Research Master Thesis
(EBM897A30)
Academic year 2015-2016
Does Credit Growth Increase Financial Fragility?
Peng Yiqing ID-S2443600
August.15, 2016
Faculty of Economics and Business University of Groningen, the Netherlands
CONTENT
ABSTRACT 1
1. Introduction 2
2. Credit growth and financial fragility 4
2.1 Credit and economic growth: theory and evidence 4
2.2 Credit, debt and financial fragility 5
3. Methodology 7
3.1 Baseline model 7
3.2 Sensitivity analysis 9
4. Data 10
4.1 Definitions of key variables 10
4.2 Description of the dataset 12
4.3 Data trends 13
5. Estimation results 17
5.1 Baseline regression 17
5.2 Robustness analysis 18
5.2.1 Arellano-Bond dynamic estimation 18
5.2.2 Credit growth and financial fragility: subsample analysis 19 5.2.3 Credit growth and financial fragility: does the type of credit matter? 20
6. Conclusion 22
APPENDIX 38
TABLES & FIGURES
FIGURES
Figure 1 Mismatch between debt and income in an economy 6
Figure 2Credit growth and financial fragility: a theoretical model 7
Figure 3 Credit stock in selected developing and developed countries 14
Figure 4 Credit growth and financial fragility: scatter plots 16
Figure 5 Balance sheet of the household sector and institutional investors 21
TABLES
Table 1 Countries included in the sample 12
Table 2 Descriptive statistics 13
Table 3 Correlations between Z-score and other independent variables 15 Table 4 Credit growth and financial fragility: fixed-effect estimation (1-period lagged) 24 Table 5 Credit growth and financial fragility: fixed-effect estimation (2-period lagged) 25 Table 6Credit growth and financial fragility: Arellano-Bond dynamic panel estimation 26 Table 7 Credit growth and financial fragility: evidence from developing countries 27 Table 8 Credit growth and financial fragility: evidence from developed countries 28 Table 9 Mortgage growth and financial fragility: fixed effect estimation 29 Table 10 Financial business credit growth and financial fragility: fixed effect estimation 30 Table 11 Nonfinancial business credit growth and financial fragility: fixed effect estimation 31
Table A.1 List of developing and developed countries 32
Table A.2 Variables and sources 33
1
Does Credit Growth Increase Financial Fragility?
Peng Yiqing*University of Groningen, the Netherlands August.15, 2016
Abstract
This paper aims to research the consequence of a weak growth-enhancing effect of credit as suggested by recent literature on finance and growth. I hypothesize that a higher growth of credit relative to GDP tends to increase financial fragility. The main argument is that credit is also debt which needs to be repaid by income. If credit grows without proportionately leading to income growth, debt repayment capabilities will be undermined, raising the risk of default and financial fragility. Using a database of 58 countries over 1998-2012, I find that financial fragility indicated by Z-score is highly correlated with credit growth in both fixed-effect panel regressions and Arellano-Bond dynamic panel regressions. This relationship remains strong in developing countries, but not for developed countries. In addition, financial fragility is found strongly associated with the growth of nonfinancial business credit, but not for mortgages and financial business credit. Empirical results highlight the potential risk of an unbalanced growth of credit and income.
JEL classification: E44, G21, O16.
Key words: credit growth, economic growth, credit stock, debt stock, financial fragility, bank fragility.
* I am very grateful to my supervisors, Prof. dr. Niels Hermes and Prof. dr. Robert Lensink, for providing useful
2 1. Introduction
Credit has long been regarded as an important determinant of economic growth by supporting investments, facilitating innovation, and improving income (Schumpeter, 1934). Since 1990s many countries have taken credit deregulation as a way of liberalizing financial markets and promoting economic growth (Bumann, Hermes, and Lensink, 2013), and strong credit growth has been observed in many economies across the world (Jordà, Schularick, and Taylor, 2016). However, recent empirical work, notably byRousseau and Wachtel (2011), suggests that the positive effect of credit on economic growth is weak and insignificant. A popular explanation is that some credit, such as mortgage or household credit, is not used to support production of goods and services in the real economy, but transactions in property and asset markets,1 which drags the overall growth-enhancing effect of credit (Beck, Büyükkarabacak, Rioja and Valev, 2012; Bezemer, 2014). This raises my concern about the consequence of unbalanced growth of credit and income, which motivates this paper.
My main argument is that credit and debt are two sides of the same coin from an accounting perspective. Credit lies on the asset side of the balance sheet of the financial sector, which constitutes a part of the liabilities of the private sector. Thus, credit stocks are also debt stocks, which need to be repaid by income. If credit fails to grow in proportion with income, it will enlarge the mismatch between debt and income in the economy so that debt servicing will be increasingly difficult to be realized. In this case, the risk of default will be high in the private sector, which will translate into a high default risk in the banking sector, leading to a situation described as financial fragility2. The purpose of this paper is to empirically test the hypothesis that a higher growth of credit relative to GDP tends to increase financial fragility.
To this end, I collected data on total credit to private sector for 58 countries over 1998-2012 from the consolidated balance sheet of monetary financial institutions from the central bank. To further test the effect of different type of credit on financial fragility, I also collected data on mortgages, credit to financial business, and credit to nonfinancial business respectively as defined by Bezemer (2014). Financial fragility is indicated by Z-score from Andrianova et al., (2015), which measures the distance to default in the banking system. A low Z-score implies high financial fragility. Country coverage and time period in the dataset are dictated by data availability.
In line with the literature on financial fragility (Demirgüç-Kunt and Detragiache, 1998; 2005; Schularick and Taylor, 2012), I find that growth of credit-to-GDP ratio is strongly related to financial fragility in both fixed-effect panel regressions and Arellano-Bond dynamic panel
1 To illustrate, Bezemer and Grydaki (2014) look at the United States and find that the share of nonfinancial
business credit in GDP has gone through a moderate increase from 100% to 110% over 1990-2006, while the share of credit to FIRE sector, i.e. finance, insurance, and real-estate sectors, has risen dramatically from 100% to 220% in the same period. Jordà, Schularick and Taylor (2015) show that the share of mortgage credit in GDP has on average increased from 30% to 70% from 1990 to 2010 in 17 developed countries. Bezemer and Muysken (2015)
look at the Dutch economy and find that mortgage-to-GDP ratio has surged from 25% in 1990 to 100% in 2012 and the share of mortgages in total credit to private sector has risen from 30% to 50%.
2 It is helpful to distinguish financial fragility from bank fragility , which are the two terms I will use frequently
3 regressions. The further subsample analysis shows that this relationship remains significant in developing countries, but not for developed countries where financial fragility is found highly correlated with the level of credit stock rather than the growth. It suggests that credit growth may increase financial fragility when credit stock is not too large, as is the case in developing countries. In addition, I do not find evidence that the growth of unproductive credit, i.e. credit that is not directly used in production, such as mortgage credit and financial business credit, makes the financial system increasingly fragile. Instead, there is a strong correlation between financial fragility and growth of nonfinancial business credit, which is expected to promote income growth and to be unrelated to financial fragility. In fact, this result is the reverse to my expectation, and I will provide several possible explanations on this finding in Section 5.2.3.
This paper is related to Rousseau and Wachtel (2011), Beck et al., (2012), Bezemer (2014), and Bezemer, Grydaki and Zhang (2016), which discuss the weak relationship between credit and economic growth and how changes in the composition of credit matters. However, I take one step further and research the consequence if credit grows faster than income improvement. This paper is closely related to Demirgüç-Kunt and Detragiache (1998; 2005, henceforth DD), where credit growth is found to have direct relationship with financial fragility. However, this paper is different in three aspects. First, DD describe financial fragility as the probability of systemic financial crisis, so they use a multivariate logit regression model to explain financial fragility in their empirical analysis. In the present paper, financial fragility is defined as the likelihood of default in the banking system. This definition allows me to use Z-score as an alternative indicator of financial fragility and study its relationship with credit growth under the panel setting. Second, DD calculate credit growth by the rate of growth of real domestic credit to the private sector (DD, 2005, p.83). In this paper, I measure credit growth by the growth rate of credit-to-GDP ratio, which captures how fast credit grows relative to GDP. It allows me to analyze whether the mismatch between debt and income can help to explain financial fragility. Third, DD look at how growth of total credit is associated with financial fragility only. However, I further explore how financial fragility is connected with the growth of different type of credit, which has not been extensively researched in existing literature yet.
4 The remainder of the paper proceeds as follows. In Section 2, I define what financial fragility means in this paper and explain why a weak relationship between credit and economic growth tends to lead to financial fragility. In Section 3, I introduce the methodology that will be used for empirical analysis. In Section 4, I provide an overview of the dataset. Section 5 discusses the estimation results. This paper concludes in Section 6.
2. Credit growth and financial fragility
2.1 Credit and economic growth: theory and evidence
The theory with respect to the relationship between credit and economic development can be traced back to the seminal publication The Theory of Economic Development by Schumpeter (1934). He believes that innovation is the driving force of economic development by creating advanced technologies, improving the process of production, and providing more products for an economy. Hence, entrepreneurs, as the initiators and leaders of innovation, play a central role in the course of economic development. For entrepreneurs to innovate, they should have new knowledge in the first place. More importantly, they should also have sufficient capital resources to purchase the inputs they need for conducting experiments or making trials before an innovation can be put into practical use. Since credit provides liquidity for entrepreneurs to finance their investment projects and bring innovation into full play, it has long been deemed as an essential for economic growth3.
Pioneered by the publication of Financial Structure and Development by Goldsmith (1969), empirical studies on finance and economic growth started to burgeon since 1990s. Consistent with Schumpeter s view, early evidence has established a positive growth-enhancing effect of credit4. For instance, King and Levine (1993) find that financial development, indicated by total credit stocks to the private sector scaled by GDP, has a strong positive effect on GDP growth based on data of 80 countries over 1960-1980. However, this study has two main limitations. First, potential simultaneity bias due to reverse causality from economic growth to financial development is not well addressed. Second, the cross-country setting implies that there might be omitted variable bias due to unobservable country-specific effects not being controlled for. In a follow-up study, Levine, Loayza, and Beck (2000)reassess the impact of finance on GDP growth under a cross-country setting, using the instrument variable (IV) approach to deal with the issue of simultaneity. In addition, they evaluate the finance-growth relationship under the panel setting, where the persistence (lagged effect) of GDP growth and country fixed effects are taken into account. Based on a dataset of 74 countries over 1960-1995, they again find that the level of financial development is positively associated with GDP growth.
3 In empirical research (also the present paper), credit refers to domestic credit issued to the private sector, as it
is believed to have most direct relationship with economic growth compared with credit to the government sector.
5 However, recent empirical evidence shows that the positive link between credit and economic growth becomes weaker over time and too much credit tends to even hinder economic growth (for example, Deidda and Fattouh, 2002; Rioja and Valev, 2004a; Rousseau and Wachtel, 2011; Manganelli and Popov, 2013; Law and Singh, 2014; Arcand, Berkes and Panizza, 2015). A popular explanation is that credit in the early literature is implicitly assumed to be productive credit, which helps to improve output and income growth. However, in reality there are different types of credit, each of which exerts different impact on economic growth. Hence, the credit-growth relationship may be influenced by the changes in the composition of credit, i.e. the share of different types of credit in total credit. Specifically, the literature suggests two perspectives on how credit can be distinguished. On the one hand, credit can be distinguished according to which sector finally uses credit, i.e. households or firms. Empirical evidence suggests that enterprise credit helps to promote economic growth, while the effect of household credit is insignificant or even negative5(Büyükkarabacak and Valev, 2010; Beck et al., 2012; Sassi and Gasmi, 2014).
On the other hand, credit can be distinguished according to how it is used from a functional perspective, i.e. to facilitate the production of goods and services in the real economy or to support transactions in the property and asset markets (Bezemer, 2014). Credit in the former case is the typical Schumpeter-type credit, which is expected to boost technological progress and create income (Goldsmith, 1969; McKinnon, 1973; Shaw 1973). On the contrary, credit in the latter case is expected to have little growth-effect, because its main job is to facilitate asset-market activities, rather than enhance GDP creation. Based on a dataset of 46 countries over 1990-2011, Bezemer et al., (2016) find that total credit stock has no significant effect on GDP growth. However, by further examining the effects of different types of credit, they find that credit flows to nonfinancial business has a positive impact on GDP growth, but not for mortgage credit and credit flows to financial business6.
2.2 Credit, debt and financial fragility
The consequence of a weak growth-enhancing effect of credit is that rising credit stock makes the financial system increasingly vulnerable to default. The reason is that credit and debt are actually two sides of the same coin. A (Bezemer, 2014, p.940) and all
(Bezemer et al., 2016, p.654), since increased liquidity created by the banking system, by definition, implies that aggregate liabilities of firms and households in the economy have also increased. Hence, credit will become a problem for the banking sector, when debt repayment becomes a problem for the private sector. This may occur when income
5 Following Büyükkarabacak and Valev (2010, p.5) and Beck et al., (2012, p.1249), household credit is defined as
6 Bezemer et al., (2016, p.655) distinguish four types of credit mortgages to households, household consumption
6 fails to grow as much as credit does so that the gap between debt and income in the economy gets larger over time. Since debt will ultimately be repaid from income, rising debt-to-income ratio implies that debt repayment is increasingly difficult to be realized and the likelihood of default will be higher (Sutherland and Hoeller, 2012). This also implies that consequently the banking system will have to face a higher risk of default. Figure 1 graphically illustrates the above argument.
FIGURE 17
Mismatch between debt and income in an economy
Suppose that from t=t0 credit starts to grow at a rate faster than GDP in an economy, which is depicted by the steeper slope of the credit (debt) curve. It can be expected that as credit stock continues to increase, the gap between the two curves, which measures the mismatch between debt and income, will get larger over time. It will become even worse if the economy is later hit by a negative income shock that leads to a drop in GDP (for example at t1). In both cases, debt servicing out of income will become increasingly difficult. A higher chance of default by the private sector implies that risk of default also increases in the banking system, making the financial system increasingly .
Financial fragility is a term often used in the literature. However, due to the fact that financial fragility by nature is not explicitly observable, this concept can be understood in many ways without a universally-agreed definition (Bezemer, 2016). Lagunoff and Schreft (2011) suggest two perspectives from which financial fragility can be described. First, financial fragility can be regarded as an ex-post concept and described by how likely a fragile financial system will lead to undesirable economic outcomes. For example, Demirgüç-Kunt and Detragiache (1998, 2000, and 2005)refer financial fragility to the likelihood of systemic banking crisis8. Loayza
7 For simplicity, I assume that debt stock is always larger than income in an economy. In this case, the gap between
debt and income will be positive and become large over time. It is also possible that the initial debt stock is smaller and it exceeds income at a certain point of time later. In this case, the debt-income gap will also increase, but from negative to positive.
8 the authors refer to the probability of system-wide crises in the banking sector, rather than
localized crises. Demirgüç-Kunt and Detragiache (1998, p.91) consider an episode systemic crisis, if it satisfies at least one of the following four conditions: i) the ratio of nonperforming assets to total assets in the banking system exceeded 10 percent; ii) the cost of rescue operation was at least 2 percent of GDP; iii) Banking sector problem resulted in a large scale of nationalization of banks; iv) extensive bank run took place or emergency measures, such as deposit freezes, prolonged bank holidays, or generalized deposit guarantees being enacted by the government.
Mismatch
Time
Debt (credit)
7 and Rancière (2006) define financial fragility as the frequency of systemic banking crises. Meanwhile, financial fragility can also be regarded as an ex-ante concept and described as to what extent a financial system can absorb economic shocks and prevent undesirable economic outcomes from being realized. Jappelli, Pagano and Di Maggio (2013)look at households and describe financial fragility as the sensitivity of household defaults to macroeconomic shocks. Tymoigne (2014) employs the definition by Minsky and Kaufman (2008, p.233), which states that financial fragility lies in the fact that the effect of an unexpected negative shock tends to be amplified throughout the financial system. In the present paper, I adopt the first approach, but define financial fragility as the likelihood of default in the banking system, rather than the likelihood of crisis, so as to capture the most direct consequence of debt-income imbalance. By this definition, a financial system becomes fragile when it is facing a high risk of default.
To sum up, debt is the other side of credit. When the banking system creates credit for the private sector, it also creates debt which needs to be repaid by income at certain point of time. Therefore, credit growth will not necessarily become a problem, if it can contribute to income improvement that allows for debt obligation. However, if income fails to grow hand in hand with credit, as is suggested by recent finance-growth literature, credit growth tends to enlarge the mismatch between debt and income and raise the default risk in the private sector, which will finally translates into the fragility in the banking system. This is the key argument of this paper (Figure 2), which will be empirically tested in Section 5. Before that, I first explain the methodology and survey the dataset in Section 3 and Section 4.
FIGURE 2
Credit growth and financial fragility: a theoretical model
3. Methodology
The purpose of the empirical analysis is to test the relationship between credit growth and financial fragility, taking into account the effects of other covariates. The analysis is based on an unbalanced panel dataset which covers 58 countries over 1998-2012.
3.1 Baseline model
First, I start with a fixed-effect panel regression as the baseline specification: Credit growth with weak
income-enhancing effect
Increased mismatch between debt and income
8 The dependent variable is financial fragility, which is measured by Z-score collected from Andrianova et al., (2015). As will be discussed in Section 4.1, Z-score measures the distance to default in the banking system, which is negatively correlated with financial fragility. A low Z-score indicates a small distance to default and thereby a high risk of default in the banking system. The key explanatory variable is Credit-GDP Growth ( ), which is calculated by the growth rate of credit-to-GDP ratio in order to capture how fast credit grows relative to income. It enters the regression with a lag of periods ( ) to make sure that causality runs from credit growth to financial fragility only 9. The coefficient of interest is expected to have a negative sign. A high growth of credit relative to GDP implies that debt is growing faster than income in the economy, which increases the risk of default in the banking system and therefore financial fragility.
is a vector of control variables. Following the literature on financial fragility (for example, Demirgüç-Kunt and Detragiache, 1998), the first set of control variables includes a number of macroeconomic indicators. To start with, since financial fragility rises when debt repayment becomes difficult, I control for GDP growth and real GDP per capita. High income growth and a high level of economic development are expected to reduce the risk of default on debt and to be negatively correlated with financial fragility. Following Komp and De Haan (2015), I also control for inflation and current account balance to capture monetary stability and net financial flows. Demirgüç-Kunt and Detragiache (2005) argue that high inflation raises the probability of banking crisis. Persistent current account deficit, which implies a high reliance on borrowing from foreign countries, makes the financial system vulnerable to sudden capital outflows (Obstfeld, 2012; Frankel and Saravelos, 2012).
The second set of covariates includes financial market indicators. I control for credit-to-GDP ratio, which measures the level of financial development. Under a developed financial system, it is more likely for banks to take excessive risk by overextending credit to the private sector, sowing the seeds of financial fragility (Brunnermeier, 2009). Demirgüç-Kunt and Detragiache (2011, p.73)also argue, Countries where the banking sector has a larger exposure to private sector borrowers are more vulnerable, perhaps as a result of mismanaged liberalization . Hence, it can be expected that a high credit-to-GDP ratio is correlated with higher financial fragility, i.e. lower Z-score. In addition, following Beck, Demirgüç-Kunt, and Levine (2006), I control for short-term real interest rate. Since interest is a component of debt obligation, a high interest rate increases the debt burden, the likelihood of default, and financial fragility (Bezemer, 2016).
The last set of covariates is concerned with bank characteristics. Andrianova et al., (2015) provides four indicators on i) equity that reflects bank capitalization; ii) return on assets that reflects bank profitability; iii) cost that reflects bank managerial efficiency; and iv) impaired loans that reflects the quality of bank assets. Given that equity and return on assets have been used to calculate Z-score, which enters equation (1) as the dependent variable, I do not control
9 As sensitivity check, I also estimate the baseline model using three-period lagged credit growth (s=3) defined in
9 for these indicators,but for banking efficiency and the quality of banking assets as suggested by Demirgüç-Kunt et al., (2008). Definitions are reported in Table A.2. A high cost-to-income ratio implies a low efficiency in the banking system which is expected to increase financial fragility. A larger share of impaired loans indicates poorer asset quality in the banking system which is expected to increase financial fragility as well.
Finally, is unobserved time-invariant country-specific fixed effects; is time fixed-effect; and is white-noise error term with zero mean. Correlations between main variables in the baseline specification are reported in Table 3.
3.2 Sensitivity analysis
In order to test the sensitivity of the relationship between credit growth and financial fragility obtained from baseline model, I conduct additional robustness checks. First, followingKlomp and De Haan (2015), I take the persistence of Z-score into account, where Z-score is believed to be correlated with its past value, which should enter the baseline equation as an additional independent variable. Then the new level equation becomes:
It is obvious from equation (2) that lagged dependent variable is correlated with the country-specific fixed effects by construction. While taking the first difference removes , lagged dependent variable in equation (3) is still correlated with the error term. Standard fixed-effect estimators will then become inconsistent due to endogeneity problem.
To address the endogeneity problem, I employ the approach developed by Arellano and Bond (1991), where the endogenous variable in the difference equation is instrumented by lagged dependent variables from the level equation, on the condition that there is no autocorrelation in the residuals:
and explanatory variables are weakly exogenous, i.e. they are only correlated with past errors:
This suggests that the following moment conditions should be satisfied:
10 To test model specification, I perform Sargan test to check if over-identifying restrictions are valid. I also conduct Arellano-Bond AR(1) and AR(2) test to check if there is first-order and second-order autocorrelation in the residuals from the difference equation (Blundell and Bond, 1998). It is expected that there exists first-order autocorrelation by construction, but not at the second order, if the assumption of no autocorrelation in the level equation is warranted.
Second, since institutional quality and the level of financial development may be different across countries (Rioja and Valev, 2004b; Rousseau and Wachtel, 2011), I make a sub-sample analysis where I split the baseline sample into developing and developed countries and repeat the baseline estimation. The purpose is to examine if the effect of credit growth on financial fragility varies across country groups. Finally, I extend the baseline model by studying how financial fragility is linked to different type of credit. Following Bezemer (2014), I distinguish three types of credit: i) mortgages; ii) financial business credit; and iii) nonfinancial business credit. I repeat the baseline estimation, but use the growth of each type of credit stock (scaled by GDP) as key explanatory variable. Definitions are introduced in equation (14)-(16). Since mortgage credit and financial business credit are mainly used by households and institutional investors, such as insurance companies or pension funds, to invest in houses, bonds, or shares in asset markets, rather than support production in the real economy, their growth is expected to enlarge the mismatch between debt and income and leads to financial fragility. In contrast, credit to nonfinancial business is in principle used for innovative and entrepreneurial activities, which are supposed to promote income growth. Hence, growth of nonfinancial business is expected to be unrelated to financial fragility.
4. Data
4.1 Definitions of key variables
The dependent variable is financial fragility, which is defined as the likelihood of default in the banking system. Following the literature on bank fragility (Demirgüç-Kunt et al., 2008; Demirgüç-Kunt and Detragiache, 2011; Klomp and De Haan, 2015), I measure system-wide financial fragility by the country-level Z-scores ( - ), i.e. the Z-score of country i in period t. According to Andrianova et al., (2015), this indicator is constructed by taking the weighted average of the bank-level Z-scores, i.e. the Z-score of each bank b in country i in period t ( - ), with the weight ( ) being the relative size of the bank in terms of bank asset ( )10. Specifically,
10 Statistic rationale of bank-level Z-score and how it is aggregated into country-level is explained in more details
11
-
In equation (12), is the equity-to-asset ratio that reflects bank capital; is the average return-on-asset of the bank that reflects bank profitability; and is the standard deviation of return-on-asset that reflects the volatility of returns. Bank-level Z-score reflects the number of standard deviations that a particular bank s return-on-asset has to fall below its mean before bank capital is completely depleted. Therefore, a higher bank Z-score implies a larger number of standard deviations that return-on-asset can drop below its mean before the bank defaults. In other words, bank Z-score is inversely correlated with bank fragility. Similarly, a country-level Z-score measures the distance to default in the banking system of a particular country, which is inversely correlated with system-wide financial fragility.
The key explanatory variable is Credit-GDP Growth ( ), which is defined as the growth rate of (private) credit-to-GDP ratio. Since mathematically the growth rate of a ratio can be approximated to the difference between the growth rate of numerator and the growth rate of denominator, measures how fast credit stock grows relative to GDP, which captures the relationship between debt and income. A high growth rate of credit-GDP suggests that debt-income mismatch is increasing.
In a similar fashion, I define Mortgage Growth ( ), FB Growth ( ), and NFB
12 4.2 Description of the dataset
The dataset covers 58 countries over the period 1998-2012. Countries included in the sample are reported in Table 1. Following Bezemer (2014) and Bezemer et al., (2016), I collect data on total credit to private sector, mortgage credit, credit to financial business and credit to nonfinancial business from the consolidated balance sheet of monetary financial institutions from the central bank. Data on financial fragility is collected from the New International
Database on Financial Fragility by Andrianova et al., (2015). One feature of this new dataset
is that it covers information of investment banks and mortgage banks, rather than commercial banks only. In addition, they construct country-level financial fragility indicators using bank-level data from Bankscope database (See Appendix for more details). This allows me to study system-wide financial fragility at the country-level. To my knowledge, this new dataset has not been widely used in existing empirical research, the most recent one being Fielding and Rewilak (2015).
TABLE 1
Countries included in the sample
Albania Egypt Japan Philippines
Australia Spain Kenya Poland
Austria Estonia Kyrgyz Republic Portugal
Argentina Finland Korea Romania
Belgium France Kazakhstan Russia
Bulgaria United Kingdom Latvia Singapore
Belarus Georgia Lithuania Sri Lanka
Botswana Greece Morocco Sweden
Brazil Hong Kong Mexico Thailand
Canada Hungary Malaysia Turkey
Switzerland Indonesia Norway Ukraine
Chile India Netherlands Uruguay
Czech Republic Ireland New Zealand USA
Denmark Israel Pakistan
Germany Italy Peru
13 TABLE 2 Descriptive statistics - -- - -- - -- -4.3 Data trends
Figure 3 displays the development of credit stock in 8 selected developing and 20 developed countries over 1998-2012. The first observation is that total credit stock in general is smaller in developing countries. It is about half as much as that in developed countries, which may be explained by the differences in the level of financial development. Nevertheless, both country groups have experienced a substantial growth in credit stock during this period. Specifically, credit stock has risen from about 40% to 50% in developing countries, and from 90% to 120% in developed countries.
When it comes to the composition of credit, it can be observed from the graph that financial business credit constitutes the smallest part of total credit in both groups of country, while the dominant position is taken by credit to nonfinancial business. This trend becomes particularly prominent in developing countries where nonfinancial business credit accounts for 60-70% of total credit, while this figure is only around 45% in developed countries. The most significant change is witnessed in mortgage credit, whose share has risen from 3% to 10% in developing countries. In developed country, mortgage share even exceeded that of nonfinancial business credit, reaching 50% in 2011.
14 concentrated between -0.15 and 0.15. Univariate analysis shows a large coefficient of -24.548 between Z-score and lagged credit growth. In short, the above scatter plots show that overall there is a negative correlation between credit growth and financial fragility. This relationship may differ across developed and developing countries.
FIGURE 3
(a) Credit stock in selected developing economies
0 10 20 30 40 50 1998 2000 2002 2004 2006 2008 2010 2012
Total private credit Financial business Mortgages Nonfinancila business
Note: Panel (a) presents the unweighted average credit stock (scaled by GDP) over 1998-2012 from a
balanced panel of 8 developing countries: Brazil, Kenya, Kazakhstan, Malaysia, Philippines, Poland, Sri Lanka, and Turkey. Credit is distinguished by total credit to private sector, credit to financial business, mortgages, and credit to nonfinancial business respectively.
(b) Credit stock in selected developed economies
0 20 40 60 80 100 120 140 1998 2000 2002 2004 2006 2008 2010
Total private credit Financial business Mortgages Nonfinancial business
Note: Panel (b) presents the unweighted average credit stock (scaled by GDP) over 1998-2012 from a
16 FIGURE 4
Credit growth and financial fragility: scatter plots
(a) Full sample
Notes: Pane (a) plots the observations on financial fragility indicator Z-score and one-period lagged credit
growth defined in equation (13) from all 58 countries in the sample. Univariate analysis gives the red fitted line and a correlation of -13.549 between financial fragility and credit growth. ***p<0.01, **p<0.05, and *p<0.1.
(b) Developing countries
Notes: Pane (b) plots the observations on financial fragility indicator Z-score and one-period lagged credit
growth defined in equation (13) from 28 developing countries in the sample. Univariate analysis gives the red fitted line and a correlation of -9.626 between financial fragility and credit growth. ***p<0.01, **p<0.05, and *p<0.1.
Coef = -13.549 ***
-.5 -.25 0 .25 .5 .75 1
Credit-to-GDP growth (lag1)
Coef = -9.626 **
-.5 -.25 0 .25 .5 .75 1
17 (c) Developed countries
Notes: Pane (c) plots the observations on financial fragility indicator Z-score and one-period lagged credit
growth defined in equation (13) from 30 developed countries in the sample. Univariate analysis gives the red fitted line and a correlation of -24.548 between financial fragility and credit growth. ***p<0.01, **p<0.05, and *p<0.1.
5. Estimation results
5.1 Baseline regression
Table 4 reports the results of the baseline model in equation (1). Credit-GDP growth is lagged by one period to make sure that causality runs from credit growth to financial fragility only. All specifications include time fixed-effects. First, I control for key macroeconomic indicators. The result in Column (1) shows that the coefficient of lagged credit-GDP growth is -2.172, which is not statistically significant. The coefficients of real GDP per capita and GDP growth are both positive and significant at the 10% and 1% level (3.859 and 0.203). A higher level of income and higher economic growth make debt default less likely to take place, indicated by a higher Z-score. Next I control for financial market indicators. The result in Column (2) shows that the coefficient of lagged credit-GDP growth is -2.86, which becomes significant at the 5% level. This suggests that a 10% increase in lagged credit-GDP growth tends to reduce Z-score by 0.286. A lower Z-score indicates a higher risk of default and thus higher financial fragility. The coefficient of GDP growth is still positive and significant at the 5% level (0.15), but not for real GDP per capita. Moreover, financial market indicators have their expected (negative) sign, but not significant.
In Column (3) I control for indicators of bank characteristics, i.e. costs and impaired loans. In general, this gives similar results as reported in Column (2). The coefficient of lagged credit-GDP growth is -2.508 and significant at the 10% level. The coefficient of impaired loans is negative and significant at the 5% level (0.053), implying that low asset quality is correlated
Coef = -24.548 ***
-.4 -.3 -.2 -.1 0 .1 .2 .3 .4
18 with high financial fragility. Given that the hit of 2008 financial crisis may affect the above estimation results, in Column (4)-(6) I repeat the previous exercise based on a smaller sample that only covers the pre-crisis period (2000-2008). The results again suggest a significant and negative correlation between lagged credit growth and Z-score (-3.245 and -3.753), except for Column (6).
Following Demirgüç-Kunt and Detragiache (2002), I re-estimate the baseline regression using two-period lagged credit-GDP growth as the key explanatory variable, taking into account the fact that the effect of credit growth on financial fragility may take a longer time to be realized. Likewise, the sample period is distinguished by the pre-crisis period (2000-2008) and the full period (2000-2012) to capture potential crisis effect. Estimation results are reported in Table 5. In all specifications, the coefficients of lagged credit-GDP growth are negative and significant, suggesting that lagged credit growth tends to increase financial fragility, indicated by a lower Z-score. Covariates in general have the same expected signs as those reported in Table 4.
In short, the results from the baseline regressions suggest that credit growth is associated with increased financial fragility, which is in line with my hypothesis and the finding of Demirgüç-Kunt and Detragiache (1998). As a robustness check, I also estimate the baseline specification using three-period lagged credit growth and an alternative indicator of credit growth as the key explanatory variable for financial fragility. These results are reported in Table A.3-A.6.
5.2 Robustness analysis
5.2.1 Arellano-Bond dynamic estimation
19 In line with Klomp and De Haan (2015),Z-score is highly associated with its past value. In all specifications, the coefficients of lagged Z-score are positive and statistically significant at the 1% level, which suggests that a high current Z-score tends to lead to a higher Z-score value in the next period. The results in Column (1)-(3) show that one-period lagged credit-to-GDP growth does not have significant influence on financial fragility, though their signs are all negative as expected. However, when credit-to-growth enters the regression with a two-period lag in Column (4)-(6), it is found to have strong relationship with Z-score. The coefficients of twoperiod lagged credittoGDP growth are all negative and significant (0.187, 0.263, and -0.255 respectively). This implies that the effect of credit growth on financial fragility may take a longer period to be realized, when the persistence of Z-score is taken into account.
When it comes to covariates, GDP growth is found again an important macroeconomic factor that may alleviate financial fragility. In all specifications, the coefficients of GDP growth are positive and significant. High income growth tends to reduce the default risk and contribute to lower financial fragility. In addition, banking costs is also a determinant of financial fragility, which measures managerial efficiency in the banking system. The results in Column (3) and (6) show that the coefficients of banking cost are negative and significant (-0.004 and -0005). Higher banking cost tends to reduce Z-score and increase financial fragility. Other covariates, such as current account balance, short-term real interest rate, and contemporary credit-to-GDP, have the expected signs and their coefficients are found significant in different specifications. Overall, the results from Arellano-Bond dynamic estimation corroborate the finding from the baseline regression that lagged credit-to-GDP growth is positively related to financial fragility as is hypothesized.
5.2.2 Credit growth and financial fragility: subsample analysis
As is illustrated in Figure 4 (b)-(c), the scatter plots of Z-score and credit-to-GDP growth are more widely spread in developing countries than those in developed countries, which suggests that the relationship between credit growth and financial fragility may differ across these two country groups. Motivated by this observation, I conduct a subsample analysis as the second robustness check. According to the classification by the International Monetary Fund11, I split the baseline sample into developing and developed countries. Next I re-estimate the baseline model based on each subsample. Estimation results are reported in Table 7-8.
The results from Table 7 show that lagged credit growth is strongly correlated with financial fragility in developing countries. In all specifications, the coefficients of lagged credit-GDP growth are negative and statistically significant, except for column (3). In stark contrast, when the same exercise is repeated in developed countries, the results in Table 8 suggest that lagged credit growth does not have significant influence on financial fragility. Instead, credit-to-GDP ratio is found highly correlated with Z-score. This suggests that it may be the level of credit stock, rather than the growth, that matters in terms of financial fragility in developed countries.
20 One possible explanation is that developed countries may have already accumulated a high level of debt stock, as is seen from Figure 3. In this case, the amount of debt which needs to be repaid may be more closely related to financial fragility rather than the speed at which debt is growing. For one thing, a larger debt stock indicates the presence of a larger debt-to-income mismatch. As debt will finally be repaid by income, too much debt relative to income makes debt repayment difficult, leading to a high chance of default. For another, when debt level is high, the banking system will be particularly vulnerable to adverse income shocks, which may jeopardize debt repayment capabilities and cause severe defaults throughout the economy.
5.2.3 Credit growth and financial fragility: does the type of credit matter?
Recent literature on finance and growth suggests that different types of credit impose different effect on economic growth. Following this finding, it is expected that the growth of different types of credit may have different impact on financial fragility by the debt-income mechanism discussed in Section 2.2. The growth of credit that is used for productive purposes is expected to be unrelated to financial fragility, as income improvement helps to reduce the likelihood of debt repayment problem and thereby the risk of default in the economy. On the contrary, the growth of credit that is used for speculative purposes in (real-estate) asset markets is expected to increase financial fragility, because it simply raises the debt level in the economy without proportionately creating income that allows for debt repayment. Hence, the purpose of the last robustness check is to test if the type of credit matters with regard to financial fragility. First, I separate private credit into three subcategories: mortgage credit, financial business credit, and nonfinancial business credit. Next I re-estimate the baseline regression using the growth of each type of credit defined in equation (14)-(16) as the key explanatory variable for financial fragility. Similar to the previous exercise, I distinguish sample period by the pre-crisis period (2000-2008) and the full period (2000-2012).
Table 9 presents the results of mortgage growth and financial fragility. The results from specification (1)-(6) show that the coefficients of one-period lagged mortgage growth are all negative, but not significant in both full and pre-crisis period, except for specification (3). Similar results are obtained when mortgage growth is lagged by two periods in specification (7)-(12). Hence, I do not find evidence that mortgage growth significantly increases financial fragility. In terms of financial business credit, I do not find evidence either that its growth is significantly related to financial fragility. Coefficients of financial business credit from most specifications in Table 10 are negative and insignificant, except for specification (7)-(8). The most striking finding is that growth of nonfinancial business credit is strongly associated with financial fragility. The results in Table 11 show that the coefficients of nonfinancial business credit are negative and statistically significant in all specifications. This suggests that higher growth of nonfinancial business credit tends to drag Z-score, i.e. higher financial fragility.
21 through positive asset price movements, not necessarily dependent on income. As long as asset prices keep rising, selling as
repayment. Thus, despite that the growth-enhancing effect of asset-market credit is relatively weak, it would not explicitly increase the risk of default and financial fragility due to asset price effect. Figure 5 shows this mechanism by the balance-sheet approach12. Suppose that asset-market credit flows to two main destinations: households and institutional investors. Households take mortgages from the banks to purchase housing assets in real-estate market, while institutional investors obtain credit from the banks to invest in bond and stock market13.
Panel (a) presents the balance sheet of household sector. Initially (t0) households have a claim of 50 on their deposits in the banks and own housing assets of 100 which are fully financed by mortgage loans borrowed from the banks. Then the household sector has an initial net worth of 50. Suppose that asset prices start to rise and the housing assets are worth 200 in the next period (t=t1). Since households have the same claim on bank deposits (50) and same mortgage debt to the banks (100), the net worth rises to 150 in t1 simply due to an upward housing price movement. In this case, as long as housing prices continues to grow, mortgages can always be repaid by selling housing assets in the real-estate market. Panel (b) shows the balance sheet of institutional investors. Initially (t0) they hold bonds and other financial assets. Each of them is worth 50 and partly financed by loans (50) from the banks. They have an initial net worth of 50. If asset prices rise that bonds and other financial assets are valued 60 and 90 respectively in the next period, they earn an additional net worth of 50 out of nothing. Hence, if need be, institutional investors can sell part of their financial assets to repay their debt to the banks (50) on the expectation that asset prices will continue to rise.
FIGURE 5
(a) Balance sheet of the household sector
Household sector (t t0) Household sector (t t1)
Assets Liabilities Assets Liabilities
Bank deposits (50) Mortgages (100) Bank deposits (50) Mortgages (100)
Housing assets (100) Housing assets (200)
Net worth=50 Net worth=150
(b) Balance sheet of institutional investors
Institutional investors (t t0) Institutional investors (t t1)
Assets Liabilities Assets Liabilities
Bond securities (50) Loans (50) Bond securities(60) Loans (50)
Other assets (50) Other assets (90)
Net worth=50 Net worth=100
12
For expository purpose, the balance sheets below are highly simplified.
13
22 Second, it can be seen from Figure 3 that mortgage stock, compared with other types of credit, accounts for a smaller proportion in total credit in both developing and developed countries. Thus, growth of mortgage credit may not significantly affect the overall risk of default in the banking system. Similarly, this explanation may hold for financial business credit, which accounts for even a smaller share in total credit stock in both country groups. Third, it should be noticed that the interpretation that growth of asset-market credit does not have significant effect on financial fragility is based on how fragility is defined in the present paper, i.e. the likelihood of default in the banking system. This interpretation may not be valid, if financial fragility is defined in a different manner, such as the vulnerability of the banking system to negative shocks to asset prices . In this case, an alternative opposite interpretation might be expected: the higher the growth of asset-market credit, the more likely the banking system will be affected by asset price fluctuations, and more fragile the banking system will become. Thus, financial fragility should be carefully defined before analyzing its relationship with credit growth.
Finally, it is often taken for granted that nonfinancial business credit is equivalent to credit in the theory of Schumpeter (1934) or Levine (1993; 1997), which helps to facilitate innovation and promote output and income growth. However, this assumption may not be true in reality. It may be possible that firms fail to use credit for innovation and production as is supposed to be, but instead invest (some) of the credit in the purchase of financial assets. In this case, nonfinancial business credit will also enlarge the mismatch between debt and income in the economy, and contributes to financial fragility. In addition, it may also be possible that firms borrow credit more than prudently, i.e. too much credit, due to over-optimism about the real economy. Consequently, the chance that firms fail to achieve their expected profits and fulfill their debt obligation will be higher, translating to a higher likelihood of default in the banking system and thus higher financial fragility.
6. Conclusion
Motivated by the finding of a weak relationship between credit and economic growth from the recent literature, this paper aims to examine if credit growth will lead to financial fragility by increasing the risk of default in the banking system. Based on a database of 58 countries over 1998-2012, I find that financial fragility is highly associated with credit growth in fixed-effect panel regressions and Arellano-Bond dynamic panel regressions. This result gives us a kind warning that we should not only see the bright side of credit as the fuel for innovation and economic development, but also the dark side that credit is also debt and credit should not grow at the cost of sowing the seeds of debt problems and fragility in the banking system.
23 to growth of nonfinancial credit. This may occur because asset-market credit accounts for a relatively small proportion of total credit so that its impact on system-wide fragility is not significant. Meanwhile, nonfinancial business credit may not be effectively used to support production as expected, creating debt repayment problem for firms and a high risk of default for the banking system.
One limitation of this paper is that country coverage and time period in the sample are limited by data availability. It would be better to have data on Z-score for more countries in a longer period. Second, I do not take bank regulation into account in this study, mainly because of the absence of country-level data which covers the sample period, though bank-level studies has suggested that bank regulation is also a key determinant of bank fragility. Third, I find that the relationship between credit growth and financial fragility varies from developing to developed countries. I suggest that credit growth may increase financial fragility when credit stock is not too large (for instance, below a certain threshold), and this proposition merits further research. Fourth, this paper provides one of the many possible ways to define financial fragility. In fact, financial fragility can be understood from different perspectives. Future studies may test the relationship between credit and financial fragility using alternative definitions and measures of fragility.
24 TABLE 4
Credit growth and financial fragility: fixed-effect estimation [1-period lagged] (1) (2) (3) (4) (5) (6) 2000-2012 2000-2008 Credit/GDP growth(-1) -2.172 -2.860** -2.508* -3.245** -3.753* -1.668 (1.353) (1.363) (1.444) (1.557) (1.873) (1.266) GDP per capita 3.859* 2.743 0.971 4.198 4.574 0.461 (2.297) (2.643) (2.502) (4.251) (5.303) (5.281) GDP growth 0.203*** 0.150** 0.133* 0.183* 0.077 0.124 (0.062) (0.074) (0.075) (0.095) (0.107) (0.111) Inflation 0.001 0.002 -0.054 -0.314 -0.533 -0.509 (0.219) (0.219) (0.202) (0.332) (0.349) (0.362) Current account -0.030 -0.049 -0.050 -0.023 -0.038 -0.029 (0.068) (0.071) (0.065) (0.107) (0.090) (0.079) Interest rate -0.023 -0.033 -0.112 -0.123 (0.060) (0.058) (0.082) (0.091) Credit-to-GDP -0.031 -0.015 -0.041 -0.017 (0.034) (0.038) (0.050) (0.059) Cost 0.006 0.003 (0.040) (0.042) Impaired loans -0.053** -0.024 (0.022) (0.018)
Time fixed effect YES YES YES YES YES YES
Observations 644 501 477 420 346 322
R-squared 0.085 0.073 0.089 0.081 0.080 0.088
Number of countries 58 53 51 55 50 48
Notes: This table presents the results of baseline regression in equation (1), using one-period lagged credit-to-GDP
25 TABLE 5
Credit growth and financial fragility: fixed-effect estimation [2-period lagged] (1) (2) (3) (4) (5) (6) 2000-2012 2000-2008 Credit/GDP growth(-2) -2.992** -3.636** -2.304* -4.422** -5.026** -2.526** (1.310) (1.365) (1.164) (1.895) (2.332) (1.073) GDP per capita 5.115** 4.890* 2.539 7.065 8.682 3.303 (2.165) (2.443) (2.596) (4.640) (5.489) (5.721) GDP growth 0.212*** 0.146* 0.124 0.190* 0.046 0.116 (0.059) (0.080) (0.090) (0.102) (0.132) (0.150) Inflation 0.135 0.238 0.060 -0.165 -0.321 -0.533 (0.250) (0.281) (0.266) (0.487) (0.470) (0.537) Current account 0.009 -0.008 -0.007 0.015 -0.014 -0.038 (0.050) (0.061) (0.056) (0.093) (0.089) (0.090) Interest rate -0.003 -0.014 -0.085 -0.091 (0.063) (0.062) (0.098) (0.114) Credit-to-GDP -0.045* -0.032 -0.067* -0.058 (0.024) (0.028) (0.037) (0.040) Cost -0.021 -0.034* (0.026) (0.019) Impaired loans -0.075* -0.059 (0.038) (0.052)
Time fixed effect YES YES YES YES YES YES
Observations 588 449 430 367 297 278
R-squared 0.120 0.118 0.152 0.127 0.146 0.179
Number of countries 57 50 48 54 48 46
Notes: This table presents the results of baseline regression in equation (1), using two-period lagged credit-to-GDP
26 TABLE 6
Credit growth and financial fragility: Arellano-Bond dynamic panel-data estimation
(1) (2) (3) (4) (5) (6) Z-score(-1) 0.413*** 0.457*** 0.407** 0.431*** 0.504*** 0.383*** (0.138) (0.171) (0.172) (0.145) (0.162) (0.147) Credit/GDP growth(-1) -0.145 -0.033 -0.040 (0.096) (0.114) (0.107) Credit/GDP growth(-2) -0.187* -0.263*** -0.255*** (0.111) (0.098) (0.077) GDP per capita -0.115 -0.034 -0.102 0.106 0.260 0.071 (0.232) (0.220) (0.234) (0.305) (0.230) (0.241) GDP growth 0.018*** 0.012** 0.012** 0.012* 0.011* 0.010* (0.006) (0.006) (0.006) (0.007) (0.006) (0.006) Inflation -0.001 0.013 0.021 0.002 0.011 0.012 (0.021) (0.023) (0.021) (0.023) (0.025) (0.022) Current account -0.009** -0.007* -0.007* -0.006 -0.006 -0.008* (0.004) (0.004) (0.004) (0.004) (0.005) (0.004) Interest rate -0.001 -0.002 -0.002 -0.002 (0.002) (0.002) (0.002) (0.002) Credit-to-GDP -0.005 -0.005 -0.006*** -0.007*** (0.003) (0.003) (0.002) (0.002) Cost -0.004*** -0.005** (0.001) (0.002)
Time fixed effect YES YES YES YES YES YES
Observations 573 442 442 520 395 395 Number of countries 57 50 50 57 50 50 Number of instruments 30 32 33 28 30 31 Sargan test 0.749 0.873 0.701 0.682 0.726 0.622 AR(1) 0.022 0.043 0.057 0.040 0.055 0.068 AR(2) 0.451 0.801 0.274 0.181 0.502 0.151
27 TABLE 7
Credit growth and financial fragility: evidence from developing countries
(1) (2) (3) (4) (5) (6) Credit/GDP growth(-1) -2.119* -3.334** -1.730 (1.221) (1.364) (1.491) Credit/GDP growth(-2) -3.503** -4.837*** -3.250** (1.301) (1.347) (1.457) GDP per capita 4.455 7.603 7.123 5.938 11.421* 11.613 (4.890) (6.199) (5.465) (5.057) (6.630) (7.024) GDP growth 0.163*** 0.146 0.054 0.147*** 0.087 -0.012 (0.056) (0.089) (0.092) (0.051) (0.096) (0.114) Inflation -0.034 0.038 0.099 0.045 0.195 0.181 (0.286) (0.213) (0.222) (0.387) (0.308) (0.299) Current account -0.083 -0.092 -0.086 -0.047 -0.058 -0.033 (0.100) (0.104) (0.092) (0.070) (0.086) (0.093) Interest rate -0.039 -0.040 -0.015 -0.012 (0.074) (0.065) (0.070) (0.060) Credit-to-GDP -0.022 -0.0001 -0.045 -0.009 (0.050) (0.054) (0.041) (0.043) Cost 0.100 0.034 (0.086) (0.056) Impaired Loans -0.108 -0.179* (0.071) (0.091)
Time fixed effect YES YES YES YES YES YES
Observations 301 256 245 273 231 222
R-squared 0.066 0.078 0.133 0.105 0.136 0.176
Number of countries 28 25 24 28 25 24
Notes: This table presents the estimation results of credit growth and financial fragility in developing countries,
28 TABLE 8
Credit growth and financial fragility: evidence from developed countries
(1) (2) (3) (4) (5) (6) Credit/GDP growth(-1) -2.238 0.116 -0.018 (2.770) (3.006) (3.393) Credit/GDP growth(-2) -2.691 1.374 0.804 (2.969) (2.754) (2.353) GDP per capita 8.485 7.341 3.764 11.158* 9.530 7.349 (6.167) (5.512) (5.634) (6.142) (6.904) (7.108) GDP growth 0.286*** 0.145 0.199* 0.288** 0.145 0.183 (0.103) (0.113) (0.114) (0.107) (0.121) (0.134) Inflation -0.059 -0.248 -0.386 -0.065 -0.216 -0.452 (0.297) (0.342) (0.342) (0.294) (0.362) (0.346) Current account 0.046 0.012 0.023 0.056 0.031 0.023 (0.064) (0.076) (0.069) (0.061) (0.073) (0.080) Interest rate -0.076 -0.065 -0.083 -0.063 (0.090) (0.094) (0.105) (0.108) Credit-to-GDP -0.081*** -0.073*** -0.087*** -0.086*** (0.024) (0.022) (0.023) (0.024) Cost -0.047* -0.047* (0.023) (0.023) Impaired Loans -0.049 -0.042 (0.031) (0.035)
Time fixed effect YES YES YES YES YES YES
Observations 343 245 232 315 218 208
R-squared 0.173 0.236 0.308 0.198 0.259 0.330
Number of countries 30 28 27 29 25 24
Notes: This table presents the estimation results of credit growth and financial fragility in developed countries,
32 TABLE A.1
(a) List of developing countries
(b) List of developed countries
Australia Hungary
Austria Ireland
Belgium Israel
Canada Italy
Switzerland Japan
Czech Republic Korea
Denmark Latvia
Germany Lithuania
Spain Norway
Estonia Netherlands
Finland New Zealand
France Portugal
United Kingdom Singapore
Greece Sweden
Hong Kong USA
Albania Morocco Argentina Mexico Bulgaria Malaysia Belarus Pakistan Botswana Peru Brazil Philippines Chile Poland Egypt Romania Georgia Russia
Indonesia Sri Lanka
India Thailand
Kenya Turkey
Kyrgyz Republic Ukraine
34 TABLE A.3
Credit growth and financial fragility: fixed-effect estimation [3-period lagged] (1) (2) (3) (4) (5) (6) 2000-2012 2000-2008 Credit/GDP growth(-3) -1.996 -1.845 -0.423 -2.795* -2.516 0.784 (1.358) (1.533) (1.349) (1.595) (2.334) (1.857) GDP per capita 4.821** 3.908* 1.217 5.786 6.797 -0.300 (2.091) (2.324) (2.504) (4.173) (4.643) (4.733) GDP growth 0.145*** 0.080 0.071 0.060 -0.101 -0.016 (0.043) (0.067) (0.080) (0.074) (0.110) (0.137) Inflation 0.068 0.112 0.017 -0.148 -0.353 -0.334 (0.237) (0.274) (0.271) (0.497) (0.501) (0.564) Current account 0.010 -0.006 -0.003 -0.037 -0.047 -0.095 (0.040) (0.051) (0.057) (0.075) (0.075) (0.095) Interest rate -0.018 -0.025 -0.169* -0.161 (0.067) (0.064) (0.099) (0.115) Credit-to-GDP -0.044 -0.036 -0.058 -0.065 (0.028) (0.031) (0.041) (0.043) Cost -0.031 -0.046** (0.027) (0.019) Impaired loans -0.071** -0.060 (0.033) (0.040)
Time fixed effect YES YES YES YES YES YES
Observations 534 401 390 316 251 240
R-squared 0.108 0.096 0.137 0.116 0.143 0.199
Number of countries 57 50 48 53 47 45
Notes: This table presents the results of baseline regression in equation (1), using 3-period lagged credit-to-GDP
35 TABLE A.4
Credit stock growth and financial fragility: fixed-effect estimation [1-period lagged]
(1) (2) (3) (4) (5) (6)
2000-2012 2000-2008
Credit stock growth (-1) -0.652 -0.839 -0.550 -1.214 -1.441 -0.615 (0.695) (0.710) (0.692) (0.779) (0.970) (0.755) GDP per capita 2.673 1.877 0.059 1.518 2.453 -0.881 (2.308) (2.664) (2.592) (3.523) (4.385) (4.420) GDP growth 0.181*** 0.111 0.101 0.138 0.034 0.081 (0.062) (0.076) (0.072) (0.098) (0.114) (0.109) Inflation -0.030 -0.040 -0.108 -0.355 -0.577* -0.621 (0.194) (0.197) (0.210) (0.292) (0.335) (0.404) Current account -0.015 -0.041 -0.046 0.004 -0.022 -0.020 (0.060) (0.064) (0.059) (0.091) (0.082) (0.079) Interest rate -0.024 -0.033 -0.107 -0.110 (0.055) (0.054) (0.074) (0.081) Credit-to-GDP -0.038 -0.023 -0.044 -0.020 (0.030) (0.035) (0.041) (0.047) Cost 0.003 0.001 (0.040) (0.041) Impaired loans -0.037 -0.012 (0.031) (0.035)
Time fixed effect YES YES YES YES YES YES
Observations 679 534 505 454 378 349
R-squared 0.074 0.068 0.074 0.067 0.068 0.075
Number of countries 58 54 52 56 52 49
Notes: This table presents the results of baseline regression in equation (1), using one-period lagged credit stock
36 TABLE A.5
Credit stock growth and financial fragility: fixed-effect estimation [2-period lagged]
(1) (2) (3) (4) (5) (6)
2000-2012 2000-2008
Credit stock growth (-2) -1.226 -1.237 -0.962 -2.236** -2.295* -1.469** (0.734) (0.774) (0.696) (0.985) (1.208) (0.685) GDP per capita 3.573 2.452 0.944 3.359 3.081 -0.119 (2.301) (3.020) (2.865) (4.077) (5.432) (5.476) GDP growth 0.197*** 0.132 0.117 0.169 0.040 0.086 (0.068) (0.083) (0.085) (0.107) (0.128) (0.137) Inflation 0.062 0.044 -0.042 -0.121 -0.322 -0.406 (0.204) (0.207) (0.195) (0.289) (0.300) (0.337) Current account -0.016 -0.040 -0.039 -0.028 -0.056 -0.046 (0.071) (0.077) (0.071) (0.121) (0.108) (0.100) Interest rate -0.025 -0.037 -0.137 -0.155 (0.064) (0.061) (0.099) (0.109) Credit-to-GDP -0.030 -0.014 -0.040 -0.016 (0.035) (0.039) (0.050) (0.058) Cost 0.007 0.003 (0.041) (0.043) Impaired loans -0.054** -0.025 (0.022) (0.018)
Time fixed effect YES YES YES YES YES YES
Observations 626 484 462 403 330 308
R-squared 0.084 0.066 0.081 0.082 0.079 0.089
Number of countries 57 50 48 54 48 46
Notes: This table presents the results of baseline regression in equation (1), using two-period lagged credit stock
37 TABLE A.6
Credit stock growth and financial fragility: fixed-effect estimation [3-period lagged]
(1) (2) (3) (4) (5) (6)
2000-2012 2000-2008
Credit stock growth (-3) -0.202 0.299 -0.048 -0.589 0.273 0.139
(0.713) (0.681) (0.802) (0.723) (0.870) (0.933) GDP per capita 4.193* 3.821 1.792 3.851 5.316 1.137 (2.180) (2.564) (2.730) (4.356) (5.560) (5.900) GDP growth 0.222*** 0.158* 0.126 0.225** 0.084 0.122 (0.057) (0.080) (0.093) (0.105) (0.141) (0.160) Inflation 0.117 0.180 0.061 0.014 -0.206 -0.359 (0.260) (0.277) (0.250) (0.477) (0.445) (0.497) Current account 0.036 0.021 0.020 0.025 0.002 -0.029 (0.050) (0.063) (0.054) (0.098) (0.096) (0.094) Interest rate 0.004 -0.013 -0.090 -0.113 (0.064) (0.063) (0.101) (0.115) Credit-to-GDP -0.045* -0.031 -0.064 -0.055 (0.027) (0.031) (0.040) (0.044) Cost -0.021 -0.034* (0.027) (0.020) Impaired loans -0.081* -0.059 (0.043) (0.053)
Time fixed effect YES YES YES YES YES YES
Observations 571 435 418 351 283 266
R-squared 0.114 0.105 0.149 0.111 0.119 0.173
Number of countries 57 50 48 53 47 45
Notes: This table presents the results of baseline regression in equation (1), using 3-period lagged credit stock
38 APPENDIX
Statistical implication of Z-score
Bank-level Z-score
Bank Z-score is widely used in the literature as an indicator that measures the likelihood of default by a bank, or bank fragility (for example, Demirgüç-Kunt et al., 2008; Demirgüç-Kunt and Detragiache, 2011; Klomp and De Haan, 2015). Following Boyd and Runkle (1993), the statistical rationale of this indicator is summarized below.
Suppose that is the equity-to-asset ratio that measures a bank s capitalization; ROA is the return on asset that measures the s profitability; is the bank s average return on asset; and is the standard deviation of the bank s return on asset. Assume that a bank will default if . In addition, assume that ROA is a random variable that is normally distributed with a mean of and a standard deviation of , i.e. ). Then the probability that a bank goes default can be written as:
where , is the cumulative distribution function of the standard normal distribution, and Z-score is defined as:
Hence, Z-score implies how many standard deviations bank s return on assets can drop below its mean value before bank s equity is completely depleted and it goes bankruptcy. A higher Z-score suggests a lower probability that a bank will default and therefore lower financial fragility.
Country-level Z-score
Andrianova et al., (2015)construct system-wide Z-score by aggregating the weighted average of bank-level Z-scores in a country, which they obtain from Bankscope Database. The weight is calculated as the size of a bank relative to the other banks in the country:
b denotes individual bank, i denotes country, t denotes year. is the weight used in the
