1
Master thesis:
Fluctuations in the financial cycle explained
Name: Nikki Quérine Rupert
Student number: 10391819
Date: 17-12-2014
Supervisor: prof. dr. H. Jager
MSc Economics, International Economics and Globalization
2
Table of contents
1. Introduction 3
2. Literature review 4
2.1. The financial cycle 4
2.2. Variables that impact the size of financial fluctuations 7
3. Data 9
4. Methodology 11
4.4. Frequency-based filter analysis 11
4.2. Regression analysis 13 4.3. Expected results 14 5. Results 17 5.1. Regression 1 17 5.2. Regression 2 20 6. Conclusion 22 6.1. Summary 22 6.2. Implications 22 6.3. Further research 23 6.4. Caveats 24 7. References 25 8. Appendix 27 A. Autocorrelation 27 B. Cross-sectional dependence 27 C. Unit root 27
3
1.
Introduction
The low inflation rates of the advanced world in the 1990s and 2000s inclined central banks to think that the economy was on a never ending stable path. The global credit crisis of 2008/09 showed the opposite: that low inflation rates are not enough to assume stability. It was realized that a better understanding of the financial system was needed in order to work towards a better indirect measure of financial stability than inflation.
From the crisis onwards the economic science has started looking for ways to give a bigger prominence to the role of finance in macroeconomic theory. One line of research tries this by getting a better understanding of the financial cycle. The financial cycle captures the cyclical behaviour of the financial side of the economy, similar to how the business cycle describes the real economy. Then the size of the amplitude of the financial cycle can be seen as a measure of financial instability. Large deviations of, particularly, house prices and credit supply from their natural values then signal the build-up of risk of financial crises (Borio 2012, p. 5) and peaks in the financial cycle often correspond with the start of financial crises (Drehmann 2012, p. 15).
Research on the financial cycle has primarily been focused on characterization of the cycle. Especially important in this respect are Drehmann, Borio and Tsatsaronis (2012). Using a new method, frequency-based filter analysis, they derive that the financial cycle has an average duration of 16 years and is best captured by the medium-term behaviour of house prices, credit provided to the private sector and the credit-to-GDP ratio.
Yet not much is known on what determines the size of the fluctuations of the financial cycle. Drehmann et al. (2012) find that financial fluctuations have become more pronounced from the 1980s onwards and allocate this to the process of financial liberalization that started in the beginning of the 1980s. Borio (2012, p. 6) argues that the shape of the financial cycle depends on financial, monetary and real economy regimes. Little research has empirically tested the influence of such determinants on the financial cycle, let alone on the financial cycle as identified by the new method by Drehmann et al. (2012).
This master thesis tries to get a better grip on what factors impact the size of financial fluctuations, the fluctuations of the financial cycle. For that goal the absolute and the actual value of the financial cycle are regressed on a range of variables that in the literature have been linked to financial variation. In doing so this paper hopes to get a better insight in the build-up of financial instability, knowledge that can be used for improvements in devising policy aimed at financial stability.
4 In line with Drehmann et al. (2012) the financial cycle is identified by so called frequency-based filter analysis. This technique identifies the medium-term cyclical path of a variable. As possible determinants of the cycle a range of variables are used that are also studied at their medium-term cyclical path as to make a good comparison with the medium-term financial cycle possible. The research will be carried out using quarterly panel data for 20 advanced economies in the period 1980-2012.
While the existing research focuses on explaining separate characteristics of the financial cycle (e.g. amplitude and duration), this research aims to explain the full movement of the cycle. Furthermore, this research contributes to the existing literature by testing a broader range of determinants than upon till now have been related to influencing the financial cycle. Also determinants that have been related to the behaviour of credit and house prices and determinants related to the build-up of financial imbalances are considered.
A few main results stand out. Firstly, fluctuations in either financial openness, inflation, short-term interest rate, money growth or global liquidity are a cause of financial fluctuations. Furthermore, a rise in either trade openness, inflation or money growth increase the size of the fluctuations of the financial cycle and thus increase financial instability. Finally, the term interest rate is negatively related to the size of financial fluctuations. A rise in the short-term interest rate thus increases financial stability.
An important limitation to the study is that it has only been able to look at 20 years of data. This has, since a financial cycle has the average length of 16 years, reduced the chance that a financial cycle in full can be studied.
The next chapter gives an overview of the literature on the financial cycle and its
determinants. Chapter 3 discusses the data and chapter 4 the methodology the research uses. Chapter 5 presents the results of the regression analysis and the last chapter contains a summary and concluding remarks.
2.
Literature review
2.1. The financial cycle
Although research on the financial side of the economy got a huge boost through the recent global credit crisis, qualitative research on the financial cycle is still quite limited.
5 The study on the financial cycle fits in an old line of theory which argues that instability and crises are inherent to the economic system and not solely the result of exogenous forces. Already in 1933 Irving Fisher sees financial variables, and not factors related to the business cycle, as most important in shaping booms and recessions (p. 341). And Minsky (1982, p. 6) argues that financial activity should be incorporated in economic theory for instability to be shown as part of the economic system.
Later on there are studies that look at possible components of the financial cycle, without using the term ‘financial cycle’. Goodhart and Hoffman (2008), for example, look at house prices, credit provided to the private sector, money and real economy indicators and find multiple linkages between these variables. Some studies look specifically at house prices: Tsatsaronis and Zhu (2004) study the drivers of house price dynamics and Agnello and Schuknecht (2011) look at booms and busts in the housing market. Others focus on credit: Mendoza and Terrones (2008) study credit booms and how they are influenced, Humea and Sentence (2009) empirically asses the causes of the global credit boom that preceded the global credit crisis of 2008/09.
But the approach of characterizing the behaviour of these financial variables as cyclical and subsequently identifying this cyclical behaviour using a business cycle approach is relatively new. Foremost two papers are important in this line of research. Claessens, Kose and Terrones (2011) aim to characterize the financial cycle in three different financial markets that they consider as the core of financial intermediation: credit, housing and equity markets. Their measure of credit is aggregate claims on the private sector by deposit banks. For the other two markets they take respectively house prices and equity prices. They capture the financial cycle using turning-point analysis, a technique that can identify cyclical behaviour and is much used in business cycle identification. Turning-point analysis uses an algorithm to find local maxima and minima within a defined period for a certain dataset. Then it selects pairs of adjacent minima and maxima that comply with certain constraints, such as a
minimum duration. Claessens et al. (2011) take 5 quarters as the minimum duration. This is comparable to the minimum duration of a business cycle. They find that the cycles for all three markets are long and severe, highly synchronized and accentuating each other.
Drehmann, Borio and Tsatsaronis (2012) build on the paper by Claessens et al., but improve it on three important points. Rather than studying the different markets separately, they construct an integrate measure of the financial cycle using a composite of variables describing these markets. Secondly, they use, together with the turning-point analysis, another, newer technique for more robust results. This so-called ‘frequency-based filter analysis’ uses a filter, a so called band-pass filter, to identify the cyclical movement in a
6 dataset. Furthermore, Drehmann et al. study not only cycles which have the length of a business cycle, that is, between 5 and 32 quarters. Drehmann et al. call these short-term cycles. Also, they study cycles with a much longer duration. Therefore they use a band-pass filter that filters cycles that have a duration between 32 and 120 quarters (or 8 and 30 years). They do so using a method by Comin and Gertler (2006) who devised this medium-term filter to study the business cycle.
Drehmann et al. (2012) have a few important results regarding the characteristics of the financial cycle. They find that peaks in the financial cycle are closely related to systemic banking crises. Furthermore, they find that the financial cycle is best characterized by taking credit, credit-to-GDP and house prices together. They also look at equity prices, an
aggregate asset price index and GDP. But house prices, credit provided to the private sector (from now on ‘credit’) and credit-to-GDP give the closest account of the fluctuations on the financial side of the economy: financial crises occur closest to the peaks of the cycles for house prices, credit and credit-to-GDP. Borio (2012) explains that there are other ways to describe the financial cycle, but house prices, credit and credit-to-GDP are the smallest set of variables that are able to replicate the “mutually reinforcing interaction between financing constraints and perceptions of value and risk”. House prices, credit and credit-to-GDP are closely related and tend to move together as usually credit is needed to buy a house.
According to Borio (2012) the financing constraints are captured by credit and credit-to-GDP and perceptions of value and risk are reflected in house prices. Thirdly, Drehmann et al. (2012) show that the financial cycle is best described using a medium-term approach. They find that the medium-term cyclical component in the different financial variables is more volatile than the short-term one. The medium-term cycles are thus more important in shaping the behaviour of house prices, credit and credit-to-GDP. Using this approach they found the financial cycle to have an average length of 16 years, so considerably longer than a
business cycle.
The financial cycle is able to show the build-up of financial instability over time and gives insight to the behaviour of perceived risk versus actual risk. Credit booms are associated with the build-up of leverage (Dell'Ariccia et al. 2012) and the deviation of credit from its trend value (credit gap) can be seen, according to Borio (2012), as a close measure of the leverage of the economy. Furthermore Borio argues that the credit gap serves as an indirect measure of the loss of absorption capacity in the economy. The more leveraged an economy is, the less debt it can take on and the lower its absorption capacity is. The deviation of house prices from trend values (house price gap) reflects the level of imbalance in this market and measures the probability of a price reversal (Borio 2012, p. 3). The combination
7 of house prices and credit is the most efficient set of variables to capture the financial cycle (Drehmann et al. 2012, p. 23).
In de upturn phase of the financial cycle, perceived risk is low, leading to a boom in both house prices and credit. In this upturn phase, financial imbalances build up (Crockett 2000, p. 5) and risk of financial crises actually rises as house prices and credit deviate more from their trend value (Borio 2012). This process continues till there is a reversal and the cycle begins its downturn phase. The reversal often, but not always, coincides with a financial crisis (Borio 2012, p. 4).
2.2. Variables that influence the size of financial fluctuations
Research on the financial cycle suggests that the behaviour of the cycle is also dependent on outside factors. Borio (2012) argues that the amplitude and duration of the financial cycle depend on financial, monetary and real economy regimes in place. Firstly, Borio (2012) finds that financial openness reinforces the cycle since it weakens financing constraints. This argument by Borio aligns with the findings by Drehmann et al. (2012) who found more pronounced cycles after the 1980s which they appoint to the era of financial liberalization that started in this period. Monetary policy can, according to Borio (2012) play a role in curbing the financial cycle. Butif a monetary regime’s only goal is low inflation it will refrain from tightening policies during a boom as long as inflation is low. Inflation could thus have an indirect negative effect on the financial cycle. As for the real economy, Borio (2012) finds that positive supply movements boost the amplitude of the financial cycle. Lower inflation arising from the supply increase will, furthermore, reduce the need for tightening policies and remove a possible break on the boom.
Claessens et al. (2011) empirically assess determinants of the amplitude and duration of a financial downturn and find, in contrast with Drehmann et al. (2012), more financial
openness to be associated with shorter downturns with smaller amplitude, especially for house prices. They argue that this supports their intuition that the domestic financial market is better able to recover with support from global activity and it supports literature that finds that financial openness reduces the risk of financial crises (for example: Cavallo and Frankel (2008)). As for financial openness, the results from Claessens et al. show find that more trade openness is associated with shorter downturns and smaller amplitude of the cycle, again especially for house prices. This seems to contradict with Borio (2012 p. 6) who argues that the positive supply movements due to globalization fuelled financial booms. Furthermore, Claessens et al. find that high inflation in the run up to a credit downturn makes
8 the downturn longer and more severe. They suggest that this is due to the effect of the uncertainty caused by price dynamics.
Concluding from research on determinants of the behaviour of credit and house prices and determinants of the build-up of financial imbalances, this research expects the behaviour of the financial cycle to depend on more variables than upon till now have been assessed in research that focuses purely on the financial cycle. Now, some variables are discussed that in the literature are being linked to financial variation but have not yet been used in financial cycle research.
Tsatsaronis and Zhu (2004, p. 4) find from their empirical research that a declining interest rate boosts demand for residential real estate since it reduces services costs of mortgages (Tsatsaronis & Zhu (2004), p. 4). Whether this holds for short- or long-term interest rates depends on the country considered. Countries with floating rate contracts (e.g. the UK) appear to depend more on short-term rates, while countries with fixed rate contracts (the US and many European countries) depend more on the long-term rate. Hume and Sentence (2009, p. 24) arrive at the same outcome, arguing that low interest rates in the 2000s were a direct stimulus to lending and had an important role in shaping the credit boom.
Goodhart and Hoffman (2008) find a positive effect of money growth on house prices and on credit (p. 192). Agnello and Schuknecht (2011) find that an increase in global liquidity
increases the probability of experiencing a boom. This effect is not significant during bust phases, but it is during upswings.
Mendoza and Terrones (2008, p. 11) find that the current account as percentage of GDP moves in the opposite direction of credit during a credit boom. It declines to a deficit when credit is expanding and it rises to a surplus when credit is declining. The researchers do not explain this opposing movement. Obstfeld (2012) recognizes that current account
imbalances can signal financial instability. But he thinks that the ratio of gross international financial flows to GDP (which is often used as a measure for financial openness) are actually a better indicator of financial distress. Net current account positions actually reflect much larger gross financial flows that are much more insightful on the level of risk, according to Obstfeld.
9
3.
Data
This study is based on quarterly data for 20 advanced economies (Australia, Austria, Belgium, Denmark, Finland, France, Germany, Greece, Italy, Ireland, Japan, the
Netherlands, Norway, Portugal, Spain, the United Kingdom and the United States) for the period 1980Q1-2012Q4. This period is 20 years shorter than in the approach by both
Claessens et al. (2011) and Drehmann et al. (2012) which is due to data unavailability before 1980 for several variables, including for the current account. Drehmann et al. (2012) found the financial cycle to have an average length of 16 years. It is an important limitation to this study that it is only able to use 20 years of data since this decreases the chance that a financial cycle can be studied in full.
For the determination of the financial cycle this research uses data on house prices, credit provided to the private sector and the credit-to-GDP ratio. The regression analysis on the cycle copies some of the determinants Claessens et al. (2011) use, namely: financial openness, trade openness and inflation. Besides, this research adds some new
determinants that in the literature are linked to financial variation: GDP growth, long- and short-term interest rates, money growth, global liquidity and the current account.
Data for credit is from the Bank of International Settlements (BIS). The data is adjusted for breaks (not for the US) and seasonally adjusted only for Canada, France, Sweden and the US. The data on nominal GDP is from Oxford Economics. The data is in current prices and not seasonally adjusted. House prices are index numbers from Oxford Economics. They are only seasonally adjusted for Denmark and the UK.
Claessens et al. (2011) define a country’s financial openness as the sum of its total foreign assets and total foreign liabilities over its GDP. Total assets and liabilities till 2005 are from the updated and extended version of the External Wealth of Nations Mark II database developed by Lane and Milesi-Ferretti (2007) which is in current prices, and not seasonally adjusted. The yearly figures are interpolated to get quarterly data. From 2005 onwards the data for total assets and liabilities is from the IMF, International Financial Statistics (IFS). They are quarterly figures, in current prices and not seasonally adjusted. For France and Norway the IFS has only yearly figures available, so these are interpolated to quarterly figures as well. The data for financial openness is not available for both Ireland and Japan. These countries are left out of the regression when financial openness is included.
10 For trade openness this research uses, following Claessens et al. (2011), the definition of exports plus imports over GDP. Both exports and imports are from the OECD, Main Economic Indicators (OECD MEI). They are nominal figures and seasonally adjusted.
The data for inflation is from OECD MEI and is the year on year change in the consumer price index per quarter.
The short- and long-term interest rates are from Oxford Economics. For the long-term rates they are quarterly average rates of government securities: government bonds with a 10-year maturity. Both the data for the long- and short-term interest rate is not seasonally adjusted.
For money growth this research uses broad money, M3. For the UK, Norway and Portugal, data on M3 is not available; so either M4 (UK) or M2 (for Norway and Portugal) is used instead. For Portugal there is no quarterly data available so the yearly figures are
interpolated. For Denmark, Japan, Norway, Switzerland, the UK and the US data is from the OECD MEI. For the rest of the countries the data is from Oxford Economics. Data for
Canada, Denmark, Japan, Switzerland and the US are period averages, the rest is end of period. Only Canada, Germany, Japan, Sweden, Switzerland, the UK and the US are seasonally adjusted. All figures are nominal values in current prices.
The broad money data described above is also used for global liquidity. Agnello and
Schuknecht (2011) compute global liquidity as “PPP-GDP weighted average of broad money
growth for all sample countries, minus the corresponding national broad money aggregate”.
This research follows this definition with a minor change since, as for global financial
conditions, it does not PPP weigh the variable. Global liquidity, as a variable, is comparable to broad money but it is included in this to determine the role of cross-country spill overs as it is an average of money growth.
Figures for national current account imbalances are from Oxford Economics. They are quarterly, nominal figures, not seasonally adjusted.
All data is directly obtained from DataStream, except for the data from BIS and the Lane and Milesi-Ferretti database. The data sources mentioned earlier are the sources used by
DataStream.
Since a lot of the data is not seasonally adjusted, the data is deseasonalized by including dummies for every quarter in the regression analysis. These seasonal dummies remove any possible seasonal effect from the data.
11
4.
Methodology
This thesis aims to discover what factors amplify the financial cycle and thus have a destabilizing impact on the financial economy. To do so the medium-term financial cycle is regressed on the medium-term cycles of a range of variables that in the literature have been linked to financial variation. Section 4.1 describes how the medium-term financial cycle and the medium-term cycles of the regressors are constructed. Section 4.2 describes the subsequent regression analysis.
4.1. Frequency-based filter analysis
Following Drehmann et al. (2012) the financial cycle is identified using frequency-based filter analysis, a technique devised by Comin and Gertler (2006). The frequency-based filter analysis uses a band-pass filter to eliminate the trend out of the data which leaves the cyclical component. In the following text, the different steps in this approach will be explained.
To be able to regress the financial cycle on the determinants, the approach by Comin and Gertler (p. 9) is followed and for the determinants their cyclical movement is studied as well. Applying the frequency-based filter analysis not only on the financial cycle variables but also on its determinants, does have a caveat. The band-pass filter removes the trend out of the data, so it is inherent to the technique that some details on de variables are lost. The loss of detail might decrease the explanatory power of the variables. This should be kept in mind when evaluating the results. Applying the frequency-based filter on the variables also had an advantage, as it removes the problem of non-stationarity in the variables.
A Woolridge test shows the dependent and independent variables used in both of our regressions is autocorrelated. which means the variables are correlated with their previous values (see Appendix A). Furthermore a Dickey-Fuller test is used to check for unit root in the independent and dependent variables. Unit root means that there is more than one trend in the series. Since the data is autocorrelated an augmented version of the Dickey-Fuller test (ADF) is used. The data also contains cross-sectional dependence (see Appendix B). Cross-sectional dependence means that there is strong correlation between the different sections in panel data, or countries in this case. This correlation can arise for example in the
presence of global shocks (Chudik & Pesaran (2013), p. 3). The cross-sectional means are subtracted within the ADF to correct for this correlation. The data for both house prices and the credit-to-GDP ratio contains unit root and is converted to growth rates in an attempt to create stationary variables (Comin & Gertler (2003), p. 5). For credit-to-GDP four-quarter
12 differences in log levels are taken. Since house prices are index numbers, year on year growth rates suffice (see Appendix C).
Of the determining variables only financial openness contains unit root. But, for comparability with the financial cycle, not only financial openness but all variables except inflation and the long- and short-term interest rates are put in four-quarter differences (if not already a growth rate). Inflation, long- and short-term interest rates are not put in growth rates since these numbers are perunages. Putting them in growth rates would have no economic meaning.
Next, a Christiano Fitzergerald band-pass filter is used to actually detrend the data
(Drehmann et al. (2012), p. 3). “A band-pass filter is basically a two-sided moving average
filter, where the moving average depends on the frequencies of the data that one wishes to isolate” (Comin and Gertler (2003), p. 4). This filter smooths out the behaviour of the
respective variable by removing data below the minimum frequency or above the maximum frequency (Comin and Gertler (2003), p. 4). With ‘frequency’ is meant: the minimum and maximum length of a full measured in quarters of years. This makes apparent the cyclical component of the variable. The level of ‘smoothness’ depends on which frequency the data is analysed.
Usually, frequency-based filter analysis is used for business cycle research, where short-term cycles are studied with frequencies between 2 and 32 quarters (so 0.5 and 8 years) (Drehmann et al., 2012 p. 4). The present research, meanwhile, follows Drehmann et al. (2012) and investigates cycles as well as Drehmann et al. (2012) at the medium term. To do so, a band-pass filter is used with, copying Drehmann et al., frequencies between 32 and 120 quarters (or 8 and 30 years), which gives a much smoother trend. So the minimum frequency in the filter that this research uses is 32 quarters and the maximum frequency is 120 quarters.
This medium-term approach is chosen since, as discussed in Chapter 2, Drehmann et al. (2012) find that medium-term cycles are much more important in shaping the behaviour of the series of house prices and credit. The volatility of the medium-term cyclical component exceeds that of the short-term one.
For the financial cycle, to arrive at an aggregate measure for one country, Drehmann at al. (2012) simply take the average of the filtered series of its house prices, credit and credit-to-GDP ratio since the three series have comparable units of measurement. An approach that this research follows too.
13 For their measure of the financial cycle, Drehmann et al. (2012) combine the frequency-based filter analysis with turning-point analysis. Since this research is interested in the financial cycle as a whole and subsequently by what factors it is determined, the exact turning-points (peaks and troughs) of the cycle are less of interest. This research therefore solely uses the frequency analysis. Dropping the turning-point analysis means loss of an extra check on the obtained results. But Drehmann et al. (2012, p. 15) find that both
techniques give a very consistent picture and that the peaks identified by both methods align well. It is thus to be expected that using just the frequency-based filter analysis will be
sufficient to study the financial cycle.
4.2. Regression analysis
To discover what factors amplify the financial cycle and thus have a destabilizing impact on the financial economy the absolute value of the medium-term financial cycle is regressed on the medium-term cycles of a range of variables that in the literature have been linked to financial variation. The absolute value of the financial cycle is taken as to create a measure for financial instability. The further away the financial cycle is from its trend, the more positive the absolute value is, and the bigger the financial instability is. For the regressors the lagged value is taken in an attempt to avoid the problem of reversed causality.
Furthermore, a second regression is performed where the actual value of the medium-term financial cycle is regressed on the same variables. This regression will give insight into whether these variables move in the same direction as the financial cycle and cause the cyclical behaviour of the financial cycle. For this regression as well, the lagged value of the regressors is taken as to avoid the problem of reversed causality.
The first regression analysis of the medium-term financial cycle on the medium-term cycles of the determinants is based on a panel data regression with both time and country fixed effects. Regression 1:
𝑌𝑌 𝑖𝑖𝑖𝑖𝑎𝑎𝑎𝑎𝑎𝑎= 𝛽𝛽1𝑋𝑋1𝑖𝑖−1+ ⋯ + 𝛽𝛽10𝑋𝑋10𝑖𝑖−1+ 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖+ 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖+ 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖+ 𝐷𝐷𝑖𝑖𝑖𝑖
And for regression 2:
𝑌𝑌𝑖𝑖𝑖𝑖 = 𝛽𝛽1𝑋𝑋1𝑖𝑖−1+ ⋯ + 𝛽𝛽9𝑋𝑋9𝑖𝑖−1+ 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖+ 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖+ 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖+ 𝐷𝐷𝑖𝑖𝑖𝑖
Where 𝑌𝑌𝑖𝑖𝑖𝑖 is the medium-term financial cycle. 𝑋𝑋1𝑖𝑖−1 to 𝑋𝑋9𝑖𝑖−1 are the medium-term cycles of respectively growth rate of financial openness, growth rate of trade openness, inflation, GDP growth, long-term interest rate, short-term interest rate, money growth, global liquidity and
14 growth of the current account deficit. And for regression 1, also the absolute value of the medium-term cycle of the current account is added as regressor, 𝑋𝑋10𝑖𝑖−1. For regressions 2 the absolute value of the medium-term cycle of the current account is not included as the economic meaning of a possible correlation between these two absolute variables would be unclear. For all the regressors their lagged value is used to study their causal relation to the financial cycle. 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖 is the country-specific intercept, 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖 and 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑖𝑖 are the time specific intercepts for both quarter and year and 𝐷𝐷𝑖𝑖𝑖𝑖 is the error term.
Both quarter and year dummies are included. Year dummies are included to function as time fixed effects that control for omitted variables that are equal for different countries but
change over time (Stock and Watson (2012), p. 400). Quarterly dummies are included to deseasonalize the data, since a lot of the data is not seasonally adjusted.
A fixed country effect model controls for omitted variables that differ between countries but not over time and thus removes the omitted variable bias from this particular source (Stock and Watson (2012), p. 396).
The panels are cross-sectional dependent, serially correlated as well as heteroscedastic. Since all these three features can lead to seriously biased results, standard errors are needed to correct for this. Driscoll and Kraay standard errors are used here since they are robust for both cross-sectional dependence, serial correlation and heteroscedasticity (Hoechle, 2007).
4.3. Expected results
Below, the expected results for both regressions 1 and 2 are discussed.
4.3.1. Financial openness
As discussed in chapter 2 the literature is not decisive on whether financial openness has a negative of positive influence on financial stability. Both Borio (2012) and Drehmann et al. (2012) found that an increase in financial openness amplifies the cycle. From this, a positive coefficient for financial openness on the absolute value of the financial cycle would be expected. An increase in the growth rate of financial openness would then cause an
increase in the deviation of the basket of the growth rates of house prices, credit and credit-to-GDP from their trend value. And, with a decrease in the growth rate financial openness would cause a decrease in the deviation of the financial cycle variables from their trend value. Claessens et al. (2011), on the contrary, find that a higher level of financial openness moderates downturns. This could point to a negative influence of financial openness on the
15 absolute value of the financial cycle as more financial openness in their view in any case decreases the deviation of the downturns of the financial cycle.
The theories of Borio and Claessens et al. perhaps should not have to conflict with each other when assuming that financial openness amplifies the financial cycle in the upturn phase while curbing the downward movement of the cycle. A positive coefficient for financial openness in our second regression would endorse this theory.
4.3.2. Trade openness
For trade openness too, the literature is ambiguous on its possible influence on financial stability. From Claessens et al. (2011) a negative coefficient on the absolute value of the financial cycle would be expected as trade openness decreases the amplitude of the cycle and thus the gap from the financial cycle variables from their trend value. Claessens et al. dedicate this to the fact that financial markets are stronger with support from global trade activity. From Borio (2012) however, a positive influence of trade openness on financial instability would be expected as he claims that globalization through production supply movements fuels booms.
Here again, the two theories might actually fit together when it is assumed that an increase in trade amplifies the financial cycle in the upturn phase while curbing the downward movement of the cycle. Also, a positive coefficient for trade would then be expected in the second regression.
4.3.3. Inflation
For inflation, a positive direct effect as well as a negative indirect effect are expected in the first regression. Claessens et al. (2011) find that high inflation in the run up to a credit downturn makes the downturn longer and more severe, which suggests a positive effect of inflation on the gap from the financial cycle sub-variables from their trend value. Borio (2012) however claims inflation to have a negative indirect effect on financial stability (and thus a positive indirect effect on the financial cycle) as low inflation removes the need for tightening monetary policies. Since this research’ regression includes the short-term interest rate (as a proxy of monetary policy) this indirect effect can be excluded. And by also running a
regression excluding the short-term interest rate the impact of the indirect effect can be estimated.
16 As for the second regression: the fact that Claessens et al. find that inflation amplifies downturns could point to a negative coefficient for inflation. This would mean that inflation amplifies financial downturns while curbing upturns.
4.3.4. GDP
As discussed, Borio (2012) indicates that positive supply movements boost the amplitude of the financial cycle. This could also be expected from economic reasoning. Economic growth stimulates trust and risk taking, giving rise to booms. Following this line of reasoning,
negative economic growth will do the exact opposite, leading to a financial downturn. A positive coefficient for GDP growth on the size of financial fluctuation is thus expected in the first regression.
From the available literature it is difficult to predict what sign the coefficient of GDP will have in the second regression. In the upturn phase of the cycle a positive effect is expected, as well as a negative effect in the downturn phase.
4.3.5. Interest rate
Both Tsatsaronis and Zhu (2004) and Humea and Sentence (2009) find that low interest rates lead to booms in both credit and house markets. This could indicate a negative
coefficient for the interest rate in the first regression. Or, perhaps a low interest rate leads to booms of the financial cycle while curbing downturns. This theory would be endorsed by a negative coefficient in the second regression.
Both the long- and short-term interest rate are included as to assess which variable is more important in explaining financial instabilities in our group of countries.
4.3.6. Money growth
From Goodhart and Hoffman (2008) a positive coefficient is expected for money growth in the first regression. And also for global liquidity a positive sign is expected, according to Agnello and Schuknecht (2011), as they find that an increase in global liquidity increases the probability of a boom.
If an increase money growth or global liquidity increase the probability of a boom but
decrease the probability of a downturn, then positive signs for these two variables would be expected in the second regression.
17 For the absolute value of the current account a positive coefficient is expected when
regressed on the absolute value of the financial cycle, as Obstfeld (2012) recognizes that current account imbalances can signal financial instability.
A negative coefficient is expected when, in regressions 2, the current account is regressed on the actual value of the financial cycle. This is expected since Mendoza and Terrones (2008, p. 11) find that the current account as percentage of GDP moves in the opposite direction of credit during a credit boom.
5.
Results
5.1. Regression 1
In this first regression the absolute value of the financial cycle is regressed on the cycles of a range of variables that in the literature have been linked to financial variation.
Table 1 shows that the one-quarter lagged values of the regressors trade openness,
inflation, short-term interest rate, money growth and the CA are significant. These variables thus have a role in enlarging (decreasing for the short-term interest rate, as it has a negative coefficient) fluctuations of the financial cycle and thus increase financial instability.
The R2 ranges from 0.1659 to 0.2189. This is not very high value for the R2 as it means that the model is able to explain at best 22% of what determines the variance of the size of the financial fluctuations.
Now we will take a closer look at the explanatory variables with significant coefficients.
5.1.1. Trade
Table 1 shows a higher level of growth in trade positively influences size of financial fluctuations. The effect is robust, as the value of the coefficient does not change by much and is still significant in specification 5 , where the insignificant variables are left out. This is in line with Borio’s (2012) claim that trade openness increases financial instability. According to Borio this effect occurs through positive supply movements that are caused by a higher level of trade. One would thus expect that when trade is left out of the regression that the effect of more trade would now run solely through the coefficient for GDP growth. But, when trade is left out of the regression (specification 2) the coefficient for GDP growth becomes significant, but negative. This is in conflict with the literature from which a positive effect of
18 increase in trade openness would be expected on the absolute value of the financial cycle’s amplitude.
5.1.2. Inflation
Furthermore from table 1 it can be seen that inflation has a positive coefficient that is significant in four of the five specifications. This is in accordance with the literature. If the short-term interest rate is excluded from the regression, specification 3, the coefficient for inflation is no longer significant. This confirms the negative indirect effect of inflation on financial stability from the literature, as low inflation rates remove the need for tightening monetary policies (captured by the short-term interest rate as a proxy). With removal of the short-term interest rate from the regression both the direct positive effect and the indirect negative effect of inflation on financial stability are combined in the coefficient of inflation. The positive and negative effects apparently single each other out making the coefficient insignificant.
5.1.3. Interest rate
The short-term interest rate has a negative and significant effect on the size of the
fluctuations of the financial cycle. This indicates that the upturns as well as the downturns get amplified by a decreasing interest rate. This is robust, as it is significant in all columns. The long-term interest rate is insignificant, which indicates that the short-term rate is more important is this respect. This could mean that overall, floating rate contracts are more prevalent in the studied countries. If the short-term interest rate is excluded it shows that the long-term interest rate now becomes significant with a negative coefficient . This makes sense, as the long-term interest rate is influenced by the short-term rate. The coefficient for the long-term interest rate now captures part of the effect of the short-term interest rate on the absolute financial cycle.
Table 1: Regression 1
(1) (2) (3) (4) (5)
VARIABLES Abs fincyc Abs fincyc Abs fincyc Abs fincyc Abs fincyc
L financial openness -0.0262 -0.0260 -0.0294 -0.0289* (0.0163) (0.0162) (0.0183) (0.0152) L trade openness 0.0447*** 0.0367*** 0.0449*** 0.0496*** (0.0075) (0.0074) (0.0076) (0.0087) L inflation 0.4414*** 0.3982*** 0.1105 0.4107*** 0.4629*** (0.1424) (0.1381) (0.1459) (0.1335) (0.1267) L GDP growth -0.0225 -0.0557** -0.0361 -0.0246 (0.0269) (0.0270) (0.0295) (0.0256) L lt interest rate 0.0062 0.0491 -0.2939*** -0.0092
19 (0.1145) (0.1358) (0.0996) (0.1120) L st interest rate -0.5901*** -0.5379*** -0.5571*** -0.5619*** (0.0967) (0.0939) (0.0998) (0.0939) L money growth 0.1211*** 0.0887* 0.0828* 0.1202*** 0.1613*** (0.0439) (0.0481) (0.0443) (0.0426) (0.0550) L global liquidity 0.0862 0.1535 0.0092 0.0843 (0.1090) (0.1105) (0.1053) (0.1075) L current account 0.0294** 0.0326** 0.0266* 0.0410** 0.0208 (0.0142) (0.0144) (0.0153) (0.0181) (0.0154)
L abs current account 0.0532*
(0.0283) Constant 0.0186*** 0.0157*** 0.0212*** 0.0183*** 0.0286*** (0.0049) (0.0045) (0.0046) (0.0049) (0.0014) Observations 2,263 2,263 2,263 2,263 2,517 Number of groups 18 18 18 18 20 Within R2 0.2108 0.1885 0.1659 0.2189 0.1832
(Tabel 1 shows a regression of the absolute value of the medium-term financial cycle on the medium-term financial cycles of a range of variables. Year and seasonal dummies, 20 countries, period 1980Q1-2012Q4. Standard errors are in parentheses with *** p<0.01, ** p<0.05, * p<0.1)
5.1.4. Money
An increase in money growth amplifies the financial cycle, as was expected from the literature. The effect is significant in three out of the five specifications.
5.1.5. Current account
Table 1 shows that the actual value of the current account imbalance is positively correlated to the size of financial fluctuation. In table 1 specification 4 shows that the absolute value of the current account is less significant than the actual value of the current account. It seems to indicate that the size of the absolute value of fluctuations of the financial cycle are not influenced by an imbalance on the current account per se, but that it matters on whether the imbalance is positive or negative. An increasing current account balance amplifies the financial cycle and thus has a destabilizing effect on the financial economy. A decreasing current account balance deamplifies the financial cycle.
While the coefficient for the current account is significant in specification 4, it seems not to be very robust as it turns insignificant in specification 5 where the insignificant variables are left out. This outcome, combined with the fact that the sign of the coefficient conflicts with what would be expected from the literature, puts doubts on the validity of this coefficient.
20 Table 2 shows the second regression, where the actual value of the medium-term financial cycle is regressed on the same set of independent variables. This regression gives insight into whether these variables cause the cyclical behaviour of the financial cycle.
For this regression the R2 is higher than for regression 1 and ranges from 0.4902 to 0.4881. The second regression is thus better able to explain the movement of the financial cycle. The fact that the R2 of this regression is higher than for regression 1 has to be caused by the use of absolute values for the financial cycle in regression 1. Using the absolute value instead of the actual value thus seems to remove some of the explanatory value of the regressors.
The second regression is able to explain around 50% of the movement of the financial cycle. This implies that the other 50% still lacks. This could indicate that there are a lot more variables that influence size of fluctuations of the financial cycle. Another explanation for the fact that not more than 50% of the movement of the financial cycle is explained could be that information on the independent variables is lost as they are studied using a band-pass filter. The filter smooths the movement of the variables and it is inherent to the technique that some detail on de variables is lost.
5.2.1. Financial openness
In contrary to regression 1, table 2 shows that the coefficient of financial openness is significant and robust for regression 2. An upward movement in financial openness thus causes an upward movement of the financial cycle while curbing the downward movement of the financial cycle. This finding thus seems to confirm that the theories of Borio (2012) and Claessens et al. are not conflicting after all.
5.2.2. Inflation
Table 2 shows that inflation has a significant and robust positive effect on the actual value of the financial cycle, as it has for the absolute value of the financial cycle. An upward
movement in inflation causes an amplification of the upward movement of the financial cycle while making downward financial fluctuations less deep. This finding is not in accordance with the literature as Claessens et al. found that a increase in inflation amplifies downturns. Furthermore it seems at odds with the fact that the first regression showed that inflation amplifies both the up- and downturns of the financial cycle.
21 One can also notice that in this second regression the long-term interest rate is significant, whereas in the first regression the short-term rate is. These results could indicate that the long-term interest rate amplifies the upturn phase and curbs downturn phase of the financial cycle, while monetary policy amplifies both the upturn and the downturn phase. This seems to be a rather questionable result as it would unlikely for the long- and short-term interest rate to have such a different effect on the financial cycle.
5.2.4. Money
As well as in regression 1, money has a positive, significant and robust coefficient. An increase money thus increases the probability of a boom but decreases the probability of a downturn. This seems to be in conflict with the result from regression 1, which indicated that growth in money both amplifies the up- and downturn phase of the financial cycle.
5.2.5. Global liquidity
Table 2 shows that the coefficient for global liquidity is both significant and robust. In
regression 1 this was not the case. Global liquidity seems to be important in the build-up and breakdown of financial imbalances but not the magnitude of these fluctuations.
Table 2: Regression 2
(1) (2)
VARIABLES fincyc fincyc
L financial openness 0.1406*** 0.1324*** (0.0292) (0.0250) L trade openness 0.0104 (0.0093) L inflation 0.6791*** 0.7379*** (0.1618) (0.1544) L GDP growth 0.0476 (0.0580) L lt interest rate -0.4629*** -0.3913*** (0.1632) (0.1424) L st interest rate 0.1351 (0.1864) L money growth 0.4322*** 0.4384*** (0.0489) (0.0468) L global liquidity 0.4585*** 0.5508*** (0.1729) (0.1972) L current account -0.0068 (0.0178) Constant -0.0093* -0.0140*** (0.0054) (0.0053) Observations 2,263 2,263 Number of groups 18 18 Within R2 0.4902 0.4881
22
(Tabel 2 shows a regression of the medium-term financial cycle on the medium-term financial cycles of a range of variables. Year and seasonal dummies, 20 countries, period 1980Q1-2012Q4. Standard errors are in parentheses with *** p<0.01, ** p<0.05, * p<0.1)
6.
Conclusion
6.1. Summary
The objective of this thesis is to find which factors influence fluctuations of the financial cycle and, by doing so, getting a better insight into the build-up of financial instability. To do so, both the absolute and the actual value of the financial cycle are regressed on the cycles of a range of variables that in the literature have been linked to financial variation. The first regression, with the absolute value of the financial cycle, gives insight into whether the independent variables influence financial instability. The second regression shows whether these variables move in the same direction as the financial cycle and thus cause the cyclical behaviour of the financial cycle.
This thesis fits in a current tendency in economic research to give a bigger prominence to finance in macroeconomic theory. The global credit crisis of 2008/09 made apparent the impact of financial instability on the real economy and showed how little is known on the causes of financial instability. Better understanding of the financial cycle and its behavior, is crucial for a better understanding of the economy and for policy design aimed at financial stability.
A few main results stand out. Firstly, from the first regression it can be concluded that a rise in either trade openness, inflation or money growth increases the size of the fluctuations of the financial cycle and thus increases financial instability. Furthermore, the results from the second regression show that fluctuations in either financial openness, inflation, short -term interest rate, money growth or global liquidity amplify upward financial fluctuations while curbing downward fluctuations. Finally, the short-term interest rate is negatively related to the size of financial fluctuations. A rise in the short-term interest rate thus increases financial stability.
6.2. Implications
These results show that financial stability is enlarged by factors that are an integral part of our current economic system: Free trade and central banks that aim at low interest rates and, when needed, stimulate the economy through monetary easing. What could be the
23 implications of these results for economic policy and the way that the economic system of most developed countries is currently set up?
The results show that increases in the level of trade increase financial instability. This would mean that the ongoing efforts of the developed (and an ever growing part of the developing) world to move towards free trade, come at a cost. It is not to say that governments should stop lowering their trade barriers, as opening up to trade gives them much prosperity. But, it is important that the implications of free trade on financial stability are taken into account by policy makers making decisions on trade policy.
Besides trade policy, the results have implications for central bank policy.
Inflation is shown to enlarge fluctuations of the financial cycle. High levels of inflation increase price uncertainty which hampers financial stability. Monetary policy aimed at price stability thus not only tends to stimulate growth but also financial stability. But, as the results show, the positive effect of low inflation on financial stability is diminished by the negative effect of low interest rates on financial stability. Low interest rates increase financial fluctuations as it reduces costs of lending. As a monetary policy goal of low inflation is usually achieved by central banks by trying to keep the short-term interest rates low, the positive effect of this low inflation on financial stability is countered by the negative effect of a low short term interest rate.
Furthermore, a positive effect that is found for money growth on financial fluctuations is relevant for current monetary policies such as the quantitive easing programme by the Federal Reserve and the current hesitation of the ECB to do the same. Once again, the effects on financial stability should be taken into account when devising such policies.
All and all, it should be realized that the monetary policies that are currently being used to repair the damaged economies after a huge financial downturn might itself make the system less stable. This research shows that there might be limitations to trade openness, fiscal expansion and monetary easing.
6.3. Further research
The first regression showed that inflation amplifies both the upward and downward
movement of the financial cycle, while the second regression showed that inflation amplifies the upward movement while decreasing the downward movement. The same effects can be said for money, this variable also had a positive coefficient in both the first and second
24 regression. These two findings seem to contradict each other and the exact effect of inflation and money on financial stability should be further researched.
Furthermore, the results for the short- and long-term interest rate are rather confusing as they seem to indicate that the long-term interest rate amplifies the upturn phase and curbs the downturn phase of the financial cycle, while monetary policy amplifies both the upturn and the downturn phase. These results ask for more research as it seems unlikely for the long- and short-term interest rate to have such a different effect on the financial cycle.
6.4. Caveats
An important limitation to the study is has only been able to look at 20 years of data. This has, since a financial has the average length of 16 years, reduced the chance that a financial cycle in full can be studied.
Furthermore the R2 for both regressions was not above 0,50. This means the explanatory power of the used regressions is not very high. This could indicate that there are a lot more variables that influence size of fluctuations of the financial cycle. Or, that information on the independent variables is lost as they are studied using a medium-term filter. The filter smooths the movement of the variables and it is inherent to the technique that some detail on de variables is lost.
25
7.
References
Agnello, L. and L. Schuknecht (2011): “Booms and busts in housing markets: Determinants and implications”, Journal of Housing Economics, vol. 20 no. 3, pp. 171-190.
Borio, C. (2012): “The financial cycle and macroeconomics: What have we learnt?”, BIS
Working Paper, no. 395.
Borio, C. and P. Disyatat (2011): “Global imbalances and the financial crisis: Link or no link?”, BIS Working Paper, no. 346.
Borio, C. and M. Drehmann (2009): “Assessing the risk of banking crises – revisited”, BIS
Quarterly Review, March, pp. 29–46.
Cavallo, E. and A. Frankel (2008): “Does openness to trade make countries more vulnerable to sudden stops, or less? Using gravity to establish causality”, Journal of International Money
and Finance, vol. 27, no. 8, pp. 1430-1452.
Chudik, A. and M. H. Pesaran (2013): “Large panel data models with cross-sectional dependence: a survey”, Federal Reserve Bank of Dallas Globalization and Monetary Policy
Institute Working Paper, no.153.
Claessens S, M. Ayhan Kose and M.E. Terrones (2011): “Financial cycles: What? How? When?”, IMF Working Paper, no. 88.
Comin, D. and M. Gertler (2006): “Medium-term business cycles”, American Economic
Review, vol. 96, no. 3, pp. 523-551.
Crockett, A. (2000): “Marrying the micro- and macro-prudential dimensions of financial stability”, Remarks by Mr. Andrew Crockett, General Manager of the Bank for International
Settlements and Chairman of the Financial Stability Forum, before the Eleventh International Conference of Banking Supervisors, held in Basel, 20-21 September 2000.
Giovanni Dell'Ariccia, D. Igan, L. Laeven, and H. Tong, with B. Bakker and J.
Vandenbussche (2012): “Policies for Macrofinancial Stability: How to Deal with Credit Booms”, IMF staff discussion note, no. 6.
Drehmann, M., C. Borio and K. Tsatsaronis (2012): “Characterising the financial cycle: Don’t lose sight of the medium term!”, BIS Working Paper, no. 380.
26 Fisher, I. (1933): “The debt-deflation theory of great depressions”, Econometrica, vol. 1, no. 4, pp. 337-357.
Goodhart, C. and B. Hofmann (2008): “House prices, money, credit, and the macroeconomy”, Oxford Review of Economic Policy, vol. 24, no. 1, pp. 180-205.
Hoechle, D. (2007): “Robust Standard Errors for Panel Regressions with Cross-Sectional Dependence” The Stata Journal, vol. 7, no. 3, pp. 281–312.
Humea, M. and A. Sentance (2009): “The global credit boom: Challenges for
macroeconomics and policy”, Journal of International Money and Finance, vol. 28, no. 8, pp. 1426–1461.
Mendoza, E.G and M. E. Terrones (2008): “Anatomy of Credit Booms: Evidence From Macro Aggregates and Micro Data”, IMF Working Paper, no. 226.
Minsky, P. (1982): “Can “it” happen again? A reprise”, Challenge, vol. 25, no. 3, pp. 5-13.
Obstfeld M. (2012): "Does the Current Account Still Matter?", American Economic Review, vol. 102, no. 3, pp. 1-23,
Portes, R. (2009): “Global imbalances”, in Macroeconomic stability and financial regulation:
Key issues for the G20. Ed. by M. Dewatripont, X. Freixas, and R. Portes, Centre for
Economic Policy Research, London.
Stock, J. and M. Watson (2011): “Introduction to Econometrics”, 3rd edition, Pearson, Boston.
Tsatsaronis, K. and H. Zhu (2004): “What drives housing price dynamics: Cross-country evidence”, BIS Quarterly Review, March, pp. 65-78.
27
8.
Appendix
A. Autocorrelation
The Woolridge test is performed to test for autocorrelation. Regression 1:
Wooldridge test for autocorrelation in panel data H0: no first-order autocorrelation
F( 1, 17) = 1.583e+06 Prob > F = 0.0000
Regression 2:
Wooldridge test for autocorrelation in panel data H0: no first-order autocorrelation
F( 1, 17) = 73750.916 Prob > F = 0.0000
So for both regression 1 and 2 the variables are autocorrelated.
B. Cross-sectional dependence
Pesaran's test of cross sectional independence = -6.351, Pr = 0.0000 Average absolute value of the off-diagonal elements = 0.280
C. Unit root
The variables credit-to-GDP, house prices and financial openness have unit root. To eliminate unit root four-quarter differences in log levels are taken. For credit-to-GDP and financial. Since house prices are index numbers, year on year growth rates ar taken. The
28 stata outcomes below show that these three variables no longer have unit root after four-quarter differences in log levels/ year on year growth rates are taken.
29 Housing:
30 Financial openness:
