• No results found

The impact of the financial crisis on the risk-taking channel:

N/A
N/A
Protected

Academic year: 2021

Share "The impact of the financial crisis on the risk-taking channel:"

Copied!
42
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The impact of the financial crisis on the risk-taking channel:

A European setting

By

Bob Lennart Tjaarda

Supervisor: Dr. J.O. Mierau

Co-assessor: Prof. Dr. L.J.R. Scholtens

MSc International Financial Management MSc Economics and Business

MSc Finance

Faculty of Economics and Business Faculty of Social Sciences

University of Groningen Uppsala University

b.l.tjaarda@student.rug.nl Bob.Tjaarda.1214@student.uu.se

S1778889 890303-P279

Abstract

This study investigates the relationship between interest rates and bank risk-taking. The risk-taking channel of monetary policy transmission identifies the negative relationship between interest rates and bank risk-taking. The financial crisis of 2007 and the regulatory developments in the European Union provide a case in which the impact of the risk-taking channel has changed. A dataset consisting of 25,772 bank-year observations for the period 2001-2013 is constructed. The dataset consists of 3,541 banks from the 12 countries that were member of the European Monetary Union in 2001. The findings partly support the existence of the risk-taking channel but no evidence is found for a change in impact of interest rates on bank risk-taking. The results also indicate that size has a positive impact on the reduction of the impact of interest rates on bank risk-taking, with larger banks being affected less by changes in interest rates. However, the impact of size has not changed over time.

Keywords: Risk-taking channel, bank-risk taking, interest rates, monetary policy, Europe JEL classifications: G21, E43, E52

(2)

1. Introduction

At the end of the first decade of the 21st century, academics argued that a low interest rate environment increases the risk-taking within the banking sector. The interest rates are still at historically low values and in Europe there is no sign of interest rates increasing soon. The European Central Bank’s (henceforth: ECB) main refinancing rate as of 10 September 2014 is 0.05%. From mid-2008 the rate has only decreased and is now the lowest rate in the history of the ECB. Critics have argued that the low rates have led to even further increased risk-taking within the banking sector. ECB president Draghi admitted that the low interest rates will result in “some local misallocation of resources” (The Telegraph, 2015). However, the ECB does not take bank risk-taking into account while setting the refinancing rate (ECB, 2011).

The increasing interest in the linkages between interest rates and bank risk-taking led to the identification of the risk-taking channel of monetary policy. The risk-taking channel identifies the relationship that a decreasing interest rates leads to an increase in bank risk-taking (Borio and Zhu, 2012). Although recently there has been a large amount of empirical literature on the risk-taking channel, the risk-taking channel is yet to be examined over time. The financial crisis that erupted in 2007 provides an interesting case for evaluating the risk-taking channel over time. This crisis led to a global recession and also contributed to the European debt crisis that started in 2009. This led to a period of recession in the European Union that started in 2007 and is still present in 2015. The financial crisis also made clear that the financial regulation was inadequate. In Europe, it became clear that a new approach of financial regulation and supervision was required (Begg, 2009). This has led to the adoption of several directives with the aim of a sound financial system and the ability to take action when problems arise. Borio and Zhu (2012) argue that these regulatory developments will influence the risk-taking channel of monetary policy transmission. While some empirical studies, like Altunbas et al. (2009) and Drakos et al. (2014), include some of the crisis years in their sample, they do not examine possible changes in the risk-taking channel due to a crisis.

To investigate the risk-taking channel in a pre-crisis and crisis period, this study constructs a panel dataset of banks from 12 countries in the European Monetary Union (henceforth: EMU) for the period 2001-2013. A pre-crisis sample for the period 2001-2006 and a crisis sample for the period 2007-2013 will be constructed to investigate a possible change. The countries in the EMU are all subject to the same central bank, namely the ECB, and this provides a dataset for which you do not have to control for differences in central bank policies. The ratio of risk assets to total assets and the ratio of non-performing loans to total loans are used as proxies for bank risk-taking. The study will

(3)

make use of four different interest rates measures to control whether the relationship is consistent. This study will build on the research of Delis and Kouretas (2011), who investigate the relationship for all EMU countries in the period 2001-2008 and provide evidence for the existence of the risk-taking channel.

The results of this study partly support the hypothesis that interest rates have a negative impact on bank risk-taking. The negative relationship is not supported by all interest rate measures in combination with bank risk-taking proxies. By dividing the sample into a pre-crisis and crisis period, the possible change in impact of the risk-taking channel is investigated. Although the bank-level lending rate has an increased impact on the ratio of risk assets, this is not supported by other interest rate measures. Finally, the study investigates the impact of size on the relationship between interest rates and bank risk-taking. The results show that large banks are better able to cope with changes in interest rates as the impact of interest rates on both the ratio of risk assets and ratio of non-performing loans is less compared to small banks. However, this relationship has not changed over the years and new regulation that favoured larger banks does not have increased the advantage of larger banks to limit the impact of interest rates on bank risk-taking.

This study contributes to the existing literature by investigating the impact of interest rates on bank risk-taking in a pre-crisis and crisis period. Where the existing literature focusses on the identification of the relationship, this study will evaluate whether the relationship has changed over time. Taken all available literature into account, this has not been investigated at the time of writing. By constructing a large dataset, the study adds a meaningful investigation to a new area of study regarding the risk-taking channel.

The remainder of the study will be structured as follows. In Section 2, an overview of the existing literature and the hypotheses development will be presented. Section 3 will present the data collection process, an overview of the dataset and the methodology that will be employed. In Section 4, the results will be presented in combination with a discussion of the results. Finally, in Section 5, some concluding remarks will be given.

2. Literature review

The literature on the relationship between interest rates and bank risk-taking started to develop in the second half of the 20th century. This led to the identification of a several channels of monetary policy transmission.

(4)

2.1 Channels of monetary policy transmission

The traditional view on the monetary transmission mechanism is the money channel or interest rate channel. In this channel, interest rates are adjusted to clear markets and to influence the borrowing and lending behaviour in the economy. It is Bernanke (1983), who builds on the work of Friedman and Schwartz (1963) on the monetary impact of bank failures, who identifies the important role of financial institutions in monetary policy transmission and argues that there is another channel. Firms can finance their investments by issuing bonds, but also through bank loans. Therefore, banks play a significant role in providing finance for investments. This has been identified as the “credit channel”. According to Bernanke and Gertler (1995), there are two linkages that can be identified within the credit channel. Both linkages have an effect on the external finance premium in the market. First, the “balance sheet channel” describes the transmission of monetary policy through the borrowers’ balance sheets and income statements. The balance sheet is influenced by changing values of collateral due to the change in interest rates. Interest rates will also affect the income statement through the interest payments. Second, the “bank lending channel” describes the effect of monetary policy on the supply of loans. This channel appears due to the existence of firms that are dependent on banks for finance. Hence, a monetary policy that would drain reserves from the financial system would likely result in firms that have more difficulty in obtaining financing.

However, the traditional money channel and the credit channel do not describe the relationship between monetary policy and bank risk-taking. Banks could perceive borrowers as less risky when interest rates drop. The acceptance of more risky borrowers will then increase the riskiness of the bank’s loan portfolio.

2.2 Risk-taking channel of monetary policy transmission

While the “balance sheet” and “bank lending” channel are extensively studied, Borio and Zhu (2012) argue that there are elements of the monetary transmission mechanism that are under-researched. Several academics have argued that long periods of low interest rate environments could lead to excessive bank risk-taking. Decisions within the banking sector are based on models which incorporate the volatility in interest rates over multiple years. A prolonged period of low interest rates will decrease the perceived volatility in interest rates and lower the risk restrictions within banks. Therefore, the risk-taking channel will be more apparent if interest rates are at low levels for

(5)

longer periods of time. Taylor (2007) and Borio and Zhu (2012) were the first to identify the risk-taking channel.

Although Taylor (2007) did not name the risk-taking channel explicitly, he concludes that the low interest rates were responsible for the increased supply of funds which contributed to the high housing inflation in the US and increased risk-taking.

Borio and Zhu (2012) identify the link between monetary policy and the perception and pricing of risk. They argue that changes in interest rates can influence taking. The authors define the risk-taking channel as “the impact of changes in policy rates on either risk perceptions or risk-tolerance and hence on the degree or 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). They argue that there exist at least three ways through which the risk-taking channel operates.

First, changes in interest rates impact the valuations, incomes and cash flows. With a decreasing interest rate, valuations will be higher as the discount rate is lowered. This will boost the asset prices and provide potential borrowers with a higher amount of collateral. The reduction in the interest rate will reduce the probability of default and allows borrowers to increase leverage.

Second, the interaction of interest rates with fixed target rates has the capability of increasing risk-taking. Rajan (2006) identified this relationship as “the search for yield”. When interest rate changes are not taken into account with setting the target returns, managers may be forced to increase risk-taking in order to achieve the targets. Rajan (2006) argues that, although bank managers are less tied to returns than managers from other industries, the competitive pressure within the banking industry increases the incentives for banks to originate more loans to firms that may take on riskier investments to achieve the initially set targets. These risks will be moved from the balance sheet of the bank to the balance sheets of investment managers, but the bank will not be able to distribute all risks and complicated risks will remain on the balance sheet of the bank. Hence, the banking sector will provide in the search for yield by firms in other industries and therefore increase the overall risk-taking.

Third, the communication policies and reaction function of the central bank may reinforce risk-taking. A high degree of transparency in the operations and decision-making within a central bank may reduce the uncertainty about the future. This may reduce market uncertainty and allows banks to decrease the risk premia that borrowers need to pay. This is called the “transparency effect”. The reaction function may also reinforce risk-taking as it cuts off large downside risks. This can imply that

(6)

the changes in interest rates will impact the risk-taking behaviour asymmetrically. A decrease of the interest rate will increase risk-taking more than an increase will limit it. This effect is called the “insurance effect”, which could lead to moral hazard issues. In a European setting, the transparency effect can also be identified. The ECB considers transparency to be a crucial component of the monetary policy framework (ECB, 2011). Transparency is important for the ECB as it enhances the understanding of monetary policy and therefore will become more credible and effective.

2.3 Theoretical models of the risk-taking channel

The relationship between interest rates and bank risk-taking has been analysed with several theoretical models in the literature. DellʼAriccia and Marquez (2006) develop a model in which they analyse the interaction between informational structure of loan markets and banks’ strategic behaviour in setting the lending requirements. Although DellʼAriccia and Marquez do not explicitly model the risk-taking channel, they do find that the adverse selection problems, resulting from asymmetric information between banks and potential customers, decrease in a period of expansionary monetary policy. Subsequently, this will reduce the banks’ lending standards and this will result in more risk and higher vulnerability to macro-economic shocks.

To focus more on the impact of interest rates, DellʼAriccia et al. (2014) develop a model that examines the net effect of changes in the interest rate on the amount of costly monitoring of the bank. Higher monitoring means that the credit risk of the loan portfolios is reduced. They find that this relationship is influenced by three forces: interest rate pass-through, risk shifting and leverage. The two main findings of the study are that a reduction of the risk-free interest rates will lead to an increase of bank leverage and that it will lower bank monitoring and increase the bank risk-taking. The interest rate through effect is an important determinant of banks risk-taking. The pass-through effect is identified as the influence of the interest rate on the asset side of the balance sheet. With a decrease of the risk-free rate, the bank will receive a lower interest rate on bank loans and this will lower the gross return. The relative costs of monitoring will increase and therefore lower the incentive for monitoring the quality of borrowers. The second determinant is the risk-shifting problem. This problem can be found on the liability side of the balance sheet. As the costs of bank liabilities will drop with a decrease of the risk-free rate, the bank will have increased profits. Subsequently, the bank has an incentive to reduce the risk-taking on the asset side as the required return will be easier to obtain as the financing costs have decreased. The final determinant is the

(7)

capital structure of the bank and will be determining the risk-taking when banks are free to adjust their capital structure to changes in the interest rate environment. Higher bank capital will result in more monitoring of the bank’s loan portfolio because there is more to lose. Borio and Zhu (2012) also argue that the capital structure is endogenous and depend on the interest rate. They refer to the study of Adrian and Shin (2009) in which the authors find that a lower interest rate increases leverage and a higher leverage lead to greater risk taking.

Valencia (2014) develops a dynamic bank model, in which the risk-taking developments in banks are modelled when banks experience a period of low monetary policy rates. His model also shows that low rates will result in lower financing costs and makes lending therefore more profitable. The increase in profitability is boosted by the limited liability. The equity holders will not be liable for a failure of the bank and increasing the leverage, which enjoys the lower financing costs, will result in even higher returns for equity holders.

Although Valencia (2014) and Dell’ Ariccia et al. (2014) model the impact of monetary policy on bank risk-taking in different ways, both study results show that there is a negative relationship between interest rates and bank risk-taking.

Finally, Abbate and Thaler (2014) build on the model of Dell’ Ariccia et al. (2014) and identify two forces that impact the bank’s risk choice. First, the lower rate of return on loans will reduce the revenue and may force banks to consider more risky investments. Second, the lower cost of deposits will result in a higher net return which the bank can use to pay equity holders. The second force will induce banks to resort to safer investments as the higher net return will reduce the need for risky investments. The authors model both forces and the first force dominates in the equilibrium, which indicates that a bank will shift to a more risky loan portfolio with lower interest rates.

2.4 Empirical findings of the risk-taking channel

The last few years, the empirical evidence on the relationship between bank risk-taking and monetary policy has grown. Jiminez et al. (2008), Ioannidou et al. (2015) and Altunbas et al. (2009) were the first to empirically investigate the relationship.

Jiminez et al. (2008) investigate the relationship between lower interest rates and credit risk-taking by banks in Spain for the period 1984-2006. They find robust evidence that banks soften their lending restrictions and grant more loans in a low short-term interest rate environment. The authors argue that in the short-term lower interest rates reduce the total credit risk of banks and in the

(8)

medium term the lower interest rates will increase the credit risk in the economy. For small banks and commercial banks the negative relationship between interest rates and risk-taking is more pronounced.

Ioannidou et al. (2015) builds on the study by Jiminez et al. (2008) and examines the relationship between riskiness and new bank loans in Bolivia for the period 1999-2003. The Bolivian Peso was pegged to the US dollar in this period and they examine the effect of a decrease in the US interest rate. The authors conclude that expansionary monetary policy in the US encourages banks to grant riskier loans. For Bolivian banks, more access to liquidity is associated with a higher degree of risk-taking with new loans.

Both Ioannidou et al. (2015) and Jiminez et al. (2008) make a clear distinction between the riskiness of existing and new loans. However, this can only be investigated with high detailed data of loans and this poses a problem for research of other countries. Delis and Kouretas (2011) solve this problem by examining the overall riskiness of a bank. The authors examine the relationship between interest rates and bank risk-taking in the euro area for the period 2001-2008. They find a significant negative relationship that was robust for several different measures of risk-taking and interest rates. The results also showed that for banks with higher equity capital, the impact of interest rates on the riskiness of assets will be less, whereas the impact of interest rates will be amplified for banks with higher off-balance sheet items.

Altunbas et al. (2009) investigate the risk-taking channel with quarterly data for listed banks operating in the EU and US for the period 1999-Q1 2008. The proxy for risk is the expected default frequency computed by Moody’s and the monetary policy is calculated with the difference between the nominal interest rate and the rate generated with the “Taylor rule”. The Taylor rule determines how much the nominal interest rate should be adjusted by the central bank to tackle inflation, output and other economic aspects (Taylor, 1993). This relationship is significantly negative, which indicates that an interest rate that is lower than given by the Taylor rule will result in a higher probability for a bank to go into default.

Eid (2011) builds on the framework of Altunbas et al. (2009) to study the risk behaviour of French banks during the period 1998-2008. The main goal of the study is to investigate the existence of the risk-taking channel introduced by Borio and Zhu (2012). For this sample, Eid (2011) concludes that low interest rates lead to higher levels of bank risk. He also finds similar results on liquidity as Ioannidou et al. (2015), that higher levels of liquidity amplify the negative relationship between interest rates and bank risk.

(9)

Delis and Kouretas (2011) take a different approach and investigate the risk-taking channel for the euro area in the period 2001-2008. They do not have loan data for this large group of banks and therefore resort to another measure of bank risk-taking. The authors use the ratio of risk assets to total assets and the ratio of non-performing loans to total loans as proxies for bank risk-taking. They find a significant negative relationship for both measures of bank risk-taking in relation with several interest rate measures.

More recent studies by Buch et al. (2014) and Angeloni et al. (2015) study the existence of the risk-taking channel for the US market. Both studies identify the transmission channel and Buch et al. (2014) find that for small banks the negative relationship between interest rates and bank risk is larger than for large banks.

Drakos et al. (2014) examine the impact of interest rates on bank risk-taking in 10 central or eastern European countries and the Russian Federation during the period 1997-2011. They make a distinction between domestic and foreign banks and show that for both types of banks there is a negative relationship between interest rates and bank risk-taking, but the impact is higher for foreign banks.

All in all the existing empirical studies show that there is a significant negative relationship between interest rates and bank risk-taking through the risk-taking channel of monetary policy. Based on these empirical studies the following hypothesis is developed.

Hypothesis 1: Interest rates have a negative impact on bank risk-taking

2.5 Changing dynamics of the risk-taking channel

Changes in the risk-taking channel over time are yet to be examined and the financial crisis that erupted in 2007 provides a unique case for evaluating the channel over time.

Although the risk-taking channel is not examined in a crisis period, the impact of crises on the bank lending channel is empirically examined. Gambacorta and Marques-Ibanez (2011) examine the bank lending channel for 14 countries in the European Union and the United States from 1999 till 2009. They find that due to financial innovation and changes in the business models significant changes in the functioning of the bank lending channel are observed. The authors also argue that the financial crisis has contributed to the change of the transmission channel and that the channel will continue to evolve over time. This finding is supported by Ciccarelli et al. (2014), who find that the importance

(10)

of the bank lending channel in Europe grew during the financial crisis of 2007. The study also supports the interventions taken by the ECB during the crisis with almost exclusively targeting banks due to the significant role of the bank lending channel in Europe.

Aside from the findings on the changing importance of the bank lending channel, Borio and Zhu (2012) argue that there are several reasons why the risk-taking channel is likely to change over time. First, the Basel III accords include a countercyclical capital buffer, which may limit the amplification effect of the channel and increase the resilience of the banking sector (Drehmann et al. 2010). This is supported by Borio (2014) where he concludes that aside from monetary policy, prudential policy plays a key role in building capital buffers to cope with risk in the banking sector. Second, the liberalisation of the banking sector has provided banks with the incentive of external financing and the use of leverage instruments. Financial innovation made the use of external finance easier to access and has added to the existence of the risk-taking channel. Third, valuations have an essential role in the risk-taking channel and accounting practices play an important role in valuations. Although there is convergence in accounting practices throughout the world (Tarca, 2004), there are still a lot of differences which also results in different valuations.

The counter cyclical buffer, discussed by Borio and Zhu (2012), is included in the Basel III regulation which applies in the EU as of 1st January 2014. Therefore, this will not be applicable for the sample period of this study. However, major changes in EU banking regulation have been made during the sample period. An important change was the implementation of the Basel II accords on the 1st of January 2008. The new Basel accords had the objective to increase the financial stability of the EU financial system and provide strong incentives for banks to conduct robust risk modelling (Dierick et al., 2005). Elizalde (2007) compared the incentives for risk-taking in both the Basel I and Basel II accords with a theoretical model and concluded that the Basel II accords reduce banks’ risk-taking incentives. A second important regulatory change is the adoption of the Capital Requirements Directive (CRD) II and III package in respectively 2009 and 2010. The main goal of the directives was to strengthen the financial stability in the European Union and this has had an impact on the quality of a bank’s capital and risk management. Supported by the empirical findings of the bank lending channel that changes in periods of crises, the argumentation of Borio and Zhu (2012) and the significant regulatory changes in the European Union, the second hypothesis is stated as follows.

Hypothesis 2: The impact of interest rates on bank risk-taking has decreased after the 1st of January 2007

(11)

The third and fourth hypotheses are based on the risk-taking channel in combination with differences in bank size. Hakenes and Schnabel (2011) find that bank risk-taking under Basel II is different between small and big banks. They find that larger banks are able to achieve competitive advantages due to the “right to choose” the approach to weight risk under Basel II. Due to this competitive loss for small banks, smaller banks are forced to take higher risks to compensate for the loss. Delis and Kouretas (2011), in their sample period ranging from 2001 till 2008, already find that larger banks are better able to limit the impact of interest rates on risk-taking. Based on the implementation of Basel II, this study argues that larger banks are better able to reduce the impact of interest rates on bank risk-taking after the 1st of January 2008 compared with the period before. Based on the findings of Hakenes and Schnabel (2011) and Delis and Kouretas (2011), the following two hypotheses are developed.

Hypothesis 3: Larger banks are better able to limit the impact of interest rates on bank risk-taking

Hypothesis 4: Larger banks are better able to reduce the impact of interest rates on bank risk-taking after the 1st of January 2008

3. Data and Methodology

3.1 Methodology

First, the effect of interest rates on bank risk-taking using the pooled Ordinary Least Squares (OLS) method will be estimated. This will provide insight in the relationship, but it does not capture variation between banks and variation over time. The pooled OLS method estimates one constant and one coefficient per individual variable that is the best fit for all banks. To cope with the variation between banks, the next step will be to treat the dataset as panel data. The regression model that will be estimated is the following:

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖,𝑡𝑡 = ∝ + 𝛽𝛽1𝑅𝑅𝐼𝐼𝐼𝐼_𝑅𝑅𝑅𝑅𝐼𝐼𝑅𝑅𝑖𝑖,𝑡𝑡 + 𝛽𝛽2𝐵𝐵𝑅𝑅𝐼𝐼𝑅𝑅_𝐶𝐶𝐶𝐶𝐼𝐼𝐼𝐼𝑅𝑅𝐶𝐶𝐶𝐶𝑅𝑅𝑖𝑖,𝑡𝑡 + 𝛽𝛽3𝑀𝑀𝑅𝑅_𝐶𝐶𝐶𝐶𝐼𝐼𝐼𝐼𝑅𝑅𝐶𝐶𝐶𝐶𝑅𝑅𝑡𝑡 + 𝜀𝜀𝑖𝑖,𝑡𝑡,

where the 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖,𝑡𝑡 variable is the bank risk-taking for bank i at time t, 𝑅𝑅𝐼𝐼𝐼𝐼_𝑅𝑅𝑅𝑅𝐼𝐼𝑅𝑅𝑖𝑖,𝑡𝑡 is the interest

rate for bank i at time t. This could be bank specific, equal across countries or equal for all banks depending on the interest measure that is employed. 𝐵𝐵𝑅𝑅𝐼𝐼𝑅𝑅_𝐶𝐶𝐶𝐶𝐼𝐼𝐼𝐼𝑅𝑅𝐶𝐶𝐶𝐶𝑅𝑅𝑖𝑖,𝑡𝑡 are the bank-level

control variables for bank i at time t and 𝑀𝑀𝑅𝑅_𝐶𝐶𝐶𝐶𝐼𝐼𝐼𝐼𝑅𝑅𝐶𝐶𝐶𝐶𝑅𝑅𝑡𝑡 are macro-economic and structural

(12)

discussed in section 3.2.1 and 3.2.2 respectively and the control variables are discussed in section 3.2.3. Finally, interaction variables will be added to the regression model in order to investigate the third hypothesis. This will be discussed in the results and discussion section.

3.2 Sample

The relationship between interest rates and bank risk-taking is investigated with the use of a large panel data set with several different interest rates and measures of risk-taking. Annual data from the Bankscope database is collected for commercial, savings and cooperative banks for the 12 euro area countries that were part of the Economic and Monetary Union (EMU) on 1 January 20011. The starting year of the study is 2001 and this has been picked due to data availability2 and Greece joining the EMU. The last year of the study period is 2013 because the data availability for 2014 is still very limited3. Countries that joined the EMU in the study period are not included into the sample because this study wants to examine the differences between periods and banks from these countries will only have data available after the date of joining the EMU. As the financing of banks plays a key role in the risk-taking channel, it is necessary for banks to take deposits as this study wants to investigate this channel. Therefore, investment banks are not included in the sample (Delis and Kouretas, 2011). In table 1, the construction of the sample is presented.

Table 1: Sample construction

In this table the search criteria are presented that are used in the Bankscope database. The variables include a bank when there is an entry in one of the years ranging from 2001 till 2013 and for the profitability the year 2000 is added. The result indicates that 3,541 banks have data for all variables in at least one of the years. The bank-level variables consist of: equity/total assets, profit before tax, total assets, non-interest expense/gross revenues, off balance sheet items, interest income on loans, gross loans, loans and advances of banks, government securities, and cash and due from banks. Banks that did not have all the variables available in a year where excluded from the sample.

Search criteria Search result (number of banks)

Country selection: AT, BE, FI, FR, DE, GR, IE, IT, LU, NL, PT, ES 6,423

Commercial banks, Savings banks, and Cooperative banks -1,376

Bank-level variables -1,506

Result 3,541

1 The study will only include commercial banks, savings banks and cooperative banks for the period 2001-2013

that operated in Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal and Spain. These countries have had a common monetary policy during this period.

2 The Bankscope database contains data for the last 15 years and therefore excludes the full year of 1999. Data

of all bank-level variables for the year 2000 results only in observations for France, Germany, Italy, Luxembourg and Portugal.

3 For the year 2014, only 211 banks have data available. This is a fraction of the total active banks from 2013.

(13)

The resulting sample consists of 3,541 banks across the 12 countries that are studied. In appendix A, the bank distribution per country plus the average capitalization can be found. It stands out that from the distribution that both Germany and Italy are very well represented with respectively 56.42% and 20.30% of the number of banks. However, when the total average assets per country are examined, the distribution changes to Germany and Italy having 18.44% and 11.56% respectively. In Germany, this difference can be explained by the large number of savings and cooperative banks that are restricted in their region of operations (Dietrich and Vollmer, 2012) and this can also be found in Italy (Fonteyne, 2007).

The initial dataset consists of 27,805 bank-year observations for which all bank-level variables are present for that year. To cope with extreme values, all the bank-level variables are trimmed with 1% and this result in a final sample consisting of 25,772 bank-year observations. The average number of observations per bank ranges from 7.3 to 3.5 depending on the bank-risk measure and the interest rate measure that is used and this indicates that the final sample could be affected by the survivorship bias. However, banks that are no longer active, due to mergers, acquisitions or failures, are included in the sample for the bank-year observations that are available.

In the study by Delis and Kouretas (2011), they investigate the risk-taking channel with the use of annual data and control for the findings by employing quarterly data. They conclude that annual data is appropriate for studying the risk-taking channel. Similar findings can be found in Gambacorta (2005), who investigates the bank lending channel with both annual and quarterly data. Therefore, the use of annual data seems to be sufficient for analysing the risk-taking channel in this setting.

3.2.1 Dependent variable: bank risk-taking

In the literature on the risk-taking channel, several measures of bank risk-taking have been used. In this study two proxies for the risk-taking within a bank will be employed. The first proxy is the ratio of risk assets to total assets which reflects the ratio of total assets that is directly impacted by monetary and macroeconomic changes. Higher ratios will represent higher bank risk-taking. The risk assets are acquired with total assets minus cash, government securities and loans and advances to other banks (Delis and Kouretas, 2011; Drakos et al., 2014). These variables are collected from the Bankscope database. The second proxy is the ratio of non-performing loans to total loans and represents the quality of the loan portfolio of the bank. These data again is obtained from the Bankscope database and higher ratios are associated with a lower quality of the loan portfolio.

(14)

Although both proxies do not exactly measure the same risk-taking within a bank, this will provide a robustness check for the findings. Both proxies for bank risk-taking measure risk, but a bank with a high ratio of risk assets may have a stricter loan acceptance protocol which results in a lower amount of non-performing loans. However, this will not be a problem as the difference will not be investigated as the main goal is the relationship with interest rates. This means that an individual bank may have a superior screening technique, but over time the risk proxies may be significantly affected by the interest rate.

3.2.2 Interest rates

As there is a wide array of interest rates in the EMU area, four different interest rates will be examined. Again, this will provide a thorough analysis of the relationship and act as a robustness check of the results. This study follows the choice for the interest rates from Delis and Kouretas (2011) and the interest rates can be divided into a short-term interest rate, long-term interest rate, central bank interest rate and the bank-level lending rate. The short- and long-term rates are at a country level, the central bank rate at an EMU level and the lending rate at a bank-level. The short-term interest rate is obtained from DataStream and is calculated by taking the annual average of the monthly average of the three month interbank rate for each country. The long-term interest rate is also obtained from DataStream and is calculated by taking the annual average of the monthly average long term government bond yield for each country4. The central bank interest rate is the refinancing operations rate set by the European Central Bank (ECB) and is retrieved from the ECB website5. The proxy of the bank-level lending rate is defined as the ratio of interest income to total customer loans and is retrieved from the Bankscope database.

The bank-level lending rate is part of the so called “pass-through process” of monetary policy transmission. This process describes to what extent banks pass-through financing costs to their customers. Findings on the relationship between the monetary policy rate and the bank-level lending rate in Europe find a lagged adjustment in the lending rate6. The bank-level lending rate is the average lending rate that is charged to the customers by a particular bank. It will be used as a robustness check for the results of the other interest rates.

4 As there are multiple long term government bond yields series for a number of countries in DataStream, the

Maastricht definition is used for all twelve countries.

5 The refinancing operations rate set by the ECB: https://www.ecb.europa.eu/stats/monetary/rates

6 Literature on the pass-through process in Europe: see Toolsema et al. (2001), Sander and Kleimeier (2004),

De Bondt (2005) and Van Leuvensteijn et al. (2013)

(15)

Fig 1. The relationship between bank risk-taking and the bank-level lending rate is given in a scatterplot.

Bank risk-taking is defined as the ratio of risk assets to total assets and the bank-level lending rate is defined as the ratio of interest income to total loans. A regression line is fitted to the data.

In figure 1, the relationship between bank risk-taking and the bank-level lending rate is shown. A regression line is fitted to the data and it becomes clear that there is a negative relationship. Although this points to a negative relationship, this will be investigated with the other proxy for bank risk-taking and several other interest rates. Besides the multiple interest rates and risk proxies, the literature points to the inclusion of several control variables.

3.2.3 Control variables

The literature has identified a large number of variables that influence bank risk-taking and these can broadly be divided into bank-level variables, regulatory variables, macro-economic variables and structural conditions. By failing to control for these forces, it will be likely that the results will be subject to an omitted-variable bias.

.2 .4 .6 .8 1 Ri s k a s s et s t o T ot al as s e ts 0 .1 .2 .3

Interest income to Total loans

(16)

Several forces behind the changes in bank risk-taking have been identified and one of the most prominent is financial innovation7. Financial innovation has led to better risk diversification, but also to a more complex operating environment for banks. Bolton et al. (2010) argues that the complex nature of risk hedging has led to an overexpansion of credit markets, which in turn has led to increased risk-taking. Another important driver behind changes in bank risk-taking is deregulation and reregulation. Brewer et al. (2003) argue that the worldwide trend of deregulation of the banking industry has led to excess risk taking within the banking industry. The financial crisis of 2007 brought this issue to the understanding of the policy makers, which has led to a wave of reregulation (Veron, 2008).

To capture the forces that influence bank risk-taking, this study will control for five bank-level variables which are obtained from Bankscope. First, capitalization defined as the ratio of equity capital to total assets. The theoretical framework of Jeitschko and Jueng (2005) argues that the relationship depends on the incentives of three agents – the deposit insurer, the shareholder, and the manager. The deposit insurer has a very conservative approach on risk-taking, the shareholder the incentive to increase risk-taking and the manager who is generally more conservative than the shareholder. Jeitschko and Jueng (2005) argue that the relationship between capitalization and risk-taking can negative or positive and depends on the forces behind the three incentives. This mixed result is also found in empirical research on this relationship8. Second, profitability defined as the ratio of profits before tax to total assets. In good times, it sounds reasonable to think that higher profits are the result of a higher level of risk assets. These profits may be used to grant new loans in the next period. In bad times, a higher level of risk assets may result in increased losses, which will decrease the amount of new loans and lower the amount of risk assets. Hence, the risk assets in a particular year depend on the profitability of the year before. Therefore, the profitability is lagged for one year as control variable9. The expected signs will differ for the two proxies for risk-taking as higher profitability in the first year will likely increase the amount of risk assets in the following year. However, higher profitability in the first year will decrease the amount of non-performing loans. This means a positive sign for the risk assets and a negative sign for the non-performing loans. Third, the bank size is controlled for with the natural logarithm of the total assets. The reasoning behind the relationship between risk-taking and size can be explained by increased risk diversification for larger banks (Demirgüç-Kunt et al., 2008). However, Hughes et al. (1996) argue that managers could be

7 Literature on the role of financial innovation and bank risk-taking: see Santomero and Trester (1998), Bolton

et al. (2012) and Kero (2013).

8 See: Shrieves and Dahl (1992), Jacques and Nigro (1997), Jacques and Aggarwal (1998) and Rime (2001).

9 In order to include the observations from 2001, profitability data from 2000 is acquired from Bankscope.

(17)

motivated to take on more risk with other investments, as response to diversification advantages which will eliminate the risk reduction. Therefore, no relationship or a negative relationship is expected for the bank size control variable. Fourth, efficiency is defined as ratio of total non-interest expenses to total revenues. Higher efficiency could indicate superior screening capabilities and risk management techniques. This indicates a negative relationship between efficiency and bank risk-taking as higher efficiency will result in a lower ratio. However, a distinction has to be made between the two proxies of risk-taking. Higher efficiency may increase the quality of the loan portfolio, which allows a bank to increase the ratio of risk assets to total assets. On the other hand, a high quality loan portfolio may reduce the portion of non-performing loans. This reasoning points to a negative relationship with risk assets and a positive relationship with non-performing loans. Fifth, off-balance sheet items are defined as the ratio of off-balance sheet items to total assets. This variable will control for the non-traditional activities of a bank. Non-traditional activities have been expanding at a rapid pace the last two decades (Clark and Siemens, 2002; Lozano-Vivas and Pasiouras, 2014). Barth et al. (2004) argue that due to the more diverse financial activities within banks, the risk exposure of banks may increase due to complexity which decreases the monitoring capabilities. To capture this possible increase in risk-taking, this variable as proxy for non-traditional activities is added10. The bank-level control variables are similar to the study of Laeven and Levine (2009) and Delis and Kouretas (2011).

Regulation is another aspect that is controlled for in the existing empirical literature on the risk-taking channel. Delis and Kouretas (2011) and Drakos et al. (2014) employ regulatory variables that are constructed by Barth et al. (2003). They construct three indices, capital stringency index, market discipline index and the private monitoring index based on a World Bank survey. The index scores are at a country level and based on seven to fourteen questions per index. Although this may have been a useful regulatory measure at the start of the 21st century, this study argues that it has become less appropriate after the financial crisis when new problems in existing regulation were recognized. The indices by Barth et al. (2003) were updated in 2006 and 2008 and a more sophisticated version was conducted in 2013, but the questions that comprise an index are still identical to the study from 2003. This leads to the conclusion that there is not an appropriate measure for the regulatory environment at a country level. Therefore, the regulatory environment is not being controlled for. However, in order to verify if the results do not deviate when the regulatory indices of Barth et al. (2003) are included, the results with inclusion of the regulatory

10 In the study of Lozano-Vivas and Pasiouras (2010), non-interest income and off-balance sheet items are used

as proxy for non-traditional activities. They find similar results for both measures in relation to risk-taking.

(18)

variables are presented in Appendix B. The conclusions made in Section 4 are in agreement with the results with inclusion of the regulatory variables.

The study will also control for the macro-economic environment at a country level with the GDP growth rate and ratio of domestic credit provided by banks to GDP. In an environment of declining economic growth, credit risk could be triggered due to increased lending in periods of increasing economic growth (Männasoo and Mayes, 2009). GDP growth is used as measure for economic growth. Banks increase their lending in periods of GDP growth as they search for higher yields. The ratio of domestic credit provided by banks to GDP is included to control for the importance of banks within the country, a bank-based or market-based economy. Bank-based economies are likely to take on higher risks due to the inability of some firms to attract capital from other sources (Larrian, 2006). For both macro-economic variables, a positive relationship is expected with higher values indicating higher risks.

Finally, the study will control for the concentration of the banking sector at a country level. The concentration will be calculated with a 3-bank concentration ratio based on the net interest revenue obtained from Bankscope. The findings on the relationship between concentration and risk-taking are mixed. Boyd and De Nicolo (2005) confirm the mixed results in their theoretical model and argue that there are a large number of opposite effects that come into play which explains the mixed results. An overview of the variables and their sources can be found in Appendix C.

3.2.4 Descriptive statistics

Table 3 presents the descriptive statistics of the study’s variables. The mean of risk assets that a bank has on its balance sheet is 80.7% with a standard deviation of 13.4%. The non-performing loans have a mean of 6.5% and a standard deviation of 0.50%. The maximum long-term interest rate of 22.5% stands out and this can be explained with the Greece debt crisis. This also explains the GDP growth rate of -8.9%. Aside from these values, the descriptive statistics do not show odd values. In table 4, the Spearman’s rank correlation matrix between a proxy for bank risk-taking, interest rates and the control variables is given. This provides a nonparametric measure of the relationship between the variables. In panel A, the correlation coefficients are given with the bank risk-taking variable risk assets. The observed coefficients with the four measures of the interest rate all show a

(19)

Table 3: Descriptive statistics

This table states the descriptive statistics of the variables. The risk assets is the ratio of risk assets to total assets, where risk assets is calculated as total assets minus cash, government securities, and loans and advances to other banks. Non-performing loans is the ratio of non-performing loans to total loans. The short-term interest rate is the annual average of the 3-month interbank rate. The long-term interest rate is the annual average of the 10 year government bond yield. The central bank interest rate is the refinancing rate set by the ECB. The bank-level lending rate is the ratio of interest income to total loans. Capitalization is defined as the ratio of equity capital to total loans. Lagged profitability is the ratio of profits before tax to total assets. The Efficiency is the ratio of total non-interest expenses to total revenues. Off-balance sheet items is defined as the ratio of off-balance sheet items to total assets. GDP growth is the real GDP growth at country level. The importance of banks is the domestic credit provided by banks to total GDP. The concentration is the 3-bank concentration ratio based on the net interest income.

Variable Observations Mean Standard deviation Minimum Maximum

Risk assets 24,617 0.807 0.134 0.022 0.996

Non-performing loans 8,624 0.065 0.050 0.000 0.380

Short-term interest rate 25,772 2.228 1.490 0.139 4.635

Long-term interest rate 25,772 3.667 1.145 1.495 22.498

Central bank interest rate 25,772 2.148 1.247 0.553 4.288

Bank-level lending rate 23,078 0.061 0.022 0.008 0.456

Capitalization 25,772 0.076 0.038 -0.458 0.864

Lagged profitability 25,772 0.006 0.005 -0.057 0.058

Size 25,772 13.400 1.634 9.190 21.495

Efficiency 25,772 0.015 0.003 -0.013 0.045

Off-balance sheet items 25,772 0.073 0.073 0.001 2.037

GDP growth 25,772 0.804 2.534 -8.864 6.638

Importance of banks 25,772 130.906 18.055 66.791 228.378

Concentration 25,772 39.193 13.860 10.542 99.664

negative relationship, which one would expect with the existence of the risk-taking channel. The correlation between the short-term interest rate and central bank interest rate is very high with 0.973, but these variables will not be used simultaneously in the analysis. Therefore, this will not result in any problems. The other correlation coefficients will also not pose any problems for the analysis as they are sufficiently low.

In panel B, the risk assets are replaced with the non-performing loans proxy of bank risk-taking. The correlation with the interest rate measures, expect the long term interest rate, are as expected from the risk-taking channel expectations. However, this also shows the direct relationship and is not controlled for other variables that may impact non-performing loans. Aside from the interest rate correlation coefficients, the other correlation coefficients do not show any signs of possible correlation problems between the variables included in the analysis.

(20)
(21)

4. Results and Discussion

4.1 Pooled OLS regression

As discussed, the first step to obtain insight in the relationship between bank risk-taking and interest rates is by pooling the data and performing an OLS regression. The regression coefficients for the interest rates are presented in table 5. The different interest rate measures all have a significant negative relationship with risk assets as proxy for bank risk-taking. The short-term, long-term and the central bank show an increase in the ratio of risk assets between 1.1% and 2.2% with a 1.2% interest rate decrease, whereas a 1% decrease of the bank-level lending rate is associated with an increase of risk assets of 2.69%. This result supports the hypothesis that interest rates have a negative impact on bank risk-taking. However, the non-performing loans as proxy for risk-taking do not support the finding11. The discussion following the control variable coefficients can be found in the fixed effects regression section.

Table 5: Pooled OLS regression

This table states the regression coefficients and the standard errors (in parentheses). The statistical significance is indicated with ***, ** and *, which corresponds to significance at levels of 1%, 5% and 10% respectively. The regressions 1-4 have as bank risk-taking proxy the risk assets, where regressions 5-6 have non-performing loans. The risk assets is the ratio of risk assets to total assets, where risk assets is calculated as total assets minus cash, government securities, and loans and advances to other banks. Non-performing loans is the ratio of non-performing loans to total loans. The short-term interest rate is the annual average of the 3-month interbank rate. The long-term interest rate is the annual average of the 10 year government bond yield. The central bank interest rate is the refinancing rate set by the ECB. The bank-level lending rate is the ratio of interest income to total loans. The following variables are control variables and the results are not presented in this table as this table is purely for indication of the relationship between interest rates and bank risk-taking. Capitalization is defined as the ratio of equity capital to total loans. Profitability is the ratio of profits before tax to total assets The Efficiency is the ratio of total non-interest expenses to total revenues. Off-balance sheet items are defined as the ratio of off-balance sheet items to total assets. GDP growth is the real GDP growth at country level. The importance of banks is the domestic credit provided by banks to total GDP. The concentration is the 3-bank concentration ratio based on the net interest income.

(1) (2) (3) (4) (5) (6)

risk assets risk assets risk assets risk assets non-performing loans non-performing loans

Short term rate -0.011*** 0.002***

(0.001) (0.000)

Long term rate -0.022***

(0.001)

Central bank rate -0.012***

(0.001)

Bank-level lending rate -2.685*** 0.334***

(0.034) (0.035)

Observations 24,617 24,617 24,617 22,083 8,624 8,108

R-squared 0.062 0.078 0.062 0.263 0.221 0.240

11 The long-term interest rate and central bank interest rate regression coefficients are not included in table 5,

but show similar findings as those in table 5. The long-term interest rate has a coefficient of 0.007 at a significance level of 1%, whereas the central bank rate has a coefficient of 0.000 and is not significant. The signs of the control variables are similar to those in table 5 for the non-performing loans.

(22)

4.2 Fixed-effects regressions

4.2.1 Fixed-effects regression: total sample

By treating the dataset as panel data the regression allows for heterogeneity between banks, whereas the OLS regression assumes that there is homogeneity. Banks in the EMU operate in different environments, which results in different effects of variables. Therefore, panel data will be more appropriate to employ. Another advantage of panel data is that both the fixed and random effects regression allows controlling for unobserved heterogeneity and will not suffer from an omitted variable bias when an omitted variable does not change over time.

The random effect model is appropriate to use when the dataset can be considered as a random sample from a certain population. This is not the case with this study, and therefore the fixed effects model is more appropriate to use. However, there are more arguments in favour of using the fixed effects model. The fixed effects model assumes that there are omitted variables and that these variables may correlate with the independent variables. The Hausman test can be performed to statistically test which of the two models is more appropriate. The test evaluates the consistency of the regression estimation of both the fixed effects and random effects model. The null hypothesis is defined as both estimators being consistent and with accepting the null hypothesis the random effects model is preferred. Rejecting the null-hypothesis with a p-value of less than 0.05 will prefer the fixed-effects model. This study will perform this test with every regression in order to validate the choice of the regression model.

In table 6, the regression results are presented for the different fixed effects regressions with all bank-year observations available. The interest rate measure coefficients for the risk assets all have a significant negative relationship as would be expected from the risk-taking channel. However, this finding is only partly supported with the non-performing loans as a proxy for bank risk-taking12. The capitalization variable is significant in all regressions, but shows a positive sign for the risk assets and a negative sign for the non-performing loans. This finding is in accordance with the mixed results from empirical studies. The positive relationship, discussed in Jeitschko and Jueng (2005) as the shareholder approach, can be explained with shareholders willing to take on higher risks to acquire higher returns. The negative relationship between capitalization and the ratio of non-performing loans can be explained from the conservative risk approach of depositors. Due to these conflicting

12 The long-term interest rate has a small positive coefficient of 0.001 at a 10% significance level and the

central bank rate has a negative coefficient of -0.003 at a 1% significance level. The signs and significance of the control variable coefficients are similar to the coefficients in table 6 for the two regression results of non-performing loans.

(23)

findings, it is important to note that the number of observations greatly differs and that only 7,113 bank-year observations are included in both subsamples. The coefficients for the profitability are as expected from the reasoning that increased profits in one year will likely increase the risk assets as the profit can be used to grant new loans, whereas the non-performing loans are likely to decrease following a year of profit. The size variable has a negative sign across all regressions and the diversification argument presented by Demirgüç-Kunt et al. (2008) can be considered valid for this sample. The efficiency variable coefficients are in line with the reasoning in Section 3. A lower efficiency variable is associated with higher efficiency and this may indicate superior screening capabilities and risk management. This allows banks to increase their lending, which is supported by the negative relationship with the ratio of risk assets. The superior capabilities can also decrease the ratio of non-performing loans and this is again supported by the positive relationship between efficiency and non-performing loans.

The positive relationship between off-balance sheet items and risk assets supports the reasoning from Barth et al. (2004), suggesting that an increase in non-traditional banking activities is associated with a bank risk-taking increase. The negative relationship with non-performing loans, on the other hand, is harder to explain. An argument could be that banks use securitization, which represents the off-balance sheet items, to change the risk profile of their loan portfolio. Martín-Oliver and Saurina (2007) argue that securitization of loans may be motivated by shedding off credit risk. This may explain the negative relationship.

The economic growth coefficients are opposite of what is expected based on the findings of Männasoo and Mayes (2009). The results show that economic growth is associated with a decrease in risk assets and non-performing loans, while an increase is expected. This indicates that banks do not increase their lending in periods of higher economic growth. For the importance of banks, the coefficients are also opposite of what is expected. However, this is only the case for risk assets as the non-performing loans show the expected positive relationship. The positive relationship can be explained by a bank-based economy where firms are more reliant on bank financing and banks have to deal with a higher ratio of non-performing loans. However, it should be noted that the coefficients are very low with a change of one percentage point in economic growth or importance of banks only results in a change of 0.1% or less in risk assets or non-performing loans. The low coefficients can also be found with the concentration variable. The coefficients are consistently negative across all regressions. The negative relationship suggests that higher concentration is associated with a lower level of bank risk-taking.

(24)

Table 6: Fixed-effects regression total sample

This table states the regression coefficients of the fixed-effects regressions and the standard errors (in parentheses). The statistical significance is indicated with ***, ** and *, which corresponds to significance at levels of 1%, 5% and 10% respectively. The regressions 1-4 have as bank risk-taking proxy the risk assets, where regressions 5-6 have non-performing loans. The Hausman states the p-value for accepting the hypothesis that a GLS random-effects model is appropriate for the estimation of the regression coefficients. A value below 0.05 indicates that the fixed-effects model is more appropriate. The risk assets is the ratio of risk assets to total assets, where risk assets is calculated as total assets minus cash, government securities, and loans and advances to other banks. Non-performing loans is the ratio of non-performing loans to total loans. The short-term interest rate is the annual average of the 3-month interbank rate. The long-term interest rate is the annual average of the 10 year government bond yield. The central bank interest rate is the refinancing rate set by the ECB. The bank-level lending rate is the ratio of interest income to total loans. Capitalization is defined as the ratio of equity capital to total loans. Profitability is the ratio of profits before tax to total assets The Efficiency is the ratio of total revenue to total expenses. Off-balance sheet items are defined as the ratio of off-balance sheet items to total assets. GDP growth is the real GDP growth at country level. The importance of banks is the domestic credit provided by banks to total GDP. The concentration is the 3-bank concentration ratio based on the net interest income.

(1) (2) (3) (4) (5) (6)

risk assets risk assets risk assets risk assets non-performing loans non-performing loans

Short term rate -0.004*** -0.003***

(0.000) (0.000)

Long term rate -0.007***

(0.001)

Central bank rate -0.004***

(0.000)

Bank-level lending rate -1.115*** 0.050

(0.031) (0.033) Capitalization 0.295*** 0.273*** 0.302*** 0.223*** -0.228*** -0.179*** (0.029) (0.029) (0.029) (0.030) (0.026) (0.028) Lagged profitability 0.799*** 0.599*** 0.784*** 0.969*** -1.469*** -1.686*** (0.106) (0.104) (0.106) (0.110) (0.077) (0.078) Size -0.014*** -0.016*** -0.014*** -0.039*** -0.012*** -0.004 (0.002) (0.002) (0.002) (0.002) (0.003) (0.003) Efficiency -0.525** -0.408* -0.512** -0.345 1.735*** 1.607*** (0.223) (0.223) (0.224) (0.230) (0.183) (0.192)

Off-balance sheet items 0.162*** 0.157*** 0.163*** 0.191*** -0.046*** -0.036***

(0.011) (0.011) (0.011) (0.012) (0.007) (0.008) Economic growth -0.000 -0.000** -0.000 -0.001*** -0.001*** -0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Importance of banks -0.000*** -0.000*** -0.000*** -0.000*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Concentration -0.000*** -0.000*** -0.000*** -0.001*** -0.000* -0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Constant 1.045*** 1.051*** 1.040*** 1.454*** 0.148*** 0.011 (0.032) (0.032) (0.032) (0.033) (0.035) (0.038) Observations 24,617 24,617 24,617 22,083 8,624 8,108 R-squared 0.041 0.041 0.040 0.110 0.378 0.383 Number of banks Hausman p-value 3,384 0.000 3,384 0.000 3,384 0.000 3,046 0.000 2,387 0.000 2,347 0.000

The results on the relationship between bank risk-taking and interest rates are not in line with the study of Delis and Kouretas (2011), who employ similar variables and a dataset that is quite similar for the period of 2001-2008. They find larger negative relationships which are significant across all performed regressions. However, they derive the significant negative relationships between the interest rate measures and the risk-taking variables with a different statistical approach. They argue that bank-level lending rates are endogenously determined by the level of its risk taking and

(25)

therefore employ a two-stage least squares instrumental variables (2SLS-IV) method. However, the ECB’s monetary policy strategy, which is based on two pillars, has price stability as primary objective (ECB, 2011). The two pillars do not take bank risk-taking into account for setting the monetary policy rate. Based on the ECB rate setting process, this study argues that monetary policy decisions are exogenous and therefore the followed statistical method is more appropriate than the method in Delis and Kouretas (2011)13.

4.2.2 Fixed-effects regression: Equal sample risk-taking proxies

The results of the total sample do not support the risk-taking channel with all combinations of interest rates and bank risk-taking and there is a large difference in the number of observations between the risk assets and non-performing loans. Therefore, this study has constructed a sample in which bank-year observations have both data available for risk assets as well as non-performing loans. This drastically reduces the sample size to 7,113 bank-year observations, which reduces the credibility of the results. However, this may provide insight whether the results are also supported by this sample or that the differences are the result of the large number of observations that are only included in the whole sample. Table 7 presents the regression results for the sample where the bank-year observations are similar and the only difference is the risk-taking proxy.

The results are highly mixed, with both the short term rate and bank-level lending rate coefficients being significantly positive and negative. Aside from the interest rate coefficients, the efficiency coefficients in the risk assets regressions stand out. The signs have not changed, but the impact of increased efficiency has increased dramatically compared to the whole sample. An efficiency increase of 1%, which means a lower efficiency ratio, is associated with a 5.5% to 6.4% increase in ratio of risk assets. This might be extreme, but if we examine the mean and standard deviation of the efficiency ratio of the whole sample, it becomes clear how difficult it is to increase efficiency by 1%. The mean is 1.5% with a standard deviation of 0.3%. Therefore, the finding is less odd than initially perceived. However, it is important to note that for this sample there is limited data available of both bank risk-taking proxies, which reduces the credibility of this finding with only 2,140 banks and an average of 3.3 observations for a period of 13 years. Therefore, the subsample is very unbalanced and more data is necessary to make valid conclusions. To conclude, the first

13 To check whether this study finds similar results as Delis and Kouretas (2011) with their method, a 2SLS-IV

method with this dataset and also use the German short-term interest rate as instrumental variable is performed. These results are in accordance with that of Delis and Kouretas (2011) as all six regressions, like in table 6, show a significant negative relationship between the proxy for bank risk-taking and the interest rate measure.

(26)

hypothesis is partly supported, as the total sample provides strong evidence for accepting the hypothesis, but this is not consistently supported by the results of an equal sample.

Table 7: Fixed-effects regression with similar sample

This table states the regression coefficients of the fixed-effects regressions and the standard errors (in parentheses). The statistical significance is indicated with ***, ** and *, which corresponds to significance at levels of 1%, 5% and 10% respectively. The Hausman states the p-value for accepting the hypothesis that a GLS random-effects model is appropriate for the estimation of the regression coefficients. A value below 0.05 indicates that the fixed-effects model is more appropriate. Regressions 1 and 2 have the risk assets as proxy for bank risk-taking, where regressions 3 and 4 have the non-performing loans. The risk assets is the ratio of risk assets to total assets, where risk assets is calculated as total assets minus cash, government securities, and loans and advances to other banks. Non-performing loans is the ratio of non-performing loans to total loans. The short-term interest rate is the annual average of the 3-month interbank rate. The long-term interest rate is the annual average of the 10 year government bond yield. The central bank interest rate is the refinancing rate set by the ECB. The bank-level lending rate is the ratio of interest income to total loans. Capitalization is defined as the ratio of equity capital to total loans. Profitability is the ratio of profits before tax to total assets The Efficiency is the ratio of total revenue to total expenses. Off-balance sheet items is defined as the ratio of off-balance sheet items to total assets. GDP growth is the real GDP growth at country level. The importance of banks is the domestic credit provided by banks to total GDP. The concentration is the 3-bank concentration ratio based on the net interest income.

(1) (2) (3) (4)

Risk assets Risk assets Non-performing loans Non-performing loans

Short term rate 0.005*** -0.002***

(0.001) (0.000)

Bank-level lending rate -0.305*** 0.111***

(0.075) (0.038) Capitalization 0.791*** 0.721*** -0.205*** -0.177*** (0.064) (0.064) (0.032) (0.032) Lagged profitability -0.0625 0.475*** -1.520*** -1.738*** (0.173) (0.170) (0.088) (0.086) Size -0.098*** -0.110*** -0.001 0.004 (0.007) (0.007) (0.003) (0.003) Efficiency -6.408*** -5.543*** 1.972*** 1.626*** (0.419) (0.423) (0.213) (0.215)

Off-balance sheet items 0.130*** 0.124*** -0.032*** -0.030***

(0.017) (0.017) (0.009) (0.009) Economic growth 0.002*** 0.002*** -0.001*** -0.001*** (0.000) (0.000) (0.000) (0.000) Importance of banks 0.000 -0.000*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) Concentration 0.001*** 0.001*** -0.001*** -0.001*** (0.000) (0.000) (0.000) (0.000) Constant 2.075*** 2.288*** -0.005 -0.091** (0.091) (0.090) (0.046) (0.046) Observations 7,113 7,113 7,113 7,113 R-squared 0.293 0.289 0.405 0.403 Number of banks 2,140 2,140 2,140 2,140 Hausman p-value 0.000 0.000 0.000 0.000

4.2.3 Fixed-effects regression: Changing environment

The financial crisis that erupted in 2007 has led to a regulatory response from governments and international institutions in order to prevent a similar crisis (Moshirian, 2011). The Basel III accords are a direct response to the crisis, but before the crisis the Basel II accords were already adopted in the EU. The implementation was on the 1st of January 2008, but not only the Basel accords were adopted, also several directives that directly influenced the banking sector. The capital requirements

Referenties

GERELATEERDE DOCUMENTEN

The third dependent variable, loan loss provisions to total assets, shows also a significant and positive value between the short-term, long-term and bank-level lending

The variables are as follows: risk assets is the ratio of risk assets to total assets, abnormal loan growth is the difference between an individual bank’s loan growth and the

banks. Lastly, various empirical papers show a difference exists in adjustment speed towards target capital buffers for well-capitalised and undercapitalised banks 15. These

This table presents regression results in the years before the crisis (pre: 2000-2006) and during and after the crisis (post: 2007- 2014) of the effect of yield curve movements and

The variables are as follows: risk assets is the ratio of risk assets to total assets, adjusted risk assets is the ratio of adjusted risk assets to total assets, non-performing

I find that a large share of non-interest income does increase the insolvency risk for cooperative banks, but not for commercial and savings banks. The increase in insolvency risk

[r]

The variables used are as follows: the risk assets is the ratio of risky assets to total assets, the non-performing loans is the ratio of impaired loans to gross loans,