The impact of culture on the relationship between the low interest rate environment and bank risk-taking

17

July 2017

Nettelbosje 2, Groningen The Netherlands

The impact of culture on the relationship between the low interest rate environment and bank risk-taking

Master thesis submitted for:

MSc Finance

&

MSc International Financial Management

Student: Robert RUICĂ Student number: S2739941

Supervisor: dr. Mario HERNANDEZ TINOCO

Second assessor: Ryanne Marjon van DALEN

ii

The impact of culture on the relationship between the low interest rate environment and bank risk-taking

Robert Ruică

Abstract:

In the aftermath of the financial crisis (2007-09), an increasing number of evidences suggest that the persisting low interest rate environment has affected the behaviour of banks by making them more willing to increase risk-taking. This thesis investigates the relationship between interest rates and bank risk-taking on large panel data consisting of U.S., Eurozone and Japanese banks between 2001-2016. The focus of the study is on the behavioural aspect of risk-taking. The novelty of the paper is represented by the moderating effect of national cultural dimensions. The results were obtained by using a panel data analysis which involves fixed effects, random effects and the generalized methods of moments for dealing with endogeneity issues. The findings suggest that indeed there is a negative effect of low level of interest rates on bank risk-taking, and this relationship is positively moderated by individualism and negatively by uncertainty avoidance.

Keywords: bank risk, interest rates, cultural dimension, fixed effects, random effects, GMM JEL classification: E43, E52, G21, G40

Word count: 12,711

I would like to take this opportunity and to dedicate my entire work in the memory of my grandparents.

Secondly, I would like to thank to my supervisor, dr. M. Hernandez Tinoco for his guidance, encouragement and useful pieces of advice provided during our collaboration.

Personal contact: robert.ruica@gmail.com

iii

Table of contents

1. Introduction ... 1

2. Literature review ... 3

2.1. The traditional channels of monetary policy transmission... 3

2.2. A comprehensive assessment of risk-taking channel ... 4

2.3. Empirical findings ... 5

2.4. The moderation effect of cultural dimension ... 6

3. Data ... 8

3.1. Dependent variable: bank risk-taking ... 9

3.2. Independent variable: interest rates ... 11

3.3. Moderator: national cultural dimensions ... 13

3.4. Control variables ... 13

3.5. Descriptive statistics ... 15

4. Empirical framework ... 19

5. Empirical results and discussion ... 22

5.1. Fixed effects and random effects regressions ... 22

5.2. Dynamic model analysis ... 26

5.3. The moderation effect of cultural dimension ... 29

6. Conclusion ... 32

Bibliography ... 33

Appendix A. Conceptual model ... 36

Appendix B. Variables, definitions and data ... 37

Appendix C. Interest rates ... 38

Appendix D. Construction of regulatory variables ... 40

1

1. Introduction

The financial crisis is the key phenomena of the 21

Century that has opened new research areas in the field of finance and economics. It fundamentally changed the way how academics perceive now subjects as capital adequacy ratios, risk assessment or prudential policy. There is almost a general consensus that the main cause of the crisis was the combination of relaxation of lending standards which led to a credit boom and a housing bubble (Acharya &

Richardson, 2009). However, there are grounds to believe that this conclusion is too broad and does not give a precise explanation about the mechanisms behind the crisis which triggered the banks to take excessive risk. Nonetheless, more and more research articles are trying to explain the bank risk-taking from different perspectives by analyzing its relationship with corporate governance, institutional framework, regulatory environment etc.

A recent line of research is asking whether risk-taking channel

of monetary policy has induced incentives to increase the risk-appetite of the banking system during the period prior crisis. According to this hypothesis, a low interest rate environment can influence banks, ceteris paribus, to soften their lending standards and to increase the level of risk assets

in their balance sheets (Delis & Kouretas, 2011). Particularly, if each single bank raises the level of risky assets in its portfolio, the whole banking system will face an accumulation of risk. As a consequence, the financial system will become more sensitive to external shocks, increasing the likelihood of a crisis to occur (Maddaloni & Peydró, 2013). This evidence is also accompanied by the findings of Apel & Claussen (2012). Furthermore, a prolonged period of low interest rates would lead to financial imbalances like creating a bubble in the house market, a sharp rise of stock prices, high leverage etc. A proper explanation follows from the idea that investors will hesitate to leverage if short-term interest rates are low but unstable.

But if rates appear to be stable, investors will attempt to take the risk of borrowing money for higher-yielding securities. Therefore, a prolonged period of low interest rate would amplify the effect of financial instability within an economy.

The main goal of this study is to investigate the impact of low interest rates on bank risk-taking behaviour. Besides the monetary policy point of view, this research intends to explore the international dimension of financial management by including national cultural

2 The term was introduced in the academic literature by Borio & Zhu (2008)

3 Bank risk assets include all bank assets which carry on a degree of risk. Therefore, assets as cash or government securities (considered as risk-free assets) are excluded.

2 dimensions as a moderator in the main relationship. Thus, this thesis is interested in investigating the behavioural aspect of risk-taking. The research question raised in this paper is: to what extent are low interest rates affecting bank risk-taking and how is this relationship moderated by the cultural country-level effects? Li et al. (2013) argue that cultural dimension impacts risk-taking activities in a direct manner through managerial decision making and indirectly through firm- and country-level characteristics. Hence, culture affects the way how individuals, corporations and institutions perceive risk. Using the concept of cultural dimension developed by Hofstede (1980, 2001), I examine whether uncertainty-avoidance, individualism, masculinity and power distance have a positive or negative impact on the relationship between low interest rates and bank risk-taking. Furthermore, this thesis aims to find cultural cross-country differences in the banking industry regarding the risk-taking activities.

This study will contribute to existing body of literature through at least four ways.

First, the study aims to provide an empirical evidence whether the low interest rates have an impact on excessive risk exposure of the banking sector. Secondly, the novel of this thesis is the moderation effect induced by the cultural dimension. Not very explored by the academic literature in the context of risk-taking, I expect this effect will provide a new insight about cross-country differences in the way how banks deal with risk. Thirdly, a comprehensive assessment of the risk-taking channel will be made. There is evidence suggesting that monetary policy has influenced the financial crisis through this transmission mechanism.

Finally, the study will provide a management insight for the risk management of banks and also for supervisory authorities in taking into account the financial stability conditions when setting the monetary policy rate. Based on historical evidences, too accommodative monetary policy may contribute in building-up financial vulnerabilities.

The thesis is structures as follows. Section 2 presents the literature review on this topic and highlights main findings by different authors. Section 3 gives a presents in depth the employed data in this study. Section 4 discusses the methodology that I will make use of.

Section 5 is reserved for the results and discussions. Finally, section 6 will conclude the

thesis.

3

2. Literature review

In the recent past, the literature focused on this subject has flourished. Most articles suggest that in the period preceding a boom and bust cycle when the economy is rapidly developing, expansionary monetary policy interacting with financial frictions can determine financial vulnerabilities. Adrian & Liang (2014) argue that these vulnerabilities, such as a lower return on investments, excessive leverage or maturity transformation

, can increase the probability of a financial crisis and severe recession in the future.

2.1. The traditional channels of monetary policy transmission

In 1996, F. Mishkin makes an overview of the transmission mechanisms of monetary policy, giving new perspectives about the traditional channels used by the central bank to induce changes into the real economy. He starts with the interest rate channel, then goes through exchange rate channel and in the end explains the credit channel, which is the subject of research articles. The interest rate channel influences directly the funding cost in the banking system and it makes an important contribution on the investment or saving decision of firms and households. The exchange rate channel is used to deal with international trades, affecting the domestic price of goods. Finally, Bernake and Gertler (1995) describe the credit channel as an “enhancement mechanism” of the interest rate through which the central bank controls the amount of credit in the economy. Hence, banks play an important role within this channel. Therefore, a change in the monetary policy rate affects the credit channel through two secondary channels, the bank lending and the balance sheet channels. While in the former, a change in the monetary policy rate reflects an adjustment in the supply of credit, the latter is concerned about changes in the collateral values of firms, due to a modification in their net worth.

However, Appel & Classen (2012) made an important distinction on the way how credit and risk-taking channel operate. A signal of loose monetary policy, as a reaction of a decrease in the official interest rate, might be translated differently in economy. In the credit channel, an increase in the supply of credit may appear due to an improvement in the firms’

collateral and thus, a better repayment capacity which makes it less risky for banks to lend money. On the other side, in the risk-taking channel an increase in lending emerge from

4 Used by banks to transform short-term debt (e.g. deposits) into long-term investments (e.g. loans). It is also found in the literature as liquidity transformation.

4 willingness to accept higher risks. Therefore, while the credit channel is more about the quantitative side of transmission mechanism, the risk-taking channel focuses especially on the behavioural aspect of the banking system.

So, the difference between the traditional credit channel of monetary policy and the risk-taking channel is very narrow, making a thorough empirical evidence problematic. For instance, an expansionary monetary policy could lead to an increase in lending through the traditional credit channel. But, what the risk-taking channel points out is that the new amount of lending is even riskier due to changes in the banks’ risk tolerance.

2.2. A comprehensive assessment of risk-taking channel

The research in this specific field was opened by the study of Borio & Zhu (2008).

They introduced the term of risk-taking channel in the transmission mechanism of monetary policy and define it as the effect of monetary policy rate on the risk perception and risk tolerance of economic agents. However, it is observable especially in boom-boost cycles, where more extreme conditions are present. In other words, by making use of this channel the central bank can influence the behaviour of financial intermediaries. Thus, the risk-taking channel can be associated to a measure of banks’ risk sensitivity. Then, a legitimate question is what triggers the risk tolerance of banks? Few determinants which describe how the risk- taking channel operates were illustrated by Borio & Zhu (2008).

Firstly, a change in the interest rate modify valuations, incomes and future cash flows.

Lower rates can boost the value of assets in the balance sheet and in turn can reshape the measure of risk within a bank through a change in internal estimates as probabilities of default, loss-given-default, risk-weighted assets etc.

The second mechanism through which risk-taking channel influence the risk appetite

of financial intermediaries is called “search for yield” (Rajan, 2005). “Search for yield” is a

normal characteristic of the financial world because every economic agent strives to increase

the return on investments. Rajan argues that low interest rates, following an economic boom

period with high rates, create incentives for investors to search financial instruments with

higher yield. The reasons behind consists in the compensation policy of a bank or fund

manager, which is correlated with a specific expected level of return on investments, and in

the herding behaviour. The latter represent an incentive for investment managers to take more

5 risk because it creates an insurance to the manager that he will not underperform other competitors.

The third determinant refers to the effectiveness and credibility of the central bank to achieve its targets. With a high degree of transparency, the central bank can influence long- term expectations of firms through its communication policies and reaction function.

Particularly, it is able to remove uncertainty about the future, encouraging banks to engage in riskier activities. So this mechanism acts as an “insurance effect” for financial companies.

Altunbas et al. (2010) argue that this represents a classical moral hazard problem.

2.3. Empirical findings

To make an empirical analysis on the risk-taking channel is a challenging task to achieve. According to the risk-taking channel, low interest rates increase the risk appetite of banks. On the other hand, low interest rates can lead to a development in the supply of credit, which corresponds to view of credit channel. Therefore, an increase in lending activity as a consequence of low interest rate is not necessarily an evidence of risk-taking channel.

However, for a thorough analysis is important to separate the effects of risk-taking channel from the traditional credit channel. In the recent past, several empirical studies made a contribution in describing the mechanisms of risk-taking channel. Nevertheless, most of these papers agree on a negative relationship between interest rates and risk-taking. On the basis of the above arguments, the first hypothesis of the study is:

Hypothesis 1: low interest rate environment will engage banks in more risk-taking activities and will lower their risk tolerance

For instance, one of the most representative study is made by Jiménez et al. (2014) on

the Spanish banking system. Using a comprehensive dataset from the Spanish Credit Register,

which contains details of individual loans from 1984-2006 on quarterly frequency, they found

a negative impact of lower short-term rates on credit risk-taking. Therefore, the study

provides robust evidence that the lower overnight rates had triggered banks to soften their

lending standards and to grant more loans to borrowers with bad credit history. Nonetheless,

they suggest that in the short run, lower interest rates reduce the total credit risk of the

banking system due to a boost in the collateral value and an increase in repayment capacity of

firms. But on the medium term, lower interest rates may increase the credit risk within the

economy. Moreover, very similar results were obtained by Ioannidou et al. (2015). By

6 employing monthly frequency loans from the Bolivian credit register between 1999-2003, they found that lower monetary policy rate increases the risk appetite of banks.

Altunbas et al. (2010) investigates the risk-taking channel through an extensive and unique database of listed banks in European Union and the United States. The sample contains quarterly balance sheet information for 643 banks over the period 1998-2008. Particularly, they use the expected default frequency (EDF) as an indicator for credit risk. However, their conclusion supports the hypothesis that low short-term interest rates over an extended period of time have contributed to an increase in banks’ risk.

Furthermore, a particular study on the relationship between the level of interest rates and bank risk-taking is made by Delis and Kouretas (2011). The main focus of their research is the level of interest rates, which can be considered a consequence of the monetary policy stance. This has the advantage of increase the focus of the relationship being investigated.

They analyze approximately 18.000 annual observations on 16 Eurozone banks over the period 2001-2008. In building-up the model, their measure of bank risk is represented by risk assets and non-performing loans. In conclusion, there is a strong evidence of a negative relationship between the interest rates and the bank risk.

2.4. The moderation effect of cultural dimension

This paper contributes to the exiting debate on the level of interest rates and bank risk taking by including a cultural dimention as a moderator variable. The national cultural dimensions employed in this thesis derive from Geert Hofstede work pioneer work on organizational culture (for example see Hofstede 1980, 2001). A very recent strand of literature is investigating the cultural effects over the bank risk-taking behaviour (see Ashraf, 2016; Adhikari & Agrawal, 2016). They intend to deepen in the behavioural aspects of bank risk-taking and provide strong evidence that national cultural influence is an important factor in explaining “the appetite for risk”.

There is a substantial literature that investigates the bank risk-taking from a rational perspective. However, scholars overlook the influence of cultural background which is an important element in the decision making process (Li et al., 2013) and it can affect the behaviour of the banking industry. A new line of research is focusing on the behavioural aspects which can explain cross-country variation in bank risk. For example, Adhikari &

Agrawal (2016) investigate whether local religiosity has an impact in risk-taking activities of

banks. They consider that banks located in more religious area exhibit lower market risk and a

7 lower risk of default. Overall, banks located in religious areas are more risk averse. Their findings show that indeed, banks located in more religious areas are more prudent in risk- taking and are more conservative in decision making.

The moderation effect of national culture over the relationship between low interest rates and bank risk-taking is especially approached in the article written by Ashraf et al.

(2016). Based on the premise that cross-country differences in bank risk-taking occur due to differences in national cultures, they employ four dimensions from Hofstede’s cultural framework – uncertainty avoidance, individualism vs. collectivism, masculinity vs. femininity and power distance – to analyze the impact of culture on bank risk-taking behaviour.

Countries which score high on uncertainty-avoidance might be more reluctant to take on more risk. Usually their behaviour is more conservative and therefore banks located in such countries might be more risk averse.

Hypothesis 2: The relationship between low interest rates and bank risk-taking is negatively moderated by the level of uncertainty-avoidance.

Individualistic countries focus more on the individual. Literature suggests that individualistic countries might be exposed to psychological biases like overconfidence and self-attribution (Chui et al., 2010). However, overconfidence and self-attribution bias might determine managers to make irrational choices and thus to take excessive risk by being confident that they will increase profitability in this way.

Hypothesis 3: The relationship between low interest rates and bank risk-taking is positively moderated by the level of individualism

Masculine countries are more oriented towards assertiveness, competition and achievements. Nonetheless, there is a high focus on money and material things.

Hypothesis 4: The relationship between low interest rates and bank risk-taking is positively moderated by the level of masculinity

Lastly, power distance is related to the distribution of power. A high level of the index highlights a clear hierarchy, while a low level suggests equality among the members of a group. Previous literature suggests a negative relationship between power distance and risk- taking (Ashraf et al., 2016, Mihet, 2013).

Hypothesis 5: The relationship between low interest rates and bank risk-taking is negatively

moderated by the level of power distance

8 3. Data

In the aftermath of the financial crisis (2007-09), an increasing number of evidences suggest that the persisting low interest rate environment has affected the behaviour of banks by making them more willing to take on higher risks. Before testing the hypotheses developed in the previous section, I will employ a large unbalanced panel data retrieved from Datastream and Orbis Bank Focus

. A panel data contains information across both time and space (Brooks, 2008). Its use is appropriate since I intend to analyze the cross-country effects of national culture on the relationship between interest rates and bank risk-taking across various countries and over a long time frame. While this empirical study mostly relies on variables collected from Datastream, the latter is used to match the missing information for some of the variables from the first database. Datastream is an extensive database compiled by Thomson Reuters. It has a global coverage and contains historical time series data for publicly listed companies and economic variables, which makes it suitable for the purpose of this study.

Furthermore, a complete list of variables and databases used in this thesis can be found in the Appendix B.

The initial sample consists of 1,243 unique listed banks from United States, Eurozone countries and Japan between 2001-2016. Particularly, this time frame gives the opportunity to analyze the impact of the low interest rate environment over three distinct periods of the last two decades – the period before the financial crisis which is characterized by a boom cycle, the financial crisis 2007-09 and finally, the recession period post crisis. Furthermore, the choice of selecting the aforementioned three geographical regions was taken after several considerations. All the regions are known for their developed banking sector and for having solid financial markets which are interconnected through the globalization of the capital markets. Between these countries, there is a huge volume of international trading and a strong collaboration within international economic organizations like the International Monetary Fund (IMF), the Organization for Economic Cooperation and Development (OECD) or the World Trade Organization (WTO) (Yochelson, 1995). Moreover, the central banks representing these regions (i.e. Federal Reserve System, European Central Bank and Bank of Japan) are among the most powerful central banks in the world. They are directly influencing

5 Since 1^st January 2017 Bankscope database has been replaced with Orbis Bank Focus database by Bureau van Dijk. Even though this database contains a large list of variables for banks, its time frame is extremely limited for European banks.

9 the state of the economy by setting their main interest rate and are responsible for the supervision of the national banking system. However, it is interesting to note that each central bank has a different mandate to pursue in order to achieve its target. A further discussion on this topic it is out of the scope of this thesis.

Several adjustments were applied to the initial sample. Firstly, banking specialization was the first criteria taken into account. Considering that the focus of the paper is measuring how the changes in the interest rates level affect the amount of risk which banks bear, I will look for banks which collect deposits and offer loans to customers. In this regards, only commercial, savings and cooperative banks can be included. Other types of banks were left out from the sample due to this theoretical consideration. Secondly, I eliminate cross-border listings of banks in order to avoid double counting. Thirdly, an important step is to detect and remove outliers from the sample. Lastly, banks which have missing data for one of the two dependent variables are removed from the sample. However, it is important to mention that this sample may be affected by a so-called “survivorship bias”. This means that some of the banks might be no longer active due to mergers or simply because of a failure (Delis and Kouretas, 2011). Therefore, the final sample contains 803 listed banks from three different regions of the world, which result in 9,182 bank-year observation over 2001 – 2016.

3.1. Dependent variable: bank risk-taking

Two different measures for bank risk-taking will be used. The first measure is the ratio

of risk-weighted assets to total assets (named as risk assets) and the second one is the ratio of

non-performing loans to total loans (named as non-performing loans). Data for both proxies

were retrieved from Datastream. Risk-weighted assets is a standard risk measure for the

banking system imposed by the Basel Committee and is used in calculating the minimum

required capital (i.e. capital adequacy ratio). It represents the total of the carrying value of

each asset class multiplied with an assigned factor called risk weighting. In other words, risk-

weighted assets are calculated for each asset class in order to determine the bank’s level of

risk exposure to potential losses. For example, low risk assets such as government bonds, cash

and balances due from other banks are considered to be risk-free and consequently, the risk

weightings for these assets are zero. On the other hand, high risk assets like residential

mortgages or corporate loans can have a risk adjustment factor as high as 50%.

10 This risk measure is preferred by most of the monetary authorities. It represents a standardized approach in measuring risk which makes more easy to assess and compare the level of risk within banks from different geographical regions. Moreover, an important aspect to highlight about this risk measurement is that it also accounts for the risk inherent in the off- sheet balance of a bank. The increase in the off-sheet balance items has been a serious issue during the financial crisis. A large number of very complex and risky financial instruments (i.e. mortgage-backed securities, collateral debt obligation, credit default swaps etc.) were hidden in separate accounts from the balance sheet. Off-sheet balance items were designated as window dressing actions in order to artificially increase profit and to make banks look more financially stable. Therefore, risk-weighted assets can be considered an accurate proxy for bank risk-taking. The higher risk asset ratio is, the more exposed to risk the bank is.

The second risk measurement for the bank risk-taking is the ratio of non-performing loans to total loans. This ratio reflects the quality of banks’ loans and it is a proxy for the credit risk. A higher rate of non-performing loans ratio indicates a deterioration in the quality of portfolio loans. It is important to acknowledge that a part of the non-performing loans will result in losses for the bank. A loan is considered to be non-performing when its payment is past due by at least 90 days.

Although bank risk-taking is expressed in terms of two measures, these two proxies

reflect upon different risk dimensions within a bank. While the risk assets ratio explicitly

shows the amount of risk which banks hold, the non-performing loans proxy rather indicates

the internal loan acceptance protocol of banks. However, another major difference is

highlighted in the study conducted by Drakos et al. (2016). On the one hand, they explain that

risk assets capture the level of bank risk from its current operations, meaning that the bank

management team is responsible for the level of risk investments. On the other hand, the ratio

of non-performing loans is affected by the past investment decisions and is heavily influenced

by the previous monetary policy environment. The use of two measures of risk intends to

capture different dimensions of bank risk-taking that enhances our understanding of the

relationship between interest rates and risk taking. This study is interested in investigating

whether these risk proxies are significantly affected by the evolution of interest rates.

11

3.2. Independent variable: interest rates

Based on the article written by Delis and Kouretas (2011) and Drakos et al. (2016), four different type of interest rates are used to assess their impact on bank risk-taking.

Namely, I employ a short-term interest rate, a long-term rate, a key interest rate determined by the central bank and a bank-level lending rate. The interest rates were obtained from Datastream. Except the bank-level lending rate, all the other three interest rates are calculated as annual averages of relevant monthly series The short-term interest rate is the equivalent of the risk-free rate and it is measured by the monthly average of the 3-month bond in each country

. The long-term rate represents the annual average of the 10-year government bond yield in each country or geographical region (i.e. for the Euro area countries). The central bank rate is the key interest rate set by the monetary authority. It is the main tool of a central banks to influence the economic activity. The main interest rate in United States refers to fed funds rate, which represent the rate at which depository institutions lend reserve balances to other depository institutions. The European Central Bank (ECB) is the monetary authority for all the countries which have adopted Euro as a currency (i.e. Eurozone). Its key interest rate used is the main refinancing operations rate, which reflects the price at which commercial banks borrow money from ECB. Lastly, Bank of Japan (BoJ) is known for adopting a “zero policy rate” for almost two decades in order to combat the deflation. Hence, its main interest rate (i.e. uncollateralized overnight call rate) is zero or very close to this value. A summary of the interest rates movement over the past two decades is enclosed in the Appendix C section.

Moreover, the sample of banks panel data allows to use a bank-level lending rate by computing the ratio of interest income to total loans. This rate is also called as “pass-through interest rate” and it shows the average interest rate which a bank charges for its loans. This rate is particularly important since it highlights the direct effect of the monetary policy transmission. More concrete, the pass-through interest rate can be seen as a mechanism which presents the degree to which monetary policy is transmitted into the real economy (Darracq Paries et al.,2014).

6 Based on the European Central Bank recommendation, I used the German 3-month interbank rate to approximate the short-term interest rate and 10-years German bond as proxy for the long-term interest rate for the Eurozone.

12

Fig 2. The relationship between non-performing loans and the bank-level lending rate

13 In figure 1 and 2 the bank-level lending rate is presented in a relationship with risk-weighted assets ratio and non-performing in scatterplot graph. The regression line is fitted to the sample and it represents a first indication that there is a negative relationship between bank-level lending rate and both dependent variables – risk assets and non- performing loans.

3.3. Moderator: national cultural dimensions

Besides investigating the relationship between low interest rate environment and bank risk-taking, this study intends to use the moderation effect of national culture dimensions.

In this way, it is possible to test whether cross-country differences in terms of culture have an impact on bank risk-taking behaviour. The national cultural dimensions is a cross-country variable which helps to explain the main relationship in a different environment. Based on the cultural framework designed by Hofstede, four dimension are used – uncertainty avoidance, individualism vs. collectivism, masculinity vs. femininity and power distance. The data is collected from his personal website.

3.4. Control variables

Previous studies (Delis and Kouretas, 2011; Demirguc – Kunt et al., 2008; Drakos et al. 2016; Laeven and Levine, 2009) have identified several factors which can affect bank risk- taking and it is necessary to control these factors in order to have unbiased results. These factors can be divided into two main categories – endogenous and exogenous variables.

Endogenous variables are internal factors, within a bank, that might affect risk-taking.

The first control variable is the bank size. This control variable is calculated as the natural logarithm of the total assets. The size of a bank has a direct influence on risk-taking. Larger banks collect more deposits and thus, have more funds available to undertake risky investments.

Secondly, bank capitalization is an important bank-level variable which has to be taken into consideration. It is calculated as the ratio of common equity to total assets of banks.

A higher level of capitalization signals a more prudent behaviour by banks. This reasoning is

14 in line with the recent developments in terms of monetary supervision after the financial crisis. The Basel Committee has set up more restrictive rules regarding the quality and the quantity of risk-based capital requirements. Hence, a higher bank capitalization shows that bank owners have a higher “skin in the game” and it is less likely that they will increase risk- taking (Ashraf et al., 2016). Thirdly, bank profitability is another concern. However, it is difficult to predict the effect of profitability on the bank risk-taking behaviour. In good times, a higher level of risk assets might lead to a higher profitability. A part of this profit might be used to expand the loan portfolio. By contrast, in a period of financial distress, there will be an increase of non-performing loans leading to a decrease in bank profitability. Because profitability impacts the amount of risk taken by the bank next year, this control variable is lagged for one year. The ratio is calculated by dividing the pretax income to total assets.

Fourth, technically efficient banks are better able to cope with uncertainties and to manage risks. These banks have advanced risk management systems and superior capabilities in identifying and mitigating risks. The efficiency ratio is computed as the ratio of operating expenses to net revenues. A lower value of this ratio (i.e. a higher level of efficiency) denotes an increased ability of banks to turn their resources into revenues (Drakos et al., 2016;

Fiordelisi et al. 2011). The last endogenous variable consists of the level of non-traditional activities. Scholars argue that more and more banks expanded their portfolio in the recent past in order to provide non-traditional services. These type of activities such as asset securitization, trading derivatives or investment banking represent a shift from the traditional financial intermediation and rise the bank exposure towards risk (Lozano-Vivas and Pasiouras, 2010). Nonetheless, the lack of regulation regarding the off-sheet balance items enforced banks to increase its non-interest income coming from non-traditional activities (DeYoung and Torna, 2013). The level of non-traditional activities is obtained from the ratio of non-interest income to total assets.

Furthermore, it is also necessary to control for exogenous factors. These are external

factors which have an impact on bank risk-taking. In this respect, two country-level variables

are used: the macroeconomic outlook of the economy and the regulatory environment. The

first country-level variable accounts for the macroeconomic differences across countries and it

is based on the GDP growth rate. The state of the macroeconomic environment can affect the

activities of banks. In positive macroeconomic conditions, banks tend to increase their lending

to customers. Thus, it is expected that a higher GDP growth rate to be positively associated

with risk-taking behaviour measured through ratios of risk assets and non-performing loans.

15 It is not unfeasible to think that the recent financial crisis that devastated the world had been at least partly a result of flaws and drawbacks in the regulation and supervision of the banking system (Barth et al., 2013). Regulatory environment is a key aspect that needs to be controlled. Failing to control cross-country differences in terms of regulation and supervision will probably lead to an omitted – variable bias. Regulation plays an important role in insuring a sound banking system stability and has the goal to enhance the efficiency of financial intermediation. Based on the dataset

constructed by Barth et al. (2008, 2013) and on their extensive research in the field of banking supervision, I will employ three indexes to capture relevant aspects of the regulatory framework in each country. The authors’ database was built on four surveys sponsored by World Bank. Since the empirical study of this thesis is based on a long time frame, I used the average scores for every index based on the four surveys. A detailed overview on how each index is constructed can be found in the Appendix D.

The first index is the capital regulatory index which provides a guideline on the minimum amount of capital that banks must hold and also on the nature and source of the regulatory capital. The second index highlights the official supervisory power and aims to quantify the degree to which the country’s bank supervisory entity has the authority to take certain actions. The last index is the private monitoring index and reflects upon the market disclosure requirements set by the supervisory authority.

3.5. Descriptive statistics

The summary statistics for the all the variables included in this study is presented in Panel B of Table 1. The descriptive statistics provides an overview of the values taken by the variables. For instance, a mean value of 65% of risk assets indicates a moderately high amount of risk in the balance sheet of banks. Even though the extreme values have been omitted from the final sample, the high values for risk assets and non-performing loans are attributed mainly to banks from Greece and Cyprus.

Moreover, the descriptive statistics highlight an important evidence of the low interest rate environment. All the four employed variables for interest rates have extremely low

7 The dataset can be found in the following link: http://faculty.haas.berkeley.edu/ross_levine/Regulation.htm

16 values. Some of them have even negative values which draws attention to the abnormal situation of nowadays. The negative interest rates appeared as a consequence of the unconventional monetary policy practiced by the European Central Bank

. Another interesting aspect of descriptive statistics is that the mean value of bank capitalization is 9,2%. This is a promising value since it is higher than the 8% capital requirement recommended by the Basel Committee.

In addition to the descriptive statistics, I included in Panel B information regarding the origin of the banks from the sample. In total, there are 803 unique banks from 19 countries which cover three different important economic regions of the world. It should be mentioned here that the high number of U.S. banks in the sample compared to Euro area and Japan is explained by the difference in the primary source of financing in these regions. In United States, companies rely more on financing through capital markets than in the other two parts of the world.

The correlation coefficients between variables are reported in Table 2. The correlation matrix shows that the interest rates variables are highly correlated between them.

However, these variables will not be used simultaneously in the same regression. Thus, any multicollinearity issues are avoided.

8 The unconventional monetary policy of ECB consists of two strategies. The first step was to reduce the key interest rate below zero and the second one was the implementation of the asset purchase programme.

17

Panel A Panel B

Country Number of banks Variable Obs Mean Std. Dev. Min Max Kurtosis Skewness

Austria 7

Belgium 2 Risk-weighted assets 8491 0.654 0.164 0 0.999 4.951 -1.027

Cyprus 3 Non-performing loans 9182 0.023 0.027 0 0.206 12.102 2.583

Estonia 1 Bank-level lending rate 9182 0.068 0.028 0.004 0.310 10.435 1.365

Finland 3 Short-term interest rate 9182 1.314 1.588 -0.258 4.858 2.850 1.113

France 18 Long-term interest rate 9182 3.202 1.439 -0.074 5.398 2.008 -0.174

Germany 11 Central bank rate 9182 1.448 1.674 -0.100 5.250 2.987 1.170

Greece 7 Size 9182 14.841 2.182 9.513 21.948 3.140 0.752

Ireland 3 Capitalization 9181 0.092 0.045 -0.623 0.790 42.078 2.887

Italy 21 Profitability 9013 0.007 0.098 -8.329 0.091 6179.438 -76.044

Lithuania 1 Efficiency 9128 0.837 0.214 0.233 8.912 284.846 10.052

Luxemburg 1 Non-traditional activities 9176 0.011 0.010 -0.058 0.140 42.154 4.638

Malta 3 GDP growth 8695 1.582 1.872 -14.814 26.276 16.839 -0.643

Netherlands 3 Capital regulatory index 9182 6.321 0.528 4.75 8.75 7.872 -0.432

Portugal 5 Supervisory power index 9182 12.813 1.300 7 13.375 9.210 -2.638

Slovakia 4 Private monitoring index 9182 9.437 0.705 6.5 10 8.366 -2.476

Spain 8 Power distance 9182 42.898 7.941 11 100 13.702 1.787

Euro area 101 Individualism 9182 82.571 16.799 27 91 4.235 -1.698

Japan 88 Masculinity 9182 65.455 12.677 14 100 5.632 0.998

United States 614 Uncertainty avoidance 9182 54.912 17.489 35 100 3.536 1.537

Total 803

Table 1. Descriptive statistics

Note: for variables definition see Appendix B.

18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

1 Risk-weighted assets 1

2 Non-performing loans -0.154 1

3 Bank rate 0.237 -0.231 1

4 Short-term rate 0.076 -0.303 0.491 1

5 Long-term rate 0.258 -0.384 0.685 0.743 1

6 Central bank rate 0.108 -0.318 0.516 0.993 0.750 1

7 Size -0.342 0.271 -0.365 -0.165 -0.433 -0.209 1

8 Capitalization 0.195 -0.266 0.220 0.053 0.184 0.089 -0.339 1

9 Profitability 0.013 -0.039 0.010 0.013 0.020 0.013 0.023 -0.067 1

10 Efficiency -0.028 0.386 -0.003 -0.004 0.007 -0.012 -0.070 -0.111 -0.224 1

11 Non-traditional activities -0.041 0.018 0.085 0.040 0.051 0.037 0.182 0.002 0.031 -0.078 1

12 GDP growth 0.089 -0.240 0.079 0.129 0.161 0.143 -0.152 0.133 0.011 -0.243 0.013 1

13 Capital regulatory index 0.250 -0.288 0.428 0.142 0.400 0.193 -0.337 0.266 0.013 -0.038 0.019 0.182 1

14 Supervisory power index 0.364 -0.314 0.269 0.008 0.279 0.063 -0.561 0.276 0.008 -0.086 -0.161 0.190 0.455 1

15 Private monitoring index 0.349 -0.323 0.240 -0.008 0.265 0.050 -0.559 0.300 0.009 -0.095 -0.164 0.192 0.385 0.809 1

16 Power distance -0.350 0.258 -0.344 -0.106 -0.369 -0.156 0.451 -0.253 -0.009 0.040 0.040 -0.153 -0.447 -0.530 -0.529 1

17 Individualism 0.422 -0.385 0.522 0.165 0.530 0.236 -0.595 0.418 0.017 -0.078 0.003 0.199 0.591 0.554 0.670 -0.667 1 18 Masculinity -0.192 0.235 -0.460 -0.209 -0.445 -0.254 0.285 -0.275 -0.014 0.010 -0.152 -0.105 -0.535 -0.093 -0.125 0.327 -0.610 1 19 Uncertainty avoidance -0.453 0.371 -0.513 -0.156 -0.523 -0.228 0.621 -0.409 -0.017 0.077 0.016 -0.233 -0.624 -0.653 -0.710 0.439 -0.716 0.568 1

Note: for variables definition see Appendix B.

19

4. Empirical framework

The use of econometric analysis of panel data which integrates both time series and cross-sectional dimensions has become a frequent practice in modern empirical studies (Arellano, 2003). This thesis is not an exception. The econometrical model used in this thesis replicates the one used by Delis& Kouretas (2011) and it is represented by the following equation:

r

= α + β

ir

+ β

b

+ β

c

+ u

(1)

where r

represents the bank risk-taking for bank i at time t and it is proxied by risk weighted assets ratio and the non-performing loans ratio. This dependent variable is explained by ir

which is the interest rate for country i at time t. Moreover, this relationship is controlled for bank-specific control variables – b

and respectively for macroeconomic and regulatory variables – c

.

The empirical framework of this paper is built on three distinct steps. The first two steps involve a comprehensive analysis of the relationship between the interest rate environment and the amount of risk taken by banks. Later, in the third step the moderation effect of the cultural dimensions is included in the analysis in order to determine the effect of culture on the risk-taking behaviour. This last step reflects upon both managerial and international dimension of the thesis.

The first step represents an analysis of the final sample based on a static panel data model

which includes fixed effects or random effects. An important feature of fixed effects and random effects estimations is that it allows to control for unobserved heterogeneity. This is one of the key issues dealing with panel data. The phenomenon means that there might be direct effects of an unobserved (omitted) variable on both left- and right-hand side variables which can lead to a biased and inconsistent estimation (Wooldridge, 2002; Arellano, 2003). In this case, a source of bank risk-taking variability might be explained by the influence of an unobserved bank-characteristic variable. Hausman test specifies which of the two models would be more appropriate to use.

9 Note that different scholars might use a different naming to describe fixed effects and random effects estimations like variable-intercept models (Hsiao, 2014) or individual-specific models.

20 Subsequently, I employ a dynamic panel data model based on generalized methods of moments (GMM) to account for endogeneity

issues discussed by the related literature.

Endogeneity represents an important concern in the empirical finance research which can lead to biased and inconsistent parameter estimates. This happens because in practice it is usually difficult to find exogenous factors or natural experiments within the examined relations (Wintoki et al., 2012).

According to Delis & Kouretas (2011), there are two potential sources of endogeneity which can affect the results. The first one is the persistence of bank risk and the second source refers to endogeneity of independent and control variables (interest rate variables, profitability, capitalization etc.). Several theoretical reasons draw attention to the dynamic nature of bank risk. Firstly, persistence is caused by an intense competition environment, which tends to enhance bank risk-taking (Cordella and Yeyati, 2002). Secondly, the banking relationship with risky borrowers tends to be longer. Thirdly, banks might need a certain amount of time to adjust to the effects of macroeconomic shocks. Lastly, regulatory environment might determine bank risk to persist.

Therefore, the static panel data model fails to seize the persistence effect of bank risk.

A dynamic model which includes a lagged dependent variable is able to capture the dynamic nature of bank risk and to provide unbiased results. Such an econometric model is proposed by Arellano and Bover (1995) and Blundell and Bond (1998). The equation below is used in estimating the dynamic panel data model through generalized method of moments.

r

= α + δ(r

) + β

ir

+ β

b

+ β

c

+ u

(2)

δ is the coefficient on the lagged risk variable and can be seen as the speed of convergence to equilibrium. A value of δ equal to 0 indicates that the bank risk is characterized by a high speed of adjustment, while a value of the coefficient equal to 1 suggests that the adjustment is very slow. Values between 0 and 1 means that risk persists, but will eventually return to its normal level. Ultimately, δ takes negative values only if convergence to equilibrium cannot be achieved, which probably suggests an error with the sample (Delis & Kouretas, 2011).

10 Endogeneity issues arise when a predictor variable is partly determined by factors within the model (Zohoori and Savitz, 1997). In other words, an explanatory variable is endogenous when it is correlated with the error term.

21 In an expository article which links the econometric principles with Stata commands, Roodman (2009) briefly describes several arguments why the GMM estimator should be used for panel data analyses and what are the underlying assumptions. First, this technique is designed for situations with “small T and large N” panels, meaning a few periods and many observations. Even though, in this thesis the number of periods is not low, it is definitely outnumbered by the large number of bank-year observations. Secondly, GMM should be considered when the study process is dynamic, meaning that current realizations of dependent variable are influenced by the past ones (e.g. economic shocks, past performance etc.).

The model presents an important feature which includes the lagged value of the dependent variable in the right-hand side of the equation, among regressors. Finally, GMM is useful in solving endogeneity issues arising from the correlation between independent variables and error term.

The last step of methodology deepens the behavioural dimension of this thesis. This part involves the analysis of moderation effect of the cultural dimension. The main objective of this analysis is to investigate whether country-specific cultural dimension strengthens or weakens the relationship between interest rates and bank risk-taking.

In order to investigate the moderating effect of the cultural dimension, the following equation has been employed:

r

= α + β

ir

+ β

national cultural measures

+ β

(β

ir

national cultural measures

)

+ β

b

+ β

c

+ u

(3)

Compared to equation (1), two additional factors have been introduced in the regression (3).

Namely, the first additional factor measures the direct effect of the cultural dimension on bank

risk-taking. The second factor represents the moderator factor and it is constructed as an

interaction term between the interest rates and the cultural dimension variables. All the other

factors from equation (3) have the same meaning as specified above in equation (1).

22

5. Empirical results and discussion

This section provides an extended view about the results obtained from the econometric analysis and a discussion based on these results is made. Before starting the analysis, it is important to mention that the pooled OLS model does not represent an accurate estimation for the panel data. It is a restrictive model which implies homogeneity among the variables used in regression. Homogeneity assumes identical behaviour and similar characteristics across banks. Based on this argument, I consider that homogeneity assumption is forced. As a consequence, the pooled OLS regression model is dropped out from the analysis. I oriented the empirical analysis towards fixed effects and random effects models.

5.1. Fixed effects and random effects regressions

A more accurate analysis of panel data is provided by the fixed effects and random effects models. These two models account for unobserved heterogeneity which consists of individual-specific effects over the explanatory variables. The presence of an unobserved variable is an important issue that the econometric study has to deal with. On one hand, if the unobserved variable is not correlated with any of the independent variables then this simply

implies an omitted variable problem. In other words, there is an unobserved factor (e.g. CEO compensation, corporate governance) which can affect both dependent and/or

independent variables, but it is not included in the regression. On the other hand, if the unobserved variable is correlated with one of the independent variables this is also problematic (Wooldridge, 2002).

A discussion centered on whether to choose fixed effects or a random effects model revolves around the assumptions made regarding the unobserved effect. The main assumption behind the random effect is that the unobserved factor is random and thus, this is not correlated with any independent variable from the regression. In contrast, the rationale for fixed effects analysis is that it allows for an arbitrary correlation between the unobserved factors and the independent variables (Hsiao, 2004; Wooldridge, 2002).

In a more detailed manner, the fixed effects model allows for unobserved

heterogeneity between the banks from the sample by allocating a specific intercept to each

bank. Thus, in the fixed effects model, the intercept may differ across banks, but the intercept

23 does not vary across time. However, a side effect of this model is that it removes any time-invariant components from the model. Hence, the regression between interest rates and the risk behaviour of banks cannot be controlled for regulatory environment. On the other side, the random effects model has a common value for the intercept and it is assumed that the error term is not correlated with any variable.

I performed the Hausman test in order to determine which one of the two models is more appropriate. The null hypothesis of this test means that the random effect is more appropriate to use, while the rejection of the null hypothesis indicates the use of the fix effect.

After running the fixed effects model and the random effects model, the conclusion of Hausman test is that the fixed effects model is more suitable to the banking sample.

In Table 3 the empirical results of fixed effects model are presented. These results are obtained by estimating equation (1) in Stata software and overall indicate that the hypothesis of a negative relationship between interest rates and bank risk-taking is supported. This conclusion holds especially in the regression models in which bank risk-taking is explained by non-performing loans variable (i.e. regressions V-VIII). In this case, the coefficients for all interest rates are negative and highly significant, suggesting that a decrease in the interest rates lead to an increase in the level of non-performing loans of banking industry. However, the first hypothesis is partially supported by the regression models in which risk-weighted assets is the dependent variable (i.e. regressions I-IV). Only the bank-level lending rate presents a negative and significant effect on the dependent variable. The other three interest rate variables – short-term interest rate, long-term interest rate and central bank rate illustrate a positive and significant relationship with the bank risk-taking when risk-weighted assets ratio is used as proxy. Hence, an increase in short-term interest rate, long-term interest rate and central bank rate determine an increase in the amount of risk taken by banks, fact which is inconsistent with the theory.

Another argument which supports the first hypothesis is that the coefficient value for

bank-level lending rate is negative and significant in both regression models. Bank-level

lending rate represents the key interest rate in this study because it increases significantly the

number of interest rate observations and also highlights the pass-through effect of bank-level

interest rates to customers (Delis & Kouretas, 2011). In other words, it shows the effective

interest rate used by each bank from the sample in every year.

24 The results of this thesis differ from the most related article written by Delis &

Kouretas (2011). They obtained a more robust evidence of the negative impact of low level of interest rates on bank risk-taking. Their results present negative and significant coefficients across all interest rates for both dependent variables (i.e. risk assets and non-performing loans). By contrast, this thesis finds a robust evidence for a negative and highly significant relationship between bank risk-taking and the low interest rate environment only when the non-performing loans are used as proxy for the risky behaviour of banks. However, these differences in obtained results are normal due to the fact that this study employs a different sample of banks from more countries, and over a longer time horizon.

An interesting result is represented by the effect of (lagged) profitability on bank risk- taking. The coefficient of profitability is positive and significant for all regression models.

From a behavioural aspect, this implies that an increase in profitability triggers banks to take on more risk, whereas from management perspective, it means that an increase in the profit will determine banks to grant even more risky loans next year. Moreover, the effect of economic growth on bank risk-taking is captured by the coefficients of GDP growth. In contrast to what theory predicts, the sign of coefficients indicates a negative relationship, meaning that banks increase their lending when the outlook of the economy is not optimistic.

The outcomes of the fixed effects model are useful in investigating the risk behaviour

of banks, but should be interpreted with caution. Fixed effects model represents a static panel

data analysis. In consequence, this model fails to explain the dynamic nature of bank risk-

taking behaviour and to resolve potential endogeneity problems which can arise from right-

hand side variables of equation (1). Therefore, in order to fix these potential issues described

by the literature (Delis and Kouretas, 2011; Drakos et al. 2016) a dynamic panel data

methodology is implemented and its results and conclusions are discussed in detailed manner

in the next section.

25

FIXED EFFECTS I II III IV V VI VII VIII

VARIABLES RWA RWA RWA RWA NPL NPL NPL NPL

Bank-level rate -0.514*** -0.133***

(0.076) (0.014)

Short-term rate 0.010*** -0.005***

(0.007) (0.000)

Long-term rate 0.016*** -0.007***

(0.001) (0.000)

Central bank rate 0.010*** -0.004***

(0.001) (0.000)

Size -0.009** 0.014*** 0.025*** 0.013*** 0.009*** 0.005*** -0.001 0.005***

(0.004) (0.004) (0.004) (0.004) (0.001) (0.001) (0.001) (0.001) Capitalization -0.067 -0.081* -0.040 -0.085* -0.076*** -0.068*** -0.084*** -0.066***

(0.049) (0.048) (0.048) (0.048) (0.009) (0.009) (0.009) (0.009) Profitability 0.043*** 0.034** 0.035** 0.035** 0.006* 0.008*** 0.008*** 0.007***

(0.014) (0.014) (0.014) (0.014) (0.003) (0.003) (0.003) (0.003) Efficiency -0.021*** -0.018*** -0.019*** -0.019*** 0.046*** 0.044*** 0.044*** 0.044***

(0.006) (0.006) (0.006) (0.006) (0.001) (0.001) (0.001) (0.001) Non-trad. activities -0.620*** -0.300 -0.310 -0.311 0.371*** 0.234*** 0.234*** 0.244***

(0.224) (0.223) (0.223) (0.223) (0.042) (0.040) (0.040) (0.040) GDP -0.004*** -0.005*** -0.004*** -0.005*** -0.001*** -0.001*** -0.001*** -0.001***

(0.001) (0.001) (0.001) (0.001) (0.000) (0.000) (0.000) (0.000) Constant 0.866*** 0.466*** 0.269*** 0.479*** -0.133*** -0.075*** 0.029*** -0.084***

(0.060) (0.056) (0.063) (0.056) (0.011) (0.010) (0.011) (0.010)

Observations 7,671 7,671 7,671 7,671 8,485 8,485 8,485 8,485

R-squared 0.097 0.062 0.062 0.085 0.339 0.302 0.302 0.291

Number of banks 803 803 803 803 803 803 803 803

Fixed effects YES YES YES YES YES YES YES YES

Table 3.

Fixed effects model

Note: Standard errors are presented in parantheses. The statistical significance is indicated with ***,

** and * , which respresents that the significance is lower than 1%, 5% and 10% respectively.

26

5.2. Dynamic model analysis

A panel GMM estimator controls for the dynamic nature inherent in bank risk variables by including a lagged dependent variable as a predictor. Moreover, it accommodates potential sources of endogeneity. I estimate equation (2) using the two step Arellano- Bover/Blundell-Bond GMM model. The dependent variables are the same as in the fixed effects model, risk-weighted assets ratio and non-performing loans ratio. Besides the lagged dependent variable, the novelty of this method is that it distinguishes between three different types of variables in the right-hand side of equation – endogenous, exogenous and predetermined variables. The following variables entered in the equation as endogenous variables: interest rate variables, capitalization, profitability, efficiency and the non-traditional activities of banks. The only exogenous variable in this model is the macroeconomic outlook of the economy proxied by the GDP growth. In addition, size and regulatory environment indexes (regulatory capital index, official supervisory power index and private monitoring index) are considered predetermined variables. Predetermined variables are the ones that are not strictly exogenous. Furthermore, according to Roodman (2009) and Wintoki et al. (2012), I employed a standard treatment of one lag for predetermined variables and two lags for the endogenous variables

.

The results of the dynamic model are presented in Table 4. If we specifically look to the relationship between interest rates and bank risk-taking variables, the results are highly significant and in line with the ones obtained by using the fixed effects model. Although several assumptions have been added to the GMM model, this provides a robustness test for the results established in the fixed effects model. Thus, the main conclusion is that, in general, the low interest rate environment triggers banks to increase their risk. This conclusion holds especially when the non-performing loans ratio is used to proxy the bank risk. In this case, all the four interest rates variables indicate a highly significant and negative relationship (see Table 4, regressions V-VIII). However, the conclusion is weakened when risk-weighted assets ratio is employed to measure bank risk because the bank-level lending rate is the only interest rate which posits a significant and negative relationship (see Table 4, regression I).

11 Except profitability variable which has been lagged only once, because it is already lagged by construction.