Low interest rates and risk-taking behavior by banks: Has
anything changed after the financial crisis?
Empirical evidence from the United States
Tiemen van Hezel
19445762 ABSTRACT
This paper examines the relationship between low interest rates and bank risk–taking, and the influence the financial crisis has had on this relationship. A dataset of 142 US banks is analyzed for the time period 2003 to 2016. The fixed effects model predicts a positive relationship between interest rates and bank risk-taking, contrasts to most studies, which predict increased risk-taking by banks in a low interest rate environment. This relationship is examined in terms of four independent interest rates (three-month Libor, the federal funds rate, the Taylor-rule rate and long-term interest rate) and two bank risk-taking proxies (risk assets ratio and abnormal loan growth). The risk-taking proxies in this thesis measure the risk appetite of banks instead of the ex-post loan performance. The results are robust for the dependent variables and different specifications of the model. Furthermore, it is demonstrated that the relationship between interest rates and bank risk-taking has been significantly influenced by the financial crisis. The relationship is insignificantly negative for the pre-crisis period, and significantly positive for the post-crisis period, meaning that in the post-crisis period low interest rates have lowered bank risk-taking.
Key words: bank risk-taking, monetary policy, low interest rates, financial crisis, interest rates, panel data
3 TABLE OF CONTENTS
1. INTRODUCTION 4
2. LITERATURE REVIEW 5
2.1 Risk-taking channels by banks 5
2.2 Empirical evidence 6
2.3 Proxies for risk 8
3. METHODOLOGY 9 4. DATA 11 4.1 Dependent variables 12 4.2 Independent variables 12 4.3 Control variables 13 5. RESULTS 16
5.1 Fixed effects regression 16
5.2 Fixed effects regression including control variables 17
5.3 Fixed effects regression on the three subsamples 18
5.4 Fixed effects regression and the influence of the financial crisis 20
6. CONCLUSION 20
7. LITERATURE 23
APPENDIX A 25
APPENDIX B 26
4
1. Introduction
The relationship between interest rates and bank risk-taking has been extensively studied in recent years, the financial crisis of 2008 made this relationship of even greater interest. After the dotcom bubble, the central banks offered low interest rates in order to stimulate the economy. These low interest rates have often been blamed as the cause of the 2008 financial crisis (Taylor, 2009). Now, more than ten years later, the world is still recovering from the financial crisis, and central banks stimulate the economy by cutting the costs of borrowing. The interest rates are lower than pre-crisis, and therefore it is crucial to gain comprehensive knowledge about the relationship between interest rates and bank risk-taking.
The traditional role of banks has been to take deposits and provide loans, where the interest charged on loans is greater than the interest paid on deposits (Hull, 2012). This business is called maturity transformation, and profitability increases as the difference between short-term and long-term interest rates increases; in other words, when the slope of the yield curve steepens. Low interest rates reduce the net interest margins of banks. Products were developed to increase profits and to enable mortgage originators to transfer credit risk to investors (Hull, 2012). The risk of these products was underestimated by the rating agencies, because of a lack of competition, poor accountability, or, most likely, an inherent difficulty to assess risk due to the complexity (Taylor, 2009).
In October 2008 market conditions deteriorated precipitously and rapidly (Taylor, 2009). The housing market’s collapse was a consequence of risk-taking behavior by the banks. While many blame the low interest rates as a major cause of the crisis, the literature is not unanimous as to whether low interest rates led to greater risk-taking by the banks. Most studies conclude that there is a significantly negative relationship between interest rates and bank risk-taking (Delis & Kouretas, 2011; Altunbas, Gambacorta, and Marques-Ibanez, 2010; Maddaloni & Peydró, 2011; Dell’Ariccia, Laeven, and Marquez, 2014; Andries, Cocriş, and Pleşcău, 2015). Low interest rates created an incentive for banks, hedge funds, and other investors to hunt for riskier assets that offered higher returns. Maddaloni & Peydró (2011) argue that low interest rates lead to securitization, and that higher securitization leads to softer lending standards and increased bank risk. Today’s low interest rates are the reason Andries et al. (2015) consider the likelihood of a new financial crisis to be increasing. Moreover, Denning (2013) argues that another financial crisis is inevitable, since banks are bigger and more complex than ever and still operate with the same unknown risks, but on a larger scale since the crisis. The Basel committee on banking supervision developed a third Basel accord in response to the financial inconsistencies in the banking sector revealed by the recent financial crisis. The introduction of this Basel III agreement in 2013 was intended to strengthen capital requirements for banks. The overall effect of regulations such as this one should have led to a less risky post-crisis banking sector in general, and therefore could have changed the interaction between banks’ risk-taking and low interest rates. However, stricter capital requirements can also induce banks to increase the risk profiles of their portfolios. The main questions this thesis seeks to answer are the following;
5
To study this relationship I use an extensive dataset of commercial, savings and cooperative banks from the US over a prolonged period, drawing a distinction between risk exposure and risk-taking. The existing literature tends to use risk proxies that measure risk exposure rather than actual risk-taking, for example with proxies such as non-performing loans, the Z-score, and loan loss provision. These measures quantify the sensitivity of the value influenced by factors independent from bank behavior. Risk assets ratio and abnormal loan growth are pursued in this thesis because these measures reflect ex-ante changes of risk tolerance of banks when making decisions, and by using two measures I can test for robustness. As explanatory variables I employ the federal funds rate, the three-month Libor rate, the Taylor-rule rate, and the long-term rate. To the best of my knowledge, research such as conducted for this thesis has never been done before, so this thesis makes a valuable contribution to the existing literature.
The main results of this thesis show a positive relationship between interest rates and bank risk-taking. First, a regression without any control variables is conducted to obtain a first impression about the relationship. Subsequently, a second attempt is conducted including bank- and macroeconomic control variables. These results support the same conclusion, namely that low interest rates decrease bank risk-taking.
The relationship between interest rates and bank risk-taking is influenced significantly by the financial crisis. For the period pre-crisis an insignificant negative relationship is found, while for the period post-crisis a significant positive relationship is found. This significant influence of the crisis on the relationship between interest rates and bank risk-taking is supported by including a dummy variable into the model that separates the pre-crisis period (2003-2007) from the post-crisis period (2008-2016).
The remainder of the paper is organized as follows: chapter 2 gives an overview of the literature and chapter 3 presents and explains the methodology. I explain how data were obtained in chapter 4, while chapter 5 provides the results. Finally, chapter 6 presents the conclusion and answers the research question.
2. Literature
review
In this chapter the existing literature about the relationship between low interest rates and bank risk-taking is discussed. Subsequently, several ways in which interest rates can influence bank risk-taking are detailed, followed by an outline of the existing empirical literature. The final part considers the different proxies to measure bank risk-taking in the literature.
2.1 Risk-taking channels by banks
The relationship between low interest rates and the risk-taking behavior of banks is a prevalent subject of research in the literature, and there are several ways in which banks are exposed to risk. Monetary policy has an influence on the risk-taking activities of banks, the so-called risk-taking channel (Borio & Zhu, 2008), which is supported by three factors.
6
perception and increase risk tolerance. The common assumption that risk tolerance increases with wealth points to an influence on risk-taking (Borio & Zhu, 2008).
Second, communication problems regarding the transparency of the decision of the monetary authority may also influence risk-taking activities of banks (Andries et al. 2015). High transparency by the central bank can reduce uncertainty about future expectations and compress the risk premia. Moreover, interest rates set by the central bank have an asymmetric impact on investing behavior. Lower interest rates encourage investing, while higher interest rates curtail investing (Borio & Zhu, 2008). Furthermore, an expansionary monetary policy decreases the required risk premium on equity, which makes investors and financial institutions more risk tolerant.
Third, the “search for yield” is discussed by Rajan (2006), who argues that in times of low interest rates the yields on loans are low. Financial institutions need to match yields of their long-term assets and short-term liabilities. This causes decreased earnings from assets and the mismatch with liabilities (Dell’Ariccia et al. 2014). In addition, investors are willing to take on more risk to meet promised future liabilities, especially when there are nominal return targets, such as those of certain pension funds or life insurers. This search for yield is often realized by securitization of loans, which is a process of creating a financial instrument by combining loans and selling it on the market. It results in more liquidity; however, it has the negative effect that it increases the risk of the banking sector (Altunbas et al., 2010).
2.2 Empirical evidence
A recent paper by Dell’Ariccia, Leaven, and Suarez (2017) presents evidence of the risk-taking channel of monetary policy of US banks. The authors examined the internal rating of loans over the period 1997 to 2011, concluding a negative relationship between the quality of new loans and low short-term interest rates. This finding is supported by Andries et al. (2015); after analyzing 571 commercial banks from the Eurozone for the period 1999 to 2011, they determined a negative relation between interest rates and bank risk-taking. Moreover, research conducted by Amador, Gómez-Gonzalez, and Pabón (2013) employs a similar method to that of Andries et al. (2015). They used a rich dataset from Colombian financial institutions between 1990 and 2011 and their results show that abnormal loan growth is positively and significantly associated with non-performing loans.
Dell’Ariccia et al. (2014) conducted research into the risk behavior of banks in a low interest rate environment. They obtained quarterly data from US banks for the second quarter of 1997 to the third quarter of 2009, and concluded that the relationship between a low interest rate environment and risk-taking behavior by banks depended on their capital structure. Well-capitalized banks will take on more risk and soften their lending standards if interest rates fall, but low capitalized banks will lower their risk and increase monitoring activities.
7
overnight rate and the granting of loans. When overnight rates were lower, virtually all banks granted more loans to less creditworthy firms.
Altunbas et al. (2010) used an extensive dataset of listed banks from the European Union and the US for the period 1998 to 2008. Contrary to most papers, they used quarterly data instead of annual data. The authors found that banks increased their risk-taking behavior in periods of low interest rates.
Two key papers in the field of interest rates and bank risk-taking behavior are those by Maddaloni and Peydró (2011), and by Delis and Kouretas (2011). Maddaloni and Peydró (2011) used a dataset from US and European banks, from 1991 and 2002, respectively, until the crisis in 2008, and conclude that low short-term interest rates lead to lower lending standards for household and corporate loans. In other words, banks are willing to grant loans and mortgages to less creditworthy borrowers. On the other hand, the authors do not find a significant result for low long-term interest rates. Delis and Kouretas (2011) support this result, and find a significant negative relation between interest rates and the risk-taking behavior of banks. In the search for yield, banks will soften their lending standards as a consequence of the threat of lower profit margins.
Overall, the empirical literature is not unanimous, but most studies show a negative relationship between interest rates and bank risk-taking behavior. This thesis analyzes this relationship for the period 2003 to 2016. The following hypothesis is formulated:
𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠 1: 𝐴 𝑙𝑜𝑤 𝑖𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑟𝑎𝑡𝑒 𝑒𝑛𝑣𝑖𝑟𝑜𝑛𝑚𝑒𝑛𝑡 𝑙𝑒𝑎𝑑𝑠 𝑡𝑜 ℎ𝑖𝑔ℎ𝑒𝑟 𝑏𝑎𝑛𝑘 𝑟𝑖𝑠𝑘 − 𝑡𝑎𝑘𝑖𝑛𝑔. This hypothesis is tested for the pre-crisis period (2003 to 2007), the post-crisis period (2008 to 2016), and the full period (2003 to 2016).
Maddaloni and Peydró (2011) state that interest rates that were too low for too long were a key determinant resulting in the financial crisis. They identified the low quality of granted loans as an important source for the recent financial crisis. Additionally, countries with lower lending standards before the crisis performed worse after the crisis. Dell’Ariccia et al. (2014) claim that the financial crisis would have been less serious if the central banks had raised interest rates sooner.
Andries et al. (2015) researched the relationship between low interest rates and bank risk-taking behavior, in addition to the influence of the financial crisis. A novelty of their paper is the significant influence the crisis had on the relationship between interest rates and bank risk-taking. However, the results are ambiguous and dependent on the risk proxies they used.
8
This thesis aims to contribute to the knowledge about the influence the financial crisis had on bank risk-taking and the low interest rate environment. The second hypothesis can be formulated as follows:
𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠 2: 𝑇ℎ𝑒 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛𝑠ℎ𝑖𝑝 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑙𝑜𝑤 𝑖𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑟𝑎𝑡𝑒𝑠 𝑎𝑛𝑑 𝑏𝑎𝑛𝑘 𝑟𝑖𝑠𝑘 − 𝑡𝑎𝑘𝑖𝑛𝑔 𝑖𝑠 𝑙𝑒𝑠𝑠 𝑠𝑒𝑣𝑒𝑟𝑒 𝑝𝑜𝑠𝑡 − 𝑐𝑟𝑖𝑠𝑖𝑠 𝑐𝑜𝑚𝑝𝑎𝑟𝑒𝑑 𝑡𝑜 𝑝𝑟𝑒 − 𝑐𝑟𝑖𝑠𝑖𝑠.
2.3 Proxies for risk
The literature on proxies for measuring the risk-taking behavior of banks varies. I discuss some proxies and argue which proxies to use, and draw a distinction between risk-taking behavior and risk exposure. Risk exposure measures the risk banks face because of macroeconomic changes, which they face independent of their behavior. Many studies use proxies that are more concerned with risk-exposure, for example non-performing loans. These proxies measure ex-post loan performance affected by exposure to poor market conditions outside the bank’s control. I want to measure the increased risk appetite of banks at the time they make the decision to take additional risk.
Most papers follow the example of Delis and Kouretas (2011) and use two proxies to measure risk-taking behavior, namely the ratio of risk assets to total assets and the ratio of non-performing loans to total loans. Risk assets reflect a banks’ behavior towards risk and its exposure implied by the riskiness of the bank’s portfolio. It is a broad measure of risk, because it does not allow for the clear separation between risk-taking and risk exposure. Nevertheless, the ratio of risk assets to total assets give a reflection of the bank’s behavior towards risk and is used in this thesis as the dependent variable to measure bank risk-taking.
The non-performing loans (NPL) ratio reflects the quality of banks’ assets. It measures credit risk, because non-performing loans are loans that are in default or close to being in default. This measure is also used by Jiménez, Lopez, and Saurina (2013), who conclude thatthe NPL ratio is an ex-post measure of credit risk. Credit risk refers to the risk that a borrower may not repay his or her loan and default. This risk is closely associated with risk exposure, as it is influenced by macroeconomic factors and is independent of the behavior of banks. Furthermore, it shows the deterioration of the loan quality, where a higher value for this ratio is associated with a higher credit risk.
Three proxies to measure bank risk-taking are used by Andries et al. (2015), namely non-performing loans, loans loss provision, and the Z-score. The authors use the loan loss provision to try to measure bank risk-taking, which is an allowance for uncollectable loans and loan payments, and call it an innovation of their study. Loan loss provision is a kind of cushion that protects banks against unexpected losses. When the loan loss provision lowers in times of low interest rates, the bank has a higher level of risk. Loan loss provision measures post-loan performance, because it reflects the risk that a loan may default. This thesis does not include this proxy, because loan loss provision measures risk exposure.
9
losses exceeds equity (Laeven & Levine, 2009). This measure covers risk exposure rather than risk-taking, as the values are more driven by external market movements. Insolvency arises through unfavorable macroeconomics factors instead of an increased risk appetite on the part of the banks. This proxy is not included in the empirical analysis of this thesis.
Altunbas et al. (2010) make use of the expected default frequency and loan loss provision as proxies to measure risk-taking behavior by banks. The expected default probability is the probability that a bank will default within a given time period. This measure is commonly used as a measure of credit risk by financial institutions, including central banks and regulators. It measures the ex-post loan performance as the NPL does, and therefore reflects risk exposure instead of risk-taking behavior.
A study by Maddaloni and Peydró (2011) used the lending standard from a unique dataset of answers from bank lending surveys. This is also suggested by Foos, Norden, and Weber (2010), who used abnormal loan growth as a proxy to measure bank risk-taking. They used this measure to test three hypotheses: the positive relation between loan growth and loan loss provisions, the negative relation between loan growth and profitability, and the negative relation between loan growth and the ratio equity to total assets. For all three hypotheses they found a significant result. I agree with the conclusion by Foos et al. (2010) that loan growth rate is an important factor in determining bank risk-taking. This is also in line with the paper by Amador et al. (2013) that exploits the intertemporal relation between loan growth and banks’ risk-taking behavior. This measure reports the exact moment when the risk is acquired and not the ex-post loan performance. This measure is also used in a recent paper by Dell’Ariccia et al. (2017), who determine loan growth as the quality of new granted loans. They want to focus on the risk attitude of banks at the time the loan is issued, instead of the ex-post loan performance that could be affected by external events. In this way, they focus on the attitude of banks towards risk that is fully under control, rather than on the overall riskiness of a bank’s portfolio. Abnormal loan growth is included in the empirical analysis of this thesis, because it ensures a separation between risk exposure and risk-taking.
In conclusion, empirical evidence is not consistent about the proxies to measure risk-taking. Most studies use proxies related to risk exposure to the market, rather than the actual risk appetite of banks. This thesis tries to solve this measurement error, and uses two proxies as dependent variables to measure bank risk-taking, namely the ratio of risk assets to total assets and abnormal loan growth. Abnormal loan growth reflects credit risk, which usually covers the primary source of risk for banks. Risk assets ratio is a broader measure of risk-taking, and measures risk outside the loan portfolio as well. These proxies reflect bank risk-taking rather than risk exposure. The following chapter describes the research method used to answer the two research questions.
3. Methodology
10 This thesis discusses the following hypotheses:
𝐻1: A low interest rate environment leads to higher bank risk-taking.
𝐻2: The negative relationship between low interest rates and bank risk-taking is less severe post-crisis compared to pre-crisis.
The general empirical model used for analysis has the following form:
𝑟𝑖,𝑡 = 𝑐𝑖,𝑡 + 𝛽1∗ 𝑖𝑟𝑖,𝑡+ 𝛿 ∗ 𝑖𝑟𝑖,𝑡 ∗ 𝐶𝑅𝐼𝑆𝑡+ 𝛽2∗ 𝑏𝑐𝑖,𝑡+ 𝛽3 ∗ 𝑚𝑐𝑖,𝑡 + 𝑢𝑖,𝑡 (1) The bank risk-taking 𝑟 of bank 𝑖 at time 𝑡, proxied by the risk assets ratio and abnormal loan growth, is a function of 𝑖𝑟, the interest rate, proxied by four alternative measures: the three-month Libor rate, the federal funds rate, the Taylor-rule rate, and the long-term interest rate. The 𝐶𝑅𝐼𝑆 term is a dummy variable which takes the value 1 for the years after 2007 and the value 0 before 2007, 𝛿 measures the impact of the financial crisis on bank risk-taking, 𝑏𝑐𝑡 are a number of bank
characteristics controls, 𝑚𝑐𝑡 are macroeconomic controls, and 𝑢𝑖,𝑡 is the error term.
The redundant fixed effects test is performed to test whether a pooled OLS is sufficient to analyze the relationship between interest rates and bank risk-taking. Consequently, the redundant fixed effects test with the null hypothesis: the bank fixed effects are all equal to zero, is rejected on a 1% significant level. This result is in favor of the use of a panel data model. I perform a Hausman test to decide whether I have to use a fixed effects model or a random effects model. The null hypothesis of the Hausman test states that the random effects model is the preferred model to use, and cannot be rejected for the tests of all independent variables. There are no fixed effects present, so the random effects model is preferred. This result conflicts with the theoretical background for the use of a random effects model or a fixed effects model. A random effects model has a more stringent assumption, because it is used when the variation across entities is assumed to be random and uncorrelated with the predictor or independent variables included in the model. In the fixed effects model, on the other hand, time-invariant variables are absorbed by the intercept. As I want to perform a test where the intercept differs across sections but not over time, the fixed effects model is preferred.
For a first impression of the relationship between bank risk-taking and interest rates, a fixed effects regression is performed, ignoring all potential estimation issues. The fixed effects regression is performed eight times, that is, two risk-taking measures are performed with four interest rate variables each. The results are summarized in Table 3 in chapter 5. The fixed effects regression has the following form:
𝑟𝑖,𝑡 = 𝑐𝑖,𝑡 + 𝛽1∗ 𝑖𝑟𝑖,𝑡+ 𝑢𝑖,𝑡 (2)
The null hypothesis and alternative hypothesis of the first hypothesis are formulated as follows:
11
The first hypothesis is also tested, taking into account the bank characteristics and macroeconomic characteristics. Equation (1) is restricted by 𝛿 =0 and has the following form:
𝑟𝑖,𝑡 = 𝑐𝑖,𝑡 + 𝛽1∗ 𝑖𝑟𝑖,𝑡+ 𝛽2∗ 𝑏𝑐𝑖,𝑡 + 𝛽3∗ 𝑚𝑐𝑖,𝑡+ 𝑢𝑖,𝑡 (3)
The same hypothesis as with equation (2) is tested: 𝐻0: 𝛽1 = 0, 𝐻𝑎: 𝛽1 ≠ 0.
Equation (3) is performed for three different subsamples. I compare the pre-crisis period with the post-crisis period. The period before the financial crisis consists of the subsample 2003 to 2007, and the period post-crisis consists of the subsample 2008 to 2016. The results presented in Table 6 in chapter 5 are used to gain a first impression of the differences between the pre-crisis and post-crisis periods. In addition, the results of the full sample period are listed for comparison. The three subsamples are only tested with the Taylor-rule rate, because the other interest rate variables are static over time for the post-crisis period, which can be seen in Appendix C, and the use of the Taylor-rule rate is in line with most studies, for example, the study conducted by Andries et al. (2015).
The second hypothesis is tested with the use of a slope dummy variable. The 𝐶𝑅𝐼𝑆 dummy has a value of 0 in the period 2003 to 2007, and a value of 1 in the period 2008 to 2016. The model has the following form:
𝑟𝑖,𝑡 = 𝑐𝑖,𝑡 + 𝛿 ∗ 𝑖𝑟𝑖,𝑡 ∗ 𝐶𝑅𝐼𝑆𝑡+ 𝛽1∗ 𝑖𝑟𝑖,𝑡+ 𝛽2∗ 𝑏𝑐𝑖,𝑡+ 𝛽3∗ 𝑚𝑐𝑖,𝑡+ 𝑢𝑖,𝑡 (4)
To test the second hypothesis with the model in equation (4), I specify the null hypothesis and alternative hypothesis in the following notation:
𝐻0: 𝛿 = 0, 𝐻𝑎: 𝛿 ≠ 0
The next chapter describes which data were needed and how the data were obtained to perform the empirical analysis.
4. Data
This chapter presents the explanation of the dataset used for the empirical analysis. Data were obtained from Bankscope, a database reporting balance sheet items of more than 30,000 financial institutions worldwide provided by Bureau van Dijk. I make use of panel data of commercial banks, saving banks, and cooperative banks. Investment banks are excluded as they do not take deposits, in line with the paper by Delis and Kouretas (2011), and American banks are used because more accurate data are available compared to European banks. Another reason for using American banks is that the financial crisis emerged in the US. The starting date of 2003 is used because of the availability of data. I used a dataset covering the period 2003 to 2016.
12
actual size of the banking market. Only banks with a minimum of one billion USD in total assets were included, because these banks are economically relevant and representative of the banking sector.
The use of annual data is in line with Delis and Kouretas (2011), who conducted their study using monthly and annual data, and concluded that annual data were sufficient to analyze the relationship between interest rates and bank risk-taking. In addition to the dependent and independent variables, some control variables are used, namely four bank characteristics and two macroeconomic characteristics. Only banks with values for all variables in the full sample period were selected, in order to obtain a balanced dataset. The final dataset consisted of 142 banks and 13,916 bank year observations. Table 1 gives an overview of the statistics included in this thesis, after winsorizing at 90%, which is used to tackle the problem of outliers and non-normality. Some variables have extreme outliers driven by yearly exceptions, for example liquidity (maximum 81.42) has a significant dispersion around the mean (6.61). Appendix A summarizes the descriptive statistics before winsorizing is applied. I used the same method as Andries et al. (2015), and applied the outlier labeling rule (Hoaghin & Iglewicz, 1987), because trimming would have led to a loss of information.
4.1 Dependent variables
As mentioned in section 2.3, two dependent variables are selected to measure bank risk-taking. The first is the ratio risk assets to total assets, where risk assets are defined as all assets minus cash, government securities (at market values), and cash balances due from other banks (Delis & Kouretas, 2010).
The second is the abnormal loan growth, which is defined as the difference between an individual bank’s loan growth and the median loan growth of all banks in that year (Amador et al., 2013). 4.2 Independent variables
The most important independent variable to use is the short-term interest rate, because banks fund their operations mainly with short-term liabilities. Maddaloni and Peydró (2011) conclude that low short-term interest rates soften lending standards, in contrast to low long-term interest rates where they did not find significant results. I have adopted the method of a recent study by Dell’Ariccia (2017), and used the federal funds rate as short-term interest rates1. The federal funds rate is the
rate at which banks lend reserve balances to other banks on an overnight basis.
In line with Delis and Kouretas (2011), I will also use another short-term interest rate to test for robustness. The three-month Libor is an alternative short-term interest rate. It is the interbank rate, so, the rate that some of the world’s leading banks charge each other on the London money market for short-term loans.
The Taylor-rule residual initiated by Taylor (1993) is used in many studies, including Dell’Ariccia et al. (2017); Andries et al. (2015); Altunbas et al. (2010); Dell’Ariccia et al. (2014); and Maddaloni and Peydró (2011). The Taylor-rule is used as guidance for central banks to set interest rates in response to economic changes. The use of the Taylor-rule rate solves the problem of pro- or
13
countercyclical movements from economic circumstances, because the interest rate is treated as an exogenous variable. The general formula of the Taylor-rule is as follows:
𝑖𝑡 = 𝑟 + 𝜋𝑡+ 𝛽𝜋 (𝜋𝑡− 𝜋∗) + 𝛽
𝑦 (𝑦𝑡− 𝑦𝑡∗) (5)
In this formula, 𝑖𝑡 is the nominal federal fund rate, 𝑟 is the real federal funds rate, 𝜋𝑡 is the rate of
inflation, 𝜋∗ is the target level of inflation and is set at two percent, 2 𝑦
𝑡 is the real GDP growth,
and 𝑦𝑡∗ represents the target level of GDP. The variables 𝛽
𝜋 and 𝛽𝑦 are equal, and have a value of
0.5 according to the conventional Taylor-rule estimation. These values represents the weight that is given to, the inflation gap and the output gap, respectively. The same value indicates an equal importance of the realization of inflation and economic targets. The target level of GDP growth has unobservable values, therefore, in line with Elias et al. (2014), Okun’s law is used to convert GDP growth to an unemployment rate. Okun’s law is a popular rule of thumb that relates changes in the unemployment rate to GDP growth at an approximate two-to-one ratio. The last part of equation (5) can be rewritten in the following way:
𝛽𝑦 (𝑦𝑡− 𝑦𝑡∗) = −2 ∗ 𝛽𝑦(𝑢𝑡− 𝑢𝑡∗) (6)
where 𝑢𝑡 is the unemployment rate and 𝑢𝑡∗ the non-accelerating inflation rate of unemployment
(NAIRU). The NAIRU is regularly estimated by the Congressional Budget Office (CBO).3
The long-term interest rate is the last independent variable, consisting of the market yield on US Treasury securities at a 10-year constant maturity. The use of the long-term interest rate is in line with studies by Delis and Kouretas (2011), and by Dell’Ariccia et al. (2017).
Appendix C shows the development of the four interest rate variables during the time period analyzed in this thesis (2003 to 2016).
4.3 Control variables
In my thesis I control for some bank characteristics used in previous studies that may influence risk-taking behavior. Firstly, bank size is measured as the natural logarithm of total assets (Altunbas et al., 2010; Delis & Kouretas, 2011; Ioannidou et al. 2014; Dell’Ariccia et al., 2017). Bigger banks have the ability to diversify risks, and are consequently assumed to take on more risk. Bhagat et al. (2015) indeed find that bank size is positively correlated with risk-taking measures, mainly because bigger banks increase leverage. A positive relation between bank size and bank risk-taking is therefore expected. Moreover, bank profitability is measured by the ratio profits before taxes to total assets. Banks with higher profitability are assumed to take on more risk, as a higher level of risky assets may lead to higher profits, and higher profits in a previous period can be used to extend more loans (Andries et al., 2015; Altunbas et al., 2010; Dell’Ariccia et al., 2017). Profitability is included in the model as a lagged variable to reflect implications of management decisions towards risk of a previous year. In addition, Delis and Kouretas (2010) identify efficiency as a potentially important variable, which is measured as a ratio of total revenues to total expenses (Andries et al., 2015; Delis & Kouretas, 2011). Due to technological changes it may influence risk-taking. More efficient banks are more capable of managing risk, and additionally higher risk may also imply
14
more efficiency. A positive relation between efficiency and bank risk-taking is expected. Furthermore, liquidity is the ratio of liquid assets to total assets (Altunbas et al., 2010; Dell’Ariccia et al., 2017). In a period of low interest rates, funding is easy to access and banks are willing to take on more risk, because they underprice the downside risk. This result is more evident for less liquid banks. Dell’Ariccia et al. (2017) find that more liquid banks lower risk-taking when interest rates are low. I expect a negative relation between liquidity and bank risk-taking, because more liquid banks are less affected by interest rate changes.
I control for these four bank characteristics, because too many control variables could lead to overfitting of the model. Next to the four bank characteristics, two macroeconomic characteristics are included in the empirical analysis. These variables control for risk not arising from bank risk-taking but rather from overall macroeconomic conditions.
Inflation is measured as the consumer price index4 and can affect bank risk-taking through
margins and total costs; a positive relation with bank risk-taking is expected. Additionally, the economic growth is measured as annual GDP growth. A positive relation between bank risk-taking and economic growth is expected, since more favorable economic conditions are associated with a growth in loans in search for higher yield (Delis & Kouretas, 2010).
The relationship between interest rates and bank risk-taking is the main topic this thesis addresses. Appendix B provides eight regressions, that is, the two bank risk-taking measures regressed with four interest rate variables each. The regressions give a first impression regarding the relationship between interest rates and bank risk-taking.
15
Table 1 gives a summary of the statistics for all the variables used in the empirical analysis. Table 2 comprises the correlation coefficients for respective variables. Appendix A summarizes the descriptive statistics before winsorizing is applied.
Table 1 Descriptive Statistics (after winsorizing at 90%)
Variable Observations Mean Median
Standard
Deviation Minimum Maximum
Risk Assets 1988 0.81 0.82 0.09 0.60 0.94 Loan Growth 1988 0.09 0.07 0.09 -0.05 0.30 Size 1988 5.91 5.94 0.18 5.50 6.17 Profitability 1988 0.01 0.01 0.00 0.01 0.02 Efficiency 1988 1.60 1.57 0.23 1.25 2.10 Liquidity 1988 6.06 4.72 4.24 1.50 17.49 Inflation 1988 0.02 0.02 0.01 0.00 0.04 Economic Growth 1988 0.04 0.04 0.02 -0.02 0.06
Three-month Libor rate 1988 1.66 0.72 1.77 0.23 5.30
Federal Funds Rate 1988 1.23 0.24 1.64 0.03 4.85
Taylor-rule Rate 1988 0.25 0.13 2.97 -3.76 5.75
Long-term Interest Rate 1988 3.26 3.24 1.00 1.80 4.80
This table reports the descriptive statistics used in the empirical analysis. Two independent variables to measure bank risk-taking are used: Risk assets is the ratio risk assets to total assets, and loan growth is the difference between an individual bank’s loan growth and the median loan growth of all banks in that period. Five bank control variables are used: capitalization is the ratio equity to total assets, size is the natural logarithm of total assets, profitability is the ratio of profits before tax to total assets, efficiency is the ratio of total revenues to total expenses, and liquidity is the ratio of liquid assets to total assets. Two macroeconomic control variables are used: inflation is measured as the change in the consumer price index (CPI), and economic growth is measured as the change in GDP. Four independent variables are used: the three-month Libor rate and the federal funds rate represent the two alternative short-term interest rates, the Taylor-rule rate represents the gap between the federal funds rate and an implied policy rate, and the long-term interest rate is measured as the US Treasury yield on securities at a constant 10-year maturity.
Table 2 Correlation Matrix
RA LG SIZE
LAG_
PROF EFF LIQ INF ECOG LIBOR FED TAYLOR LONG- TERM Risk Assets 1.000 Loan Growth 0.134 1.000 Size 0.015 -0.140 1.000 Profitability (-1) -0.026 0.022 -0.331 1.000 Efficiency 0.103 0.036 -0.090 0.075 1.000 Liquidity -0.463 -0.064 -0.027 0.001 -0.114 1.000 Inflation -0.021 0.112 -0.388 0.055 0.077 -0.035 1.000 Economic Growth -0.068 0.199 -0.259 0.131 0.048 -0.036 0.580 1.000 Libor Rate 0.027 0.155 -0.400 -0.065 0.052 -0.106 0.613 0.340 1.000 Federal Funds Rate 0.007 0.160 -0.414 -0.031 0.057 -0.097 0.586 0.445 0.982 1.000
16
5. Results
This section summarizes the results of the empirical analysis. The results of a fixed effects regression to estimate the relationship between four different interest rate variables and two bank risk-taking measures are given in Table 3. The entity fixed effects model is also performed, including bank- and macroeconomic control variables, the results of which are listed in Table 4. Table 5 summarizes the results of the relationship between interest rates and bank risk-taking for different subsamples, where the interest rate is measured with the Taylor-rule rate. The fixed effects model is performed for three periods: the pre-crisis period (2003 to 2007), the post-crisis period (2008 to 2016), and the whole sample period (2003 to 2016). The last table presents the influence of the financial crisis on the relationship between interest rates and bank risk-taking. The results of a fixed effects model in interaction with a slope dummy for the crisis are given in Table 6.
5.1 Fixed effects regression
Table 3 reports the results of the fixed effects regression where all potential estimation issues are ignored. The results are obtained from the estimation with equation (2), and give a first impression about the relationship between the two risk-taking proxies and the independent interest rates. The regression is performed with the dependent variables multiplied by ten to influence the scaling, because otherwise most estimated coefficients are around 0.
Table 3 Interest Rates and Bank Risk-taking: Fixed Effects Regression
Risk Assets Loan Growth
Variable I. II. III. IV. V. VI. VII. VIII.
Constant
8.016* 8.042 8.061* 7.997* 0.717* 0.741* 0.831* 0.530*
(0.002) (0.001) (0.001) (0.004) (0.003) (0.002) (0.002) (0.006)
Three-month Libor Rate 0.030* 0.081*
(0.001) (0.001)
Federal Funds Rate 0.020* 0.089*
(0.001) (0.001) Taylor-rule Rate 0.018* 0.081* (0.000) (0.001) Long-term Rate 0.021*** 0.098* (0.001) (0.002) Cross-section 142 142 142 142 142 142 142 142 Observations 1988 1988 1988 1988 1988 1988 1988 1988 R-squared 0.726 0.724 0.726 0.724 0.187 0.188 0.234 0.173 Durbin Watson 0.670 0.672 0.668 0.665 1.554 1.571 1.633 1.536 F-statistic 34.474* 34.119* 34.499* 34.005* 2.990* 3.011* 3.980* 2.721*
This table reports the coefficients of the pooled OLS regression and the standard errors in parentheses. Columns I-IV report the results with risk assets ratio as the dependent variable and Column V-VIII with abnormal loan growth as the dependent variable. For both dependent variables there are four regressions, each with their own dependent variable. The dependent variables used are the Libor rate, the federal funds rate, the Taylor-rule rate, and the long-term rate. The R-squared indicates the proportion of the variance in the risk measures that is predicted from the interest rate. The Durbin-Watson statistic detects the presence of autocorrelation. The F-statistic measures the quality of the fit of the model. *, **, and *** indicate significance at 1%, 5%, and 10%, respectively.
17
rates and bank risk-taking. All four interest rates show a significant positive relation with the two bank risk-taking measures. The relationship between risk assets and the long-term interest rate is significant at a 10% level; furthermore, all other results are significant at a 1% significance level. The result that risk-taking is less affected by the long-term interest rate compared to short-term interest rates confirms expectations. The low r-squared for the regressions with loan growth as dependent variable indicate that only a modest proportion of the variance in the loan growth is explained by the interest rates. In short, the results of Table 3 reject the first null hypothesis: 𝐻0: 𝛽1 = 0. The alternative hypothesis, nevertheless, is accepted: 𝐻𝑎: 𝛽1 ≠ 0. I expected the
alternative hypothesis to be negative, 𝐻𝑎: 𝛽1 < 0, but the regression estimates a positive coefficient, 𝐻𝑎: 𝛽1 > 0.
The redundant fixed effects test is performed for formula (3) with the null hypothesis, 𝛽2 = 𝛽3 =
0. The F-statistic can be rejected at a 1% significant level. This means that a panel approach is necessary. First I regressed a panel model with random effects, and performed a Hausman test. The Hausman cannot be rejected, which means that the random effects model is preferred. However, I wanted to perform a test where the intercept differed across sections but not over time, as I explained in chapter 3, and the fixed effects model is such a model. Results of the fixed effects regression model are listed in Table 4. The regression is performed with the dependent variables multiplied by ten to influence the scaling, so that not all estimated coefficients are around 0.
5.2 Fixed effects regression including control variables
The results for both risk measures show a significant positive relation with the interest rates, implying that as interest rates rise, banks increase their risk-taking. This finding is in contrast to expectations and earlier studies. The results reject the first null hypothesis: 𝐻0: 𝛽1 = 0; the alternative hypothesis is accepted: 𝐻𝑎: 𝛽1 ≠ 0. Even though, the expected hypothesis stated that low interest rates lead to more bank risk-taking (𝐻𝑎: 𝛽1 < 0), the results show the opposite, namely that low interest rates lead to less bank risk-taking ( 𝐻𝑎: 𝛽1 > 0). These results confirm the results in Table 3, which means that the coefficients for the interest rates retain their respective impacts on bank risk-taking after including bank characteristics and macroeconomic characteristics.
18
Table 4 Interest Rates and Bank Risk-taking: Fixed Effects Model
Risk Assets Loan Growth
Variable I. II. III. IV. V. VI. VII. VIII.
Three-month Libor Rate 0.027* 0.093*
(0.008) (0.014)
Federal Funds Rate 0.023* 0.084*
(0.008) (0.015) Taylor-rule Rate 0.021* 0.083* (0.004) (0.008) Long-term Rate 0.034** 0.136* (0.016) (0.030) Size 0.336* 0.319* 0.321* 0.359* 0.708* 0.673* 0.704* 0.863* (0.072) (0.073) (0.069) (0.087) (0.135) (0.136) (0.127) (0.164) Profitability (-1) 0.031 0.023 0.018 0.005 0.146** 0.126** 0.114** 0.061 (0.031) (0.031) (0.030) (0.030) (0.058) (0.058) (0.055) (0.057) Efficiency 0.088 0.088 0.106 0.076 0.214 0.211 0.275** 0.157 (0.076) (0.076) (0.075) (0.077) (0.141) (0.142) (0.138) (0.144) Liquidity -0.090* -0.091* -0.088* -0.091* -0.035* -0.037* -0.027* -0.037* (0.004) (0.004) (0.004) (0.004) (0.007) (0.007) (0.007) (0.007) Inflation 2.799** 3.722* 3.431* 3.981* -4.984** -2.173 -3.649*** -1.508 (1.252) (1.188) (1.156) (1.189) (2.336) (2.221) (2.127) (2.227) Economic Growth -4.404* -4.742* -5.484* -4.446* 8.149* 6.921* 3.882* 7.998* (0.580) (0.591) (0.618) (0.582) (1.084) (1.106) (1.137) (1.090) Hausman Test 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Observations 1988 1988 1988 1988 1988 1988 1988 1988 R-squared 0.805 0.805 0.806 0.804 0.239 0.234 0.265 0.230 F-statistic 51.404* 51.228* 51.767* 51.109* 3.899* 3.793* 4.480* 3.709*
These are the results of the empirical analysis with fixed effects. The values represent the coefficients and the standard errors are given in parentheses. Bank risk-taking is proxied by two measures: risk assets (I-IV) and loan growth (V-VIII). As control variables, I include bank characteristics and macroeconomic controls. There are four independent variables used: the Libor rate (I &V), federal funds rate (II &VI), Taylor rate (III &VII), and the long-term rate (IV& VIII). Eight tests are performed using a fixed effects model. The Hausman test examines whether to use a fixed effects model or a random effects model. The R-squared shows the quality of the fit of the model. The F-statistic reports the results of the redundant fixed effects test. *, **, and *** indicate significance at 1%, 5%, and 10%, respectively.
The two macroeconomic control variables give conflicting results between the risk measures of loan growth and risk assets. Inflation has a positive relation at a 1% significant level when risk assets are used as dependent variable. Higher inflation is associated with an increase in the riskiness of a banks’ portfolio and a higher default risk. On the other hand, the negative coefficient of inflation when loan growth is used as dependent variable is contrary to expectations. The last control variable, economic growth, is expected to have a positive coefficient with bank risk-taking. More favorable economic conditions are associated with a growth in loans in the search for higher yield. The positive coefficient with loan growth at a 1% significant level confirms this theory. The results with risk assets as dependent variables contradict to expectations. For regressions V-VIII, only a modest proportion of the variance in the loan growth is explained by the interest rates, which implies a low quality of the fit of the model.
5.3 Fixed effects regression on the three subsamples
19
relationship. The pre-crisis period, post-crisis period, and full period are compared. The regressions are only performed with the Taylor-rule rate as dependent variables.
Table 5 Bank Risk-taking and the Taylor-rule Rate: Three Different Periods
Independent Variable: Taylor-rule rate
Dependent Variable: Risk Assets Dependent Variable: Loan Growth Full Period Pre-crisis Post-crisis Full period Pre-crisis Post-crisis
Taylor-rule Rate 0.021* -0.013 0.045* 0.083* -0.003 0.138* (0.004) (0.025) (0.007) (0.008) (0.062) (0.015) Size 0.321* -0.707* 0.142 0.704* 2.205* 1.159* (0.069) (0.257) (0.093) (0.127) (0.631) (0.206) Profitability(-1) 0.018 -0.175* -0.840 0.114** 0.064 14.999** (0.030) (0.052) (2.851) (0.055) (0.127) (6.280) Efficiency 0.106 0.086 0.125 0.275** 0.307 -0.197 (0.075) (0.123) (0.091) (0.138) (0.303) (0.199) Liquidity -0.088* -0.095* -0.088* -0.027* -0.040** -0.021** (0.004) (0.006) (0.004) (0.007) (0.016) (0.008) Inflation 3.431* 8.478 2.716* -3.649*** -25.425 -0.802 (1.156) (11.861) (1.004) (2.127) (29.185) (2.211) Economic Growth -5.484* -15.835* -4.769* 3.882* 26.430** -0.418 (0.618) (4.712) (0.621) (1.137) (11.594) (1.368) Hausman Test 0.000 0.000 0.000 0.000 0.000 0.000
Entity FE yes yes yes yes yes yes
Observations 1988 710 1278 1988 710 1278
R-squared 0.806 0.921 0.890 0.280 0.455 0.380
This table shows the coefficients and standard errors in parentheses of the entity fixed effects panel data model. Two independent variables, risk assets and loan growth, are analyzed with the Taylor-rule rate as independent variable. Five bank control variables are used: capitalization, size, lagged profitability, efficiency and liquidity. Two macroeconomic control variables are used: Inflation and economic growth. The R-squared shows the quality of the fit of the model. *, **, and *** indicate significance at 1%, 5%, and 10%, respectively.
20
5.4 Fixed effects regression and the influence of the financial crisis
Table 6 Interest Rates and Bank Risk-taking and the Influence of the Crisis: A Fixed Effects Panel Model
Independent Variable: Taylor-rule Rate Dependent variable: Risk Assets Dependent variable: Loan Growth Taylor-rule Rate*CRIS 0.041* 0.163* (0.011) (0.020) Taylor-rule 0.003 0.013 (0.006) (0.012) Size 0.229* 0.337** (0.073) (0.133) Profitability (-1) -0.027 -0.065 (0.032) (0.059) Efficiency 0.148*** 0.445* (0.076) (0.138) Liquidity -0.087* -0.020* (0.004) (0.007) Inflation 2.994* -5.394** (1.158) (2.101) Economic Growth -5.333* 4.488* (0.617) (1.119) Hausman 0.000 0.000 Observations 1988 1988 R-squared 0.808 0.291
This table shows the relation between bank risk-taking proxied by risk assets and loan growth and interest rate proxied by the Taylor-rule rate. The values represent the coefficients and the standard errors are in parentheses. The coefficient on the dummy CRIS captures the effect of the crisis, and has a value of 1 for the period 2008 to 2016 and of 0 otherwise. The R-squared measures the percentage of the variance of the dependent variable that is explained by the model. *, **, and *** indicate significance at 1%, 5%, and 10%, respectively.
The effect of the crisis on the relationship between interest rates and bank risk-taking is significant for both risk proxies, because the results indicate a substantial positive influence of the financial crisis. The results are in line with my expectations, as I expected the relationship to be less negative after the financial crisis. Summarizing the results of Table 6 leads to a rejection of the second null hypothesis, 𝐻0: 𝛿 =0, and acceptance of the alternative hypothesis, 𝐻1: 𝛿 ≠ 0. This means that the
relation between interest rates and bank risk-taking became significantly positive for the post-crisis period compared to the pre-crisis period. Since the financial crisis, banks take less risks compared to the situation before the financial crisis in a low interest rate environment.
6. Conclusion
This thesis examined the relationship between low interest rates and bank risk-taking and the influence of the financial crisis of 2008, using a balanced panel dataset of 142 US banks for the period 2003 to 2016.
21
and therefore this thesis differentiates between the pre-crisis period and the post-crisis period. The results reveal an insignificant negative relation between interest rates and bank risk-taking for the pre-crisis period, and a significant positive relation for the post-crisis period. This relationship was further examined by including a dummy variable. The effects of the crisis, reviewed over the period 2008 to 2016, have a significant positive influence on the relationship between interest rates and bank taking. For the post-crisis period the relationship between interest rates and bank risk-taking is significantly positive, meaning that in times of low interest rates, banks lower risk-risk-taking. It is economically relevant to ask whether this change is due to new regulations or bank’s risk-taking behavior. It may be due to increased regulatory stringencies with tighter capital ratios which limited lending activities of banks and constrained their risk behavior. The other reason could be that banks behave more responsibly since the crisis and have learned from their excessive risk-taking behavior prior to the crisis. Nonetheless, the results show decreased bank risk-risk-taking since the crisis, at least in terms of granting new loans and their stance towards risky assets.
The contribution of this thesis to the existing literature is the use of risk measures that describe the ex-ante risk-taking, and not the risk exposure that is related to economic conditions. Most studies rely on proxies that measure ex-post loan performance influenced by economic conditions outside the control of banks. This thesis studies the relationship between low interest rates and bank risk-taking, and the influence of the financial crisis. The results deepen the understanding of the causes of the crisis, and may help in preventing a financial crisis in the future.
The dataset comprises a rather limited set of banks, as I preferred to the use of a balanced dataset. Sample selection results in a survivorship bias, as some banks with missing data or banks that went into default during the examined period are not selected. The sample is also affected by a selection bias, because the selection criteria explained in Chapter 4 are subjectively chosen for this thesis. The same is true for the definition of the time periods, which are in line with Andries et al. (2015), however, other authors can choose different subsamples. Another limitation is the insignificant result of the Hausman tests. Conducting the tests concluded the use of a random effects model, but these results were in contrast to the theoretical background of the tests I wanted to perform. I wanted to test where the intercept differ across sections but not over time, therefore I performed the fixed effects model.
As a last limitation, the potentially lagged effect of interest rates on bank risk-taking is not integrated in the model. Delis and Kouretas (2011) argue that bank risk-taking of the previous period may influence the bank risk of the current period. It might also be that risk-taking by banks needs a certain amount of time to adjust to macroeconomic changes. A dynamic model by including a lagged dependent variable captures the persistent character of bank risk and provides unbiased results (Andries et al., 2015). However, fixed effects model estimations are liable to seriously biased coefficients in dynamic models (Nickell, 1981). The correlation between transformed lagged variable and the error term creates a bias in the estimate of the coefficient of the lagged dependent variable, called the Nickell bias. For this reason, this thesis did not include a dynamic model.
22
predict. A research setup with a richer dataset or other risk measures would broaden our knowledge and may help in preventing a future credit crisis. Another challenge for future research would be to examine whether tighter regulations have lowered bank risk-taking or if banks learned from the past and changed their behavior towards risk.
23
7. Literature
Altunbas, Y., Gambacorta, L., & Marques-Ibanez, D. (2010). Does monetary policy affect bank risk-taking?
Altunbas, Y., Gambacorta, L., & Marques-Ibanez, D. (2010). Bank risk and monetary policy. Journal of Financial Stability, 6(3), 121-129.
Amador, J. S., Gómez-González, J. E., & Pabón, A. M. (2013). Loan growth and bank risk: new evidence. Financial Markets and Portfolio Management, 27(4), 365-379.
Andries, A. M., Cocriş, V., & Pleşcău, I. (2015). Low interest rates and bank risk-taking: Has the crisis changed anything? Evidence from the Eurozone. Review of Economic and Business Studies, 8(1), 125-148.
Bhagat, S., Bolton, B., & Lu, J. (2015). Size, leverage, and risk-taking of financial institutions. Journal of Banking & Finance, 59, 520-537.
Borio, C., & Zhu, H. (2012). Capital regulation, risk-taking and monetary policy: a missing link in the transmission mechanism? Journal of Financial Stability, 8(4), 236-251.
Delis, M. D., & Kouretas, G. P. (2011). Interest rates and bank risk-taking. Journal of Banking & Finance, 35(4), 840-855.
DellʼAriccia, G., Laeven, L., & Marquez, R. (2014). Real interest rates, leverage, and bank risk-taking. Journal of Economic Theory, 149, 65-99.
Dell'Ariccia, G., Laeven, L., & Suarez, G. A. (2017). Bank Leverage and Monetary Policy's Risk‐ Taking Channel: Evidence from the United States. the Journal of Finance, 72(2), 613-654.
Denning, S. (2013). Big banks and derivatives: why another financial crisis is inevitable. Forbes Magazine.
Duprey, T., & Lé, M. (2016). Bankscope dataset: getting started.
Foos, D., Norden, L., & Weber, M. (2010). Loan growth and riskiness of banks. Journal of Banking & Finance, 34(12), 2929-2940.
Hull, J. (2012). Risk management and financial institutions, Web Site (Vol. 733). John Wiley & Sons.
24
Jiménez, G., Lopez, J. A., & Saurina, J. (2013). How does competition affect bank risk-taking? Journal of Financial Stability, 9(2), 185-195.
Jiménez, G., Ongena, S., Peydró, J. L., & Saurina, J. (2014). Hazardous Times for Monetary Policy: What Do Twenty‐Three Million Bank Loans Say About the Effects of Monetary Policy on Credit Risk‐Taking? Econometrica, 82(2), 463-505.
Laeven, L., & Levine, R. (2009). Bank governance, regulation and risk taking. Journal of financial economics, 93(2), 259-275.
Maddaloni, J. Peydró. (2011). Bank Risk-taking, Securitization, Supervision, and Low Interest Rates: Evidence from the Euro-area and the U.S. Lending Standards, The Review of Financial Studies, Volume 24, Issue 6, 1 June 2011, Pages 2121–2165
Nickell, S. (1981). Biases in dynamic models with fixed effects. Econometrica: Journal of the Econometric Society, 1417-1426.
Rajan, R. G. (2006). Has finance made the world riskier? European Financial Management, 12(4), 499-533.
25 Appendix A
Table 7 Descriptive Variables (Initial Sample)
Variable Observations Mean Median Standard Deviation Minimum Maximum
Risk Assets 1988 0.80 0.82 0.11 0.13 0.99 Loan Growth 1988 0.10 0.07 0.26 -0.51 9.45 Size 1988 5.89 5.94 0.24 4.45 6.27 Profitability 1988 0.01 0.01 0.01 0.00 0.10 Efficiency 1988 1.62 1.57 0.30 1.04 4.61 Liquidity 1988 6.61 4.72 6.89 0.48 81.42 Inflation 1988 0.02 0.02 0.01 0.00 0.04 Economic Growth 1988 0.04 0.04 0.02 -0.02 0.06
Three-month Libor Rate 1988 1.66 0.72 1.83 0.23 5.30
Federal Funds Rate 1988 1.23 0.24 1.71 0.03 4.85
Taylor-rule Rate 1988 0.25 0.13 3.08 -3.76 5.75
26 Appendix B
Figures 1-4 present the pooled regression between bank risk-taking proxied by the risk assets on the vertical axis and the four different interest rates on the horizontal axis. The trend line predicts a positive correlation for all regressions.
.55 .60 .65 .70 .75 .80 .85 .90 .95 0 1 2 3 4 5 6 LIBOR RA .55 .60 .65 .70 .75 .80 .85 .90 .95 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 LONG_INTEREST RA .55 .60 .65 .70 .75 .80 .85 .90 .95 -4 -2 0 2 4 6 TAYLOR RA .55 .60 .65 .70 .75 .80 .85 .90 .95 0 1 2 3 4 5 FED RA
Figure 3 Risk Assets and the Federal Funds Rate
Figure 2 Risk Assets and the Long-term Interest Rate
27
Figures 5-8 present the pooled regression between bank risk-taking proxied by loan growth on the vertical axis and the four different interest rates on the horizontal axis. The trend line predicts a positive correlation for all regressions.
-.10 -.05 .00 .05 .10 .15 .20 .25 .30 -4 -2 0 2 4 6 TAYLOR LG -.10 -.05 .00 .05 .10 .15 .20 .25 .30 0 1 2 3 4 5 6 LIBOR LG -.10 -.05 .00 .05 .10 .15 .20 .25 .30 0 1 2 3 4 5 FED LG -.10 -.05 .00 .05 .10 .15 .20 .25 .30 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 LONG_INTEREST LG
Figure 5 Loan Growth and the Taylor-rule Rate
Figure 8 Loan Growth and the Three-month Libor Figure 7 Loan Growth and the Federal Funds Rate
28 Appendix C
Figure 9 Development of the Four Interest Rates Used in this Thesis
This figure present the short-term interest rate measured by the three-month Libor and the federal funds rate, the Taylor-rule rate and the long-term interest rate for the period 2003 to 2016.
