at time t}
The concordance index uses turning points to identify expansion and recession phases. Expansion phase means a part of the cycle leading up to a peak, whereas the recession phase is a part of the cycle leading up to a trough. Using the concordance statistic allows us to assess the degree of synchronization between financial cycles and offers an alternative method to the dispersion measure. The combination of the two synchronization measures will provide useful information on synchronicity of financial cycles in the core euro-‐area countries.
Harding and Pagan (2002) also developed a measure to test the statistical significance of the concordance index. This measure uses the correlation between 𝐶)# and 𝐶)$to test for no concordance. The null hypothesis means no concordance between the two series and thus has a correlation coefficient r4 of zero (Male, 2011). The method uses the following regression to estimate the correlation coefficient:
456 s76s78 = a + r4 458 s76s78+µ)
Here a is a constant term and µ) is an error term. The terms s4# and s4$ are the respective estimated standard deviations of 𝐶)# and 𝐶)$. It is then possible to use the t-‐statistic of the correlation coefficient estimate to determine the statistical significance. This method must use robust standard errors to counter possible serial correlation in the residuals (Avouyi-‐Dovi and Matheron, 2005).
The second statistical method will be a dispersion measure. A dispersion measure makes it possible to assess whether cycles converge or diverge (Gächter et al., 2012). This allows us to assess whether financial cycles diverge or converge during stress periods. Dispersion is measured by the standard deviation of the cyclical components across the examined country sample (Gächter et al., 2013). A higher level of dispersion means lower synchronicity and a lower level of dispersion means higher synchronicity. Stremmel (2015) uses a similar dispersion method and finds that financial cycles are less synchronized during good times (high dispersion).
3.2 Variable selection and data
This section will address the variable selection for the individual time series of the core euro-‐ area countries and describe the selected data set. As described and motivated in the literature review, this thesis will focus on housing prices and credit variables to construct synthetic financial cycles. In line with the methodology of Stremmel (2015), this thesis will use four variables:
1. Credit-‐to-‐GDP ratio 2. House price index
3. Annual growth rates of the nominal bank credit to households and non-‐financial counterparties
4. Annual growth rates of house prices
Credit-‐to-‐GDP ratio is defined here as total credit to households and to the non-‐financial sector. All variables will be standardized to ensure comparability of their units. The two growth rate variables will be four-‐quarter differences in log-‐levels (Stremmel, 2015). This thesis will analyse the behaviour of the financial cycle for the core euro-‐area countries over the period 1980Q1 – 2015Q4. The choice for this period was based on two considerations. First consideration being that the period must be sufficiently long to allow for the occurrence of one or more complete financial cycles. The average period of financial cycle in the academic literature ranges from 3 to 18 years. Secondly, the choice for the period should be in line with the dominant literature on financial cycles. Stremmel (2015) uses the period 1980Q1 – 2012Q4, however the author admits that the relatively short period is a caveat because of the
limited number of full cycles incorporated. Claessens et al. (2012) and Drehmann et al. (2012) use a dataset starting in 1960Q1. However, due to availability of the data it was not possible to find a dataset for the selected variables starting in 1960Q1. Therefore, the time period of the dataset will be 1980Q1 – 2015Q4. The choice for 2015Q4 was made to incorporate the most recent data available for the selected variables.
The Credit-‐to-‐GDP data is available via the Bank for International Settlements (BIS) Statistics database. It is quarterly data from the BIS long series on total credit. The data on house prices was obtained via Datastream and is also a BIS dataset. It is quarterly data from the BIS long series on residential property prices. There was no data on residential property prices in Austria in this dataset. Therefore, this thesis uses a dataset from Oxford Economics which contains the house price index for Austria. As mentioned earlier, data on the annual growth rates was constructed by taking the four-‐quarter differences in log-‐levels. Data for the business cycle was obtained from an Oxford Economics dataset and contains nominal GDP per capita.
4. Financial cycles in the core euro-‐area countries
This section of the thesis revolves around the construction of the country-‐specific financial cycles and discusses any findings and results that were found in the process. After the construction of the financial cycles, a few key characteristics will be discussed.
4.1 Country-‐specific financial cycles
As mentioned in the methodology section, this thesis will construct individual financial cycles for the core euro-‐area countries. To explain this process, one country will be singled out to illustrate the construction of the synthetic cycle. This thesis will use the Netherlands as an example for the construction of the financial cycle. The choice for the Netherlands is arbitrary and has no further motivation. After the process is described for this individual country, the results for the other core euro-‐area countries will be discussed. All regressions and tests in this thesis were performed with Stata, a statistical software program.
The first step in the construction of the country-‐specific financial cycles is to use the CF band pass filter on the four variables. Before we can do this, it is necessary to assess whether the time series are stationary. The filter gets different parameters if a series is stationary. To
test for stationarity in time series, augmented Dickey-‐Fuller tests will be performed. The augmented Dickey-‐Fuller test is a regression-‐based test for a unit root in an autoregressive model (Stock and Watson, 2015). Under the null hypothesis, the series has a stochastic trend. The alternative hypothesis states that the series is stationary. Table 1 shows the outcomes of the augmented Dickey-‐Fuller tests.
Table 1 – Augmented Dickey-‐Fuller tests
Credit/GDP Credit/GDP
growth House prices House prices growth
Austria -‐1.066 -‐3.803*** 1.315 -‐5.307*** Belgium 2.120 -‐2.882* 2.463 -‐3.098** France 2.273 -‐2.899** 0.093 -‐1.909 Germany -‐2.022 -‐3.158** -‐0.006 -‐2.867* Italy 0.263 -‐2.766* -‐3.143** -‐5.844*** Netherlands -‐1.146 -‐3.333* 0.9572 0.1952 Spain -‐0.176 -‐1.591 -‐1.165 -‐1.480
* 10% significance, ** 5% significance, *** 1% significance
Throughout the analysis, this thesis will use a 5% significance level. The results of the augmented Dickey-‐Fuller tests are used to decide which series are stationary.
After performing the augmented Dickey-‐Fuller test, the next step is to use the CF band pass filters on the individual time series. The CF band pass filter isolates the cyclical pattern of individual time series. The cyclical component is then standardized in order to ensure comparability. Because individual filtered time series are additive, it is possible to construct a synthetic cycle. Once the CF band pass filters are applied and the individual time series are standardized, we can plot the series in a graph. Figure 1 shows the four series for the Netherlands. Figures for the other countries are available in appendix A.
Figure 1 – Individual series for the Netherlands
In figure 1 the cyclical components of the four financial variables are illustrated. A graphical investigation of figure 1 does not yield any clear results. The amplitude of the variables seems relatively stable over time. We see in appendix A that this is not the case for countries like Spain and Italy. Credit-‐to-‐GDP and house prices also appear to move together over time, but this is not a clear result. Looking at appendix A, one interesting observation stands out. There appears to be a reduction in amplitude of the variables shortly after 2000Q1 for most of the countries. Belgium and the Netherlands show this to some extent, but more clearly Germany, Spain and Italy. This seems to coincide with the implementation of the Euro, which was implemented in 1999 and in final circulation in 2002. Such convergence (we do still need to verify if there is a higher degree of synchronization in this period), is in line with theory regarding OCAs. Frankel and Rose (1998) state that the degree of business cycle synchronization is an important requirement for a successful OCA. The business cycle and the financial cycle are closely related, so this observation fits in the OCA theoretical framework. The graphical investigation does not provide any conclusive evidence about the cause of this convergence. Later in the analysis, we will assess whether there is actual convergence of financial cycles during the period of the implementation of the Euro.
-4 -2 0 2 1980q1 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 2015q1 Quarter
Credit-to-GDP Credit-to-GDP growth rate House prices House prices growth rate
The final step in the construction of the country-‐specific financial cycles is to add the four individual filtered series. Adding the series creates one synthetic cycle per country. Figure 2 shows the constructed financial cycle for the Netherlands. Financial cycle figures for the other core euro-‐area countries can be found in appendix B.
Figure 2 – Financial cycle for the Netherlands
A graphical investigation of figure 2 reveals a relatively short financial cycle, with cycle lengths varying around 3 to 6 years. This seems in line with the short-‐term cycle length of 3 to 5 years found by Drehmann et al. (2012). Looking at appendix B, we see a similar cycle length for the other core euro-‐area countries. Although the amplitude of the financial cycle in Austria and Italy seems far less stable than in the other countries. The following section will use a more formal method than a graphical investigation to assess some of the financial cycle characteristics of the individual countries.
4.2 Financial cycle characteristics
According to Claessens et al. (2011) the main characteristics of cyclical phases are duration, amplitude, slope and frequency. This section will analyse and compare the financial cycles of
-4 -2 0 2 4 Amp lit ud e fin an ci al cycl e 1980q1 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 2015q1 Quarter
the core euro-‐area countries using these characteristics. Analysing these key characteristics helps identifying certain differences and similarities between the cycles.
Before looking at the main characteristics of the financial cycles, we take a short look at the descriptive statistics of the financial cycles. Analysing the descriptive statistics enables us to make a first scan of the cycle characteristics and possible detect any unexpected results. Table 2 shows the descriptive statistics for the financial cycles of the core euro-‐area countries.
Table 2 – Descriptive statistics
Std. Dev. Min Max
Austria 1.3117 -‐3.0212 4.3746 Belgium 1.1439 -‐2.5857 2.3887 France 0.9943 -‐1.9006 1.6002 Germany 1.1379 -‐2.7184 2.5123 Italy 1.9642 -‐3.3936 9.2955 Netherlands 1.3595 -‐2.8290 2.9642 Spain 1.1980 -‐3.1833 3.2728
The column with standard deviations shows us that all values are relatively close to each other, expect for Italy. The high standard deviation of Italy coincides with a relatively high minimum and maximum value of the cycle. Later in the analysis it is confirmed that the Italian financial cycle is something of an exception compared to the other cycles. The financial cycles of the other countries do not seem to contain any unexpected outcomes.
We now look at the duration of the individual financial cycles. We can split the duration of the cycle in average duration of the expansion phase and average duration of the recession phase. Adding these two phases gives us the average duration of the financial cycle. The following definition is used by Harding and Pagan (2001) for the average duration of the expansion phase: 𝐷
=
:5;<45 (,= 45><)45 :?< 5;<In this equation, the numerator measures the total duration of the expansions of a series. The denominator measures the number of peaks in the series. Table 3 shows the average duration of the expansion, recession and full cycle in quarters.
Table 3 – Average duration financial cycles
Average duration
expansion Average duration recession Average duration full cycle
Austria 5.62 5.92 11.54 Belgium 6.25 5.75 12 France 7 6.7 13.7 Germany 6.55 7.2 13.75 Italy 4.24 4.5 8.74 Netherlands 7 8.22 15.22 Spain 4.8 5.14 9.94
From the above table, we see that the average duration of the full cycle ranges from 8.74 to 15.22 quarters. This formal outcome confirms the shorter cycle length found during the graphical investigation. There is no clear pattern to be found in the average duration of the expansion and recession phases. If we take the total averages of both the expansion and recession duration, we find that this is 5.92 quarters in the case of expansion and 6.22 quarters in the case of recession. This is not in line with the article of Claessens et al. (2012), where upturns are often longer than downturns. As for the duration of the full cycle, only Italy seems to be something of an outlier with a much shorter average duration of just over 2 years. This is confirmed if we look at the figure of the Italian financial cycle, here we see high frequency and unstable amplitude. Looking at the Netherlands, we find the longest average full cycle of 15.22 quarters. Another observation that stands out is the difference between the average duration of the expansion and recession in the Netherlands. The average recession duration is 1.22 quarters longer than the average expansion duration. This is the largest difference between expansion and recession duration of all core euro-‐area countries.
The second characteristic up for review, is the average amplitude of the financial cycle. It is possible to calculate both the average amplitude of the expansion and the recession. The following equation is used by Harding and Pagan (2001) to define the average amplitude of the expansion. 𝐴 = *)+,𝐶)D𝑦) (1 − 𝐶)B,)𝐶) *=, )+,
The numerator of the above equation measures the total change during the expansions. The denominator measures the number of peaks in the series. This measure only incorporates full cycles. Table 4 shows the average amplitudes for the core euro-‐area countries.
Table 4 – Average amplitude financial cycles
Average amplitude
expansion Average amplitude recession Average total amplitude
Austria 1.7961 2.1770 3.9731 Belgium 1.6155 1.5974 3.2129 France 1.7918 2.0499 3.8417 Germany 1.9103 2.1063 4.0166 Italy 1.9126 1.8557 3.7683 Netherlands 3.0155 3.5075 6.523 Spain 1.7891 1.8543 3.6434
In the above table, we see that the average total amplitude of the financial cycles lies between 3.2129 and 6.532. The above values in isolation do not mean much, because we are dealing with synthetic cycles. However, the values are useful for comparing the amplitudes of the different financial cycles. We see that in most countries (except Belgium and Italy), the average amplitude of the recession is larger than the average amplitude of the expansion. This could suggest that recessions are associated with larger drops in value. If we take the total average amplitude of the expansion, we find a value of 1.9758. This is smaller than the total average amplitude of the recession, with a respective value of 2.1640. The size of the different average amplitudes of the core euro-‐area countries are all relatively close to each other, except for the Netherlands. Looking at the Netherlands, we find an average amplitude that is twice as large as Belgium’s average amplitude. The relatively large amplitude of the Netherlands is in line with the longer average duration of the full cycle.
The slope of the recession/expansion is defined by Claessens et al. (2011) as the amplitude divided by the duration. To measure the average slope of a cycle, we divide the average amplitude by the average duration. However, the average slope of a cycle might not be very revealing, so we are more interested in the average slopes of the recession and expansion phase. Table 5 shows the average slopes for the core euro-‐area countries.
Table 5 – Average slopes financial cycles
Average slope
expansion Average slope recession
Austria 0.3196 0.3677 Belgium 0.2585 0.2778 France 0.2560 0.3060 Germany 0.2916 0.2925 Italy 0.4511 0.4124 Netherlands 0.4308 0.4267 Spain 0.3727 0.3608
According to Claessens et al. (2011), the slope tells us something about the violence (speed) of a cyclical phase. A steep slope in the case of a recession would thus means rapid deterioration and possibly severe effects to the economy. If we again take the averages of the expansion phase and the recession phase for the core euro-‐area countries, we find respective values of 0.3400 and 0.3491. The average slope of the recession is slightly steeper than the average slope of the expansion. This steeper average slope of the recession is in line with the results of Claessens et al. (2011). However, it should be noted that the difference is very small. Looking at the individual countries we see that Italy and the Netherlands have steeper slopes for both expansion and recession. In the case of the Netherlands this is not surprising if we look at the earlier obtained characteristics. Italy again seems to be something of an outlier. The final characteristic we will analyse is the frequency of full cycles. A full cycle simply means a combined expansion and recession phase up until the next expansion phase. Table 6 shows the amount of full cycles for each of the core euro-‐area countries. For analytical purposes, the table also contains a column with full cycles before 2000Q1 and after. In the case of synchronization, we would expect the amount of full cycles after 2000Q1 to converge. The choice is for 2000Q1 is deliberate and is chosen to reflect the implementation of the Euro.
Table 6 – Frequency of financial cycles
1980Q1 – 1999Q4 2000Q1 – 2015Q4 Total frequency Austria 7 5 12 Belgium 8 4 12 France 6 4 10 Germany 6 4 10 Italy 10 6 16 Netherlands 6 3 9 Spain 10 5 15
