Monetary policy, capitalization and bank risk-‐taking:
Evidence from Europe
a, bAnna-‐Sophie Steinebach* June 2017 Abstract
This paper investigates whether bank leverage drives the risk-‐taking channel of monetary policy in the euro area by examining the impact of monetary policy on financial intermediaries risk appetite. By constructing a panel of 255 commercial, cooperative and savings banks from 16 European Monetary Union countries over the period 2005-‐2015, I find robust evidence that a reduction in the policy rate decreases bank risk-‐taking. Bank risk-‐taking is measured by the abnormal loan growth rate. The link between the short-‐term interest rate and bank risk-‐taking depends on the degree of bank capitalization. This positive relationship is more pronounced for lower capitalized banks. Overall, the results suggest that monetary policy loses some of its effectiveness in a low-‐interest rate environment.
Keywords: Monetary policy, risk-‐taking, banks, leverage, abnormal loan growth rate JEL-‐Classification codes: E43, E44, G20, G21, G28
a This paper is a thesis for the MSc. Economics and MSc. Finance for the Faculty of Economics & Business of the University of Groningen. Course code: EBM877A20.
b. I would like to thank David-‐Jan Jansen (DNB) and Jochen Mierau (RuG) for their input during the process of writing this research. The views expressed in this paper are those of the author and does not reflect the official view of the DNB.
* University of Groningen, Faculty of Economics and Business, student number: s2368935, email: a.j.m.steinebach@student.rug.nl.
1. Introduction
In the aftermath of the dot-‐com bust, a number of central banks eased monetary policy to stimulate the economy. This resulted in a historically low-‐interest rate environment around 2005. Many academic researchers and business press blame that in the run-‐up to the crisis, the low-‐interest rates were held too low for too long. As a result of the abundant liquidity, banks took excessive risks by increasing leverage and fueling asset prices (Borio and Zhu, 2008; Adrian and Shin, 2009; Taylor, 2009). Since the global financial crisis, central banks try to boost demand and inflation by extraordinary measures. The low-‐interest rates can support asset prices and reduce non-‐performing loans. However, when low-‐interest rates are persisting for a long period of time, this might increase banks’ risk-‐taking behavior (Jiménez et al., 2014). Hence, the recent existence of the environment of exceptionally low-‐ interest rates raises the question whether this will lead to the next financial crisis (Dell’Ariccia and Marquez, 2013).
There are several ways in which a low-‐interest rate environment can influence bank risk-‐taking. First, monetary policy might influence the risk-‐taking behavior due to valuations, cash flows and incomes, which consequently modify banks’ estimations of expected risk. For example, the price of financial assets would normally increase due to the low-‐interest rates. Hence, this can influence banks’ estimates of volatility, the probability of default and the loss in case of default. Due to the increase in risk tolerance, an expansion of a bank its balance sheets will arise (Adrian and Shin, 2009; Borio and Zhu, 2008). Second, banks have a higher incentive to increase their risk-‐taking due to lower returns on their investment.
In contrast, recent literature finds that low-‐interest rates have a negative impact on the profitability of banks’ lending business. Consequently, in a low-‐interest rate environment, this could reduce the responsiveness of the supply of loans when the interest rate decreases. Additionally, monetary policy might lose some of its traction due to the impaired banking system and the existing debt overhangs (Andrés et al., 2017; Amador and Nagengast, 2015; Borio et al., 2016).
degrees of capitalization. Overall, I find a positive relationship between monetary policy and bank risk-‐taking. Lower interest rates lead to a reduction of the abnormal loan growth rate. The relationship between monetary policy and bank risk-‐taking is more pronounced for lower capitalized banks. In other words, I do not find evidence of capitalization driving the risk-‐taking channel of monetary policy, where lower short-‐term interest rates increase bank risk-‐taking, in the euro area. There are several possible reasons for this outcome, for instance, the effect that low-‐interest rates have on banks’ profitability, the large debt overhangs, the impaired banking system, and zombie lending in the euro area. The core results tend to hold after several robustness tests are performed. The results have policy implications since it suggests that monetary policy loses some of its effectiveness in a low-‐ interest rate environment. Since lower capitalized banks decrease their risk-‐taking more than higher capitalized banks, it is advisable for the monetary authority to focus on the effect of monetary policy on banks with different capitalization ratios when deciding about the monetary policy stance.
banks (Gambacorta and Mistrulli, 2004; Altunbas et al., 2010). Additionally, the sample period includes the pre-‐crisis, the financial crisis, the sovereign debt crisis and the post-‐crisis period. This gives the opportunity to investigate whether and to what extent risk-‐taking behavior of banks with different degrees of capitalization varies over the business cycle. The financial crisis emphasizes the importance of analyzing the different types of risks that the industry faces and which could induce financial imbalances. Most of the literature about this topic focuses on the (pre)-‐crisis period. Less research focuses on the link between monetary policy and bank risk-‐taking for the euro area after the crisis period. Since there exist a low-‐ interest rate environment again, it is important to include the post-‐crisis period in this research.
The remainder of the paper is organized as follows. The next section discusses the existing and relevant literature about this topic. Section 3 describes the data, the process of obtaining the sample and the construction of the variables. Section 4 gives a description of the model specification. Section 5 presents the empirical results, it discusses the limitations of the study and includes robustness checks to verify the findings. The final section, Section 6, gives the conclusion, the policy implications of the main findings and it gives recommendations for future research.
2. Literature review
2.1. Theoretical foundation: Monetary policy and bank risk-‐taking
when a bank its capital structure is fixed exogenously. The risk-‐shifting problem could operate via the liability side of a banks’ balance sheet. Highly capitalized banks will monitor less, when they face a drop in the policy rate. The opposite is true for poorly capitalized banks. Hence, a monetary easing will increase monitoring and consequently reduce their risk-‐taking behavior. This means that net effect of a change in the interest rate on bank monitoring and its interaction with bank leverage dependents on the structure and contestability of the banking industry.
2.2. Risk-‐taking channels of monetary policy: Traditional risk-‐shifting effect or
search for yield effect
mediation effect by the degree of capitalization tend to offset each other, higher capitalized banks are more sensitive to changing their risk-‐taking behavior to changes in monetary policy.
Alternatively, there might exist a search for yield effect for financial intermediaries with short-‐term assets and long-‐term liabilities. The monetary easing reduces the margin that exists between the short-‐term assets relative to their long-‐term liabilities. In turn, these financial intermediaries might switch to assets with a higher risk and return. In the search for yield effect, lower capitalized financial institutions would be more effected by decreases in their margins.
2.3. Empirical findings of the risk-‐taking channel of monetary policy
Dell’Aricca et al. (2016) test the prediction of the theoretical framework that risk-‐taking is a function of bank capital (Dell’Aricca and Marquez, 2013) by using confidential data on U.S. banks’ loan ratings from the Federal Reserve’s Survey of Terms of Business Lending over the period 1997-‐2011. They study the link between the short-‐term interest rate, bank leverage, and bank risk-‐taking. They focus on ex-‐ante risk-‐taking by loan-‐level data on newly issued loans and thereby differentiate from other studies that rely on ex-‐post loan performance, which could be affected by subsequent events. Their findings support the theoretical model of Dell’Ariccia and Marquez (2013). Dell’Ariccia et al. (2016) find a negative correlation between ex-‐ante risk-‐taking by a bank and the short-‐term interest rates. This result gives evidence of a risk-‐shifting channel of monetary policy. It displays that the inverse relationship between bank risk-‐taking and interest rates is increasing in bank capital. They find that this effect depends on the degree of capitalization, where the negative correlation between risk-‐taking and the short-‐term interest rate is more pronounced for highly capitalized banks. Moreover, the relationship is more pronounced during periods of financial distress, and in regions that are less in sync with the national wide business cycle.
capital and more off-‐balance sheet items. Albunas et al. (2014) investigate the relationship between monetary policy and bank risk-‐taking through panel data including quarterly information for listed banks operating in the U.S. and the EU. They focus on the existence of a risk-‐taking channel of monetary policy through a cross-‐country analysis. They find evidence that unusually low levels of the policy rate over a longer period of time contribute to an increase in bank risk-‐taking.
Existing literature, using detailed data from credit registers, focuses mostly on single-‐ country analyses (i.e. Austria, Bolivia and Spain). For example, Jiménez et al. (2014) use a variety of duration models and include detailed monthly information of borrower quality from the Credit Register of the Bank of Spain. They find the same average relationship between the policy rate and bank risk-‐taking as Dell’Ariccia et al. (2016). However, in contrast, they find that the least capitalized banks are the most sensitive to changes in the policy rate. When there exists a monetary policy easing, these banks take on more risk. The results of Jiménez et al. (2014) are more consistent with a search for yield channel. A similar study, but with a different perspective is taken by Ionnidou et al. (2015) who focus on the impact of monetary policy rates on loan prices. They study the impact of the U.S. federal funds rate on the riskiness and pricing of new loans granted in Bolivia between 1999 and 2003. The monthly probability of default on individual bank loans increases when the U.S. federal fund rate decreases. A more pronounced negative relation exists between the interest rate and risk-‐taking. This result is consistent with the model of Dell’Ariccia and Marquez (2013) since banks with a higher liquidity ratio are less leveraged and the liquid assets can be used to reduce debt.
2.4. Diminishing effectiveness of monetary policy
that might explain the results is the impact that low-‐interest rates have on the profitability of a bank its traditional lending business. This is another version of the pushing-‐on-‐a-‐string which states that monetary policy may lose some of its traction at very low interest rates (Karras, 1996; Gambacorta and Rossi, 2010).
Additionally, monetary policy might lose some of its effectiveness, due to the impaired banking system and existing debt overhangs (Andrés et al., 2017; Borio et al., 2016). Besides this, the recent problems that the euro area experiences are similar to the problems that Japan faced in 1990. Authorities in both Japan and the euro area fail to adequately recapitalize the banking sector. Consequently, weakly capitalized banks have the opportunity to evergreen loans to low-‐quality zombie firms to avoid additional losses in the short-‐run. Albertazzi and Marchetti (2010) find evidence that these “evergreening” policies have taken place during the global financial crisis in Italy. Moreover, zombie banks do not grant new loans to more profitable projects, but loans are misallocated to unproductive and weak firms. Acharya et al. (2016) argue that zombie lending is an explanation for the slow recovery of the economy.
2.5. Hypotheses
The insights from the literature provide direction to my analysis. The research question of this paper is: Does there exist a risk-‐taking channel of monetary policy driven by capitalization in the euro area?
• Hypothesis 1: Lower levels of short-‐term interest rates lead to greater bank risk-‐ taking.
• Hypothesis 2: The relationship between the short-‐term interest rates and bank risk-‐ taking depends on the degree of capitalization. Consistent with the risk-‐shifting channel of monetary policy, the negative relationship between short-‐term interest rates and bank risk-‐taking is more pronounced for higher capitalized banks.
3. Data
3.1. Sample
The primary data is obtained from the SNL Financial Banking database. McGraw Hill Financial maintains this new global database, which includes banks’ ownership structure, financial statements and industry-‐specific financial market data of both private and public companies in the world. Since the first of January 2015, the European Monetary Union consists of 19 countries. The sample includes 16 EMU countries, namely Austria, Belgium, Cyprus, France, Finland, Germany, Greece, Ireland, Italy, Luxembourg, Malta, the Netherlands, Portugal, Slovakia, Slovenia and Spain.1 Before any selections are made, the
unadjusted sample consists of 768 banks located in Europe. In this study, I only include banks that are loan making and deposit taking financial institutions. Banks that do not fall in the group of commercial, cooperative and savings banks are excluded from the analysis. Hence, the sample does not include investment banks, since these banks do not take deposits. The sample starts with an unbalanced panel dataset of 433 individual commercial, cooperative or savings banks that are joining the EMU. The paper uses annual data for the panel analysis in the period 2005-‐2015.2 This period is interesting since it encompasses the
financial crisis alongside the Eurozone sovereign debt crisis. Besides this, the sample period makes it possible to test whether the relationship between the policy rate and bank risk-‐ taking is different after the financial crisis.
3.2. Ownership status and structure
In contrast to the literature (see, e.g., Altunbas et al., 2011; Leaven and Levine, 2009; Demirgüc-‐Kunt and Huizinga, 2010), the sample includes both listed and unlisted banks. It is important to include unlisted banks since they represent the majority of banks in the euro area. In this sample, around 77,25% of the banks are unlisted. In comparison to listed banks, unlisted banks are smaller, have a more traditional business model and these banks focus more on lending activities. Listed banks are often more active in non-‐lending activities. Consequently, unlisted banks are primarily exposed to credit risk. The risk-‐taking of banks
1 The dataset does not include the three EMU countries Estonia, Latvia and Lithuania. For these countries there is no data
available in SNL financial.
2 Delis and Kouretas (2011) support the validity of reporting annual data when studying the risk-‐taking channel of
through lending activities will be underestimated when focussing on listed banks only since, in that case, the risk of non-‐lending activities will be overstated (Köhler, 2015). Therefore, a more representative picture is given by including these banks in the sample.
Besides this, the sample consists of savings, cooperative and commercial banks. From these banks, savings and cooperative banks are often unlisted. These banks differ in terms of ownership structure and business structure from commercial banks (Beck et al., 2009; Hesse and Cihák, 2007; Köhler, 2015). Shareholders have the aim to maximize profits. Stakeholders, who are the owners of savings and cooperative banks, aim to improve their financial access in certain selected geographical areas and provide financial services to specific sectors. The differences between the ownership structures of banks suggest that there might exist differences in their risk-‐taking behavior (Köhler, 2015). Therefore, both commercial and savings/cooperative banks are included in the sample.
3.3. Sample selection and potential problems
Several selections are made in order to deal with problems that can arise when using the unadjusted dataset. The problem of double counting will arise when both the consolidated account of the banks’ parent and the unconsolidated account of the banks’ parent are reported.3 In the sample, the consolidated statement of the parents’ bank is included
because a significant part of the balance sheet items of the parents’ countries are related to the activities of the subsidiaries or banks operating in different countries. To capture all activities of multinational banks, including consolidated statements is necessary. Unconsolidated balance sheets could lead to a biased measurement of informational asymmetries of banks. Besides this, unconsolidated statements are not available in the dataset or only available for the time period 2010-‐2014. Using consolidated balance sheets will create problems as well since balance sheets could potentially exaggerate the size of the loans.
The data includes both the bank holding company, the bank part of the holding company and the subsidiary of the parent bank. To avoid double counting, only the bank part of the holding company is included in the sample, except from the cases where the bank holding company was equal to the bank part of the holding company. In the adjusted
3 The unconsolidated account includes statement of banks that do not include the statements of the controlled
sample, banks are included at their institutional level and not at the level of the bank holding company since these may also include other activities besides banking. The decisions, which banks to include in the sample, are made on a bank-‐by-‐bank basis. After selecting the bank part of the bank holding company, a sample of 285 banks is left. To mitigate survivorship bias, all operating and acquired/defunct banks with at least two years of financial statement accounts between 2005-‐2015 are included in the sample.4
The sample period (2005-‐2015) includes a period of many mergers and acquisitions. The database does not adjust for mergers and acquisitions. In the case of a merger, the merged banks are treated as two separated entities until the merger takes place. Thereafter, only one bank is reported. When a bank is merged, acquired or when a bank changes its name, the unique identifier for each bank remains unchanged. A new bank will obtain a new identifier in the case when a merger or acquisition intrinsically changes the bank. The SNL merger and acquisition database include information about 537 mergers and acquisitions for banks in the EMU during the period 2005-‐2015. Both databases can be crossed to obtain a better picture of the M&A environment and the decisions made by banks during the sample period.5 This is useful since it might be that some variables are affected by acquisitions during this time period. For example, the abnormal loan growth rate might be affected by acquisitions in this period, which increases the amount of loans. To ensure that the acquisitions that take place during the sample period do not drive the results, a dummy variable for acquisitions is included. The dummy variable is equal to one for the buyer in the year when an acquisition takes place and zero in the other years. Unfortunately, the SNL merger and acquisition database does not include enough information for the relative size of the M&A deal, therefore, this is not taken into account.
The disadvantage of the database is that there are limitations in data availability. Therefore, the bank risk-‐taking variables and the control variables, included in the regression, contain several missing values. Analyzing the variables display that outliers, when included, might influence the results while having little economic meaning. Therefore, the influences of the outliers, which are a few extreme observations, on the results need to
4 There might still exist a possibility that the sample is affected by survivorship bias. Some banks might be excluded from
the sample, due the fact that these banks are not included in the database.
5 The structure of the SNL Mergers and Acquisition database is followed, where a complete M&A means that a new entity
be reduced. Hence, both the risk-‐taking variables and the control variables are winsorized at the 1st and 99th percentiles of their sample distributions. Furthermore, historical information
about the bank its ownership and status is not included in the dataset. It only includes the ownership structure and status of the current year. A potential change in the specialization is not accounted for since it is expected that this limitation will not bias the results.
Table 1. Sample
Table 1 shows the number of banks and observations by country for the whole sample period. The table shows the results for listed, unlisted banks, cooperative/saving banks and commercial banks, separately. The sample period goes from 2005 to 2015. Source: SNL Financial.
Country
Total number
of observations Total number of banks of which listed of which unlisted
of which commercial banks of which savings/cooper ative banks Austria 128 16 1 15 15 1 Belgium 53 6 0 6 6 0 Cyprus 16 2 1 1 2 0 Germany 342 45 9 36 24 21 Finland 29 4 2 2 3 1 France 505 66 14 52 5 61 Greece 45 6 5 1 6 0 Ireland 65 8 2 6 8 0 Italy 261 33 12 21 32 1 Luxembourg 34 5 0 5 5 0 Malta 34 4 4 0 4 0 Netherlands 85 10 0 10 10 0 Portugal 56 8 1 7 6 2 Slovakia 21 3 3 0 3 0 Slovenia 26 3 0 3 3 0 Spain 150 36 4 32 13 23 Total 1850 255 58 197 145 110
banks might represent the behavior of banks in these countries. Nevertheless, the fixed effect model includes time and bank fixed effects. Therefore, no major deviations will be expected. However, as a robustness check Germany and France are excluded from the sample to check whether this affects the results.
3.4. Variable construction
3.4.1. Bank risk-‐takingThe abnormal loan growth rate is included as the main measure of banks’ risk-‐taking. This paper focuses on a cross-‐country analysis. To capture the relationship between monetary policy and banks’ risk appetite, it is important to use a risk-‐taking measure that captures real risk-‐taking and not the exposure to risk. The abnormal loan growth rate can be defined as the difference between an individual bank its loan growth and the median loan growth of all banks from the same country and year (Foos et al., 2010). This rate measures banks’ lending activity. When using the abnormal loan growth rate, higher growth rates do not necessarily reflect excessive risk-‐taking. If all banks have high growth rates, then this is taken into account. Foos et al. (2010) find new and comprehensive evidence of the inter-‐temporal relationship between the riskiness of individual banks and the abnormal loan growth rate. They find a significant positive correlation between the past abnormal loan growth rate and subsequent loan losses with certain lags. Higher growth rates of loans are associated with greater risk appetite when the bank increases lending by lowering the lending standards, when collateral requirements are softened or when there exist a combination of both (Foos et al., 2010). As a result, banks start to grant new loans with rates that are not according to the associated default risk. Furthermore, Foos et al. (2010) find a negative correlation between the abnormal loan growth rate and bank solvency. In conclusion, the abnormal loan growth rate is a useful measure of risk-‐taking since it measures the real risk-‐taking behavior of banks and not only banks’ exposure to risk.
Table 2. Descriptive statistics for the abnormal loan growth rate
The table summarizes the descriptive statistics for the main measure of bank risk-‐taking, namely the abnormal loan growth rate. The table displays the results for the total number of banks, listed and unlisted banks, and commercial banks and savings/cooperative banks, separately. Obs. is the number of observations of the abnormal loan growth rate. The mean value is the average value of the variable. The highest value of the variable is represented by the maximum (Max.), while the minimum (Min.) represents the lowest value of the variable during the period 2005-‐2015. P25: 25th percentile; P75: 75th percentile. The overall standard deviation is divided in within banks over time and between banks over time. All variables are winsorized at the 1% and 99% level. Source: SNL financial and own calculations.
Obs. Mean Overall Within Between Min. Max. P25 P75
All banks 2301 1.164 13.952 12.247 7.175 -‐41.350 70.001 -‐3.280 3.865 Listed banks 523 1.991 12.650 11.357 5.936 -‐41.350 70.001 -‐2.823 4.176 Unlisted banks 1778 0.934 14.260 14.262 7.456 -‐41.350 70.001 -‐3.372 3.640 Commercial banks 1279 1.265 17.225 15.110 8.728 -‐41.350 70.001 -‐2.097 2.747 Savings/cooperative banks 1022 1.037 8.165 7.220 4.743 -‐31.033 70.001 -‐5.894 5.894
Table A1 includes information about the variables, there adjoining definitions and data sources. Figure B1 shows graphs of the development of the risk-‐taking measures over time. The descriptive statistics of the abnormal loan growth rate are reported in Table 2. This table shows an average abnormal loan growth of 1.16%, with an adjoining standard deviation of 13.95%. Separating the descriptive statistics according to bank types shows significant differences between the abnormal loan growth rates. While listed banks in this sample report an average abnormal loan growth rate of 1.99%, unlisted banks have on average an abnormal loan growth ratio that is equal to 0.93%. The higher average abnormal loan growth rate of listed banks indicates that these banks are riskier than unlisted banks. Comparing the average abnormal loan growth rate of commercial banks and cooperative/savings banks shows that the average abnormal loan growth rate of commercial banks is 1.27% and 1.04% for savings/cooperative banks, respectively. In all the cases, the abnormal loan growth rate has the highest standard deviation within banks over time. Overall, it seems that outliers affect the minimum and maximum of the abnormal loan growth rate.6 This is possible, as explained, due to the mergers and acquisitions that might
affect this variable in the sample period. Therefore, it is necessary to control for this in the regression. Focusing on the average abnormal loan growth rate over time shows that it has the highest mean value around 2006-‐2008 and that the rate faces a drop in 2009. After 2010, the value of the abnormal loan growth rate decreases below 1% (Figure B1).
The descriptive statistics suggest that the differences in the ownership status and structure of a bank result into differences in the risk-‐taking behavior of these banks.
6 When separating the sample, the minimum and maximum are the same due to the winsorizing of the variables at the 1%
Therefore, it is important to take the differences in business models across banks into account to investigate whether this has effect on the link between the short-‐term interest rate and bank risk-‐taking.
3.4.2. Interest rate
This study focuses on the question whether there exists a relationship between the policy rate and bank risk-‐taking. The data for the short-‐term interest rate variables is obtained from Thomson Financial Reuters. The quarterly averages of the interest rates are transformed to annual average to combine the interest variable with the bank risk-‐taking variable and the other control variables. The overnight interest rate (EONIA) is the short-‐ term interest rate that is charged among banks and is based on the rate at which banks are able to borrow from the ECB. The ECB focuses on the maintenance of price stability, by using the interest rate channel. Before and during the crisis period, the overnight interest rate is a useful measure to monitor the effectiveness of the monetary policy transmission. Researchers focusing on het risk-‐taking channel commonly employ the overnight interest rate (see, e.g., Jiménez et al., 2014; Maddaloni and Peydró, 2011). However, it is questionable whether the use of the overnight interest rate (EONIA) is still representative as a measure of monetary policy for the whole euro area in the post-‐crisis environment.7
Hence, in the section of robustness checks, another proxy for monetary policy is used.
3.4.3. Control variables
Bank risk-‐taking might be affected by factors other than monetary policy. Hence, several bank-‐specific control variables are included to control for the fact that these variables might affect banks’ risk appetite.8 The bank-‐specific variable size (size), defined as the natural
logarithm of total assets, accounts for the potential of the ‘too-‐big-‐to-‐fail’ phenomenon. Larger banks might be too large to fail. Hence, these banks have a higher risk appetite due to safety net policies. However, Ionnotta et al. (2009) and Mohsni and Otchere (2014) find
7 In the period September 2010 to October 2015, the amount of reporting banks fell from 42 to 24. This fall in reporting
banks causes the loss of representativeness of EONIA since this fall led to a decrease in the turnover of 40 billion. Besides this, the EONIA is biased towards the northern banks in the European Union (Heijmans et al., 2016).
8 This paper intended to include the control variables profitability (profit before taxes/total assets) and liquidity (liquid
that the sign of the relationship between bank risk-‐taking and bank size is well documented but ambiguous. In a similar vein, some studies find that well-‐capitalized intermediaries have a more prudent behavior (Delis and Kouretas, 2011), while other studies find that these banks have a higher risk-‐taking behavior. To control for this, the ratio of total equity to total assets (capitalization) is included. This variable does not adjust for risk and measures book leverage. Overall, empirical literature supports the view that higher capitalized banks reduce risk-‐taking and increase bank soundness (see, e.g., Gambacorta and Mistrulli, 2004; Wheelock and Wilson, 2000; Demirgüc-‐Kunt and Huizinga, 2010; Berger and Bouwman, 2009). Recent studies (see, e.g., Coval and Thakor, 2005; Mehran and Thakor, 2011) focus on moral hazard considerations and find a negative relation between bank risk-‐taking and capital. Banks with higher levels of capital will increase the screening process of borrowers. Consequently, this results in a lower risk-‐taking behavior. Nonetheless, it might be that the managerial rent-‐seeking channel leads to higher risk-‐taking behavior due to the agency problems between managers and shareholders. This agency conflict is reduced when leverage increases since informed debt holders encourage a bank its managers to work efficiently (Diamond and Rajan, 2006). This results in a positive relationship between capitalization and bank risk-‐taking. Besides this, in principle, adequately or highly capitalized banks have a higher buffer to absorb losses which can result in higher risk-‐taking (Berger and Bouwman, 2009). Moreover, it could also be the case that both banks with very high or low capitalization increase risk-‐taking. This will happen when there exists a non-‐linear relationship (Calem and Rob, 1999). Besides this, the country control variable of the importance of the banking sector (importance), calculated by the ratio of domestic credit to GDP, is included. To meet the demand for credit, a higher ratio of domestic credit by banks to GDP, in other words a higher credit constraint, increases banks’ risk-‐taking (Männasoo and Mayes, 2009). At last, a control variable for acquisitions (acquisition) is included. This dummy variable is equal to one in the case that a bank acquires another bank during the sample period and this dummy is equal to zero when no acquisition has taken place in a specific year in the sample period. This dummy variable is included to control for the fact that acquisitions might result in a higher loan growth rate which is not related to an increase in banks’ risk-‐taking.
there adjoining definition and data sources. Table 3 reports the descriptive statistics of the proxies for monetary policy and the main control variables. Table 4 shows the correlation table of the control variables and the abnormal loan growth ratio as a risk-‐taking measure.
Table 3. Descriptive statistics for the main variables
The table reports the descriptive statistics for the main regression variables. The mean value is the average value of the variable. The highest value of the variable is represented by the maximum, while the minimum represents the lowest value of the variable during the period 2005-‐2015. The number of observations refers to the observations that are included in the sample. St. Dev. is the standard deviation. P25 is the 25th percentile and P75 is the 75th percentile. The sample period goes from 2005 to 2015. A description of the variables is included in table A1. All variables are winsorized at the 1% and 99% level. Source: SNL financial and own calculations.
Variable Observations Mean St. Dev. Minimum Maximum P25 P75 Abnormal loan growth (%) 2301 1.164 13.945 -‐41.350 70.001 -‐3.280 3.865 Overnight interest rate (%) 11 0.716 1.459 -‐0.107 3.864 0.095 2.836 Size 2609 16.862 1.284 13.542 21.016 16.099 17.519 Capitalization (%) 2608 7.149 3.579 0.845 17.798 4.563 9.054 Efficiency (%) 2558 62.224 17.770 15.287 142.756 53.001 70.503 Complexity derivatives (%) 2174 2.112 4.776 0.000 33.490 0.039 1.782 Bank importance (%) 176 94.818 23.586 33.808 253.574 82.069 104.874 Acquisition 3135 0.027 0.162 0.000 1.000 0.000 0.000
Table 4. Correlation matrix
4. Methodology
4.1. The model
This paper builds upon the empirical panel model used by Delis and Kouretas (2011). This model is augmented by several theoretical and empirical elements to investigate the link between the interest rate and the level of bank risk-‐taking through bank leverage in more detail. The empirical model is estimated as follows:
𝑅!,! = 𝛼 + 𝛽!𝑖𝑟!+ 𝛽!𝑏𝑎𝑛𝑘!,!+ 𝛽!𝑚!,!+ 𝑢!,! (1)
The variable Ri,t is the risk variable of bank i, at time t and depends on the short-‐term
interest rate, 𝑖𝑟!, at time t. Where i= 1,…,N, t= 1,…, T, and N is the number of banks, and T is the numbers of years. The empirical model includes a vector of bank-‐level control variables as described by bank of a bank i at time t and a macroeconomic control variable, mj,t, in
country j at time t, which is common to all banks in a particular year and in a specific country. The fixed effect models are estimated with robust standard errors clustered by bank-‐level.
To test whether the relationship between the interest rate and bank risk-‐taking depends on bank capitalization, an additional interaction term is included in equation (1), namely:
𝑅!,! = 𝛼 + 𝛽!𝑖𝑟!+ 𝛽!𝑖𝑟!𝐾!,! + 𝛽!𝑏𝑎𝑛𝑘!,!+ 𝛽!𝑚!,!+ 𝑢!,! (2)
The main coefficient of interest is the interaction term between the short-‐term interest rate (irt) and bank capitalization (Ki,t). Ki,t is the capitalization ratio, measured by
total equity to total assets. When there exists a negative relationship between monetary policy and bank risk-‐taking (negative value for 𝛽!), then on the one hand, a positive coefficient on the interaction term between the short-‐term interest rate and bank capital would be consistent with a “search for yield” channel. On the other hand, a negative coefficient 𝛽! on the interaction term would be consistent with a traditional risk-‐shifting channel.
4.2 Endogeneity problem
A number of identification challenges arise with the empirical model since there might exist a problem of endogeneity of the interest rate variable and the bank control variables. Additionally, the bank risk-‐taking variable might be dynamic and persistence in nature. During the global financial crisis period, central banks changed the monetary policy as a response to financial stability concerns. Hence, it is more likely that there exists an endogeneity problem of monetary policy during this period. Moreover, there might exist unobservable, but omitted bank-‐control variables. As a result, a pooled OLS regression might result in biased and inconsistent results.
The Hausman test, which examines whether random or fixed effects should be included, displays that a fixed effects model is preferred over a random effects model. This paper uses a fixed effect model, including both bank and time fixed effects, to overcome the problem of the omitted variable bias. The panel regression model includes unobserved factors that change both across banks and over time. This is important since the different regulatory environment can bias the results. The time fixed effect captures macroeconomics variables as well (i.e. GDP growth and inflation). The panel regression model includes robust standard errors clustered by bank-‐level.
5. Empirical results
In this section, the paper examines whether there exists a relationship between the short-‐ term interest rate and bank risk-‐taking. First, a regression between the short-‐term interest rate and bank risk-‐taking without control variables is estimated. Second, the paper reports the main results (equations (1) and (2)). Third, the rest of the results are reported which include the specification of a low-‐interest rate environment (equations (3) and (4)), a regression controlling for the crisis period (equation (5)), and a regression separating the sample between banks with a different ownership status and structure. At last, this section includes robustness checks to verify the findings.
5.1. Bivariate regression
Table 5 includes the estimation results of the regression between the short-‐term interest rate and bank risk-‐taking without the control variables. When the short-‐term interest rate decreases with 1%, this results in a reduction of the abnormal loan growth rate of 1.438%.
Table 5. Estimation results bivariate regression
The table reports the estimation results of the regression without control variables. The dependent variable is the abnormal loan growth rate. A description of the variables is included in Table A1. Both bank and time fixed effects are included. Fixed effect estimates are clustered by bank level. The robustness of the standard errors is included in the parenthesis. *** denote the statistical significance at the 1% level, ** at the 5% level and * at the 10%.
(1)
Short-‐term interest rate 1.438**
(0.585)
Observations 2,301
R-‐squared 0.014
Number of banks 280
Bank fixed effects Yes
Year fixed effects Yes
5.2. Baseline model
Table 6 presents the result of the baseline model where the abnormal loan growth rate is used as a measure of bank risk-‐taking. The result displays a positive relationship between the short-‐term interest rate and bank risk-‐taking. When the interest rate decreases with 1%, this leads to a reduction of the abnormal loan growth rate of around 2.46% (column (1)). This finding is in contrast with most of the existing literature (see, e.g., Delis and Kouretas, 2011; Jiménez et al., 2014; Dell’Ariccia et al., 2014). These researchers find a negative link between banks’ risk appetite and the policy rate. However, most of these studies focus on the period before the financial crisis. Since the period 2005-‐2015 contains the global financial crisis and the sovereign debt crisis, a different response of banks’ risk-‐taking to changes in monetary policy over time is expected. In these periods, banks increase their risk aversion and the amount of liquidity in financial markets (Acharya et al., 2013). Additionally, monetary policy has faced stiff headwinds in the post-‐crisis period. Moreover, monetary policy might lose some of the effectiveness due to the effect on profitability, the impaired banking system and the existing debt overhangs (Andrés et al., 2017; Amador and Nagengast, 2015; Borio et al., 2016).
increase in the abnormal loan growth rate of 1.14% (column (1)). This result is similar to the findings of Gambacorta and Shin (2016) and Borio and Gambacorta (2017). Larger capitalized banks face decreasing absolute risk aversion due to lower-‐funding cost associated with the higher capitalization ratio.
Table 6. Estimation results baseline model
The table reports the estimation results of equations (1) and (2). The dependent variable is the abnormal loan growth rate. A description of the variables is included in Table A1. Fixed effect estimates are clustered by bank level. The robustness of the standard errors is included in the parenthesis. *** denote the statistical significance at the 1% level, ** at the 5% level and * at the 10%.
(1) (2)
Short-‐term interest rate 2.463*** 3.146***
(0.703) (1.000)
Short-‐term interest rate x Capitalization
-‐0.088 (0.080) Size 11.00*** 10.69*** (2.578) (2.579) Capitalization 1.138** 1.239** (0.552) (0.553) Complexity -‐0.682** -‐0.649** (0.287) (0.285) Efficiency -‐0.076* -‐0.076* (0.044) (0.044) Bank importance -‐0.026 -‐0.035 (0.035) (0.037) Acquisition 4.369* 4.342* (2.478) (2.473) Observations 1,850 1,850 R-‐squared 0.071 0.072
Number of banks 255 255
Bank fixed effects Yes Yes
Year fixed effects Yes Yes
Column (2) in Table 6 shows the estimations results of equation (2). The baseline model is enriched with an additional interaction term given by the product between the short-‐term interest rate and capitalization. This interaction term displays a negative, but insignificant coefficient. Figure 1 gives a further intuition for the interaction between the interest rate and capitalization. Based on a calculation of the predictive margins of the abnormal loan growth rate, I find that the effect of this interaction term is economically significant. Based on the estimation coefficients reported in column (2) of Table 6, a reduction in the short-‐term interest rate from its 75th percentile of 2.84% to the 25th