1
Master Thesis
The Influence of Households and Firms’ Debt Service Ratios
on Financial Instability
By Mats de Vries
MSc International Economics and Business Faculty of Economics and Business
University of Groningen
Abstract: Following Minsky’s Financial Instability Hypothesis this paper aims to prove a
positive influence of debt payments as a share of income, measured at household and non-financial corporations, on non-financial instability. Debt payments as a share of income are introduced as an indicator of financial fragility and are measured by the recently released Debt Service Ratio (DSR). To capture financial instability the Financial Stress Index (FSI) is introduced which is an aggregate composite variable. In line with Minsky’s theory an increase in debt service ratios increases financial instability. Focussing on the components of the DSR it appears that an increase in Household DSR increases financial instability, while the influence of non-financial corporation DSR is insignificant.
Keywords: financial instability, financial fragility, Financial Stress Index, Debt Service Ratio JEL: C15, C23, O16, E50
2
1. Introduction
In 1963 Hyman Minsky released the Financial Instability Hypothesis, which is with hindsight a theory explaining the origination of the financial crisis of 2007-2008. According to the theory financial crises are unavoidable because economic prosperity causes borrowers and lenders to be reckless. In fact, reckless to such an extent that borrowers are unable to meet their debt obligations anymore. Financial institutions, however, keep lending money in the hope that asset prices rise to enable repayment. This optimistic behaviour creates assets bubbles, increases financial fragility, and eventually leads to financial instability. Since its release and up until the latest financial crisis Minsky’s theory has been largely ignored by mainstream economists. Currently, however, economists start drawing inspiration from Minsky’s work, as did I for this paper.
In this paper I will investigate if an increase in debt obligations as a share of income indeed increases instability in the financial system. The role of the debt payments, consisting of interest payments and amortizations, as a cause of instability is currently underexposed in the literature. Investigating how the financial system is affected by debt obligations was difficult since extensive data on debt payments was not available. However, Drehmann et al. (2015) recently constructed the Debt Service Ratio (DSR), which is a variable measuring debt payments for both households and non-financial firms on country-level for 18 advanced economies. This allows me to use the build-up of debt obligations as an indicator of financial fragility at household and non-financial firms. To measure financial instability the Financial Stress Index composed by Cardelli et al. (2011) will be used. The DSR and FSI allow me to test the impact of debt payments as a share of income on instability in the financial system.
3
2. Literature review
In this paper I will investigate if an increase in debt service ratios as an indicator of for financial fragility has an impact on instability in the financial sector. Key to understanding this relationship is to define the elusive concepts of “financial fragility” and “financial instability”. In the literature the two terms are often used interchangeably which can cause confusion. The next section provides a review of available literature on definitions and measurements proving that the two terms are distinctive.
Financial fragility
An adequate definition of financial fragility is given by Allen and Gale (2004):
“A situation where a small aggregate shock in the demand for liquidity leads to disproportionately large effects in terms of default or asset-price volatility.”
Also Minsky (1986, p. 233) describes the phenomenon financial fragility:
“The robustness or fragility of the financial system depends upon the size and strength of the margins of safety and the likelihood that initial disturbances are amplified.”
In other words, financial fragility is a situation in which the financial system is vulnerable to small shocks developing into full-blown financial crises. According to Kindleberger and Aliber (1978, pp. 104-105), these shocks or disturbances can be trivial: a refusal of credit to some borrower, a credit flight, a bankruptcy, et cetera.
Whereas definitions of financial fragility are available, a straightforward measure or indicator is not. Consequently, directly observing and measuring fragility in the financial system is challenging. The financial system is constantly hit by small shocks, since every day businesses file for bankruptcy or banks refuse credit to customers. In general, this would not harm the financial sector on a great scale and fragility would not be observed. Only when a small shock tips the financial sector over the edge into a financial crisis we know we were experiencing financial fragility1. It seems that financial fragility is only
observable in hindsight, after the start of a financial crisis.
While directly measuring financial fragility seems impossible, studies are available discussing indicators or indirect measures of financial fragility. Bezemer and Grydaki (2014) argue that a situation of macroeconomic stability (they look at the Great
4 Moderation) combined with strong growth of assets prices, or credit to asset markets or credit relative to output, are all potential indicators of rising financial fragility. The Great Moderation (1984-2008) fits this description and is used as a period facing financial fragility which ultimately led to the financial crisis of 2007-2008. Using Granger causality tests they find that credit growth towards financial and real estate sectors (FIRE-sector) is more driven by past credit growth than by output growth. Credit to the FIRE-sector is considered to be non-productive because it does not cause economic growth. Bezemer and Grydaki (2014) eventually conclude that financial deepening as measured by credit growth causes the economy to be more vulnerable to shocks. Hence, they argue that credit growth is an (indirect) measure of financial fragility.
Adrianova et al. (2015) constructed a new database on financial fragility covering 124 countries over the period 1998 to 2012. The database uses bank-level data to create eight different measures of financial fragility each focusing on a different aspect of vulnerability in the financial system. These measures are capitalisation, asset quality, managerial efficiency, earnings, liquidity, risk exposure and the Z-score (bank insolvency measure). Loayza and Rancière (2006) find a negative relationship between financial intermediation and growth in the short run, which can be explained by the occurrence of financial fragility. They capture the effect of financial fragility by banking crises and financial volatility (the standard deviation of the growth rate of the private credit to GDP ratio). Their findings include that in the short run countries with more banking crises and higher financial volatility experience in the short-run a more negative effect of financial intermediation on output growth than for countries with less banking crises and lower volatility. Tymoigne (2011) constructs an index to measure financial fragility for residential housing in the United States, United Kingdom and France. Based on Minsky’s Financial Instability Hypothesis, his index aims to measures financial fragility by detecting the Ponzi finance schemes in the three countries. According to Tymoigne, Ponzi finance schemes can be detected by analysing interaction of asset prices and debt. An increase in debt increases asset prices, which, in turn, increases the amount of debt; this snowball effect is an indicator of financial fragility.
5 literature which measure fragility on a country-level have in common that their indicator of financial fragility is related to credit. As will become clear from the next section the extensive amounts of credit provided is observed as one of the main causes of the financial crisis of 2007-2008 and thus proves to be a good indicator of financial fragility. In this paper the Debt Service Ratio, which is also related to credit, is introduced as an indicator of fragility.
Financial instability
The concept financial instability is best defined by Allen and Wood (2005), who defined episodes of financial instability as:
“episodes in which a large number of parties, whether they are households, companies, or (individual) governments, experience financial crises which are not warranted by their previous behaviour, and where these crises collectively have seriously adverse macro-economic effects.”
In other words, financial instability is the outcome when a fragile financial system unexpectedly tips over and turns into a crisis2.
Financial instability is in comparison to financial fragility easier to measure since financial crises are observable: stock markets plummet, interbank rates increase tremendously, and currencies start depreciating at a fast rate. This enumeration directly shows that financial instability is a widely spread phenomenon which makes it difficult to capture it with one variable. Gadanecz and Jayaram (2008) argue that the measurement of financial instability is difficult due to complex interactions of different elements of the financial system among themselves and with the real economy. For six sectors they summarize financial instability measures commonly used in the literature3. Besides sector-specific
measures there are also overarching aggregate measures of financial instability. Illing and Liu (2003) were the first authors to construct a composite indicator of financial stability. Their indicator contained variables reflecting financial instability in the banking system, the foreign exchange market, and the equity market. This work has been expanded by Cardarelli et al. (2011) who created the Financial Stress Index (FSI). The FSI is a composite
2 In this paper financial crises are sometimes denoted as financial instability, since a financial crisis is a period featuring high financial instability.
6 variable identifying episodes of financial stress based on high-frequency price variables that can signal instability in the banking, securities, or foreign exchange markets. In this paper the FSI is used as a proxy for financial instability.
Summarising, it seems that financial instability emerges from financial fragility. A situation of financial fragility does not imply, however, that the economy is already in a crisis. It only implies that the system is vulnerable to shocks which may cause a financial crisis. We speak of financial instability when a fatal shock hits the economy starting a financial crisis. It is also quite possible that a country has a very fragile financial sector while financial instability is low. Before the financial crisis of 2007-2008 few people were worried about the occurrence of a financial crisis. In retrospect can be concluded that credit levels and credit growth before the crisis were unsustainable and eventually led to financial instability. In this paper I will investigate if a similar effect on financial instability, as measured by the Financial Stress Index, is observable when focusing on the debt service ratios of household and non-financial corporations instead of credit.
The role of credit as a cause of financial instability
The influence of debt service ratios on financial instability is underexplored. The role of credit, which is related to debt service ratio, however, has been studied extensively. Through time the conception of the effects of financial intermediation has shifted. Mainstream economists argued that financial intermediation had a positive influence on the real economy. After the outbreak of the crisis of 2007-2008, the idea that financial intermediation is a potential cause of financial crises and the subsequent recessions gained awareness. I will show, however, that already long before the latest crisis, scholars were arguing the negative effects of too much credit.
7 successful innovation and thus enhances growth. Beck et al. (2001) updates Goldsmith’s work for the period 1980-1995 and find that economies grow faster in countries with higher financial development and a legal system that effectively protects the rights of outside investors. Rousseau and Wachtel (2011), however, state that after 1990 the positive credit-growth correlation diminishes due to the occurrence of financial crises. They find a positive relationship between finance and growth if credit lending is not done excessively. A credit boom, contrarily, can lead to an unstable banking sector and inflationary pressures.
Financial deepening and accompanying credit growth could thus lead to output growth, however simultaneously it could lead to financial imbalances and an economy that follows a bust and boom cycle. Already in the early 20th century Mises (1912) used Wicksell’s
(1898) idea of a distinction between the natural interest rate and the market interest rate to explain busts and booms in the economy. Like Mises, Schumpeter (1939) believes that the economy follows a cycle. After the bust, resources will be relocated from old businesses to new and more productive businesses. Schumpeter emphasises the important role of credit creation in this process by enabling entrepreneurs to innovate. More recently, Allen and Gale (2000) argued that the booms originate because of agency relationships in the banking sector since lenders cannot observe how risky borrowers’ investments are. This risk shifting leads to bubbles in asset prices since more credit is invested. Moreover, they show that financial instability occurs when central banks do not meet disproportionate credit expansions expectations of investors.
8 the banking sector growths heavily. Jorda et al. (2013) find a negative influence of credit on the real economy. Higher credit intensity during the boom leads to more severe recession after the bust for both business-cycle recessions and financial crises. Brunnermeier and Sannikov (2014) include the financial sector in a macroeconomic model. They show that when total risk goes down, due to financial innovations as securitization, and equilibrium leverage increases, a situation of financial fragility is created. This is also known as the volatility paradox, since measures to improve risk management lead to an increase instead of a drop in systematic risk.
Besides the extensive literature about the role of credit in causing financial fragility, also more recently the composition of credit in causing financial fragility has been explored. Overall it is argued that credit for productive use is sustainable and that credit to asset markets and housing markets (mortgages), also known as the Financial, Real Estate and Insurance (FIRE) sector, is unsustainable (Bezemer and Grydaki, 2014; Kalemli-Ozcan et al., 2012). Büyükkarabacak and Valev (2010) show that an increase in household debt has no influence on long-term income. Instead, it causes financial fragility which can cause a banking crisis. An increase in enterprise credit can bring about the same effect, however, it is also a source for an increase in long-term income growth. Beck et al. (2012) observe that there is a relative increase in household credit over non-financial firm credit. They find that credit to non-financial firms has a positive influence on economic growth while credit to households has no effect. Jorda et al. (2014) observe that the share of mortgages loans in banks’ total outstanding loans has increased from 30% in 1990 to 60% in 2014. Traditional banking activities as channelling household savings to productive non-financial investments are replaced by extensive mortgage lending. The real estate credit boom over the years has become an important indicator of financial fragility. Zhang and Bezemer (2015) build on Jorda et al. (2014) and find a negative effect of the share of household debt (in the form of mortgages) in total credit before a financial crisis on the depth and growth loss of economic recessions after a financial crisis.
Debt payments as an indicator of financial fragility
9 progressively reckless. This causes a fundamentally unstable economy that moves from boom to bust. To explain this Minsky distinguishes three income-debt relations for economic units, namely hedge finance, speculative finance and Ponzi4 finance. Hedge
finance is most common after a financial crisis when borrowers are still cautious, outstanding loans are low, and amortizations and interest rates are paid easily. During the speculative stage confidence rises and banks increase lending to business. Speculative finance units are units that are able to meet their interest payments commitment but, cannot pay their amortizations out of their current cash flows. In the Ponzi stage banks make loans to firms and households which are not able anymore to fulfil their principal and interest payments. Business still receive loans because banks assume that asset prices (of the collateral) to go up and that this value can cover up the initial loan and missed interest payments. During a boom, when the economy is near full employment the ratio of speculative and Ponzi finance schemes increases, asset markets seize up, and cash-flows of financial institutions become more sensitive to interest rates. This leaves the financial system very fragile. Then at a certain moment, like during the financial crisis of 2007-2008, an exogenous shock causes the asset bubble to burst. This is now known as the “Minsky-Moment” (Lahart, 2007), and marks the start of a new financial crisis. The Financial Instability Hypothesis shows that, theoretically an increase in debt payments indicates an increase in financial fragility. The role of debt payments as an indicator of financial excesses and vulnerabilities is still underexposed despite the relevance of Minsky’s paper in explaining the financial crisis of 2007-2008.
Dynan and Kohn (2007) show that up to 2008 U.S. households were increasingly accumulating debt. Factors explaining this rise are demographic shifts, increases in house prices, and financial innovation. The consequence was an increasing sensitivity of U.S. households to economic shocks. Dynan et al. (2013) continue that when a large share of household income is spent on debt repayment, households have fewer funds available to spend on goods and services. When households have high debt levels to income they are also more likely to default on their obligations when facing misfortune such as unemployment or prolonged illness. They also highlight that due to financial innovation U.S. households have increased their debt payments to income since financial innovation makes borrowing easier. Johnson and Li (2010) find that households with relatively high
10 debt service ratios are more likely to be turned down for new credit applications. They argue that therefore the debt service ratio is a good indicator for measuring household borrowing constraints.
The literature review shows us that the role of credit as an indicator of financial fragility and as a cause of financial crisis has been studied extensively. In the 1980s and 1990s financial liberalisation measures allowed banks to create an abundance of cheap credit which drove up asset prices which led to the creation of asset bubbles. More recently, scholars start investigating the distinctive effect of several types of credit. It appeared that credit to enterprises is productive and creates economic growth. While credit to households, mostly in the form of mortgages, leads to the build-up of asset bubbles and eventually to the outbreak of a financial crisis. However, the role of debt payments in explaining financial instability is still underexplored. Minsky (1978) showed that the build-up of debt obligations are an important source of financial instability. More recently, Dynan and Kohn (2007) and Dynan et al. (2013) showed for the US that an increase in debt payments can create financial vulnerabilities. By means of this study I will contribute to and extend the existing literature in several ways. I will prove that there is a positive and significant link between the role of debt service ratios, measured at the level of households and non-financial corporations, and financial stability. Additionally, statistical proof will be established, showing that the distinctive effect between household and firm credit is also observed with regards to their debt service ratios.
3. Methodology
The aim of this paper is to investigate the influence of debt payments measured at households and non-financial corporations on instability in the entire financial sector. First, I will introduce the Financial Stress Index (FSI) as a measure for financial instability and the Debt Service Ratio (DSR) as an indicator for financial fragility at households and non-financial corporations. Next, an overview of the econometric framework is introduced including several fixed-effects baseline models. Lastly, several methods to test the robustness of my results are presented.
Measuring financial instability
11 introduced by Cardelli et al. (2011)5. The FSI is constructed on a quarterly basis as a
composite of variables indicating stress in the banking sector (banking sector beta, TED
spread, and the slope of the yield curve), the securities market (corporate bond spreads, stock market returns, and return volatility) and the foreign exchange market (return volatility)67. The advantage of using the FSI is, according to Cardelli et al. (2011), that the
index facilitates the identification of four main events related to financial crises, namely large shifts in asset prices (stock markets returns; and slope of the yield curve, the flatter or more negative the slope becomes the less profitable bank activities are), an abrupt increase in risk/uncertainty (volatility on stock and foreign exchange markets, volatile markets are considered risky by investors), sudden shifts in liquidity (TED spreads; higher TED spreads indicate higher perceived counterparty risk by banks), and the health of the banking system (slope of the yield curve, affecting the profitability of banks by lowering the difference between their income (long-term loans) and their costs (short-term deposits); and the beta of banking-sector stocks, a higher beta indicates more systemic risk indicating an unhealthy banking-sector). The index is a trend variable with a mean of zero. A value of the FSI above its trend indicates an increase in financial instability, and vice versa.
The FSI is favoured over other measures of financial instability for several reasons. Gadanecz and Jayaram (2008) argue that measurement of financial instability is difficult due to complex interactions between and within several components of the financial sector. The FSI overcomes this problem since it has good coverage of the entire financial sector. Cardelli et al. (2011) argue that many indicators of financial instability are dummy variables only capturing full-blown financial crises. The FSI is a continuous variable, it does not only report full-blown crises but also “near misses”. Another advantage over indicator variables is the ability to identify the beginning and peaks of financial instability episodes. Lastly, since the FSI is a composite of sectors it is possible to analyse the sectors included separately.
5 In view of Cardelli et al. (2011) financial stress can be interpreted as financial instability.
6 “The FSI for each country is constructed as a variance-weighted average of seven variables, grouped into three subindices which are associated with the banking, securities and foreign exchange markets. The index is constructed by taking the average of the components after adjusting for the sample mean and standardizing by the sample standard deviation. Then, the index is rebased so that it ranges from 0 to 100. An increase in the FSI equals higher financial fragility (Cardelli et al., 2011).”
12
Measuring financial fragility at households and non-financial corporations
To measure the effect of debt payments as a share of income on instability in the financial system I will use the Debt Service Ratio (DSR), constructed by Drehmann et al. (2015). On country-level it is measured for the total non-private financial sector and its components households and non-financial corporations. It is measured as the imputed amount of income used for interest payments and amortizations8. The DSR is dependent on the
principal, the interest rate, and the maturity: lower (higher) interest rates and longer (shorter) maturity means a lower (higher) DSR since fewer (more) payments are dispersed over an extended (shortened) period9.
Introducing main research topics
The construction of the DSR allows me to investigate Minsky’s (1978) Financial Instability Hypothesis empirically. He argued that due to an excessive increase in debt accumulation borrowers became unable to fulfil their interest and amortization payments, hence increasing financial fragility. During this build-up of fragility households and firms become more exposed to shocks in asset prices through the greater leverage in their balance sheets, and more exposed to unexpected changes in income and interest rates because of higher debt payments relative to income. Ultimately, such a situation potentially leads to a financial crisis. Therefore, this paper investigates if an increase in debt service ratios leads to an increases in financial stability (model 1).
As mentioned before, Büyükkarabacak and Valev (2010) argue there is a clear distinction between uses of credit and their role in creating financial instability. Credit to non-financial firms, increases investments in capital goods and increases productivity and thus enhances economic growth permanently. Household credit which consists mainly of mortgages, however, does not lead to permanent economic growth. Rapid expansions of household credit lead to an increase in asset prices generating financial vulnerabilities. Moreover, Dynan and Kohn (2007) and Dynan et al. (2013) showed that U.S. households which spend a large share of income on debt repayments have fewer funds available to
8 Data on amortizations are not available. Therefore, the authors used an approach to measure the amortisations indirectly. The aggregate DSR for sector j at time t is calculated as follows:
Dj,t denotes the total stock of debt, Yj,t denotes quarterly income, ij,t denotes the average interest rate on the existing stock of debt per quarter and sj,t denotes the average remaining maturity in quarters.
13 save or to spend on goods and services, increasing their sensitivity to financial shocks. The main aim of this paper is to investigate if an increase in the DSR, as an indicator of financial fragility at households and firms, indeed increases financial instability. Since the DSR is constructed separately for households and non-financial firms I will also explore if the distinctive effect between households and firms is observable when considering debt payments instead of credit as an indicator of financial fragility (model 2).
Econometric framework
For baseline model 1 a panel regression is estimated controlling for unobserved country fixed effects to assess the impact of debt service ratios on financial instability. Equation 1 shows the baseline model:
𝐹𝑆𝐼𝑖,𝑡 = 𝛼 + 𝛽1𝑇𝑜𝑡𝑎𝑙 𝐷𝑒𝑏𝑡 𝑆𝑒𝑟𝑣𝑖𝑐𝑒 𝑅𝑎𝑡𝑖𝑜𝑖,𝑡+ 𝛾′𝐶(𝑖)𝑖,𝑡+ 𝜑𝑖 + 𝜙𝑡+ 𝜖𝑖,𝑡 (1) where β1 captures the relation of the total DSR10 with financial instability (as measured
by the FSI) of country i in quarterly period t, Ci,t is a set of control variables, φi controls for
unobserved country-specific but time-invariant effects, and ϕt controls for the
unobserved country independent time-specific effects. Lastly, ϵi,t is the error term.
Equation 2 shows model 2:
𝐹𝑆𝐼𝑖,𝑡 = 𝛼 + 𝛽1𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝐷𝑒𝑏𝑡 𝑆𝑒𝑟𝑣𝑖𝑐𝑒 𝑅𝑎𝑡𝑖𝑜𝑖,𝑡 + 𝛽2𝑁𝑜𝑛
− 𝑓𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝑐𝑜𝑟𝑝𝑜𝑟𝑎𝑡𝑖𝑜𝑛 𝐷𝑒𝑏𝑡 𝑆𝑒𝑟𝑣𝑖𝑐𝑒 𝑅𝑎𝑡𝑖𝑜𝑖,𝑡+ 𝛾′𝐶(𝑖) 𝑖,𝑡
+ 𝜑𝑖+ 𝜙𝑡+ 𝜖𝑖,𝑡
(2)
where β1 captures the relation of the household DSR and β2 captures the relation of the
non-financial corporation DSR with financial instability (as measured by the FSI) of country i in quarterly period t, respectively.
To assess the strength of the link between the debt service ratios of households and firms and financial instability, several control variables are added to the regression. The first control variables is GDP growth (I) to control for the effect of economic growth on financial stability. Moderate economic growth is received positively on financial markets leading to stability. Next, inflation (II) is included. Boyd et al. (2001) find that inflation negatively influences financial sector performance. House price appreciation (III) is added to control for rapid expanding real estate markets putting pressure on financial markets
14 and creating instability. Financial openness (IV) is added to control for international shock transmission. Claessens and van Horen (2012) mention that shocks to parent banks can be transmitted to their foreign subsidiaries and vice versa; this contagion effect is expected to increase financial instability. The model includes Trade openness (V) since according to Claessens et al. (2012) international trade is a transmission channel of financial instability. To control for financial liberalisation, the variable credit market
deregulation (VI) is added to the specification11. The effect of financial liberalisation on
financial instability is ambiguous. According to Dell'Ariccia and Marquez (2004) less regulation lowers the screening of potential borrowers which increases financial fragility and eventually causes a financial crisis. De Meza and Webb (1987) argue that financial liberalisation lowers the cost of borrowing, making it easier to repay loans, thus lowering instability. Following the work of Claessens et al. (2010) who also used FSI as an independent variable, GDP per capita (VII) is included to complete the baseline specification.
Next, tests will be performed with two time-related dummies to control for country invariant events that took place over the timespan of my sample. A major event in the period studied in this paper is the financial crisis of 2007-2008. To control for this period two dummy variables are added, namely Financial Crisis and Pre-crisis. Financial crisis is created using Laeven and Valencia’s (2012) systemic banking crises database which contains all systemic banking, sovereign debt, and currency crises appearing in the period 1970-201112. Pre-crisis is added to examine if the effect of the DSR on the FSI before the
crisis is different than after the crisis. After a financial crisis financial instability falls; however, debt service ratios tend to stay stable or even increase slightly. Usually interest rates do not drop immediately after a crisis, income on the other hand does fall which could lead to an increase in the DSR (Drehmann et al., 2015).
Instead of looking at the sole impact of the DSR on the FSI, it is also interesting to observe whether this impact is different when interacted with other indicators of financial fragility. In the run-up to the financial crisis house prices appreciated quickly (Claessens et al., 2010). Secondly, high credit levels, credit composition, and credit growth have been argued to be indicators of financial fragility (Büyükkarabacak and Valev, 2010; Jorda et
11 Appendix Table A.2 provides definitions and sources for all control variables.
15 al., 2013; Bezemer and Grydaki, 2014). The previous measures indicate vulnerabilities in households and firms their balance sheets. Interacting the DSR with the financial fragility indicators allows me to observe under which circumstances the debt obligations increase the FSI.
Robustness
To test the robustness of the Financial Stress Index as a proxy of financial instability I will replace the DSR with credit-to-GDP ratios. An increase in credit-to-income ratios have proven to be a cause of financial crisis and economic recessions (Büyükkarabacak and Valev, 2010; Jordà et al., 2013; Jordà et al., 2014). If the FSI is true indicator of financial instability then it is expected that an increase in credit-to-GDP ratios should lead to an increase in financial instability. To test this total DSR will be replaced by total debt/GDP, Household DSR will be replaced by Household debt/GDP, and Non-financial corporation DSR will be replaced by non-financial corporation debt/GDP.
This paper tries to determine the causal effect of debt obligations in the private sector on instability in the financial sector. However, it could be possible that it is not an increase in the DSR that explains an increase in the FSI, but an unobserved variable correlating with the DSR and determining the FSI. Another possibility could be the existence of a loop of causality, so that the DSR not only determines the FSI, but also vice versa. The former two results are endogeneity problems, which occurs when one of the explanatory variables is correlated with the error term. The problem of reversed causality is also known as a simultaneity bias. As the DSR and FSI are observed in the same period it is possible that the FSI influences the DSR. The same goes for the other explanatory variables and their link with the FSI. To overcome this, all independent variables will be lagged with one period in all specifications.
To overcome other endogeneity problems several methods are available and in this paper I will introduce two of them13. The first method is the instrumental variable (IV) two stage
least squares (2SLS) fixed effects estimator. To perform this method I need to find instrument variables for the endogenous DSR that satisfy two conditions: (1) the IVs must be uncorrelated with the error term, (2) the IVs must be correlated with the DSR. In other
16 words, the instrument variables only affect the FSI through their effect on the DSR. Based on the necessary conditions, lagged values of the DSR and first differences are used as independent variable. To test for over-identification of the instruments the Sargan-Hansen test is applied (Hill et al., 2012).
The second method applied to solve endogeneity problems is a Generalized Method of Moment (GMM) dynamic panel model developed by Arellano and Bover (1995) and Blundell and Bond (1998). By combining both regressions in levels and in differences, it yields unbiased estimators. One-period lagged level variables are used as instruments in the differenced equations and one-period lagged differences are used in the level equations. However, in this paper only the difference equation will be applied. Following Roodman (2009) it is advised for a paper with many time observations and a small number of countries to only use the difference GMM estimators. Using both equations would double the number of instruments since two equations have to be estimated. To test the validity of the instruments a Hansen test for over-identifying restrictions will be applied. Furthermore, an Arellano-Bond test to detect autocorrelation in the error term is carried out.
4. Data
Data collection and descriptive statistics
An extensive database is constructed containing quarterly data for 17 advanced countries for the period 1999Q1 – 2014Q414. The choice of the country sample and the observed
timespan is restricted by the availability of data on the Financial Stress Index and the Debt Service Ratio15. I received an updated version of the Financial Stress Index database from
Dr. Lall from the IMF. Data on the DSR is publicly available at the website of the BIS. The data for all control variables is collected at publicly accessible sources. Data on financial indicators are collected from databases provided by the Bank for International Settlements (BIS), complemented by data from the IMF: Joint External Debt Hub, the Fraser Institute’s Economic Freedom Indicators, and a database on systemic banking crises from Laeven and Valencia (2012). Data on macroeconomic indicators is collected
14 Appendix D Table A.2 lists the countries in the sample.
17 from statistics provided by the OECD and the International Financial Statistics (IFS) database from the IMF. To my knowledge, the database constructed for this paper is unique since no database with similar coverage and detail has been collected and reported before.
Table 1 reports the descriptive statistics of the variables used for the baseline model specification. The mean of the FSI in the sample is -0.283, indicating that the period 1999-2014 was relatively stable.16 This seems odd since the financial crisis of 2007-2008 took
place in this period, however, the financial crisis was preceded and followed by a long financially stable period. Only during the crisis years of 2007-2008 instability in the financial system was high. The FSI reached its maximum of 17.45 in 2008Q4 in the Netherlands, the minimum score of -6.79 was reported in Italy in 2011Q4.17 Total private
non-financial sector DSR has a mean of 18.5%. Households spent on average 10.2% of their income on debt payments, while non-financial corporations spend a much higher share, of 42.8%, of their income on fulfil debt obligations. The private non-financial sector in Denmark has the highest DSR which was recorded in 2008Q4, while in 1999Q1 in Italy the lowest DSR is reported. Remarkable is the maximum non-financial corporations DSR
16 The complete FSI which covers the period 1979-2014 has a mean of 0.
17 The low Italian FSI is almost solely driven by a low score on the corporate bond spread component (corporate bond yields minus long-term government bond yields). In late 2011, long-term government bond yields were very high in Italy, while corporate bond yields were low which resulted in low FSI scores.
Table 1: Descriptive Statistics
VARIABLES N mean sd. min max
FSI 1,088 -0.283 2.934 -6.790 17.45
FSI Banking 1,088 -0.224 1.387 -4.007 6.629
Total private non-financial sector DSR 1,088 0.185 0.0472 0.0910 0.326
Non-financial corporation DSR 1,024 0.428 0.107 0.182 0.834
Household DSR 1,024 0.102 0.0420 0.0290 0.237
House price appreciation 1,024 196.9 76.19 60.62 413.8
Trade openness 1,088 0.723 0.327 0.180 1.694
GDP per capita 1,088 39,436 7,474 24,763 62,152
GDP Growth (Q-on-Q) 1,088 0.00393 0.00840 -0.0690 0.0350
Financial openness 1,088 1.262 0.964 0.215 5.472
Credit market deregulation 1,088 9.228 0.814 5.700 10
18 score of 83.4% recorded in Norway in 1999Q2. An explanation could be that in 1999 the Norwegian economy featured high interest rates, increasing debt service payments and a slight drop in asset prices.
The control variables show that the sample countries have high on average a high score for financial openness (foreign financial assets and liabilities as a share of income), indicating that the financial systems of sample countries are well connected with the rest of the world. Noteworthy is that trade openness (measured as the share of exports plus import of GDP) is on average high with 72.3%. The sample contains trading nations like the Netherlands and Belgium, while Japan and the USA are more dependent on their domestic economy. Since the sample only contains advanced economies, real GDP per capita is high with an average of $39,436, and with Portugal being the poorest and Norway the richest country, respectively. GDP growth is low with an average growth of 0.39% per quarter. Moreover, inflation is low with an average of 0.44% per quarter. Credit markets are highly deregulated with on average almost a perfect score of 9.228 out of 10. Low values were reported in Portugal and United Kingdom in 2009 indicating tighter regulations after the financial crisis of 2007-2008. House price appreciation is an index variable with 1995 as base year. On average the sample countries saw an increase of 96.9%. Housing prices appreciated the most in Norway, where prices since 1995
-5
0
5
10
1999q4 2004q4 2009q4 2014q4
FSI Total FSI Banking
1999Q1-2014Q4
Financial Stress Index
19 increased up to 2014Q3 with 313.8%. Japan, on the other hand, saw its housing prices fall with a minimum reached in 2009Q2.
Trends in the data
Figure 1 depicts the average trends of the independent variables total Financial Stress Index and the banking sector Financial Stress Index for the period 1999Q1 until 2014Q4, respectively. It clearly shows that the period 2007-2008 featured unusual high instability. Both variables start increasing around mid-2007, first mainly due to an increase in instability in the United States and United Kingdom. As of 2008 the financial crisis spread to Europe and Japan displayed by an even sharper increase in both indexes. The average FSI reached its peak in the fourth quarter of 2008 with an index score of 9.9. Figure A.1 in appendix G shows that many countries experienced even more stress in 2008Q4, like Japan (14.66), The Netherlands (17.45), Belgium (14.76), United Kingdom (14.78), and the United States (15.08). On the contrary, Finland (2.52) and Switzerland (2.58) exhibited almost no instability. The period before the crisis exhibited low instability, a crisis was not expected. The period of low financial instability before the crisis may even have been a catalyser for the eventual financial crisis of 2007-2008. After the crisis, governments worldwide intervened in capital markets to ensure stability. Another spike
0 .1 0 0 .2 0 0 .3 0 0 .4 0 0 .5 0 1999q4 2004q4 2009q4 2014q4 TPNFS DSR Household DSR Non-financial corporation DSR 1999Q1-2014Q4
Debt Service Ratio
20 noticeable is around the end of 2000, probably indicating the burst of the dotcom-bubble. The FSI banking indicator follows nearly the same pattern since FSI banking is one of the composites of the total FSI. Moreover, the correlation of 0.71 between the two variables also indicates a joint pattern. The difference between two variables is largest during the crisis years, since other composites of FSI also have extreme values driving the FSI score up.
In figure 2, the average trends of the several Debt Service Ratio variables are displayed. The non-financial corporations DSR is on average 32.6% higher than household DSR indicating that households spend less income on their obligations than non-financial firms. Firms need more debt to finance their activities than households, resulting in higher debt obligations. When comparing the FSI trend lines with the DSR trend lines the existence of the financial crisis of 2007-2008 is hard to observe. However, when zooming in on the period preceding the crisis of 2007-2008, a noticeable increase in all three variables can be noticed. Especially the non-financial corporation DSR increased tremendously, with an increase over 6% in the period 2006Q1-2009Q2. This corresponds with Minsky’s theory which suggested that in periods of prosperity, and financial stability firms increase their debt levels. Figure A.3 in appendix G shows that especially firms in Denmark (16%) and Norway (25%) saw a sudden increase in their debt payments in the
-5 0 5 10 F in a n ci a l St re ss I n d e x 0.17 0.18 0.19 0.20 0.21
Debt Service Ratio
1999Q1-2014Q4
Total FSI and TPNFS DSR
21 period preceding and during the financial crisis. Firms in Spain and Portugal have high DSRs during the entire sample period. Figure 2 also shows that household DSR shows a more gradual increase up until 2008Q3, after which it falls again. Figure A.5 in appendix G depicts that especially households in Australia, Denmark, The Netherlands, and Norway have high debt payments when compared to their incomes. Remarkably, in Germany both households and non-financial corporations saw a drop in their debt payments as a share of income in the period examined. Households saw their DSR decrease from 9.4% to 6.7% and non-financial corporations from 25% to 20%.
Lastly, figure 3 displays a scatter plot of the sample wide quarterly average of the TPNFS DSR and total FSI. As the red fitted line shows, there clearly is a positive linear relation between TPNFS DSR and total FSI, which was already confirmed by the positive correlation between two variables. The points resemble an exponential relation, however, based on figure A.6 in appendix G displaying a scatter plot containing all observations I discard this possibility in the regressions.
Heteroskedasticity, autocorrelation and multicollinearity
To check whether there is multicollinearity among the independent variables I will use a correlation matrix which can be found in Table A.3 in the appendix. High correlations that could potentially bias the results are highlighted. Moreover, also the Variance Inflation Factor (VIF) is used which can detect the severity of multicollinearity. The VIF score measures to what degree the variance of the estimated specification is increased due to collinearity among independent variables. Table A.4 in the appendix contains the VIF scores, the mean VIF scores are 1.56 and 1.64. A model suffers from severe multicollinearity if the VIF is higher than 10, therefore I conclude that there is no multicollinearity among the independent variables in the baseline model specifications. When including the interaction terms presented in the previous section there obviously appears to be multicollinearity18. The interaction terms are also as individual variables in
the model and will thus correlate with the interaction terms. According to Allison (2012) this can be safely ignored since multicollinearity does not affect the p-values of the interaction terms.
22 To test if the model suffers from autocorrelation a Wooldridge test is performed (Drukker, 2003; Wooldridge, 2010). Appendix F contains the outcomes for both models. For both models the null hypothesis of no first-order autocorrelation is rejected. A likelihood-ratio test is performed to check for the presence of heteroskedasticity. The outcome in appendix F show that the null hypothesis of no heteroskedasticity is rejected. Since the model suffers from autocorrelation and heteroskedasticity robust standard errors need to be applied in the regressions, otherwise the outcomes of hypothesis tests would be misleading.
Next, a Hausman test is conducted to test whether a fixed effects model is actually preferred over a random effects model. Table A.11 and A.12 shows the result of the Hausman test. With a p-value of 0.0000 for both tests I reject the null-hypothesis that there is no systematic difference in the coefficients estimation. This means that the alternative hypothesis is accepted that there is a difference between both methods. As a result, fixed effects estimation methods are used.
23
5. Results
Baseline model specification
Table 2 and 3 present the results of the country-fixed effects estimations to identify if an increase in debt service ratios measured at households and non-financial corporations lead to an increase in financial instability. In both tables the Financial Stress Index is used as a proxy for financial instability19. In table 2 the total private non-financial sector
(TPNFS) DSR is the key independent variable. Table 3 shows the results with household (HH) DSR and non-financial corporation (NFC) DSR as the key independent variables to check whether there is a distinguishable effect noticeable between households and firms. Column (1) of table 2 shows that there is a positive and significant correlation between higher debt payments in the private sector and financial instability. This is in line with Minsky’s Financial Instability Hypothesis that higher debt payments increase financial instability. When debt service ratios increase households and non-financial corporations become more exposed to shocks in asset prices, and more exposed to unexpected changes in income and interest rates. This increases financial instability. Furthermore, the coefficient should be interpreted as follows: since the DSR is measured in percentages a 1%-point increase in the DSR increases the FSI with 0.2569 points. In columns (2) to (7) the inclusion of several control variables changes the significance of the impact of TPNFS DSR on financial instability for some specifications. In column (2) GDP per capita, GDP growth and inflation are added as control variables. GDP per capita has for most specifications a positive and significant impact. This result is consistent with findings of Claessens et al. (2010) who also investigated the link between GDP per capita and the FSI. The next control variable added is GDP growth which has a negative and significant influence on instability in the financial sector. This is logical since GDP growth is received by financial markets as a positive development which causes the FSI to fall. Inflation increases financial instability significantly at the 1% significance level in column (2) and the 10% significance level in column (5). This is in line with Boyd et al. (2001) who argue that inflation negatively influences financial sector performance. Moreover, inflation is unwanted by financial markets since in general it slows down the economy, therefore, leading to financial instability.
24 In column (3) financial openness and trade openness are added which leaves the coefficient of TPNFS DSR insignificant. Trade openness does not enter the regressions results significantly. Financial openness, however, has a positive and significant influence on the FSI which is in line with results of Claessens and van Horen (2012). More outstanding financial assets and liabilities increases the interconnectedness of financial systems. When countries are relatively more active on international financial markets they are also more vulnerable to adverse financial shocks. Greater interconnectedness increased systemic risks and the likelihood of contagion. In column (4) house price appreciation and credit market deregulation are added. Controlling for these two effects shows a positive and significant impact of the DSR on financial instability. The effect of house price appreciation on financial instability is negative and significant. This may be unexpected, however it is fairly easy to explain why it is not. Up until the financial crisis of 2007-2008 the increase in housing prices was received as something positive by financial markets, thus lowering instability in the financial system. The variable credit market deregulation enters the equation in both cases insignificantly. Column (5) denotes a model containing all control variables in which the DSR has a positive and significant coefficient at the 1% level of 41.64, meaning that a 1%-point increase in the DSR increase the FSI with 0.4164 points.
In column (6) time fixed effects are added to control for country-invariant time-specific events. The effect of the DSR on financial fragility is not significant anymore. By adding time-fixed effects a dummy is included in the model for every quarter. Adding many dummy variables can be disadvantageous for the power of statistical tests, since it reduces the signal (smaller changes are compared) but the noise (the standard deviation) remains the same (Lobell, 2012).
25 (2013) who argue that higher debt repayments as a share of income for households make them vulnerable to economic shocks which could cause financial instability. In the run-up to the financial crisis of 2007-2008 households increased their debt levels (mainly mortgages) which were invested in already existing assets which drove up the asset prices. To be able to buy houses in a booming asset market, households needed to take on more debt, leading to an accompanying growth in debt payments. Following Minsky’s Financial Instability Hypothesis this eventually leads to financial instability. With regards to the control variables there are no noteworthy differences between the outcomes of table 2 and table 320.
Regression results with interaction terms
Table 4 shows the results for both models including interaction terms with pre-crisis controlling for the period before the financial crisis of 2007-2008 and financial crisis (based on the systemic banking database of Laeven and Valencia, 2012) controlling for the financial crisis of 2007-200821. For model 1 the influence of total DSR on financial
instability remains positive and significant when controlling for both events. The financial crisis dummy enters the equation positive and significant at the 10%-level. This means that during financial crises the FSI is significantly higher. The pre-crisis dummy, on the other hand, is insignificant as are the interaction terms for model 1, implicating that the effect of the DSR was not significantly different during the specified time-periods. Looking at model 2 we observe that the coefficient of Household DSR also remains positive and significant when including the time-related event dummies and interaction terms. Again the financial crisis dummy has a positive and significant impact on the FSI, this time at the 1%-level. The pre-crisis dummy and the interactions terms remain negative.
Table A.15 in appendix H shows the results when the key independent variables are interacting with other well-known indicators of financial fragility. There are some interesting results. The interaction between total DSR and Household debt, and the interaction between total DSR and total debt are both negative and significant at the 10%-level. Indicating that an increase in household debt or total debt decreases the impact of total DSR on instability in the financial sector. The opposite is expected, high debt levels
20 Regressions were also performed with the Banking sector FSI as dependent variable, however, no noteworthy differences were observed. Results are available upon request.
26 have proved to be indicators of high financial fragility. When debt service ratios increase with already high debt levels it is expected that financial instability increases. For model 2, adding the interaction terms leaves the Household DSR only significant in column (6). Remarkable is that when interacting with total debt the non-financial corporation DSR becomes suddenly positive and significant while the household DSR is insignificant.
Robustness
To test whether the FSI is a good indicator of financial instability, I have replaced the DSRs with credit-to-GDP ratios, the results are shown in table A.13 for model 1 and A.14 for model 2. Table A.13 shows for some specifications a negative and significant relationship between total private credit as share of GDP and the Financial Stress Index. Same goes for table A.14 which shows a negative relationship between Household credit as a share of GDP and the Financial Stress Index. This is surprising if the FSI would be a true indicator of financial instability. Büyükkarabacak and Valev (2010) for example show that an increase in household credit increases banking crisis probability. An explanation could be that actors on financial markets react positively to an increase in debt levels which is reflected in a lower FSI. The suitability for the FSI as a proxy for financial instability should, however, is questionable. These results also explains the negative coefficient for the interaction terms discussed in the previous section.
27 with respect to table 3. Also the coefficient of Household DSR increases remarkably and is significant at the 5%-level for the GMM-estimation.
28
Table 2: Financial Stress Index and Total private non-financial sector Debt Service Ratio (Model 1)
(1) (2) (3) (4) (5) (6)
VARIABLES FSI FSI FSI FSI FSI FSI
Total private non-financial sector DSR 25.69** 6.991 0.437 44.99*** 41.64*** -7.097 (10.39) (10.11) (10.25) (13.17) (12.92) (12.96) GDP per capita 0.000290** 0.000132 0.000729*** 0.000662*** 0.000205 (0.000103) (0.000112) (0.000161) (0.000178) (0.000202) GDP Growth -75.11*** -67.30** -79.64** -78.35** 3.816 (25.35) (25.20) (27.45) (26.84) (14.23) Inflation 38.87*** 25.91* -2.873 (13.07) (12.67) (17.67) Financial openness 1.084*** 1.380*** 0.451 (0.355) (0.345) (0.320) Trade openness 2.616 -0.872 -1.858 (1.671) (2.175) (2.429)
House price appreciation -0.0302*** -0.0338*** -0.00541
(0.00646) (0.00652) (0.00739)
Credit market deregulation -0.0396 0.0919 -0.450
(0.399) (0.358) (0.295) Constant -5.066** -12.91*** -8.604** -31.35*** -29.78*** -1.025 (1.924) (2.929) (3.325) (7.217) (6.693) (8.402) Observations 1,071 1,071 1,071 1,008 1,008 1,008 R-squared 0.032 0.114 0.129 0.185 0.220 0.665 Number of countries 17 17 17 16 16 16
Country FE Yes Yes Yes Yes Yes Yes
Year FE No No No No No Yes
29 Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Table 3: Financial Stress Index and Household DSR and Non-financial corporation DSR (Model 2)
(1) (2) (3) (4) (5) (6)
VARIABLES FSI FSI FSI FSI FSI FSI
Household DSR 61.69*** 32.35* 26.80 61.76** 53.18** 6.241
(11.33) (17.93) (21.50) (22.34) (24.35) (11.19)
Non-financial corporation DSR -5.421 -4.149 -4.687 3.450 3.296 -2.812
(3.106) (2.569) (3.239) (3.671) (4.315) (3.340)
GDP per capita 0.000213 9.00e-05 0.000738*** 0.000644** 0.000278
(0.000144) (0.000155) (0.000204) (0.000239) (0.000234) GDP Growth -73.66** -66.02** -84.02** -82.24** 7.188 (26.59) (26.04) (28.80) (27.78) (13.75) Inflation 37.35** 24.73 2.081 (13.77) (14.15) (16.08) Financial openness 0.680 1.182** 0.304 (0.465) (0.528) (0.368) Trade openness 3.546 0.447 -0.970 (2.191) (2.678) (2.150)
House price appreciation -0.0306*** -0.0324*** -0.00804
(0.00745) (0.00783) (0.00776)
Credit market deregulation 0.00782 0.137 -0.348
(0.464) (0.416) (0.276) Constant -4.245** -9.963** -7.489 -30.84*** -28.82*** -5.214 (1.743) (4.299) (4.579) (8.950) (8.989) (9.348) Observations 1,008 1,008 1,008 945 945 945 R-squared 0.080 0.138 0.145 0.213 0.232 0.706 Number of countries 16 16 16 15 15 15
Country FE Yes Yes Yes Yes Yes Yes
30
Table 4: Interactions with time-related events (Model 1 and 2)
Model 1 Model 2
(1) (2) (3) (4)
VARIABLES FSI FSI FSI FSI
Total private non-financial sector DSR 36.86** 39.21*** (14.08) (11.36)
Financial crisis 3.526*
(1.772) Financial Crisis*Total private non-financial sector DSR -3.294
(8.819)
Pre-crisis -0.140
(1.396) Pre-crisis*Total private non-financial sector DSR -8.152
(7.615) Household DSR 53.24** 52.04* (24.01) (27.56) Non-financial corporation DSR 2.487 1.066 (4.047) (5.128) Financial crisis 4.912*** (1.597) Financial Crisis*Household DSR -7.670 (15.51)
Financial Crisis*Non-financial corporation DSR -2.516
(3.990) Pre-crisis -1.995 (1.956) Pre-crisis*Household DSR -8.948 (11.34) Pre-crisis*Non-financial corporation DSR 2.766 (3.487) Constant -24.50*** -25.37*** -24.00** -23.35** (6.874) (6.515) (8.416) (9.058) Observations 1,008 1,008 945 945 R-squared 0.253 0.251 0.270 0.271 Number of countries 16 16 15 15
Country FE Yes Yes Yes Yes
Year FE No No No No
31
Table 5: Two-stage least squares and Generalized Method of Moments (Model 1 and 2)
TSLS Difference-GMM
(1) (2) (3) (4)
VARIABLES FSI FSI FSI FSI
Total private non-financial sector DSR 38.91*** 207.5***
(6.984) (49.40) Household DSR 51.39*** 234.7** (9.956) (109.6) Non-financial corporation DSR 2.834 20.29 (1.754) (24.45) GDP per capita 0.000673*** 0.000657*** 0.000782* 0.000780 (8.68e-05) (9.69e-05) (0.000407) (0.000487) GDP Growth -81.32*** -85.38*** -48.56** -71.50** (16.15) (16.47) (19.11) (28.15) Inflation 25.36 25.32 1.873 6.117 (21.32) (22.42) (10.29) (12.11) Financial openness 1.418*** 1.171*** 0.273 -0.980 (0.247) (0.322) (1.188) (1.685) Trade openness -0.511 0.716 14.15** 11.56* (1.233) (1.282) (5.457) (5.795) House price appreciation -0.0328*** -0.0314*** -0.0770*** -0.0691***
(0.00371) (0.00383) (0.0181) (0.0181)
Credit market deregulation 0.130 0.162 2.054*** 1.142
(0.187) (0.193) (0.560) (1.105)
Observations 992 930 992 930
R-squared 0.229 0.239
Number of countries 16 15 16 15
Country FE Yes Yes Yes Yes
Year FE No No No No
Number of Instruments 69 131
AR2 Test (p-value) 0.00325 0.00426
Hansen Test (p-value) 1 1
32
6. Concluding remarks
Based on Minsky’s financial instability hypothesis theory this paper empirically investigates the relation between debt service ratios, as an indicator of financial fragility at households and non-financial corporations, and financial instability. This paper contributes to the existing literature about financial instability in several ways. A fruitful discussion is provided covering the definitions and measurements of the concepts of “financial fragility” and “financial instability” which are often used interchangeably in the existing literature. As I showed, there is a distinction between the two terms: a situation of financial instability emerges from a situation in which financial fragility was high. Moreover, a situation of high financial fragility does not imply that the economy is already in a crisis. It only implies that the system is vulnerable to shocks which can cause a financial crisis. To carry out the research an extending database has been constructed covering 17 advanced countries for the period 1999Q1-2014Q4. The database contains the Financial Stress Index (FSI) as a proxy for financial instability and the Debt Service Ratio (DSR) as a proxy for financial fragility at households and non-financial corporations. The empirical results show that for most specifications of model 1 there is a positive and significant influence of total debt service ratios on the Financial Stress Index. This confirms Minsky’s financial instability hypothesis that an increase in debt payments increases vulnerabilities which eventually lead to financial instability. In the second model the composites of total DSR are considered as key independent variables. A significant and positive influence of household DSR on the Financial Stress Index is found, while the non-financial corporation DSR enters the equation insignificant. The results appear to be robust when controlling for endogeneity problems by using both TSLS and GMM estimation methods.
Policy implications
Generally speaking, this paper proves that an increase in debt service ratios increases financial instability. Since financial instability is an undesired situation some policy implications are suggested. Already in 1986 Minsky wrote a book called stabilizing an
unstable economy which is still highly relevant today. He suggests that modern economic
33 reform agenda is needed which aims for an economy with low-investment, high-consumption and full-employment governed by small organizations to minimize bureaucracy. However, after the reforms are done Minsky argued that financial stability will only be temporary as capitalist finance will find new ways to create financial fragility. New reforms should follow because there is no possibility to do change things once and for all, financial instability will always return.
More practically, relating Minsky’s reform agenda to the findings of this paper, policy makers should lower the incentives to invest in already existing assets. High debt service ratios are a result of high credit levels which fuel the asset price bubble. Governments should therefore aim at reducing the incentives for households and non-financial corporations to borrow unsustainable amounts of credit. The results showed that Household DSR increases financial instability. A suggestion would, therefore, be to abolish mortgage interest reduction schemes. Not only the incentives at the borrower side should be reduced, also lenders (banks) their incentives to lend money should be reduced. However, currently the opposite is observed which is worrisome. The main central banks in the world, the ECB and the FED, lowered their policy rates to almost 0% or in case of the ECB even to a negative rate. On top of that both institutions engaged in quantitative easing (QE). Both measures are flooding capital markets with cheap credit again fuelling asset markets worldwide (i.e. the Dutch stock index the AEX reached in 2015 pre-crisis levels). This will very likely increase debt service ratios and, therefore, increase financial instability as my results showed. Therefore, central banks are advised to increase their policy rates and stop their quantitative easing programs.
Limitations
34 however most European countries scored very low on the FSI. In the FSI the bond market is captured by the difference between corporate bonds and government bonds, since government bonds were high in this period the FSI produced a low score. Furthermore, the robustness tests show that the Financial Stress Index used as a proxy of financial instability is questionable.
Second, in computing the DSR Drehmann et al. (2015) make some assumptions since micro-level data on debt payments is not available for a long range of countries and time periods. Data for total debt, income and interest payments are fairly easily accessible, for calculating average maturity they make assumptions. They compose an average maturity which is different for households and non-financial corporations, however, which is the same for all the countries. Cross-sectional differences in the DSR between countries are, therefore, lower than when not using an average maturity. This can lead to under- or overestimation of the DSR.
Third, due to limited data availability only a short period of time (1999Q1-2014Q4) and a small number of countries (17 countries) are covered. This was mainly due to restrictions in the availability of the DSR. However, a broader and more historical database by Drehmann et al. (2015) is proposed. This is important since it is difficult to present the results found in this paper as bold claims. The results could be driven by events peculiar for this period (i.e. the financial crisis of 2007-2008).
Fourth, the regression results are subjected to change when adding several control variables in some of the specifications. Moreover, when adding time-fixed effects all the variables are left insignificant. As explained before adding many dummy variables can be disadvantageous for the power of statistical tests. It reduces the signal (smaller changes between variables are compared), however the noise (the standard deviation) remains the same.
35
Future research
Due to time limitations several topics have not been covered in this study, however, which would be very interesting for future research. Following the work of Zhang and Bezemer (2015) it would be interesting to investigate the influence of debt service ratios on crisis duration and severity. They focus on especially on household mortgage credit, due to composition of the DSR this study can also be carried out using the DSR. Next, it would be interesting to test a threshold model to see if the DSR only influences the FSI after a certain threshold. Hansen (2000) could be used as a starting point for this research. A last suggestion would be to extent the database of this study with emerging countries. It would be interesting to observe whether the mechanisms differ between advanced and emerging countries. Both the FSI and the DSR are available for 15 extra developing countries above the 17 advanced countries in this study, however, quarterly data availability for control variables is limited.
Acknowledgment: I would like to thank Dr. D.J. Bezemer for supervising my thesis and
providing me with helpful comments and suggestions.
References
Allen, F., and D Gale. “Financial Fragility, Liquidity, and asset prices.” Journal of the
European Economic Association, December 2004, v. 2, iss 6,, 2004: 1015-1048.
Allen, F., and D. Gale. “Bubbles and Crises.” American Economic Review, 2000: 236-255. Allen, W.A., and G. Wood. “Defining and achieving financial stability.” Journal of Financial
Stability 2 (2006), 2005: 152-172.
Allison, P. When can you safely ignore multicollinearity? 10 September 2012. http://statisticalhorizons.com/multicollinearity.
Andrianova, S., et al. A new international database on financial fragility. Working Paper 15/18, University of Leicester, 2015.
Arellano, M., and O. Bover. “Another look at the instrumental variable estimation of error-components models.” Journal of econometrics, vol. 68, issue 1, 1995: 29-51.
