Banking and the risk-taking channel of monetary policy:
Empirical evidence from the U.S.
Thilo Kristen*
Master Thesis for the MSc Finance 8th of June 2017
Abstract
This paper examines how monetary policy’s decisions shape the risk-taking behavior of banks through the level of short-term interest rates. Motivated by recent contributions in this field, this analysis assesses the existence of a risk-taking channel of monetary policy and whether it is persistent for a pre-crisis and post-crisis setting. In this regard, a data set of 391 U.S. banks for a period between 2002 and 2016 is conducted. Empirical results support the hypothesis that decreasing interest rates induce banks to increase portfolio risk. Additionally, the analysis shows that the recent excessive monetary policy, where policy rates have deviated significantly from those implied by the Taylor rule, has contributed to the increase of risk associated with bank operations. As such, the results support the claims of other researchers to incorporate fi-nancial stability considerations into monetary policy’s decision making to account for the risk-taking channel through the transmission mechanism.
JEL classifications: E43, E52, G21
Keywords: bank risk-taking, transmission mechanism, monetary policy, interest rates, Taylor rule
Student number: s3233731 Supervisor: Dr. J.O. Mierau Email: t.kristen@student.rug.nl
2 1. Introduction
The financial turmoil, beginning in the early 2000s, has caused the Federal Reserve to decrease heavily the policy rates throughout the years 2000 to 2004 to prevent the economy from slipping into a period of deflation as happened in Japan in the 1990s. For many research-ers, the prolonged period of unconventional monetary policy in this years has been a main cause of the outburst of the financial crisis and the recession that followed thereafter (Taylor, 2009; Brunnermeier, 2009). In particular, it has affected bank’s behavior in terms of their risk aware-ness: Studies as Dell’Ariccia, Laeven, and Marquez (2014), Maddaloni and Peydró (2011), Delis and Kouretas (2011), and Altunbas, Gambacorta, and Marquez-Ibanez (2010) argue that expansionary monetary policy might encourage banks to increase the risk associated with their positions.
In this regard, Borio and Zhu (2012) have addressed these concerns and introduced the risk-taking channel of monetary policy as complementing channel of the transmission mecha-nism of interest rates. They propose that low interest rates have a substantial impact on “either risk perceptions or risk-tolerance and hence on the degree of risk in the portfolios, on the pricing of assets, and on the price and non-price terms of the extension of funding” (Borio and Zhu, 2012, p. 242). As such, monetary decisions, taken in advance to stabilize the economy, have even induced the financial sector to increase risk-taking throughout portfolio allocations and contributed to the build-up of the financial crisis.
Motivated by recent contributions, the purpose of this paper is to examine the relationship between interest rates, monetary policy, and the risk-taking behavior of banks in the U.S., based on a panel data set for the period 2002 until 2016. As such, the analysis allows to obtain a more general understanding by expanding existing studies towards a greater time horizon. More spe-cifically, this paper addresses two major areas of concern. Firstly, how the mere level of interest rates affects the risk-taking behavior of banks. And secondly, on how monetary policy contrib-utes to the risk-taking channel. In this regard, a measure is introduced that accounts for the distance between real interest rates and what can be implied from the Taylor rule, as a measure of conventional monetary policy (Taylor, 2009). Since information about portfolio allocations are mostly available on aggregated level, three measures of risk are introduced which enable the identification of the risk-taking behavior in a low interest rate environment. Those are the abnormal loan growth, the risk assets, and the Z-score.
3
policy in terms of the level of the policy rates, intensifies the risk-taking of banks. All results remain persistent when including different bank-level and macroeconomic variables that might affect the risk-taking behavior of banks, as well as when controlling for distributional effects for certain bank-level characteristics.
As such, this work contributes to the existing literature by providing a more profound understanding of how banks respond to expansionary monetary policy and how they adjust their portfolio risks in periods of decreasing interest rates. The analysis implies that the risk-taking effect cannot merely be attributed to an overall weaker risk awareness prior to the financial crisis. Moreover, results confirm that the risk-taking channel works persistently through the transmission mechanism. Therefore, this paper supports the claims of other researchers that monetary authorities need to put greater emphasis on financial stability considerations when exerting their policy (Borio and Zhu, 2012; Diamond and Rajan, 2012).
The remainder of the paper is organized as follows: Section 2 provides a theoretical as-sessment on how banking depends on the level of interest rates, reviews the relevant contribu-tions in this field, and formulates hypotheses based on recent empirical studies. In section 3 the empirical strategy, the data set, and employed variables are introduced, while section 4 presents the results and their discussion. Finally, section 5 summarizes the main findings and addresses implications for monetary authorities and financial supervision.
2. Literature review and hypotheses development
2.1 Maturity transformation and the transmission mechanism
4
their positions and manage the related risk actively to maintain constant net interest margins (Dell’Ariccia and Marquez, 2006).
In this regard, banks are strongly interconnected with the monetary policy exerted by the central banks, in what is conceptualized in the literature as the transmission mechanism (Beck, Colciago, and Pfaifar, 2014). As monetary authorities intend to stimulate the economy, they influence the aggregated demand and supply of credit through the transmission mechanism by affecting the yield curve through open market transactions or changing yields on reserves. In times of stress, the central bank might want to loosen monetary policy and encourage invest-ments by providing cheap liquidity to the economy. This can be done by either purchasing short-term bonds or lowering the prime rate for reserves held at the central bank (federal funds rate in the U.S.). In turn, the shift in the yield curve has direct implications for the profitability of banking business (Diamond and Rajan, 2006).
Monetary authorities are mainly concerned with the overall economic performance, rather with implications of interest rate changes for the banking sector (Adrian and Shin, 2009). How-ever, some recent publications argue that the link between monetary policy and the risk-taking of banks is not well understood and requires further study (Borio and Zhu, 2012).
2.2 The risk-taking channel of monetary policy
The traditional literature of the transmission mechanism provides two major perceptions: The interest rate channel and the credit channel of monetary policy (Beck, Colciago, and Pfaifar, 2014). The former emphasizes the role of banks as pure intermediary, where policy makers intend to stimulate the economy by influencing the yield curve. In response, researchers introduced the credit channel of monetary policy which accounts for distortion of the borrower-lender relationship, especially due to agency problems.1
However, both perspectives fail to recognize that monetary policy might have a direct link towards the risk-taking behavior of banks (Beck, Colciago, and Pfaifar, 2014). In what has followed the financial crisis, researchers claim that low interest rates might have played a cru-cial role. Especru-cially the prolonged period of monetary easing in the early 2000s might have impinged banks to increase their appetite for risk (Taylor, 2009). In the following, many re-searchers have studied what is called the risk-taking channel of the monetary policy and ex-panded the existing channels of the transmission mechanisms discussed above. In particular,
1 Main references for the credit channel of monetary policy are Bernanke and Gertler (1995) and Kashyap and
5
the risk-taking channel promotes an increase in risk appetite of financial institutions in perspec-tive of monetary easing [major papers in this area are from Adrian and Shin (2010), Borio and Zhu (2012), Diamond and Rajan (2012), and Dell’Ariccia, Laeven, and Marquez (2014)]. The following summarizes several ways on how the risk-taking channel of the transmission mech-anism can be rationalized.
One argument in favor of the risk-taking channel of monetary policy addresses the effects of yield curve changes and how banks adjust their portfolio allocation (De Nicolò et al., 2010; Adrian and Shin, 2010). The literature identifies three major arguments in this matter: (i) asset substitution, (ii) “search for yield”, and (iii) procyclical leverage. When the yield curve shifts downward, banks need to align their lending rates and experience a decrease in interest income on newly issued loans. To compensate for the decrease in interest income, financial institutions adjust their portfolios and (i) substitute for overall riskier assets (Rajan, 2006; De Nicolò et al., 2010). A similar channel, which is discussed in the literature, is termed as (ii) “search for yield”. To meet commitments made about future cashflows and fixed costs, financial institutions need to match the yields of their assets and liabilities. In times of low interest rates, banks need to increase asset risk to maintain profitability to meet promised future cashflows (Rajan, 2006; Borio and Zhu, 2012; Buch, Eickmeier, and Prieto, 2014).
Another argument in this regard is put forward by Adrian and Shin (2010) and addresses how banks adjust their (iii) capital ratio in response to business cycle fluctuations. They argue, that banks target constant capital ratios and adjust their portfolios accordingly. When interest rates trend downwards, assets gain more in value relative to liabilities, given their longer ma-turities, leading to an overall increase of the capital ratio. Consequently, banks increase total assets to maintain the capital level constant. Nevertheless, they assume that an increase in asset quantity leads to an overall decrease in portfolio quality in terms of its riskiness. Accordingly, banks grant more loans to effectively less creditworthy borrowers.2
Another line of reasoning of the risk-taking behavior of banks is concerned with corporate governance mechanisms, in particular with implications derived from reviewing the ownership structure of financial institutions. De Nicolò et al. (2010) argue that the collaboration between managers and shareholders might spur risk-taking of banks, since their relation is subject to asymmetric information and management acts under limited liability, where all losses on failed projects are distributed to shareholders. Accordingly, the management can be expected to max-imize their private utility and engage in those projects that might generate the maximum payoff,
2 Banks prefer to increase their balance sheets in response to an erosion of leverage, rather than distributing
6
rather than in projects that generate the highest utility from shareholder perspective (highest NPV), since shareholders cannot efficiently monitor management’s behavior (Keeley, 1990). Thus, in the presence of asymmetric information, investors are expected to fail when pricing the risk of banks operations leading to an underestimation of funding costs (De Nicolò et al., 2010). Now, when banks operate in a low interest rate environment, managers might act ac-cording to the asset substitution and “search-for-yield” effect described above, when their re-muneration is linked to the overall profitability. In order to compensate for compressed returns, managers increase asset risk disregarding of shareholder wealth.
2.3 Empirical evidence and hypothesis development
Most recently, some papers have provided testable models which incorporate the risk-taking channel of monetary policy. One attempt by Dell’Ariccia, Laeven, and Marquez (2014) shows that, if a bank’s capital structure is fixed, the capitalization ratio determines the monitor-ing effort exerted over operative decisions. They argue that the monitormonitor-ing effort is negatively related to the capital ratio of banks, saying that well capitalized banks decrease monitoring in response to low interest rates. This view is compatible with the asset substitution effect, where banks increase asset risk in response to monetary easing.
Another testable model by Valencia (2014) focuses on the limited liability of banks as main driver of excessive risk-taking. When monetary authorities perform excessive policy, banks adjust their portfolios towards more risky assets (more than what the social planner or the regulator defines as optimal), since losses on failed projects (loans) are bounded due to limited liability.
Still, taking recent contribution into account the empirical evidence is somehow limited. One contribution by Maddaloni and Peydró (2011) supports the existence of the risk-taking channel, by concluding that banks lower their lending standards in response to low interest rates. They review the credit supply of U.S. (from Q2 1991 until Q3 2008) and EU banks (from Q4 2002 to Q3 2008) prior to the financial crisis and find a negative relation between low interest rates and the quality of granted loans, whereas banks are willing to grant loans and mortgages to overall less creditworthy borrowers in response to a low interest rate environment.
7
Peydró (2015) confirm these findings, when analyzing credit data of borrowers in Bolivia for a period between 1999 and 2003.
In their work, Buch, Eickmeier, and Prieto (2014) use a different approach to examine the risk-taking channel. They employ a FAVAR-model (factor-augmented vector autoregres-sive model) to analyze the response of banks to changes in interest rates. Based on data from the Federal Reserve’s Survey of Terms of Business Lending (STBL) from Q2 1997 until Q2 2008, they find that small banks adjust their loan portfolios towards a higher fraction of riskier borrowers, while larger banks maintain the composition, when granting new loans. Further-more, they find that banks do not charge higher risk premiums when increasing the fraction of risky borrowers in their loan portfolios.
In conclusion, banks are expected to respond with an increase in their risk-appetite throughout periods of low interest rates. Considering the above publications which are mainly focused on a pre-crisis setting, the first hypothesis aims to extent on prior findings for a greater time horizon and can be formulated accordingly:
Hypothesis 1: Banks respond with an increase in risk-taking when interest rates are decreasing.
In addition to the mere dependence between risk-taking and real interest rates, this paper focuses on how deviations from conventional monetary policy relate to the risk-taking channel through the transmission mechanism. In this regard, Maddaloni and Peydró (2011) use a bench-mark rate, derived from what is known as the Taylor rule (Taylor, 1993), in comparison with real interest rates. They find that deviating negatively from the benchmark rate induces banks to soften lending standards and thus increase portfolio risk. Similar findings are produced by Altunbas, Gambacorta, and Marquez-Ibanez (2010), who use the gap between a benchmark rate (they employ different modifications of the Taylor rule) and the real interest rate to disentangle effects of decreasing interest rates and divergences compared to conventional monetary policy.
To expand on the contributions of those papers, this work aims to relate deviations from conventional monetary policy more closely to the risk-taking behavior of banks. As such, the Taylor framework provides a well-established tool to determine policy rates in the context of monetary policy targets as a benchmark rate for conventional monetary policy (Taylor, 1993; Taylor, 2009). Accordingly, the second hypothesis can be formulated as follows:
8
2.4 Distributional effects of bank-level characteristics
Furthermore, Delis and Kouretas (2011) discuss in their paper distributional effects of bank characteristics and their impact on the risk-taking channel. In particular, their analysis provides empirical evidence that poorly capitalized banks engage in more risk-taking compared to those with a greater capital endowment. They rationalize their findings by a greater ability of larger banks to absorb effects of interest rate changes on risk-taking. Consistent with their work are findings of Jiménez et al. (2014) for the Spanish banking sector. Especially for the loan business, they find that poorly capitalized banks grant more loans to overall riskier bor-rowers, while they require less collateral, compared to banks with greater capital levels. As such, these analyses aim to detect a distributional effect for an U.S. sample and results in the third hypothesis:
Hypothesis 3: Banks with lower capital ratios show greater risk-taking compared to banks with better capital endowments when interest rates decrease.
Additionally, this paper examines, whether the findings of Ioannidou, Ongena, and Peydró (2015) for the effect of bank size for the Bolivian banking sector are comparable with the U.S. sector. They find that larger banks grant more loans to ex-ante riskier borrowers, and hence increase credit risk more aggressively in response to decreasing interest rates compared to smaller banks. Therefore, hypothesis 4 aims to assess, whether banks size amplifies their risk-taking:
Hypothesis 4: Banks of greater size engage in more risk-taking, when interest rates decrease compared to smaller banks.
3. Methodology and data description
The following section summarizes the identification methods used to test the hypotheses, introduces indicators on how to measure risk-taking based on observational data, and discusses additional variables that might influence the risk-taking behavior of banks. So, as to test the hypotheses whether banks effectively adjust their risk perceptions in regard of low interest rates, the identification process follows the empirical model:
bank-9
specific (𝐵𝐶𝑖𝑡), economic and structural control variables (𝐸𝐶𝑡) are included (discussed in sec-tion 3.3), as well as the error term (𝜀𝑖𝑡). As such, the estimation method is analogous to what has been done by Delis and Kouretas (2011) for EU banks between 2001 and 2008. However, this analysis extends on the time horizon by facilitating an annual data set for U.S. banks for the years 2002 to 2016.
Therefore, an unbalanced panel data set of bank-level data is obtained from the Orbis Bank Focus database (provided by Bureau von Dijk).3 The initial data set has been filtered to receive a representative sample for the purpose of this analysis, which contains 391 banks and a total of 4688 bank-level observations: Firstly, banks with less than one billion USD in total
3 Data on market-level interest rates and economic indicators have been collected from the Federal Reserve Bank
of St. Louis research database.
Table 1
Descriptive statistics (winsorized data, 90%) Variable
Observa-tions Mean Median
Standard
deviation Maximum Minimum Abnormal loan growth 4688 4.699 3.031 13.983 51.228 -24.181
Risk assets 4688 88.156 90.098 7.753 97.935 66.409 Z-score 4688 2.209 1.332 2.420 11.029 -0.025 Capitalization 4688 9.756 9.589 2.189 14.679 5.701 Size 4688 14.882 14.535 1.371 18.372 12.708 Liquidity 4688 5.519 4.125 4.004 18.102 1.444 Efficiency 4688 159.624 156.193 26.536 229.674 114.187 Profitability 4688 90.742 95.300 52.053 193.000 -106.900 GDP growth 15 3.775 3.784 1.966 6.670 -2.037 Inflation 15 2.496 2.805 1.045 3.839 -0.356 Concentration 15 2.887 2.734 0.501 3.681 1.994
Federal funds rate 15 1.228 0.176 1.626 5.021 0.089
Libor rate 3-month 15 1.530 0.692 1.669 5.297 0.234
relative Taylor rule gap 15 7.928 2.651 9.841 29.024 -5.861
10
assets are excluded to compile a data set of economically relevant banks with the mere purpose of maturity transformation. Then, only consolidated accounts have been selected to avoid dou-ble counting and to exclude specialized subsidiaries.4 Thirdly, the sample contains only listed institutions to assure comparability across banks in terms of their accounting standards (Demirgüç-Kunt and Huizinga, 2010). Fourthly, investment banks have been excluded, since they are focused on fee-based operations, rather than the supply of credit (Delis and Kouretas, 2011). Further, central banks and specialized governmental institutions have been excluded. Lastly, inactive, dissolved, and banks without Orbis Bank Focus accounts have been excluded from the data set.5
Appendix A provides descriptive statistics of the data set after all filters have been ap-plied. Since the bank-level variables show some significant dispersion around the mean, a win-sorizing technique is used to smooth out the distribution of the variables and to ease the effect of outliners (Bollinger and Chandra, 2005).6 Thereby, extreme values of both tails are replaced by realization of a predefined percentile of the distribution. In this case, the data have been winsorized by 10%, according to which 5% of the lower and upper tail are replaced with the realization at the 5% percentile for the lower tail and the 95% percentile for the upper tail re-spectively. This method is preferred, since the data set has been relatively small already and outliners are mainly yearly exceptions, rather than driven by single banks for consecutive years. Thus, the winsorizing is preferred over the mere trimming of data, since it retains more infor-mation of the initial data set and accounts for heavily skewed distributions due to overly ex-treme realizations. Descriptive statistics for the winsorized bank-level data set used throughout the empirical analysis are presented in Table 1, while Table 2 comprises the correlation coeffi-cients for respective variables. The following sections motivate the different variables em-ployed in this study.
4 The consolidation codes C1, C2, and U1 have been used to filter for consolidated accounts at the level of the
parent company.
5 Disregarding of the availability of quarterly data, this study relies on the use of annual data to capture changes in
risk-taking in respond to interest rate changes. Contributions by Delis and Kouretas (2011), Gambacorta (2005) and Ashcraft (2006) show that applying an analysis on either frequency leads to similar results in the respective context.
6 Bollinger and Chandra (2005) discuss the use of a winsorizing technique in their paper. Clearly, manipulating
11 Table 2 Correlation matrix Variable Abnormal loan growth Risk assets Z-score
Capitaliza-tion Size Liquidity Effi-ciency Lagged Profitabil-ity GDP growth Inflation Concentra-tion Federal funds rate Libor rate 3-month relative Taylor rule gap Abnormal loan growth 1
Risk assets 0.120 1 Z-score -0.020 -0.161 1 Capitalization 0.127 0.002 0.024 1 Size 0.004 0.324 -0.031 0.180 1 Liquidity -0.034 -0.141 -0.120 0.010 0.132 1 Efficiency 0.048 -0.039 0.242 0.148 0.103 -0.197 1 Lagged Profitability 0.043 -0.172 0.314 0.215 0.029 -0.115 0.485 1 GDP growth -0.093 -0.147 0.223 0.025 -0.071 -0.054 0.109 0.220 1 Inflation 0.022 -0.026 -0.063 0.018 -0.028 0.030 -0.056 -0.008 0.150 1 Concentration -0.179 0.128 -0.231 -0.114 0.053 0.036 -0.116 -0.270 -0.457 0.044 1
Federal funds rate -0.201 -0.144 0.160 -0.135 -0.114 -0.127 0.149 0.299 0.416 0.005 0.078 1
Libor rate 3-month -0.212 -0.137 0.144 -0.151 -0.115 -0.130 0.146 0.296 0.345 -0.019 0.125 0.993 1
12 3.1 Measuring bank’s risk-taking
In order to determine the riskiness of a bank’s operations, three different measures are introduced, (i) the abnormal loan growth, (ii) the ratio of risk assets to total assets (henceforth: risk assets), and (iii) the Z-score. All three measures have been employed by other researchers to approximate the risk-taking of banks in combination with aggregated bank-level data.
One concern in this context relates to the distinction between the two dimensions of risk-taking and risk-exposure. Firstly, quantifying risk-risk-taking aims to assess the likelihood or prob-ability that a position leads to an undesired outcome, isolated from what the outcome will be. Accordingly, measuring risk-taking reflects ex-ante changes in the perception of risk or the risk tolerance when making operational decisions.7 In contrast, risk-exposure aims to quantify the sensitivity with which the value of a position might change. In this regard, measuring risk-taking might overlap to some extent with the measuring of risk-exposure. Nevertheless, to as-sess the effect of interest rates on the risk-taking of banks, all three measures are employed throughout this analysis to provide a more profound understanding. The use of respective measures is explained more detailed in the following.
Throughout the analysis the main measure is the (i) abnormal loan growth which is cal-culated as the difference between institution i’s annual loan growth rate at period t and the mean of the annual loan growth of all banks in the sample (Foos, Norden, and Weber, 2010; Amador, Gómez-González, and Pabón, 2013).8 The use of this measure relates to the portfolio effect explained in section 2.1. Accordingly, banks increase the riskiness of their loan portfolios in a low interest-rate environment by granting loans to overall less creditworthy borrowers. Thus, an excessive increase in loan quantity is associated with a decrease in the quality of individual loans, resulting in higher expected losses given a higher probability of default of borrowers (Jiménez et al., 2014). Among others, this view is empirically supported by Maddaloni and Peydró (2011), who show that banks lower their lending standards when interest rates are de-creasing. Clearly, this measure gives an indication about the credit risk associated with the loan portfolio of the bank and relates more closely to the measuring of taking rather than risk-exposure.
7 The change in lending standards, discussed by Maddaloni and Peydró (2011), is one example on how the
risk-taking behavior might change.
8 Performing the analysis with either the mean or median of the annual loan growth results in similar findings for
13
As a second measure, the (ii) ratio of risk assets relative to total assets is used, which is also employed by Delis and Kouretas (2011) and Drakos, Kouretas, and Tsoumas (2016) in a similar context. In contrast to the abnormal loan growth, this measure indicates riskiness in a much broader sense. As such, the risk assets are calculated as total assets less cash (inclusive balances with the central bank), government securities, and balances due from other banks. Clearly, higher values for risk assets are associated with greater risk and a greater exposure implied by a bank’s operations. Hence, this measure does not allow to distinguish clearly be-tween risk-taking and the risk-exposure of the bank balance sheet.
Lastly, the (iii) Z-score is employed, which is closely related to the risk of insolvency of a financial institution and has been used in similar ways by Laeven and Levine (2009) and Demirgüç-Kunt and Huizinga (2010).9 It measures the distance to default, where higher values indicate higher stability of the financial institution. It is calculated as:
𝑍𝑆𝐶𝑂𝑅𝐸𝑖𝑡 = 𝑅𝑂𝐴𝑖𝑡 + 𝐶𝐴𝑅𝑖𝑡
𝜎𝑖𝑡𝑅𝑂𝐴 (2)
where 𝑅𝑂𝐴𝑖𝑡 is the return on assets at moment 𝑡, 𝐶𝐴𝑅𝑖𝑡 determines the ratio of equity capital to total assets at time 𝑡, and 𝜎𝑖𝑡𝑅𝑂𝐴 indicates the standard deviation of return on assets at time 𝑡 for entity 𝑖 (Laeven and Levine, 2009). The three last periods are used to determine the standard deviation of returns for the purpose of this study. However, the Z-score is subject to some debate among researchers, whether risk of insolvency is closely related to the risk-taking of banks. Delis and Kouretas (2011) argue against the Z-score, since insolvency is not necessarily linked to an aggressive increase in risk-taking of banks rather than unfavorable economic movements or external shocks. Hence, the Z-score as a measure of exposure seams more appropriate rather than to be a measure of risk-taking. Nevertheless, both dimensions are closely related and dif-ficult to separate when using aggregated balance sheet data. Therefore, the Z-score as a measure of insolvency risk is used as a complement to assess changes in portfolio risk associated with decreasing interest rates.
3.2 Explanatory variables
In a first step, this paper concentrates on short-term interest rates when assessing the ef-fect of the level of interest rates on banks’ risk-taking behavior, since banks fund their opera-tions mainly with short-term deposits (Adrian and Shin, 2009). In this regard, Maddaloni and Peydró (2011) provide empirical evidence, whereas short-term rates affect banks perception of
9 The Z-score was introduced by Roy (1952). In addition to the above equation, other estimation methods have
14
risk and their risk-tolerance, while long-term interest rates show no significant effect. This view relates to the critique of Adrian and Shin (2009), who state that monetary authorities oversee dependencies between the monetary policy rate and risk-taking of banks, since their policy is predominantly concerned with the level of long-term interest rates, rather than implications of their policy for short-term interest.
Therefore, the main interest rate is the federal funds rate, which reflects the costs of over-night borrowing and lending between financial institutions. It is targeted by the central bank and actively influenced by open market transactions to stimulate the economy (Bruno and Shin, 2015). Thus, the rate accounts for a direct link between monetary policy’s decision making and the funding costs of the bank (throughout this paper, the federal funds rate is also termed as the policy rate). Furthermore, the 3-month Libor rate is used as a second short-term interest rate. It is the interbank rate, which is charged for unsecured borrowing between banks for the respec-tive time horizon. Besides introducing another short-term interest rate as a form of robustness check, the main reason is to capture changes in the main determinants of the funding costs for the banking sector.
The second stream of this analysis concentrates on how banks adjust their risk appetite in response to monetary policy changes, especially on how a significant deviation of interest rates from conventional monetary policy effects the risk-taking behavior of banks. As such, conven-tional monetary policy is determined by the superior objectives: price stability, sustainable em-ployment, and moderate long-term interest rates (Adrian and Shin, 2010). The Taylor rule al-lows policy makers to determine an implied policy rate, based on these objectives (Taylor, 1993). It is a function of the difference of real inflation and respective target levels and the economic performance measured as the output gap (derived as the difference of real GDP and its potential output) and it is calculated as:
𝑇𝑅𝑡 = 𝜋𝑡+ 𝜋∗+ 𝛽
𝜋 (𝜋𝑡− 𝜋∗) + 𝛽𝑦 (
𝑦𝑡− 𝑦𝑡∗
𝑦𝑡∗ ) (3)
where 𝛽𝜋 and 𝛽𝑦 determine the weighting between price stability targets and targets for the economic output. The real inflation is denoted by 𝜋𝑡 and 𝜋∗ denotes the target inflation, while 𝑦𝑡 measures the real GDP and 𝑦𝑡∗ its long-term potential.10
Now, in order to relate real interest rates to the implied policy rate, a new measure is introduced, the relative Taylor rule gap (𝑅𝑇𝐺𝐴𝑃𝑡). It aims to capture the gap between both rates and is calculated as:
10 In this analysis, the target inflation 𝜋∗is set to two percent, which is in line with monetary policy targets and
15
𝑅𝑇𝐺𝐴𝑃𝑡 = 𝑇𝑅𝑡− 𝐹𝐹𝑅𝑡 𝐹𝐹𝑅𝑡
(4)
with 𝐹𝐹𝑅𝑡 as the federal funds rate and 𝑇𝑅𝑡 as the rate derived from Equation (3). Measuring the distance between the two rates, relative to the federal funds rate, allows to determine the relative distance between real interest rates and the implied interest rate.
Figure 1 reflects the divergence between the federal funds rate and the implied policy rate for 2002 and 2016. Some papers emphasize that real interest rates have been to accommodative in the years prior to the financial crisis, or “too low for too long”, and induced banks to increase risks of their operations [among those are Taylor (2009), Buch, Eickmeier, and Prieto (2014), and Bruno and Shin (2015)]. However, the divergence between the real interest rate and the implied rate for conventional monetary policy has become even greater starting in late 2009, when the Federal Open Market Committee (henceforth: FOMC) of the Federal Reserve lowered target rates for the federal funds rate to a span between 0% and 0.25%. Since then, the FOMC maintains targets for the policy rate at the lower zero bound. While in 2004 the benchmark rate was 2.4 times greater than the real interest rate, this ratio increased to 20.4 in 2011 and remained in the range between 14.2 and 29.0 until 2015.Only recently, in December 2015, when the FOMC announced an increase of the federal funds target rate, the gap between both rates de-creased.
While recent publications present evidence for enhanced risk-taking of banks, for “too low” monetary policy rates for the pre-crisis periods (Maddaloni and Peydró, 2011), this paper aims to assess the relation for an extended time horizon. As such, it is expected to provide more general understanding on how real short-term interest rates affect the risk-taking.
Figure 1. Comparison of federal funds rate and Taylor rule rate for the period 2002 to 2016.
-1% 0% 1% 2% 3% 4% 5% 6% 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 In tere st rate (in % )
16 3.3 Control variables
To address any concern regarding the omitted variable bias, several bank-specific, eco-nomic, and structural control variables are introduced, which might affect the risk-taking be-havior. As a first bank-specific control variable the ratio of equity capital to total assets is in-cluded. From regulators perspective, capital requirements are the main device to overcome moral hazard problems and force banks to operate in a more prudent fashion, since they would experience higher private losses in case of failure (known as “skin-in the-game effect”, De Nicolò et al., 2010). Delis and Kouretas (2011) find empirical support for a negative relation between the level of capitalization and risk-taking. However, the impact of the capital ratio remains ambiguous, when taking the work of Calem and Rob (1999) into account. They show that an increase in capital requirements tends to induce more risk-taking by ex-ante well-capi-talized banks, due to a U-shaped relationship between capital and risk-taking, depending on deposit insurance and strictness of capital requirements.
As a second measure, the logarithm of the banks total assets is employed. Following the reasoning of Ioannidou, Ongena, and Peydró (2015), where bigger banks are expected to take overall riskier positions, since they have overall greater ability to diversify those risks, one can expect a positive impact on risk-taking.11
Furthermore, the ratio of liquid assets over total assets as an approximation of banks li-quidity (those assets which can be liquidized quickly: cash balances, reserves, marketable se-curities as bonds and stocks) is used. Including a measure of liquidity, relates to the model of Acharya and Naqvi (2012), who claim that banks exert higher risk throughout their operations, when cheap funding results in excessive availability of liquidity. The authors argue that, when banks operate under high levels of liquidity, “bank managers behave in an overly aggressive manner by mispricing downside risk” (Acharya and Naqvi, 2012, p. 361), and take on more risk than socially optimal. Clearly, this line of reasoning is in line with the portfolio effect discussed in section 2.2 and the empirical findings of Maddaloni and Peydró (2011), Jiménez et al. (2014), and Ioannidou, Ongena, and Peydró (2015).
The fourth bank-specific control gives an indication about the banks efficiency and is calculated as the ratio of total revenues to total expenses, where higher values indicate more efficient operations. Regarding risk, banks with higher ratios are expected to have better
11 Similar implications for risk-taking can be derived from the work of Boyd, Jagannathan, and Kwak (2009). They
17
ties and techniques (from technological and knowledge perspective) to manage their risk expo-sure more efficiently, allowing for overall greater risk of individual positions than banks with lower efficiency ratios (Delis and Kouretas, 2011).
Lastly, the profitability is used to indicate the overall return from the banks operations, including all profits resulting from non-interest earning activities and return generated from traditional banking business. Profitability is calculated as the ratio of earnings before taxes over total assets and is included in the empirical model as lagged value to reflect implications of business decisions from the previous year. Drakos, Kouretas, and Tsoumas (2016) discuss an ambiguous relation between a bank’s profitability and their risk-taking behavior, since higher ratios of profitability could have been realized either by increasing the riskiness of positions leading to higher returns, or by acting more prudently and avoiding costs in forms of non-per-forming loans and write-offs.
Further, to account for economic influences on banks risk-taking decisions, two economic indicators are considered. These indicators control for risk that do not arise from a shift in bank’s risk-taking decisions, rather than an increase of the overall riskiness given economic circumstances. In this regard, the lending business is assumed to be riskier in times of recession, since failure to pay and default on borrower level are highly correlated (De Nicolò et al., 2010; Borio and Zhu, 2012). Therefore, the growth of GDP is included as a control variable to reflect effects of business cycle fluctuations, because banks are expected to adjust their operations to maintain constant profitability (Adrian and Shin, 2010). Additionally, the inflation is expected to affect banks through margins and overhead costs (Demirgüç-Kunt and Huizinga, 2010; Mad-daloni and Peydró, 2011).
18 4. Empirical results and discussion
4.1 First results of a pooled OLS regression
To begin with, a pooled OLS regression of the risk measures and either of the short-term interest rate should give a first indication on how changes of interest rates relate to the risk-taking of banks (ignoring all potential estimation issues). As can be observed from Table 3, the results support the existence of a risk-taking channel of monetary policy through interest rates. Firstly, columns I to IV imply that abnormal loan growth and risk assets are negatively related to changes in interest rates, either for the federal funds rate and the 3-month Libor rate. The findings for both measures support the expectation that banks increase their risk appetite by adjusting their portfolios towards riskier positions, either by increasing the credit risk inherent of the loan portfolio or by shifting towards more risky operations.
Table 3
Interest rates and bank risk-taking: pooled OLS regression
Abnormal loan growth Risk assets Z-score
I II III IV V VI
Constant 6.738 *** 7.331 *** 89.032 *** 89.164 *** 1.928 *** 1.904 *** (0.251) (0.271) (0.140) (0.152) (0.044) (0.047) Federal funds rate -1.66 *** -0.713 *** 0.229 ***
(0.123) (0.069) (0.021)
Libor rate 3-month -1.72 *** -0.659 *** 0.2 ***
(0.120) (0.067) (0.021) Observations 4688 4688 4688 4688 4688 4688 Cross-sections 391 391 391 391 391 391 Adjusted R-squared 0.037 0.042 0.022 0.020 0.024 0.019 F-statistic 181.418 *** 206.098 *** 107.25 *** 96.248 *** 113.815 *** 90.492 *** Durbin-Watson statistic 1.404 1.407 0.178 0.179 0.899 0.898
This table reports coefficient estimates and their standard errors (in parentheses). Regressions I and II have the abnormal loan growth as dependent variable, while regression III and IV have risk assets, and regressions V and VI have the Z-score. The abnormal loan growth is calculated as the banks loan growth less the mean of the country's total loan growth, the risk assets are calculated as total assets less cash, balances with the central bank, government securities, and balances due from other banks, the Z-score is calculated as the sum of return on assets and the capi-talization ratio divided by the standard deviation of returns over the last three periods. The federal funds rate and the 3-month Libor rate indicate changes of the short-term interest rate. The adjusted R-squared measures the degree to which the model explains the variance of the dependent variable. The F-statistic indicates the goodness of fit. The Durbin-Watson statistic indicates the degree of first-order autocorrelation. The statistical significance of the coeffi-cients is indicated as *** (at the 1% level), ** (at the 5% level), and * (at the 10% level).
19
However, these results do not allow for any generalization, since no additional variables have been included that might affect the risk-taking of banks, as well as any distortions that may result from unobserved dependencies of these variables.
4.2 Interest rates and risk-taking: the fixed and random effects model
In addition to the pooled regression, fixed and random effects can be used to account for unobserved heterogeneity across firms. The presence of fixed or random effects implies that there is an effect explaining the variation in the dependent variable that is unobserved by the variables included in the model (Arellano, 2003). Therefore, to control for unobserved hetero-geneity, both estimation methods are applied separately to the data set. To assess the adequacy of a random effects model, the Hausman test is applied to test, whether the entity specific effect is uncorrelated with the explanatory variable of the estimation model (Hsiao, 2014). When ap-plying the test on the final sample, the results support a rejection of the null hypothesis.12 Ac-cordingly, the results of random effect estimation method might produce inconsistent estimators for the purpose of this study. Since there is no certainty whether the Gauss-Markov properties are sufficiently satisfied, the use of a random effect estimation method is refused. Similarly, the redundant fixed effect test is used to test for the presence of fixed effects. It tests the null hy-pothesis for the joint significance of the entity fixed effects using a F-test.13 The use of fixed effects appears to be appropriate for the purpose of this study, since the null hypothesis of the redundant fixed effects test gets rejected. In order to include entity fixed effects into the analy-sis, Equation (1) can be rewritten as:
𝑅𝐼𝑆𝐾𝑖𝑡 = 𝛼 + 𝛽 𝐼𝑅𝑡+ 𝛾 𝐵𝐶𝑖𝑡+ 𝛿 𝐸𝐶𝑡+ 𝜇𝑖+ 𝜀𝑖𝑡 (5) where 𝜇𝑖 accounts for entity fixed effects of entity 𝑖.
Further, arising problems from the presence of heteroskedasticity and autocorrelation are addressed to avoid biased standard errors throughout the estimation process. A Breusch-Pagan test can be performed to test, whether the model composition is somehow subject to heteroske-dasticity (Breusch and Pagan, 1979), whereas the null hypothesis assumes homoskeheteroske-dasticity
12 The Hausman test results in rejection of the null hypothesis across all combination of the risk measures as
dependent variables. When testing for the abnormal loan growth, risk assets and Z-score, including all control variables and the federal funds rate as interest rate, the test reports 𝜒𝐴𝐿𝐺2 = 231.5, 𝜒𝑅𝐴2 = 68.0, and 𝜒𝑍𝑆2 = 80.4
respectively, whereas the p-values equal zero for all three specifications.
13 The redundant fixed effects test examines the null hypothesis, whether there is a homogenous relation across
entities in the data set. When the test rejects the null hypothesis, the entity fixed effects are significantly different from zero. When testing for the abnormal loan growth, risk assets and Z-score, including all control variables and the federal funds rate as interest rate, the test reports 𝐹1= 3.749***, 𝐹2= 26.074***, and 𝐹3= 3.723***
20
among the variables. Running the test results in rejecting the null hypothesis, hence, the esti-mation needs to be corrected to account for biased standard errors.14 Additionally, the Durbin-Watson statistic indicates the presence of serial correlation within the estimation process.15 In accordance with Petersen (2009), a White period standard error estimation method, which as-sumes cross-sectional correlated errors, is used to account for heteroskedasticity and autocor-relation simultaneously. The results for the fixed effects regression are presented in Table 4.
Results for the abnormal loan growth support the expected relationship towards interest rates. Columns I to VI indicate a negative relation between low interest rates and the risk-taking measure, implying that as interest rates decline, banks increase their lending activity exces-sively. These findings compare to those of Maddaloni and Peydró (2011), who conclude that banks decrease their overall lending standards when issuing new loans to borrowers in response to decreasing interest rates. Accordingly, the overall quality in terms of risk and return of the loan portfolio decreases, since relatively less creditworthy borrowers are provided with funding. The increase of portfolio riskiness materializes in an increase of future loan losses, as shown by Foos, Norden, and Weber (2010), and has significant effects on banks profitability and sol-vency.
Further, the results for the risk asset in columns VII to XII support the first hypothesis, since coefficients for the different interest rates are negative and significant for all model spec-ifications. Consequently, banks tend to adjust their portfolios towards more risky assets and decrease their holdings of safer assets as cash and government bonds. The findings are similar to those of Delis and Kouretas (2011) and are in line with the “search for yield” argument, presented in section 2.2, where banks increase their demand for riskier assets when interest rates do not yield sufficient returns. Furthermore, the results for the Z-score support the theoretical expectations that decreasing interest rates translate into a higher risk of default implied by the positive coefficients.
The coefficients for the interest rate variables retain their respective impact on the risk measures when including macroeconomic and bank-level control variables. For the GDP growth, the findings confirm the theoretical expectations, represented by negative coefficients for the abnormal loan growth and risk assets, and positive coefficients for the Z-score. Banks
14 Results for the Breusch-Pagan test, when running the regression with the abnormal loan growth as dependent
variable, all controls, and the federal funds rate, leads to 𝜒12= 6.32 with a probability of 0.01. Hence, the null
hypothesis gets rejected. Running the test for all other combination leads to the same conclusion.
15 The Durbin-Watson statistics are presented for all regressions in the following tables. They confirm that serial
21
Table 4
Interest rates and bank risk-taking: fixed effects regression
Abnormal loan growth Risk assets Z-score
I II III IV V VI VII VIII IX X XI XII XIII XIV XV XVI XVII XVIII
Constant 6.388 *** 6.91 *** 30.041 *** 29.914 *** -46.266 *** -43.395 *** 88.813 *** 88.886 *** 82.865 *** 83.066 *** 58.614 *** 59.062 *** 1.943 *** 1.923 *** 4.57 *** 4.484 *** -7.13 *** -7.251 *** (0.179) (0.217) (1.598) (1.582) (12.267) (12.340) (0.093) (0.113) (0.660) (0.647) (6.101) (6.120) (0.038) (0.045) (0.271) (0.271) (1.819) (1.835)
Federal funds rate -1.375 *** -0.563 *** -0.286 * -0.535 *** -0.543 *** -0.348 *** 0.217 *** 0.183 *** 0.155 ***
(0.146) (0.161) (0.174) (0.075) (0.084) (0.085) (0.031) (0.031) (0.034)
Libor rate 3-month -1.445 *** -0.645 *** -0.411 ** -0.477 *** -0.518 *** -0.331 *** 0.187 *** 0.169 *** 0.143 ***
(0.141) (0.155) (0.169) (0.074) (0.082) (0.086) (0.029) (0.029) (0.032) GDP growth -1.25 *** -1.236 *** -1.35 *** -1.321 *** -0.087 ** -0.118 *** -0.127 *** -0.149 *** 0.105 *** 0.118 *** 0.108 *** 0.119 *** (0.120) (0.118) (0.120) (0.116) (0.036) (0.035) (0.035) (0.034) (0.019) (0.018) (0.020) (0.019) Inflation 0.879 *** 0.847 *** 1.096 *** 1.061 *** -0.162 *** -0.174 *** -0.041 -0.048 -0.145 *** -0.143 *** -0.111 *** -0.109 *** (0.158) (0.158) (0.167) (0.168) (0.042) (0.042) (0.052) (0.052) (0.028) (0.028) (0.029) (0.029) Concentration -7.663 *** -7.508 *** -7.494 *** -7.25 *** 2.317 *** 2.342 *** 1.694 *** 1.711 *** -0.907 *** -0.909 *** -0.79 *** -0.79 *** (0.493) (0.498) (0.546) (0.552) (0.222) (0.226) (0.251) (0.259) (0.085) (0.086) (0.094) (0.095) Capitalization 0.829 *** 0.808 *** -0.007 -0.012 -0.005 -0.003 (0.219) (0.220) (0.107) (0.107) (0.033) (0.033) Size 4.307 *** 4.086 *** 1.996 *** 1.979 *** 0.589 *** 0.59 *** (0.843) (0.846) (0.422) (0.425) (0.121) (0.122) Liquidity -0.202 ** -0.212 ** -0.283 *** -0.285 *** -0.009 -0.009 (0.091) (0.091) (0.055) (0.055) (0.015) (0.015) Efficiency 0 0.013 0 ** -0.015 ** 0.014 *** 0.014 *** (0.016) (0.016) (0.007) (0.007) (0.002) (0.002) Lag. Profitability 0.02 *** 0.022 *** -0.002 -0.002 0.005 *** 0.005 *** (0.006) (0.006) (0.002) (0.002) (0.001) (0.001) Observations 4688 4688 4688 4688 4513 4513 4688 4688 4688 4688 4513 4513 4688 4688 4688 4688 4513 4513 Cross-sections 391 391 391 391 386 386 391 391 391 391 386 386 391 391 391 391 386 386 Adj. R-squared 0.189 0.193 0.245 0.247 0.267 0.268 0.692 0.690 0.720 0.720 0.741 0.740 0.221 0.216 0.285 0.284 0.314 0.313 F-statistic 3.797 *** 3.875 *** 4.868 *** 4.899 *** 5.174 *** 5.189 *** 27.974 *** 27.704 *** 31.612 *** 31.626 *** 33.687 *** 33.67 *** 4.408 *** 4.312 *** 5.737 *** 5.723 *** 6.234 *** 6.221 *** Durbin-Watson stat. 1.810 1.816 1.790 1.794 1.831 1.839 0.608 0.607 0.648 0.651 0.674 0.649 1.230 1.227 1.323 1.323 1.410 1.410
22
respond to an economic downturn by adjusting their portfolios towards riskier assets to maintain constant profit margins (Adrian and Shin, 2010). Similar results are found for the effect of inflation, where growing inflation rates are associated with an increase of credit risk (measured by the abnormal loan growth) and default risk (measured by the Z-score). Given the findings of Demirgüç-Kunt and Huizinga (2010), where an increasing inflation relates positively to the return on assets, Table 4 provides evidence for the “search-for-yield” channel discussed above. However, the setting with risk assets as measure provides contradicting results, when employ-ing the macroeconomic variables only, and insignificant results when addemploy-ing the bank-specific control variables.
As Table 4 presents, the concentration variable produces mixed results for the different measures. For the abnormal loan growth and the Z-score, the results point towards a negative effect of higher market concentration. Hence, less competition among banks (higher relative market share of individual banks) diminishes their incentive to increase asset risks and results in more conservative banking. Or differently, banks are prone to increase risks when the market becomes more competitive. On the contrary, the results in the setting with risk assets suggest a positive relation between higher concentration and higher risk-taking, which is in line with what has been suggested by Boyd and De Nicolò (2005). They rationalize the positive relationship by an increase of borrower default risk, since banks pass-through higher lending rates.
The existing literature provides mixed findings for the impact of the capitalization ratio on risk-taking. However, this analysis provides results that confirm assumptions made by Ca-lem and Rob (1999). They hypothesizethat well capitalized banks are induced to take riskier positions, which is supported by the results related to the abnormal loan growth. Findings of Delis and Kouretas (2011), who present a negative relation between the capitalization ratio and risks assets, cannot be verified for the U.S. sample. Both outcomes for risk assets and the Z-score are insignificant.
With regards to the size as control variable, the settings with abnormal loan growth and risk assets as risk measures imply that larger banks are less risk averse, similar to findings from Boyd, Jagannathan, and Kwak (2009) and Ioannidou, Ongena, and Peydró (2015). Reversely, as regressions XVII and XVIII illustrate, the size of banks is positively related to the Z-score, indicating that size alleviates default risk, partly supporting the ideas that bigger banks have an overall greater ability to diversify risks (Boyd, Jagannathan, and Kwak, 2009).
How-23
ever, the results appear coherent with how the variable is measured. As the liquidity term de-scribes one fraction of banks total assets, in particular while risk assets comprise the contrary part of bank’s total assets. Hence, liquidity might still have an impact on the riskiness of single positions, but measuring the impact on aggregated level conflicts with the identification strat-egy. Accordingly, the liquidity variable will be excluded in the following analyses.16
Further, efficiency (measured as ratio of total revenues to total expenses) and lagged prof-itability (ratio of earnings before taxes over total assets) do not have a substantial effect on risk-taking, since coefficients are either close to zero (for the Z-score) or statistically insignificant. The presented results work in favor of hypothesis 1, where banks respond to decreasing interest rates by exercising higher risk throughout their portfolio allocation. While the results for the abnormal loan growth clearly imply a higher risk-taking, the Z-score illustrates an in-crease of risk-exposure, as response to decreasing interest rates.17 Hence, these findings corre-spond to what has been found by Delis and Kouretas (2011) for a European sample for the years 2001 until 2008.
4.3 Deviating from conventional monetary policy
After testing the dependence of interest rate levels on the risk-taking behavior of banks, the following analysis aims to detect the relation towards deviations from conventional mone-tary policy. For that purpose, the interest rate variable is replaced with the relative Taylor rule residual in Equation (5). As discussed in section 3.2, introducing the relation of real interest rates and the implied policy rates allows to evaluate effects of their distance towards the risk measures.18 Table 5 presents the results.
The coefficients for the relative Taylor rule rate, when regressing with the abnormal loan growth and risk assets as measures for risk-taking (regressions I to VI), are positive and highly significant throughout all three specifications (including control variables). Accordingly, banks risk-taking is amplified when the real interest rate (the federal funds rate) falls below the bench-mark rate and the relative distance between both rates increases. The results for the Z-score
16 Running the regression excluding the liquidity variable does not alter the impact and significance of the
coeffi-cients of the individual variables. Appendix C provides a replication of regressions V, VI, XI, XII, XVII, and XVIII excluding the liquidity variable.
17 Some researchers raise the selection bias or survivorship bias (Delis and Kouretas, 2011) as major concern when
dealing with bank-level data. In this study, inactive and dissolved banks have been excluded from the data set which might shape the results in a certain way. Conversely, it is expected that those excluded banks went bankrupt because of their excessive risk-taking behaviour in preceding periods. Accordingly, the presented results only understate the real effect of interest rates on bank’s risk-taking and that including dissolved banks would only amplify the negative interdependence with interest rates presented in this section.
18 The analysis is performed similarly as before, using a fixed effects model and adjusting for clustered standard
24
Table 5
Relative Taylor rule gap and bank risk-taking: fixed effects regression
Abnormal loan growth Risk assets Z-score
I II III IV V VI VII VIII IX
Constant 3.024 *** 32.117 *** -14.603 87.592 *** 84.771 *** 61.49 *** 2.238 *** 3.951 *** -4.702 ***
(0.179) (1.497) (13.339) (0.097) (0.555) (6.368) (0.034) (0.280) (1.739)
rel. Taylor rule gap 0.211 *** 0.235 *** 0.223 *** 0.071 *** 0.109 *** 0.065 *** -0.004 -0.006 0.001
(0.023) (0.025) (0.029) (0.012) (0.015) (0.016) (0.004) (0.005) (0.005) GDP growth -1.632 *** -1.693 *** -0.397 *** -0.316 *** 0.195 *** 0.168 *** (0.119) (0.119) (0.045) (0.048) (0.023) (0.022) Inflation 0.176 0.332 * -0.444 *** -0.237 *** -0.153 *** -0.143 *** (0.169) (0.195) (0.053) (0.065) (0.029) (0.030) Concentration -8.159 *** -7.17 *** 1.777 *** 1.446 *** -0.709 *** -0.543 *** (0.450) (0.492) (0.183) (0.214) (0.086) (0.092) Capitalization 0.632 *** -0.031 -0.021 (0.228) (0.109) (0.033) SIZE 1.972 ** 1.695 *** 0.367 *** (0.927) (0.445) (0.115) Efficiency 0.034 ** -0.007 0.014 *** (0.016) (0.007) (0.002) Lagged Profitability 0.027 *** -0.001 0.007 *** (0.006) (0.002) (0.001) Observations 4688 4688 4513 4688 4688 4513 4688 4688 4513 Cross-sections 391 391 386 391 391 386 391 391 386 Adjusted R-squared 0.185 0.265 0.278 0.688 0.727 0.734 0.200 0.274 0.308 F-statistic 3.723 *** 5.299 *** 5.428 *** 27.378 *** 32.62 *** 32.686 *** 3.988 *** 5.486 *** 6.104 *** Durbin-Watson statistic 1.807 1.800 1.852 0.615 0.706 0.678 1.216 1.321 1.419
25
(regressions VII to IX) indicate no significant results, while the control variables remain per-sistent in terms of their coefficients and their significance compared to results in Table 4.
The results for regressions I to VI can be explained by effects, operating through the communication policy and reaction function of the central bank. In their paper, Borio and Zhu (2012) elucidate that risk-taking behavior might be influenced by what they introduce as “trans-parency effect” and “insurance effect” related to changes of monetary policy. The former de-scribes that transparency of monetary policy diminishes accompanied uncertainty surrounding future decision making, and hence affects ultimately banks behavior. Accordingly, banks are induced to increase their portfolio risk, when they anticipate a prolonged period of low interest rates, working through the channels asset substitution and “search for yield”, as discussed in section 2.2. The “insurance effect” relates to how banks perceive the commitment of central banks to intervene in times of stress. Thereafter, the attempt of authorities to mitigate downside risk throughout the policy rate might spur an asymmetric impact on the risk tolerance and fos-ters increasing risk-taking behavior (Borio and Zhu, 2012).
Conclusively, the findings provide evidence for the second hypothesis, stating that devi-ating from conventional monetary policy has adverse effects on the risk-taking decisions made by banks. These findings compare to those of Altunbas, Gambacorta, and Marquez-Ibanez (2010) and Maddaloni and Peydró (2011) in the way that unusual low interest rates over a pro-longed period have contributed to an extensive risk-taking by banks. However, both papers make these contributions for the pre-crisis period, while this paper examines effects for risk-taking disconnected from the financial crash in 2007 and 2008 for an even greater period. These findings make it inevitable for monetary authorities and bank supervision to put greater empha-sis on the risk-taking channel of the transmission mechanism.
4.4 Distributional effects of bank characteristics
Lastly, to test for distributional effects of the bank-level variables that have been used throughout the analysis, interaction variables for bank size and the capitalization ratio with the interest rates are introduced to test for hypothesis 3 and hypothesis 4. The use of the interaction variable allows to determine, whether the relation between interest rates and the risk measure is alleviated or narrowed for a certain level of the respective bank characteristic (Agoraki, Delis, and Pasiouras, 2011). Therefore, the regression gets reformulated, such that it takes the follow-ing form:
26 Table 6
Distributional effects of bank characteristics: fixed effects regression
Abnormal loan growth Risk assets Z-score
I II III IV V VI Constant -49.626 *** -14.874 54.432 *** 61.393 *** -7.115 *** -4.762 *** (12.097) (13.339) (6.256) (6.383) (1.838) (1.722) GDP growth -1.357 *** -1.704 *** -0.136 *** -0.319 *** 0.107 *** 0.169 *** (0.120) (0.119) (0.035) (0.050) (0.020) (0.023) Inflation 1.081 *** 0.341 * -0.055 -0.236 *** -0.112 *** -0.148 *** (0.168) (0.196) (0.053) (0.065) (0.030) (0.030) Concentration -7.534 *** -7.15 *** 1.666 *** 1.453 *** -0.788 *** -0.538 *** (0.546) (0.493) (0.250) (0.213) (0.094) (0.092) Capitalization 0.867 *** 0.65 *** 0.013 -0.029 -0.008 -0.03 (0.224) (0.228) (0.109) (0.108) (0.033) (0.033) SIZE 4.42 *** 1.986 ** 2.128 *** 1.701 *** 0.583 *** 0.372 *** (0.847) (0.926) (0.433) (0.446) (0.122) (0.114) Efficiency 0.014 0.033 ** -0.012 * -0.008 0.014 *** 0.015 *** (0.016) (0.016) (0.007) (0.007) (0.002) (0.002) Lagged Profitability 0.021 *** 0.027 *** 0. -0.001 0.005 *** 0.007 *** (0.006) (0.006) (0.002) (0.002) (0.001) (0.001)
Federal funds rate -0.216 -0.271 *** 0.153 ***
(0.178) (0.084) (0.033)
FFR (#) * CAR (#) 0.002 0.012 -0.011
(0.083) (0.038) (0.015)
FFR (#) * Size (#) 0.117 0.003 0.
(0.100) (0.053) (0.021)
rel. Taylor rule gap 0.225 *** 0.065 *** 0.
(0.029) (0.016) (0.005) RTGAP (#) * CAR (#) -0.013 -0.004 0.001 (0.010) (0.005) (0.002) RTGAP (#) * Size (#) 0.002 0.003 0.007 *** (0.015) (0.009) (0.003) Observations 4513 4513 4513 4513 4513 4513 Cross-sections 386 386 386 386 386 386 Adjusted R-squared 0.266 0.278 0.732 0.734 0.314 0.309 F-statistic 5.139 *** 5.407 *** 32.247 *** 32.521 *** 6.219 *** 6.115 *** Durbin-Watson statistic 1.825 1.856 0.657 0.678 1.412 1.425
27
To account for multicollinearity among the independent variable and the interaction term, “mean-centering” for the interest rate and bank-level characteristics is used. Therefore, respec-tive variables get adjusted by their means and are included as interaction variable into the em-pirical model along the unadjusted control variables (Delis and Kouretas, 2011). The correla-tions before and after the “mean-centering”, presented in Appendix B, indicate that the adjust-ment of the variables diminishes respective issues. Table 6 presents the results.
Firstly, the results indicate that the relation between interest rates and the risk measures do not change when including the interaction variables. Hence, the variation is only captured in the variation of the interaction variables. Nevertheless, only regression VI produces significant results for the interaction term when using the Z-score as risk measure, while analysis with either abnormal loan growth and risk assets produce insignificant results only. As setting VI indicates, the interaction term of the relative Taylor rule gap and bank size is positively related to the Z-score. Although, the real effect of the interaction term on the insolvency risk is negli-gible with a coefficient of 0.007 (in percent).
Thus, it can be concluded that the effects of decreasing interest rates and unconventional monetary policy on the risk-taking behavior of banks are persistent across bank’s capital en-dowment and its size. However, these results are conflicting with those of Delis and Kouretas (2011) and Jiménez et al. (2014) for the impact of the capitalization ratio on risk-taking, and with those of Ioannidou, Ongena, and Peydró (2015) for the impact of bank size, as discussed in section 2.3. Nevertheless, these findings cannot be verified from the U.S. sample over a greater time span, and hence do not support the third and fourth hypotheses.
5. Concluding remarks
Motivated by the contribution of Borio and Zhu (2012), the objective of this paper is to contribute to the understanding of the impact of monetary policy’s decisions through the trans-mission mechanism on the risk-taking behavior of banks. For this purpose, a regression analysis has been performed using a panel data set of annual observations for 391 U.S. banks between 2002 and 2016. Respective period allows to derive more general inferences independent from effects influenced by the crisis years 2007 and 2008, and thus adds to the results of prior con-tributions in this matter.
28
by the Z-score. Similar effects are also derived from the risk assets as a third risk measure which reviews risk in a broader sense (see section 4.2).
Regarding the increase of credit risk, banks are induced to lower their lending standards which results in a lower quality of the loan portfolio, and hence increases risks of future loan-losses. Conclusively, the abnormal loan growth provides a reliable measure of the risk-taking of banks. In turn, the Z-score provides an approximation on how bank’s risk-exposure relates to changes in interest rates, or differently, on how interest rates affect their distance to default. In order to test the second hypothesis, the analysis has been modified to capture effects of deviations of real interest rates from conventional monetary policy. Therefore, the gap be-tween real interest rates and the implied policy rate has been set into relation to the federal funds rate. By doing so, bank’s response to policy decisions has been separated from merely measur-ing the level of interest rates, and hence enables a better understandmeasur-ing on how monetary policy determines bank’s risk-taking behavior. The findings support the second hypothesis that signif-icant deviations from conventional monetary policy increase the risk-taking. As such, banks risk-taking relates adversely to the central banks communication and reaction of their monetary policy through the “transparency and insurance effect”, discussed by Borio and Zhu (2012).
Further, the analysis provides no evidence for distributional effects of bank’s size or the degree of capitalization. Therefore, it can be concluded that the risk-taking channel is persistent among both characteristics for the U.S. banking sector, which is equivalent to rejecting hypoth-esis 3 and hypothhypoth-esis 4. Lastly, all the presented results remain robust when including either bank-level and macroeconomic control variables that might affect the risk-taking.
The results introduce major challenges for either the monetary authorities and bank su-pervision. In particular, monetary policy authorities need to account for the interconnection between risk-taking and the monetary policy rate, since it has substantial effects for the financial stability of the economy. Especially, the current expansionary monetary policy of the Federal Reserve might trigger excessive risk-taking by banks, motivated by the effects of asset substi-tution, “search for yield” and procyclical leverage. Similarly, supervision of the banking sector must establish a framework that fosters financial stability, as such that the risks of a bank’s positions are well understood by management and investors to properly incorporate them into operative decision making. Further, associated risk of bank’s operations need to be transparent for financial markets, to mitigate their mispricing by external financial agents.
29
as presented in this paper are persistent in other economically relevant markets. In this regard, the effect of the all-time low interest rates after 2009 might be of special interest. Especially for the European setting, and whether negative short-term interest rates have an even amplifying effect on the dynamics of the risk-taking channel. Secondly, an extended examination on how emerging regulations affect the risk-taking channel might be valuable and how it interacts with the transmission mechanism. In particular, whether higher regulation standards contribute to the understanding of risk attributed to bank’s positions and whether a greater understanding results in more prudent monetary policy’s decision making to foster financial stability.
30 Appendix A
Descriptive statistics (initial sample)
Variable
Observa-tions Mean Median
Standard
de-viation Maximum Minimum Abnormal loan growth 4688 7.416 3.031 2.477 5149.123 -103.838
Risk assets 4688 87.851 90.098 9.253 99.772 25.523 Z-score 4688 2.477 1.332 4.006 89.531 -1.787 Capitalization 4688 10.183 9.589 5.187 85.373 -4.686 Size 4688 14.958 14.535 1.617 21.668 9.926 Liquidity 4688 6.302 4.125 7.306 79.043 0.255 Efficiency 4688 159.851 156.193 43.887 548.315 -1179.191 Lagged Profitability 4688 85.560 95.300 131.330 1926.300 -3025.400 GDP growth 15 3.775 3.784 1.966 6.670 -2.037 Inflation 15 2.496 2.805 1.045 3.839 -0.356 Concentration 15 4.935 6.072 2.201 7.326 0.876
Federal funds rate 15 1.228 0.176 1.626 5.021 0.089
Libor 3-month 15 1.530 0.692 1.669 5.297 0.234
Taylor rule rate 15 -7.928 -2.651 9.841 5.861 -29.024
