Why is it always the same? : Risk culture as a predictor for bank market performance during crises

(1)

0

Why is it always the same?

Risk culture as a predictor for bank market performance during crises

Master’s Thesis

MSc Finance – Corporate Finance and Banking Amsterdam Business School, University of Amsterdam

Author: Daniel Goudswaard, BSc. Student number: 10528709

Thesis Supervisor: Razvan Vlahu, Ph.D.

(2)

1

Statement of originality

This document is written by Daniel Goudswaard who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Acknowledgements

I’d like to thank my thesis supervisor Mr. Razvan Vlahu, Ph.D. for supporting me with my work. I would also like to thank my family and friends who supported me throughout my study and while writing this thesis. To all of you, thank you for your help.

(3)

2

Appendix list

Appendix A. Variable definitions and sources ... 54

Appendix B. Dates and descriptions of crisis events ... 57

Appendix C. Descriptive statistics of small banks ... 58

Appendix D. Descriptive statistics of medium sized banks ... 59

Appendix E. Descriptive statistics of large banks ... 60

Appendix F. Descriptive statistics of very large banks ... 61

Appendix G. Descriptive statistics of Europe and Central Asia ... 62

Appendix H. Descriptive statistics of the United States ... 63

Appendix I. Descriptive statistics of “other” countries/regions... 64

Appendix J: Regional distribution of banks ... 65

Appendix K: 2008 Crisis Return on 1998 Crisis Return – small banks ... 66

Appendix L: 2008 Crisis Return on 1998 Crisis Return – medium sized banks ... 67

Appendix M: 2008 Crisis Return on 1998 Crisis Return – large banks ... 68

Appendix N: 2008 Crisis Return on 1998 Crisis Return – very large banks ... 69

Appendix O: 2008 Crisis Return on 1998 Crisis Return – small & medium banks ... 70

Appendix P: 2008 Crisis Return on 1998 Crisis Return – large & very large banks ... 71

Appendix Q: Regressions of MES on Risk Culture proxies - 1/6 ... 72

Appendix R: Regressions of MES on Risk Culture proxies - 2/6 ... 73

Appendix S: Regressions of MES on Risk Culture proxies - 3/6 ... 74

Appendix T: Regressions of MES on Risk Culture proxies - 4/6 ... 75

Appendix U: Regressions of MES on Risk Culture proxies - 5/6 ... 76

Appendix V: Regressions of MES on Risk Culture proxies - 6/6 ... 77

Appendix W: Regression of MES on five Risk Culture proxies - Region Fixed-Effects ... 78

(5)

4

1. Introduction

Well-functioning banks are essential for our financial system. But as crises will not cease to exist, ensuring that banks can weather crises is key, not only for the banks themselves, but also regulators to protect the stability of the financial system and ultimately society. (Sundararajan and Baliño, 1991).

Since the 2007-2009 financial crisis the interest in the concept of risk culture has significantly grown. Since then it has been called the most fundamental tool for effective risk management (IIF, 2009). According to Power, Ashby and Palermo (2013) risk culture is not a static thing but rather a continuous mixture of both formal and informal processes. They state that prior to the 2007-2009 financial crisis many organizations were unaware of the concept of risk culture or indifferent to its balance between risk-taking and control.

The risk culture hypothesis from Fahlenbrach, Prilmeier and Stulz (2012) states that some banks have a persistent risk culture which causes them to have an inherently larger exposure to systemic risk. This means that banks that performed poorly in a previous crisis, will also perform poorly in new crises due to the persistence of risk culture. They find evidence for the risk culture hypothesis in their sample of US banks and relate it to greater reliance on short-term funding, higher leverage and greater growth than other banks before the crises. They conclude that banks that are most negatively affected during a crisis do not learn from the experience. They do not alter their business model or become more cautious regarding their risk culture. Fritz-Morgenthal, Hellmuth and Packham (2016) developed a score to assess risk culture and find evidence that better ECB stress test results correspond to a better risk culture for European financial institutions.

However, Weiß, Bostandzic and Neumann (2014) find no evidence that bank size, leverage, non-interest income are persistent determinants of systemic risk. Their study on a global sample of banks contradicts the risk culture hypothesis and finds that global systemic risk is predominantly driven by characteristics of the regulatory regime.

This study is most closely related to the paper of Fahlenbrach et al. (2012) who first hypnotized and found evidence for the risk culture hypothesis. Similar to Beltratti and Stulz (2012) they analyze the factors that contributed to the poor performance of banks during the 2007-2009 financial crisis and find a relation with the 1998 Russian/LTCM

(6)

5

crisis. They find that banks that relied more on short-term funding, had more leverage, and grew more in size, respectively loans, are more likely to perform poorly during both crises. They employ the same measure of bank performance, buy-and-hold returns, which I use as a dependent variable in the first part and the third part of my analysis. I use similar explanatory variables, but instead of focusing on the effect of the 2008 financial crisis on U.S. banks, I will investigate five crisis periods on a global sample of banks. My study is related to the studies of Fahlenbrach et al. (2012), Bartram, Brown and Hund (2007) and Weiß et al. (2014) in my choice of the investigated crisis periods.

This study is also closely related to the study of Weiß et al. (2014) on the factors that determine systemic risk during crises. In the second part of my study the choice of dependent variable for the Marginal Expected Shortfall from Acharya Pedersen, Philippon and Richardson (2010) is related to the study from Weiß et al. (2014), as are control variables. In the second part I use marginal expected short fall to test the relation of risk culture proxies on a bank’s equity tail risk. Furthermore, this study is also closely related to the paper of Fritz-Morgenthal et al. (2016) in which they investigated the relation between risk culture proxies and EU stress test performance. Their framework of mostly qualitative risk culture proxies served as a reference point for me to identify and group quantitative risk culture proxies.

This study consists of three parts. In the first part I investigate the risk culture hypothesis from Fahlenbrach et al. (2012). I find evidence for the risk culture using a global sample of banks and confirm that the returns of the 1998 Russian crisis predict bank performance during the 2007-2009 financial crisis. The evidence from the regional samples of banks provide less conclusive evidence. The results per region do not differ much in their support for the risk culture hypothesis, for all the results are somewhat ambiguous. The second part of this study is dedicated to testing risk culture proxies on their relation to equity tail risk as dictated by the risk culture hypothesis. I find that five from a total of eighteen risk culture proxies from literature are associated with equity tail risk. I select these risk culture proxies for further testing the third part. In the third part I test the five selected risk culture proxies on their predictive ability to poor bank performance during crises. I find there are three risk culture proxies that consistently predict poor bank performance during crises on the global sample of banks; asset growth, tier 1 ratio and performance pay. The results are robust to multiple specifications and time periods. However, since the results of the regional samples are

(7)

6

somewhat ambiguous I recommend future studies to focus regional risk culture factors. A more narrow scope could shed more light on the regional differences between risk cultures and provide more insight on its proxies. Bank executives, regulators and investors as indicators for a banks risk culture. My hope is that bank executives, investors, regulators and policy make use of risk culture proxies to address banks’ risk cultures. This could prevent bank failure, decrease the need for state aid or perhaps even prevent financial crises altogether.

2. Literature Review

In the following section, I briefly discuss the related theoretical and empirical literature on the determinants of a bank’s performance during crises.

There are many theories about the determinants of banks performance during crises. The vast majority of research has focused on the bank’s business models and its characteristics. For instance, Adrian and Shin (2010) show that broker-dealers increase leverage during credit booms. This may be because they have the best business opportunities during these booms, but it also makes them more vulnerable to market circumstances when the boom ends. Furthermore, Loutskina and Strahan (2011) show more geographically concentrated mortgage lending banks performed better than their less geographically concentrated counterparts during the global 2007-2009 financial crisis. They explain this by the bank’s better use of private information in making loans.

Beltratti and Stulz (2012) also investigated the global 2007-2009 crisis and find evidence that banks that are financed with short-term capital market funding are more sensitive to crises. They find that banks that had less leverage and lower than average returns right before the crisis performed better than banks with more leverage and higher returns before the crisis. Fahlenbrach et al. (2012) have similar outcomes when they investigate the global 2007-2009 crisis in relation to the 1998 Russian crisis. They find that if banks relied more on short-term funding, had more leverage, and grew more in size they were more likely perform bad in both crises.

The learning hypothesis from Malmendier and Nagel (2011) states that past experiences of executives and investors affect their subsequent behavior and performance. They learn from past experiences and will thus avoid behavior that previously led to negative outcomes. The same mechanism is supposed to hold for

(8)

7

organizations such as banks. According to Gennaioli, Shleifer, and Vishny (2012) a crisis could reduce a bank’s risk appetite and lead it to assess payoff probabilities differently. This means that if a bank performs poorly during a crisis it learns from the experience and performs better in the subsequent crisis.

The main finding of Fahlenbrach et al. (2012) goes against this as they find evidence that bank’s stock return performance in the 1998 Russian crisis predicts the probability of failure during the global 2007-2009 crisis. They state this is supportive of the so called risk culture hypothesis; that there is persistence in risk culture or aspects of the business model that make a bank inherently sensitive to crises. This increase in equity tail risk means that a bank that performed bad during a previous crisis is also likely to perform bad during a subsequent crisis. In their sample of 138 US banks Fahlenbrach et al. (2012) find that for each percentage point of loss in market value in the Russian crisis, a bank lost an annualized 0.66 percentage points during the 2007-2009 financial crisis. They find that the predictive power is concentrated in large banks which is consistent with Power et al. (2013) that find that larger firms have an inherently more persistent and harder to adjust risk culture due to their complexity. If bad bank performance is indeed driven by risk culture during times of financial crises, bank executives, investors, regulators and policy makers could design strategies based on risk culture to prevent bank failure during crises or perhaps prevent financial crises altogether.

Martynova, Ratnovski & Vlahu (2014, 2015) argue that banks with a high franchise value or a more profitable core business may have higher risk-taking incentives. A higher franchise value enables them to take risk on a larger scale and off sets lower incentives to take risk of fixed size. This would mean that banks with a higher franchise value may have higher risk-taking incentives. They find the effect is stronger when a bank uses inexpensive senior funding to increase its balance sheet, and when it can increase leverage due to more protection of creditor rights. They argue similarly that a more profitable core business enables banks to borrow more and take risk on a larger scale.

Their findings are supported by those of Bostandzic, Pelster and Weiss (2014). They show that bank size are associated with an increase in global systemic risk, but find no convincing evidence that non-interest income influences systemic risk. Perotti and Suarez (2009) identify the scale and speed of liquidity runs as the primary cause of

(9)

8

propagation of a crisis. They argue banks that rely excessively on short-term uninsured funding to increase systemic risk. Weiß et al. (2014) use a global sample of banks to investigate the determinants of systemic risk. They find that most crises are characterized by significant increases in moderate systemic risk, but contrary to previous research they find no evidence that bank size, leverage, non-interest income or the loss provisions are persistent determinants of systemic risk. Acharya, Santos and Yorulmazer (2010) describe systemic risk as a negative externality from a bank’s failure on other institutions and the economy. If banks are not regulated to maintain efficient levels of systemic risk widespread failure of banks can occur with significant costs for society as a result. However Benoit, Colletaz, Hurlin and Pérignon (2013) find that the Marginal Expected Shortfall from Acharya et al. (2010) as used by Weiß et al. (2014), falls short on capturing systemic risk. Marginal Expected Shortfall is a widely used measure to capture systemic risk but it actually captures a bank’s sensitivity to market crashes, a bank’s equity tail risk.

Berger and Bouwman (2013) focus on how capital affects bank performance during crises. They find that for small banks capital helps increase the probability of survival and increase market share, both during crises and during normal periods. For medium and large banks capital primarily enhances performance during the 1990-1992 credit crunch and the 2007-2009 financial crisis. However, Perotti, Ratnovski and Vlahu (2011) find that from a theoretical perspective banks that are forced to have a higher regulatory coverage ratio, may be incentivized to take more risk. This would be because they do not internalize the negative realizations of tail risk projects. Kleinow and Moreira (2016) find evidence for this in the investigation of systemic risk in Europe as they find a bank’s systemic risk contribution seems to be driven by Tier 1 capital.

Next to a bank’s business model and its characteristics other drivers of bank performance are also investigated. Barth, Caprio and Levine (2013) state that because banks have such an important function they are surrounded by an apparatus of political, legal, cultural and technological forces. I focus on the legal, political and cultural aspects as these have a broad literary base behind them and are relatively easy to identify. They are loosely represented by studies on bank laws and regulation, corporate governance, and risk culture.

La Porta, Lopez-de-Silanes, Shleifer and Vishny (1998) examine the differences in legal protection for investors between countries and from this construct an anti-director

(10)

9

index. In a subsequent paper Djankov, La Porta, Lopez-de-Silanes and Shleifer (2008) revise the anti-director index and find additional evidence for a link between the differences in legal system and the differences in financial performance. Because they see corporate self-dealing the main problem of governance in most countries they furthermore create the anti-self-dealing index. Since the effectiveness of law to deal with self-dealing is such a fundamental element of shareholder protection they state index should be preferred in cross-country empirical work.

Next to legal protection mechanisms for investors, the law also protects depositors through deposit insurance schemes and society as a whole through banking regulations. Bostandzic et al. (2014) show that deposit insurance schemes that give less certainty and put more financial risk on banks and depositors increase systemic risk. However, they find no convincing evidence that a bank’s supervision is a determinant for systemic risk. Contrary to this Weiß et al. (2014) do find evidence that that the regulatory regime drives global systemic risk.

Where laws and regulations shape the environment banks are allowed to act in, Shleifer and Vishny (1997) state that corporate governance is the mechanism through which investors try to ensure bank executives act in their interest. Laeven and Levine (2008) investigated the relation between bank regulations, corporate governance and risk taking in banks. They find that a stronger corporate governance structure reduced the variation in risk taking between banks. Furthermore they find that the relation between bank risk taking and regulations depends critically on the bank’s ownership structure. They find evidence that banks with powerful shareholders tent to take greater risks and are more susceptible to moral hazard problems. However, they also find that this relation diminishes if there are stronger shareholder protection laws are in place. Beltratti and Stulz (2012) find that in countries with more strict banking regulation the large banks did perform better and decreased loans less during the 2007-2009 financial crisis. However, generally speaking the differences in banking regulations between countries have no relation to the performance of banks during the crisis.

De Haan and Vlahu (2016) review the empirical literature on the corporate governance of banks. They highlight that banks are special because of their regulations, capital structure, and complexity and opacity of their business and structure. For instance, according to Aebi, Sabato and Schmid (2012) in non-financial firms a large board is considered not to be in the interest of shareholders, but for banks performance

(11)

10

and board size are positively related. Aebi et al. (2012) also find that if a bank’s CRO directly reports to the board of directors the bank performs significantly better during the 2007-2009 financial crisis. De Haan and Vlahu (2016) find mixed results concerning whether financial expertise of board members may be important. This is despite the fact that informational asymmetries are more pronounced for banks due to their opacity and complexity. Next to the governance mechanisms of the board De Haan and Vlahu (2016) focus on the mechanisms of ownership structures and executive compensation. They suggest to give special attention to the role of compensation and ownership of insiders and their interaction on risk-taking incentives. They also find most studies agree that when banks have concentrated ownership by government the bank’s governance and bank performance are generally negatively impacted. They conclude with three explanation for the mixed results they found; firstly the differences in time period covered, secondly, the often ignored interaction between corporate governance variables and thirdly, the variation in national regulations and governance systems.

O’Reilly and Chatman (1986) define culture as a system of shared values and norms that define appropriate attitudes and behaviors for organizational members. According to Power et al. (2013) risk culture is not a separate kind of thing to culture. They see it as culture with a focus on the concerns its poses to risk-taking and risk control activities. Fritz-Morgenthal et al. (2016) developed a score to assess risk culture based on nine risk score indicators; Regulatory requirements, Business strategy, Governance, Portfolio, Employees, Risk strategy, Reputation, Other effects and Cultural indicators. They composed these risk score indicators of publicly available information and tested them in relation to ECB stress test indicators and find evidence that better stress test results correspond to a better risk culture for European Financial institutions.

Since there are large cultural differences between regions it seems logical to assume there are also regional differences in risk culture. For instance Albert (1991) finds that in Europe there is a broader focus on society, this stems from a more “social” model of capitalism. Contrary to this is the US, there they have the more individualistic Anglo-Saxon model of capitalism; here the focus is more singularly and directly on the shareholder. These cultural differences can also be seen in laws and regulations. Generally speaking, in Europe there are more laws and regulations for firms, so one it could be that these regulations already lower risk appetite, thus diminishing the basis for the risk culture hypothesis. However, at the time of writing European banks are

(12)

11

doing worse than their US counterparts and did not fare better during the latest crisis either. This might point to an even more persistent risk culture compared to the US.

Furthermore, previous research about bank’s business models and it characteristics could very well be explained by risk culture. Fahlenbrach et al. (2012) stated there is no clear separation between the impact of a bank’s business model and its characteristics and the impact of a bank’s risk culture. For instance, the study from Adrian and Shin (2010) showing that broker-dealers increase their leverage in good times. They state this may be because of better business opportunities during credit booms which makes them more vulnerable to crises. However, the fact that the banks are willing to be more exposed to crisis risks can differ and could be explained because of differences in risk culture. Furthermore, Loutskina and Strahan (2011) explained their finding of better performance of more geographically concentrated mortgage lending banks by their greater use of private information. But it could very well be that the more concentrated mortgage lending banks have a more protective risk culture compared to the less concentrated banks. They might not have been willing to take the on risk of losing their private information advantage at the cost of growing rapidly.

Fahlenbrach et al (2012) tested the risk culture hypothesis in the US by explaining the return of the 2007-2009 financial crisis with the return of the 1998 Russian crisis. To confirm that risk culture is the driving force behind the relation risk culture indicators should be identified that predict this relation. Fritz-Morgenthal et al. (2012) have attempted to do this, they identify risk culture indicators and explain stress test performance of European banks. However, their indicators are mostly manually collected, subjective and qualitative in nature. They cannot be easily replicated and further tested. First, I investigate the evidence of the risk culture hypothesis on a global sample of banks. From the risk culture hypothesis it follows that there is a relation between risk culture proxies and equity risk. Secondly, I test this relation to select quantitative risk culture proxies identified from literature. Thirdly, I test the selected risk culture proxies on their relation to poor bank performance during crises. I contribute to literature by finding evidence for the risk culture hypothesis as a global phenomenon and by identifying three proxies for risk culture that indicate an increased chance of poor bank performance during crises. The results can be used by executives, regulators and policy makers to design strategies based on risk culture to prevent bank failure during crises or perhaps even prevent crises before they can occur.

(13)

12

3. Methodology

In this section I first discuss the three parts of this study, secondly, I define the main dependent and independent variables used in my regressions, and thirdly, I present the empirical approach of my analysis.

This study consists of three parts, first I investigate if there is international evidence for the risk culture hypothesis from Fahlenbrach et al. (2012). To do this I follow Fahlenbrach et al. (2012) and estimate a bank’s buy-and-hold return during the 2008 financial crisis period using the return from the 1998 Russian crisis as the explanatory variable using a global sample of banks. Since Fahlenbrach et al. (2012) found there is evidence for the risk culture hypothesis in the Unites States and this a large part of my sample, my first hypothesis is that I find evidence for the risk culture hypothesis in my global sample.

In the second part I investigate proxies for risk culture and test whether they follow the risk culture hypothesis. Following the evidence of the first part of my study I assume the risk culture to be true and use this to asses risk culture proxies from literature. The risk culture hypothesis states that banks have persistent risk cultures that make some banks inherently more exposed to equity tail risk. I use Marginal Expected Shortfall as proposed by Acharya et al. (2010) as a dependent variable to assess and select risk culture proxies on their explanatory power on a bank’s equity tail risk. This leads me to my second hypothesis; risk culture proxies explain a bank’s equity tail risk as measured by marginal expected shortfall.

In the third and last part I test whether the risk culture proxies also predict poor bank performance during crisis periods. I select the risk culture proxies that test positive on the second hypothesis and use the buy-and-hold return as a dependent variable to determine whether the selected risk culture proxies predict poor bank performance during crises. This brings me to my third and main research hypothesis: poor bank performance during crises can be explained by selected risk culture proxies.

3.1 Dependent Variables: measures of bank performance

The first dependent variable I use, in the first and third part of this study, is a bank’s buy-and-hold return during crises. I use the buy-and-hold return adjusted for cash dividends and stock splits as a measure for a bank’s performance during the different

(14)

13

crises. This variable measures the stock performance of a bank, but Fahlenbrach et al. (2012) find that a bank’s buy-and-hold return also has a strong relation with a bank probability of failure during the 2007-2009 financial crisis. This is in consistent the efficient market hypothesis that market efficiency causes all relevant information to be incorporated into share prices.

In the second part of this study I employ a measure of a bank’s equity tail risk as the dependent variable, the Marginal Expected Shortfall from Acharya et al. (2010). I use Marginal Expected Shortfall to assess whether risk culture proxies as given by literature explain a banks sensitivity to crises as follows from the risk culture hypothesis. Marginal expected Shortfall is defined as the mean net equity return of a bank during times of a market crash. I follow Acharya et al. (2010) in their proposal to use the return of each bank during the 5% worst days of the market of the year as is done by other studies. I calculate the Marginal Expected Shortfall for each in the sample for this period using the DataStream World Bank Index. For banks in Asia and the Pacific, Europe and Central Asia and North America I also calculate regional Marginal Expected Shortfall to test if risk culture is a region specific phenomenon. I do this using the DataStream Far East Bank Index, Dow Jones Global Index Europe Banks, and the FTSE North America Index Banks as the relevant market proxies, respectively.

𝑀𝐸𝑆

_5%𝑖

=

1 #𝑑𝑎𝑦𝑠

∑

𝑅

𝑡

𝑖 𝑡:𝑠𝑦𝑠𝑡𝑒𝑚 𝑖𝑠 𝑖𝑛 𝑖𝑡𝑠 5% 𝑡𝑎𝑖𝑙

(1)

Marginal Expected Shortfall is a widely used measure to capture systemic risk but it actually captures a bank’s sensitivity to market crashes, a bank’s equity tail risk. As Acharya et al. (2012) show equity tail risk is an important part of systemic risk, but critics such as Benoit et al. (2013) find that it falls short in capturing the multiple facets of systemic risk. In this study I am interested in the sensitivity to market crashes so this will have no impact on the analysis. However I do find it important to note that I follow a growing body of literature (e.g. Acharya et al, 2012; Benoit et al., 2013) and will not use the term systemic risk for the measure Marginal Expected Shortfall to prevent any misinterpretation. I will use the term equity tail risk, but it will be the same measure that is called systemic risk in other studies (e.g. Acharya et al., 2010; Weiß et al., 2014).

(15)

14

3.2 Main Variable of interest: Risk Culture

The main variable of interest is risk culture. Risk culture is not a static thing but rather a continuous mixture of both formal and informal processes and therefore hard to define and measure. I use the framework from Fritz-Morgenthal et al. (2016) as guidance to determine quantitative risk culture proxies based on nine risk score indicators; Regulatory requirements, Business strategy, Governance, Portfolio, Employees, Risk strategy, Reputation, Other effects and Cultural indicators. For each of these categories I identify potential quantitative proxies for risk culture after which I regroup the indicators into three categories. For clarification, with a positive relation to risk culture I mean that I expect the willingness of a company to take risks decreases and subsequently performance during crises to increase, while a negative relation to risk culture means I expect willingness to take risk increases and subsequently performance during crises decreases.

3.2.1 Business

The first variables that reflects a banks risk culture are business characteristics. Four are expected to have a negative relation to risk culture, while for two variables, loss provisions and Tier 1 ratio, the sign of the relation is not clear. I expect higher asset growth, leverage, percentage of non-interest income and short-term funding to all be related to and increased willingness to take on risk. High asset growth could indicate that a bank wants to grow at the risk of being unable safely manage the growth, the willingness to do this indicates a more risky risk culture. A high leverage indicates that the bank is very dependent on the state of the economy. If the economy is doing well, the bank is likely to profit more, but if the economy is doing poorly it is more likely to profit less or even make loss compared to a less leveraged bank. The willingness to take on this extra sensitivity to the performance of the economy indicates a more risky risk culture. If the percentage of noninterest income increase this means the bank has a relatively larger percentage of unsteady income. Choosing to be exposed to the risk of losing this type of income indicates a more risky risk culture. Finally, an increased reliance on short-term funding can cause problems for a bank in case liquidity dries up. The willingness to take this risk indicates there is a more risky risk culture.

Following the reasoning from Weiß et al. (2014) loss provisions can have a negative relation to a bank’s risk culture in that they can signal poor quality of loans and thus the willingness of the firm to sell these loans and take more risk. Contrary to this, I pose

(16)

15

that the variable loss provisions can also be positively related to risk culture in the sense that it can be a choice of management to take larger precautionary measures and thus loss provisions. This could be the reason for the non-significance of loss provisions Weiß et al. (2014) found in their analysis.

The variable tier 1 ratio measures the how much tier 1 capital a bank has per risk weighted assets. At first it seems this should have a clear positive relation to risk culture as it increases the buffer a bank have to take on adverse events as is confirmed by Berger and Bouwman (2013). However, the negative relation with systemic risk from Perotti et al. (2011) and Kleinow and Moreira (2016) begs to question whether this is the whole story. An explanation could be that banks have to comply with a certain level or the regulatory measure, but this does not mean they want to comply. I therefore expect that relation between tier 1 ratio and a bank’s risk culture loses significance or can even turn negative with increased regulatory stringency.

3.2.2 Culture & Law

Problems of culture in organizations are often mentioned since the 2007-2009 crisis. Risk culture is not a separate kind of thing to culture in general. It is rather a looking at culture and focus on the concerns its poses to risk-taking and risk control activities (Power et al., 2013). Hofstede (1980) had an enormous impact with his framework for the measurement and consequences of culture, but there are still few finance studies examining culture as a factor of interest. Guiso, Sapienza and Zingales (2011) investigate how to best examine the value of culture in finance since it is so a hard to define. They find that the measure of trust in strangers is the most promising component of culture as it is well-founded economically, correlates with financial performance and it is relatively easy to measure.

The research from Guiso et al. (2011) points to trust as a key measure to capture culture in finance. This is further enhanced by the fact that while Guiso et al. (2011) find a very strong correlation between trust and economic development for countries with a per capital GDP above $20,000, there is no correlation below that level. They suspect this to be because trust is most needed in more sophisticated financial services transactions. In more simple businesses transactions such as running a farm there is far less reliance on trust between parties. This is in line with the general notion that trust is a key condition for the financial industry to work properly and a driver for risk taking in general. From a risk culture perspective I pose that trust will increase risk taking during

(17)

16

good times as trust assumes that there is a mindset that things, i.e. risks, will turn out okay. However, this also means that countries that are generally more trusting will have taken on more risks when a crisis presents itself. Therefore expect trust to be negatively related with performance during crises.

Culture and law are similar in that they do not change much over the years. La Porta et al. (1998) furthermore state that they interact and help shape each other over the years. Law connects mostly to risk culture in a repressive sense that limits the amount of risk a bank is allowed to take to protect society. However, it is key a supervisor is given enough authority by law to actually be able to influence the risk culture of banks. To proxy for this I use the measure of Supervisory power from Barth, Caprio and Levine (2013). It measure the supervisory authority’s power to take specific actions to prevent and correct problems on a scale from 0 to 14. I expect supervisory power to be positively related to a bank’s risk culture as a supervisory authority with more power is expected to be able to limit irresponsible risk taking more than their less powerful counterparts.

3.2.3 Governance

The first two governance variables I use to proxy for risk culture are the commonly studied variables board size and board experience. According to Aebi et al. (2012) smaller boards are considered to be more effective in non-financial because decision-making is easier in smaller groups. This would be especially important during crises when the timing of decision-making can be critical. However Aebi et al. (2012) find that bank performance and board size are positively related during the 2007-2009 financial crisis. The explanation they give for this is that corporate governance characteristics that are normally considered “good” led to boards taking calculated risks to maximize shareholder value, but which turned out poorly during the crisis. This would mean that “good” corporate governance would lead a firm to be willing to take on more risk during normal times with initial good, but poor crisis performance. This would imply that these corporate governance variables would have a negative relation to risk culture. However, Adams and Mehran (2003) explain the phenomenon simpler by stating that the positive relation between board size and bank performance is because of a banks complex business and regulatory environment. This means that a larger board is in a better position to assess the risks of a bank’s business and thus lower the risk culture as a bank is more aware of the risks taken. If this is the case I would expect board experience to have a positive relation with bank performance as this would also enable a

(18)

17

board to better asses the risk of the complex business and regulatory environment. Contrary to this, the explanation from Aebi et al. (2012) is correct I would expect board size not to have a significant positive relation with performance if I control for the return in the buildup of a crisis.

Aebi et al. (2012) also find that if a bank’s CRO directly reports to the board of directors the bank performs significantly better during the 2007-2009 financial crisis. Their explanation for this is that a CEO might have a different agenda than the CRO which’s main task is to manage risks. Similarly they find that having a risk committee alone is not sufficient, it also need to be dedicated as is measured by the number of their meetings. I also expect there is a positive relation between a dedicated role to risk and a bank’s risk culture and performance during crises. To proxy for this I use a dummy that is one if there is a person within the higher management of the bank explicitly tasked with risk management, risk role. I also use two other measure that try to capture a firms commitment to mitigating risks; a dummy that is one if a firm uses advanced risk management techniques or not and a measure to whether the firm invests in its employees through training and development on a scale from 1 to 100. All three can have a positive relation to a bank’s performance during crises as they can prepare a firm for the future risks. However, I also recognize the danger of a firm using complicated risk management techniques or potentially growing overconfident from their experience on risk management in pre-crisis times. This could give a false sense of security of being prepared for risks, while the sensitivity to equity tail risk might increase during a crisis. If this is the case I expect a non-significant or negative relation between a banks’ risk culture and having a dedicated risk role, using advanced risk management techniques, and training and development.

The next variable is dummy that equals one if the bank has a powerful shareholder. Laeven and Levine (2009) find that firms with a powerful shareholder in the form of a majority owner, one that has a board seat or who has strong family ties with the firm tend to take greater risks. The explanation for this is that they use their power to maximize short-term shareholder value. As I expect this to increases the amount of risk the company is willing to take I initially expect this variable to have a negative relation to performance. Another shareholder related risk culture proxy is the managerial ownership percentage. According to Aebi et al. (2012) managerial ownership may provide incentives to maximize shareholder value and limit the bank’s risk exposure.

(19)

18

Berger, Imbierowicz and Rauch (2012) also find evidence that higher inside ownership substantially lowers the probability of default. They explain this through the stronger incentives for diversified shareholders to take risk as compared to the manager owner which has a large portion of its wealth in the form of bank equity. As managerial ownership thus decrease the willingness to take on risk I expect it to have an initial positive relation with risk culture. However, Laeven and Levine (2009) find that the effect of a bank’s ownership structure on risk taking and performance is largely dependent on the presence of whether there are strong shareholder protection laws and bank regulations.

To follow up on this De Haan and Vlahu (2016) suggest to give special attention to the role of managerial ownership, compensation and risk-taking. There is mixed evidence on the relation between compensation and risk-taking. On one hand Erkens, Hung and Matos (2012) highlight the benefits of managerial ownership from performance-based compensation focused on shares distribution instead of options or cash. While on the other hand Fahlenbrach and Stulz (2011) find that banks which provide stronger incentives to CEOs performed worse during the crisis. This is aligned with the study on risk culture from Power et al. (2013) which finds that after the crisis there is a lot more emphasis on the repressive incentive approach to comply with certain standards with consequences for compensation, promotion, etc. I therefore add the variable performance pay and expect it to have a negative relation to a bank’s risk culture. Bebchuk and Spamann (2010) find that equity-based compensation causes executives to focus on the short-term. Because this increases managerial ownership I see a negative relation between managerial ownership and performance based compensation on one side and risk culture on the other. Since Laeven and Levine (2009) find that the influence of ownership structure is dependent on shareholder protection laws and regulation I expect this effect to be mitigated when controlled for these factors.

The last group of governance variables are proxies of the ethics of a bank. In their study on risk culture Power et al. (2013) find there is a post crisis emphasis on the need to be ‘good’ as opposed to ‘greedy’, next to the repressive incentive approach to comply with certain standards with consequences for compensation, promotion, etc. This long-term organic approach emphasizes the renewal of values in the organization, in particular an ethos of client service. To proxy for whether a bank has a poor ethical

(20)

19

footing I use the variable controversies. This variable counts the number of controversies published in the media around business ethics or tax fraud. I expect the variable to be negatively related to a banks risk culture. I also include two other variables that try to capture something similar, financial transparency and whistleblower protection. These are dummy variable that equal one if the company has a policy to improve financial transparency or a whistleblower protection mechanism in place, respectively. As both give some insight in whether a bank is open to the ethical aspect of their work and provides for an open environment to discuss (risk) policy problems I expect these variables to have a positive relation with risk culture.

3.3 Control Variables

In our regression analyses, I use a comprehensive set of control variables to try to avoid potential omitted-variable bias. To further test for omitted variable bias I perform and include the Ramsey RESET in most regression specifications. First I discuss three firm-specific control variables after which I will discuss the country-specific variables I control for. As a first control variable I use the pre-crisis return. I include this variable because a bank’s stock performance during a crisis could be explained by an excessive pre-crisis return. The crisis return could then, at least in part, be seen as a correction to the earlier return and thus the pre-crisis is expected to have a negative relation to the crisis return. Secondly I control for size, and include the market-to-book ratio and the logarithm of market capitalization, which I expect to have a negative relation to crisis returns. Brunnermeier, Dong and Palia (2012) find only weak evidence for a destabilizing effect of high values of market-to-book ratio, however from a theoretical perspective larger banks and firms have an inherently more persistent and harder to adjust risk culture. According to Power et al. (2013) this is because large firms, particularly those with large group structures, have complex systems and processes that can be inflexible and hard to enforce. Furthermore if a bank is deemed too big to fail this could incentivize managers to take more risk than socially optimal.

Laeven and Levine (2009) emphasized that laws and regulations interact with other variables such as the ownership structure and risk-taking. Therefore I control on country-specific differences in the regulatory regimes and deposit insurance schemes. I follow Barth et al. (2013) and use their Capital Regulatory Index (CRI), on the stringency of capital regulations, and supervisory independence measures, to capture

(21)

20

independence of the supervisory authority from the government and independence through legal protection from the banking industry.

To control for the differences in deposit insurance scheme I use the databases compiled by Demirgüç-Kunt, Kane and leaven (2005, 2013). I control for the dummy variable that takes the value of one if there is an explicit deposit insurance scheme, if foreign currency deposits are covered by an explicit deposit insurance scheme, if interbank deposits are covered by an explicit deposit insurance scheme, if the deposit insurance scheme has no coinsurance, if the deposit insurance fund is a permanently ex-ante funded, and if the deposit insurance is funded partly or exclusively from government sources, respectively. Where a more stringent regulatory regime is usually expected to lead to lower risk-taking, according to Pennacchi (2009) the existence of an explicit deposit insurance scheme can increase the amount of risk bank managers are willing to take due to a moral hazard problem it creates.

Lastly I control for differences in shareholder protection laws per country. To do this I add the Anti-Self-Dealing index of Djankov et al. (2008) in my regressions. I expect that a better shareholder protection laws have a positive influence on the risk culture of banks in that country. For robustness I also perform my regressions with the anti-director-rights index from La Porta et al. (1998) as revised by Djankov et al. (2008) and Spamann (2010), their results are virtually identical to using the Anti-Self-Dealing index in my analysis. An overview of all variables used in this study including the data sources can be found in appendix A.

3.4 Empirical approach

3.4.1 Part 1 – Analyzing the risk culture hypothesis

The first step to confirm the risk culture hypothesis from Fahlenbrach et al. (2012) is to test whether hold return during the 1998 Russian Crisis predicts buy-and-hold return during the 2009 financial crisis. The dependent variable is the 2007-2009 financial crisis return, for which I follow Fahlenbrach et al. (2012) and set from July 1, 2007 to December 31, 2008. This period is chosen because from 2009 onwards the new uncertainty about whether banks would be nationalized could influence results. If banks delist or merge prior to December 2008, I put the proceeds in a cash account until December 2008. I run the regressions on the global sample, but also on regions to test whether there are regional differences in risk culture.

(22)

21

𝑅𝑒𝑡𝑢𝑟𝑛

2007−2009

= 𝛼

𝑖

+ 𝑏

1

𝑅𝑒𝑡𝑢𝑟𝑛

1998 𝑐𝑟𝑖𝑠𝑖𝑠

+𝑏

2

𝑅𝑒𝑡𝑢𝑟𝑛

𝑅𝑒𝑏𝑜𝑢𝑛𝑑 1998 𝐶𝑟𝑖𝑠𝑖𝑠

+𝑏

3

𝑅𝑒𝑡𝑢𝑟𝑛

2006

+ 𝑏

4

𝐵𝑒𝑡𝑎 + 𝑏

5

𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒

+ 𝑏

6

𝐵𝑡𝑀 (2)

+𝑏

7

𝐿𝑜𝑔 (𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒) + 𝑒

𝑖,𝑡

The main explanatory variable for confirming the risk culture hypothesis is the buy-and-hold return during the Russian crisis of 1998. Following Fahlenbrach et al. (2012), I set the start of the crisis on August 3, 1998 and the end on December 31, 1998. The return of the 1998 crisis is the buy-and-hold return from August 3, 1998 to each individual stock’s lowest stock price during the crisis period.

The rebound buy-and-hold return from the 1998 crisis is the six-month buy-and-hold return following the lowest price of 1998 and the return of 2006 is the buy-and-hold return in calendar year 2006. I follow Acharya et al. (2010) and calculate bank leverage as the quasi-market value of divided by the market value of equity. They define the quasi market value of assets as book value of assets minus the book value of equity plus the market value of equity. A complete list of variables and definitions are provided in appendix A.

The second step in confirming the risk culture hypothesis is to analyze whether there are asymmetries in the relation between crisis returns in the Russian crisis and returns during the 2007-2009 Financial crisis. To confirm the risk culture hypothesis the relation should be driven by the poorest performing banks. If this finding is driven by the well performing banks it would be more supportive of the learning hypothesis in the sense that well performing banks are better learners. I split my sample into quintiles based on a bank’s returns during the Russian crisis. Quintile 1 contains all observations whose return during the Russian crisis is among the 20% lowest. I then repeat the regressions but use the quintile indicator variables as the explanatory variable.

3.4.2 Part 2 – Risk culture proxies and equity tail risk

In the second part of this study I investigate whether proxies for risk culture can predict performance during crises. First I use the framework of Fritz-Morgenthal et al. (2016) to identify potential risk culture proxies based on their nine risk score indicators; Regulatory requirements, Business strategy, Governance, Portfolio, Employees, Risk strategy, Reputation, Other effects and Cultural indicators. I identify eighteen quantitative measures that could proxy for risk culture according to literature. I then

(23)

22

simplify the grouping and put the risk culture proxies in one of three categories; Business, Culture & Law and Governance. All variables, their definitions and sources can be found in appendix A.

Secondly, I test the risk culture proxies on whether they explain a bank’s sensitivity to crises. The risk culture hypothesis states there is a relation between a bank’s risk culture and a bank’s sensitivity to crises, so I assess and select proxies on whether they adhere to this relation. I test the relation between each individual risk culture proxy and a bank’s equity tail risk during crises. I estimate the banks Marginal Expected Shortfall (MES) using ordinary least squares (OLS) with heteroscedasticity consistent standard errors. The baseline regression model is given by

𝑀𝐸𝑆

_{𝑖,𝐶𝑟𝑖𝑠𝑖𝑠}

= β

₀

+ β

₁

× 𝑅𝑖𝑠𝑘 𝐶𝑢𝑙𝑡𝑢𝑟𝑒 𝑝𝑟𝑜𝑥𝑦 𝐴

_{𝑖,𝑝𝑟𝑒−𝐶𝑟𝑖𝑠𝑖𝑠}

(3)

+Ω × 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠

𝑖,𝑝𝑟𝑒−𝐶𝑟𝑖𝑠𝑖𝑠

+ ε

𝑖

I regress Marginal Expected Shortfall on each of the eighteen risk culture proxies for five crises periods; the Mexican crisis (1995), the Asian crisis (1997), the Russian/Long-Term Capital Management crisis (1998), the Dotcom crash and the 9/11 terrorist attacks (200-2001) and the recent financial crisis (2007-2009). All dates and descriptions of the crises periods can be found in appendix B. If banks delist or merge prior to the end of a crisis period, I set the Marginal Expected Value to the latest value until the end of the crisis. I run the regressions on the global sample, but also on three regions to test whether the importance of specific risk culture proxies differ per region. The regions I test are Asia the Pacific, Europe and Central Asia and North America. I use the results to assess which risk culture proxies to select for further testing.

3.4.4 Part 3 –Risk culture proxies and poor bank performance

In the third and final part of this study I test whether the selected risk culture proxies predict poor bank performance. The baseline regression model is given by

𝑅𝑒𝑡𝑢𝑟𝑛

𝑖,𝐶𝑟𝑖𝑠𝑖𝑠

= β

0

+ β

1

× 𝑅𝑖𝑠𝑘 𝐶𝑢𝑙𝑡𝑢𝑟𝑒 𝑝𝑟𝑜𝑥𝑖𝑒𝑠

_{𝑖,𝑝𝑟𝑒−𝐶𝑟𝑖𝑠𝑖𝑠}

(4)

+Ω × 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠

𝑖,𝑝𝑟𝑒−𝐶𝑟𝑖𝑠𝑖𝑠

+ ε

𝑖

For this part of the analysis the dependent variable is the buy-and-hold return during the five crisis periods. I run my regressions on the global sample, but also per region to test whether there are differences between regions.

(24)

23

4. Data and descriptive statistics

In this section I provide information on the construction of my sample and data sources, and give some descriptive statistics.

4.1 Sample Construction and data sources

I start selecting a global sample of all financial institutions in Thomson Reuters DataStream. All listed and delisted financial institutions in Thomson Reuters DataStream over the period 1990 through 2017 are included to correct for survivorship bias, returning 22,560 financial institutions. Following Fahlenbrach et al. (2012) and Weiß et al. (2014) I exclude all firms that are not in the traditional banking industry (e.g., online brokerages). I start the screening method of Fahlenbrach and Stulz (2011) based on primary Worldscope SIC codes (SIC), but deviate where they choose to manually correct for errors. Instead I use the Thomson Reuters Business Classification (TR5N) and DataStream level 6 industrial classification (INDM) in combination with SIC codes. To screen whether the firm is a financial institution by SIC codes is between 6000 and 6300, whether the word “bank” is in the TR5N name and whether the word “bank” is in the INDM name. I keep financial institutions that have SIC codes between and 6000 and 6300 and that have the word “bank” in either the TR5N or INDM name. I classify the financial institutions that have the word “bank” in both TR5N and INDM names as “A banks” and financial institutions that have the word “bank” in either of them as “B banks”. As a robustness check I will run my analysis on both the whole sample consisting of A and B banks and on the sample of “A banks” alone.

Financial accounting data are retrieved from the Thomson Reuters DataStream. I exclude banks that lack or have inconsistencies in share price and accounting data. Next, I follow the screening procedures proposed by Ince and Porter (2006) to correct for errors in DataStream daily returns and bank stock price data. I first exclude banks from our sample if the bank’s stock price drops below $ 1 to control for the distorting effect of rounded prices. Secondly, I exclude banks that yield a monthly stock return of more than 300% which is reversed in the following month. Next, I implement the screening procedures from Hou, Karolyi and Kho (2011); first I exclude bank-years if the number of zero return days exceeds 80% of a given month and finally I exclude all non-trading days. Non-trading days are defined as those days on which 90% or more of the listed stocks have zero returns. Similar to Weiß et al. (2014) but in contrast to

(25)

24

excess of $ 10 billion, and also consider smaller banks. This results in a total 4,459 banks in my sample. To compare differences between regions I assign each bank in our sample to one of six different regions (Asia & Pacific, Europe & Central Asia, Latin America & Caribbean, Middle East & North Africa, North America and Sub-Saharan Africa) based on the location of the bank’s headquarters.

I use the World Value Survey (2015) to get data on trust to proxy for a bank’s culturally given attitude towards risk. The data is an answer to the question “Do you trust strangers?” on a scale from 1 to 5 on which 1 meant “Completely trusting” to 5 meaning “Do not trust at all” of which I take the average of all respondents to this question per country. Regulatory data on capital regulation come from Barth et al. (2013) while deposit insurance data comes from Demirgüç-Kunt et al. (2005, 2013). For data in shareholder right I use the anti-self-dealing index from Djankov et al. (2008) and the Anti-director-rights index from La Porta et al. (1998) as revised by Djankov et al. (2008) and Spamann (2010). I use Orbis Bankscope and Compustat ExecuComp to collect and merge data on salary and total compensation. Secondly, I use Orbis Bankscope to collect data on function titles to create a dummy, risk role, which equals 1 if there is someone explicitly responsible for risk management. Thirdly, I use Orbis Bankscope to gather data on managerial ownership and on whether a firm has a powerful shareholder. The powerful shareholder data from Orbis Bankscope I merge with shareholder data from Laeven and Levine (2009) and Thomson Reuter’s Asset4. I furthermore use Thomson Reuters Asset4 database to collect data on the remaining Board Experience, Controversies, Financial Transparency, Advanced Risk Management, Training and Development, Whistleblower Protection, Average board years, Board Independence and Board Size . The complete list of variables, definitions and sources can be found in appendix A.

I use the studies from Fahlenbrach et al. (2012), Bartram, Brown and Hund (2007) and Weiß et al. (2014) to select crises events. However, I do not label the beginning of the subprime crisis in 2007 and the collapse of Lehman Brothers in 2008 or the Dotcom crash and 9/11 terrorist attacks as two different crisis periods as in Weiß et al. (2014). In both cases the two events are so close to each other that I treat them as one crisis. This brings me to investigate the following crisis periods: the Mexican crisis (1995), the Asian crisis (1997), the Russian/Long-Term Capital Management crisis (1998), the Dotcom crash & 9/11 terrorist attacks (2001), and the recent financial crisis (2008). The

(26)

25

dates we relate to each crisis as well as the corresponding brief descriptions of the events are provided in Appendix B.

4.2 Descriptive Statistics

Table 1 provides the descriptive statistics for the data of my global sample of banks. The median and mean annualized returns from July 2007 to December 2008 are -26% and -27%. The median and mean returns in the Russian crisis from August 3, 1998 to the lowest stock price in 1998 are slightly higher at approximately 24% and -22%. The rebound performance of the banks in our sample is quite good during the six months following the Russian crisis. The median and mean rebound returns are 14% and 30%, respectively. The average return from January to December 2006 is also pretty good at 8%, as is the median return at 11% in our sample. The median amount of assets at the end of 2006 is $1,500 million, only a fraction of the average amount of over $64,213 million. Assets growth was pretty high with the average bank having a growth of 12% in 2006. The average market-to-book ratio is 1.88 while the median stands at 1.67. The mean and median market capitalization are approximately $6,539 million and $307 million, with an average and median leverage of 7.63 and 6.52, respectively. The Tier 1 ratios are quite high with an average and median Tier 1 ratio of 12.14 and 11.00, respectively. The lowest reported Tier 1 ratio was 5.58, which is well above the then effective Basel II regulatory capital requirement of 4%.

To investigate the influence of banks size I’ve also the sample by 2006 assets. For comparability I used the three size classes of Berger and Bouwman (2013) and add another class used in the study from Beltratti and Stulz (2012); small banks with up to $1 billion in assets, medium banks with between $1 billion and $3 billion in assets, large banks between $3 billion and $50 billion in assets and very large banks with more than $50 billion in assets. Descriptive statistics for each size class can be found in the Appendix C to F. Perhaps the most interesting part is that asset growth is slightly increasing with size, the Tier 1 ratio was decreasing with size and that the very large banks had much higher leverage than any of the other size classes. Combined with the fact that the largest banks performed the worst during both the Russian crisis and the 2008 financial crisis this could indicate a connection. However, the close second worst performers were the small banks, which seems to invalidate any preliminary conclusions one might make by glancing the descriptive only. Furthermore, the large and very large banks recovered much better than the small and medium sized banks in the rebound

(27)

26

Table 1: Descriptive statistics of the global sample of banks

This table presents summary statistics for the global sample of banks. Variable definitions and data sources are provided in Appendix B. Absolute financial statement items are given in $ thousands with accounting data measured at the end of fiscal year 2006.

N Min. Pct.25 Median Pct.75 Max. Mean Std. Dev.

2008 crisis period return 1,004 -0.96 -0.44 -0.26 -0.09 0.70 -0.27 0.24 1998 crisis return 957 -0.78 -0.32 -0.24 -0.14 2.50 -0.22 0.21 1998 rebound return 856 -0.42 0.04 0.14 0.28 15.00 0.30 0.95 2006 return 969 -0.77 -0.04 0.08 0.22 2.21 0.11 0.30 Total assets 563 18,025 639,8910 1,500,000 8,000,000 2,200,000,000 64,213,264 262,973,756 Total liabilities 562 6,748 572,014 1,350,00 7,100,000 2,100,000,000 61,273,865 253,467,166 Asset growth 540 -0.20 0.03 0.08 0.17 1.12 0.12 0.16 Market-to-Book ratio 716 0.23 1.29 1.67 2.17 11.00 1.88 1.10 Market capitalization 571 8,620 106,153 307,185 1,700,000 250,000,000 6,539,265 22,668,950 Leverage 549 1.02 5.01 6.52 8.50 129.43 7.63 6.87 Tier 1 ratio 418 5.58 9.50 11.00 13.20 46.34 12.14 5.18

(28)

27

return after the Russian crisis. Market-to-book ratios are also higher for large and very large banks.

To compare my sample with previous studies I’ve also split the sample into banks from the Europe and Central Asia, banks from the Unites States and banks from other countries/regions. My final sample comprises of 1470 banks in 2006. Of these 159 are from Asia and the Pacific, 292 banks are from Europe and Central Asia, and 924 banks are from the United States. The remaining 95 banks are from Canada or the other regions; Latin America & Caribbean, Middle East & North Africa or Sub-Saharan Africa. Of those banks, 1011 banks are recognized by all three sources as banks. From these “A banks”; 117 banks are from Asia and the Pacific, 209 are from Europe and Central Asia, 614 are from the United States and 71 are from other countries/regions.

The most noteworthy differences between the Europe and Central Asia, Unites States and banks from other countries/regions are that European and Central Asian banks performed far worse during the financial crisis of 2008 compared to the two other regions, but that their returns in 2006 were also by far the highest. During the Russian crisis it performed approximately the same as the United Stated but recovered much faster afterwards. Banks from places other than Europe, Central Asia or the United States performed much better during each crisis but also were the only group that had an average negative return during the calm period of 2006. Leverage ratio was the highest in Europe and Central Asia while Tier 1 ratio was the highest in the United States. The descriptive statistics per country/region can be found in the tables of appendices G to I.

Overall, my sample is similar to those in other studies. The sample is heavily biased towards the Unites States banking industry as is consistent with prior studies. The regional division of banks, especially for “A banks”, is consistent with prior studies from Weiß et al. (2014) and Pelster, Irresberger and Weiß (2016). Furthermore the sample of United States banks is comparable to that of Fahlenbrach et al. (2012), again the especially the “A banks”. Variables are similar, although my sample has more banks and smaller banks. The regional division per crisis can be found in table J of the appendix. There are two explanations for the differences between samples; first, is that Fahlenbrach et al. (2012) do not correct for known DataStream errors by following the screening methods of Ince and Porter (2006) or Hou et al. (2011). Since I do follow their screening procedures I have a relatively more restricted sample for my analyses,

(29)

28

despite have a larger sample in absolute terms. Furthermore, I constructed my sample based on primary SIC codes, the Thomson Reuters Business Classification and DataStream level 6 industrial classification, contrary to Fahlenbrach et al. (2012) constructed their sample using primary SIC codes and manual adjustments. Fahlenbrach et al. (2012) first found evidence for the risk culture hypothesis in their United States sample of banks. Since the U.S. portion of my sample is similar to theirs and that my sample as a whole takes up more than half of my sample I expect to find similar results to Fahlenbrach et al. (2012). However, Fahlenbrach et al. (2012) find that the predictive power is concentrated in large banks. My sample includes a larger number of smaller banks compared to Fahlenbrach et al. (2012) so I expect my results could be less conclusive compared to their findings. Furthermore, it is important to note that for a number of variables data is lacking, especially during the first two periods, the Mexican Peso crisis and the Asian crisis, missing data is a problem. Therefore my results are biased toward the other three crises and less can be inferred about the first two.

5. Results

In this section I present the results of my analyses om the risk culture hypothesis, the relation between risk culture proxies and equity tail risk, and the relation between risk culture proxies and equity tail risk.

5.1 Analyzing the risk culture hypothesis

I now test the whether I can find evidence for the learning hypothesis or the risk culture hypothesis. The learning hypothesis from Malmendier and Nagel (2011) entails that poor performers learn from their experience during crises and thus perform better during subsequent crises. This implies that crisis returns of poor performers in 1998 do not or negatively predict return during the 2008 crisis. Opposite of this is the risk culture hypothesis from Fahlenbrach et al. (2012), banks have a persistent risk culture that causes some banks to be inherently more sensitive to crises. This implies that a bank’s return in a crisis has a positive relation with their return during a subsequent crisis that is driven by the poorest performing banks. Table 2 shows strong support for the positive relation from the risk culture hypothesis. There is a strong positive relation between the crisis return of 1998 and the crisis return during the 2008 financial crisis. The effect is both economically and statistically significant; banks that performed poorly

Why is it always the same? : Risk culture as a predictor for bank market performance during crises