The Influence of Underlying Risks on Market Betas of U.S. and European Banks

European Banks

H.L. Huzeling

*

________________________________________________________________________

A

BSTRACT

This study tests the hypothesis whether market risk of banks is determined by underlying risks approximated by accounting ratios. Analyzed through a pooled ordinary least squares model, data of European and U.S. banks for 2003-2007 indicate that market beta is significantly influenced by credit risk, management quality risk, liquidity risk and leverage risk. Unknown factors do also have substantial influence on market beta in the year 2007. Moreover, the influences of accounting ratios on market risk can not be seen as being constant, since these change annually.

Key words : Banks, Market Beta, Accounting Ratios, Risk

JEL Classification : G21

______________________________________________________________________________

* Student (s1322621) at Rijksuniversiteit Groningen, Economics and Management, Finance ; Email : H.L.Huzeling@student.rug.nl ;

1.

I

NTRODUCTION

The current collapse of several financial institutions provokes customers to be concerned about their savings accounts. For decades, such a situation did not occur and as far as people can remind, banks were very reliable companies. However, since the most recent years everything seems to be different. Collapsing industries like the U.S. housing industry are a serious threat to banking companies. Debtors may therefore not be able to pay and there may also arise huge liquidity problems caused by harmed customer confidence, which will lead to abnormal deposits withdrawals. As a result, banks suffer gigantic historical losses and some of them are already gone bankrupt. Due to this financial crisis, stock returns of financial institutions became very volatile, which concerns many investors. Since investors are therefore exposed to the risk of losing large amounts of money, they want to determine the riskiness of their investments. The sensitivity of banks’ stock returns to market movements is traditionally displayed by their market betas. These betas provide useful insights about the riskiness of banks’ stock returns, however a turbulent financial environment may require more information. Stated differently, investors may have interest to know which underlying risks1_{determine market betas. According to Beaver et al.}

(1970), these observed market betas are the result of decision processes of investors altogether, since market betas are derived from security prices. Expectations about risk and return are driven by market betas, however our knowledge about the determination of risk is limited when we are unaware what exogenous variables are determining market betas.

Although there are several factors that may explain market betas, like future cash flows, this study investigates whether market betas are influenced by underlying risks, which are readily observable from financial statements. The underlying risks which are incorporated by this study are credit risk, leverage risk, liquidity risk and management quality risk. It is assumed that these accounting determined risks capture the most2_{relevant relationships between market betas and}

underlying risks as suggested by previously done research. This study covers the period from 2003 to 2007 and is based on data of 74 European and 38 U.S. banks.

1_{Risk is often associated with downside sensitivity. This would imply that all stated risks have a positive} influence on market beta, since investors expect higher returns when risk increases. In this study, an underlying risk does not necessarily needs to be downside risk, since not all signs can be postulated a priori. However, since most underlying risks are downside and for simplicity reasons, underlying risks is the given name to sensitivity of market betas to pre-specified accounting ratios.

Most related research is performed during steady financial times, however this study incorporates the beginning of a financial crisis as well. Therefore, this study provides useful insights, especially concerning downside risk. By applying year dummies, there can be determined whether unknown factors in a certain year influence market beta as well. This study does also contain several analyses to determine the effect on market betas more specifically. Market betas will be split into downside market betas and upside market betas to determine the influence on market betas in a declining as well as in a rising market. Additionally, there will be distinguished between systematic and unsystematic risk, by using total risk as well as dependent variable. Moreover, when underlying risks appear to be significantly influencing market betas, interrelated variables will be applied as well to determine the yearly effect of these underlying risks. These very extended and specified analyses of the influence of underlying risks on market betas will confirm and extend existing literature and may provide great practical information to investors and bank managers. Since previous research is mainly been done in the U.S., the inclusion of European banks in the data set will contribute to existing literature in another way. Each individual investor might have some additional investment tools when there is better understanding of the relation between market determined and accounting determined risk. Broad knowledge about this relation may lead to an improvement in investors’ decision making. Additionally, practical information might be provided to a bank’s management when they know how to influence their sensitivity to market movements.

The study is organized as follows. Section 2 presents a review of previously done research about the market betas and about the relation between market and accounting determined risk. Section 3 contains the methodology used for this study and the data are extensively described in section 4. Empirical evidence is provided by section 5, which is divided into several subsections. The concluding remarks are pointed out by section 6.

2.

R

ELEVANT

L

ITERATURE

2.1 MARKET BETA AS A MEASURE OF RISK

captured very well by the traditional beta, which is among others suggested by Kaplanski (2004). Market beta, βiis traditionally measured by

)

(

)

,

(

m m i i

r

Var

r

r

Cov





(1),

where ri measures the rate of return of asset i, rm measures the rate of return of the market,

Cov(ri,rm) measures the covariance between the rates of return, and Var(rm) measures the variance

of the rate of return of the market. The modern portfolio theory developed by Markowitz (1952), on which traditional market betas are based, assumes that stock returns follow a normal distribution. However, it is questionable whether common market indices and stock returns actually do follow this normal distribution. Moreover, Roy (1952) argues that investors do differently care about downside losses than about upside gains. In addition, the behavioral approach of Kahneman & Tversky (1979) and the axiomatic approach of Gul (1991) suggest that investors place greater weights on losses than on gains in their utility functions. Thus, investors averse to downside losses, will demand higher expected returns for stocks with greater downside risk. Therefore, accepting a normal distribution of stock returns suggested by Markowitz (1952) is premature and probably mistaken. Arguing this, there is enough ground to state that the traditional market beta is not the most successful measurement tool for detecting risk premiums. More meaningful is to split market betas into an upside market beta and a downside market beta, like Bawa & Lindenberg (1977) and more recently Ang et al. (2005). These articles formulate downside market beta, βi–as

)

(

)

,

(

ref

r

r

Var

ref

r

r

r

Cov

m m m m i i

_









(2),

which calculates downside market beta when market returns are below a reference point, ref. Ang et al. (2005) uses three different reference points, the average market return, risk free rate and zero rate of return. Since they are highly correlated to each other, as indicated by Ang et al. (2005), this study uses only the average market return μmas a reference point which is also used

)

(

)

,

(

m m m m m m i i

r

r

Var

r

r

r

Cov













 (3).

Taking the average market return as reference point will not heavily harm the amount of necessary observations to calculate both, downside and upside market betas. Since the zero rate of return and the risk free rate are more or less fixed reference points, an extremely growing or declining market could lead to a scarcity of observations to calculate downside or upside market betas. Naturally, when market betas are split, ARCH and GARCH effects can not be measured. Upside market beta βi+ uses logically the same reference point and is given by

)

(

)

,

(

m m m m m m i i

_Var

_r

_r

r

r

r

Cov













 (4).

Although, banking activities differ between banks with dissimilar core businesses, unfortunately there can not be distinguished between them. Concerning U.S. data, 85% of all observations are bank holding companies3_{, 6 % are mortgage banks and 5% are commercial}

banks. Rearranging the dataset to subsamples dictated by these percentages will lead to analyses which are based on too little observations. Moreover, the specialization as suggested by Bankscope may be a little arbitrary, since this an actual judgment which may be altered during time. Several banks did change their specialization during the financial crisis to generate better liquidity. For example, Goldman Sachs, American Express and CIT Group have changed there specialization to bank holding companies in 2008, which may lead to some confusion, since they had other specializations in the years before. Concerning European data, 60% of all observations are commercial banks, 27% are bank holding companies and 8% are cooperative banks. Although, there can be composed better subsamples of European data than of U.S. data, they are still unable to capture the differences between types of banks, since they are of a limited size.

2.2 SYSTEMATIC AND UNSYSTEMATIC RISKS

Market betas tend to measure systematic risk, which can not be diversified. Recessions, wars and interest rates are all sources of systematic risk since they influence the entire market. It is very important to understand the difference between systematic and unsystematic risk, when it

comes to assessing risk. When the economy is in a recession, it is more likely that debtors fail to repay debt, which is a source of systematic risk. However, the same recession may cause a bank run when creditors think that banks can not fully withdraw their deposits. Since creditors do not want to bear the risk of bankruptcy, they will move their money to elsewhere. Lack of liquid assets will lead to financial difficulties and may even lead to bankruptcy. Bank runs are industry specific events so the can not be seen as systematic risk, however must be seen as unsystematic risk. Since market betas are not able to measure unsystematic risk, they can not measure banking sensitivity to bank runs and other unsystematic risks, which implicates that market betas do not capture all risks of banking activities. This study is focused on the relation between market determined risk given by market betas and accounting determined risk given by accounting ratios. Since there is reason to think that not all accounting determined risks will be reflected in market betas, the influence of accounting ratios will be measured on total risk as well. Markowitz (1959) defines total risk of a portfolio as the variance of the portfolio’s returns. Based on Markowitz (1959) and Sharpe (1964) this study gives total risk 2( )

i r



by

)

(

)

(

)

(

2 2 2 2 i m i i

r

e

r













(5),

where the first part 2 2

(

)

m i



r



reflects systematic risk and the second part 2( )

i

e

2.3 MARKET BETAS AND UNDERLYING RISKS

The most traditional article about the association between market betas and underlying risks resembled by accounting ratios is that of Beaver et al. (1970). They examine the contemporaneous association between market determined and accounting determined measures of risk of 307 U.S. non bank institutions for the period 1947 – 1965, and prove that market measures of risk can be explained by accounting measures of risk. To assess market betas, that study uses time series regressions, which point out that market betas were constant during the investigated period. Contrary to this, among others, Chen & Keown (1985) argues that betas are nonstationary. Following this theory, this study uses annually derived market betas. Consequently, if the market betas are taken annually, the data turns into an unbalanced panel data. As indicated by Baltagi & Song (2006) there exists much theoretical and empirical literature about which technique is most sufficiently analyzing unbalanced panel data. Examining which model should be used, the practical use of it is set as highest priority, since it is difficult to pick the best analyzing technique. For this reason, a pooled ordinary least squares (OLS) model incorporating year dummies will be used.

think that differences in European accounting principles will be minimal. Therefore, differences in accounting principles are not taken into account by this study. Of course, accounting variables can not be taken haphazardly, however must be driven by economic reasoning. The underlying risks that will be investigated are credit risk, management quality risk, liquidity risk and leverage risk. The theoretical background for including these underlying risks is mainly suggested by Angbazo (1997) and Mansur et al. (1993).

2.3.1. CREDIT RISK

Credit risk is calculated by dividing net charge offs (NCO) by total loans (TL), which is based on Angbazo (1997). Net charge offs are based on the difference between actually written off loans and previously classified as uncollectible loans. Thus, when banks are writing more off than expected in advance, this signalizes that less debtors were able to pay, which resembles increased credit risk. Credit risk is usually incorporated in charged interest rates, so it may be unlikely that credit risk will influence market beta. However, when there is an unexpected decrease in the solvability of debtors, market beta is expected to increase, since there can not be adjusted for unexpected decreases in solvability of debtors. Consequently, when net charge offs increase unexpectedly, credit risk is expected to increase as well, so the net charge offs to loans ratio will be positively related to market beta4_{. This positive relation is confirmed by Ramcharran}

& Kim (2003), which states that higher credit risk is associated with higher stock returns, from which market betas are calculated. Using an augmented dealership model of interest spreads, and data for different size classes of banks covering the period of 1989 to 1993, Angbazo (1997) points out that there is evidence that interest margins5_{of banks reflects default}6_risk.

The calculation of credit risk differs among articles, since Mansur et al. (1993) calculates credit risk by the loan loss (LLR) reserve to total loans ratio. Management’s estimate of credit risk exposure is reflected by the allowance for loan loss reserve. Other things remaining equal, a higher loan loss reserve reflects a higher expected loss in the portfolio of loans. This relationship

4_{Naturally, it has an influence in the same direction on downside risk as well, however downside market} beta and upside market beta are not treated separately in this section about underlying risks. It speaks for itself that a proposed positive influence on market beta, will have also an even stronger positive influence on downside market beta and a proposed negative influence on market beta will may be have also an even stronger influence on upside market beta.

5_{Angbazo (1997) uses interest margins instead of stock returns. This study approaches interest margins as a} proxy for market beta since both are firm specific indicators of risk sensitivity. However, one should beware of differences between an accounting based approach and a market based approach.

implicates that the loan loss reserve to total loans ratio should be positively related to market beta. Mansur et al. (1993) uses fifty nine U.S. banks, covering the period of January 1986 to September 1990. Results of that study point out that credit risk is an underlying risk that significantly influences market beta.

2.3.2. MANAGEMENT QUALITY RISK

Management quality risk is the risk that management activities are not in line with the best interest of the shareholders. For calculating management quality risk, there are two methods. The first method is based on Mansur et al. (1993) and calculates management quality risk by dividing net income (NI) by total assets. This ratio measures accrual income, which reflects the effects of managerial decisions and market events on profits. Johnson (1968) suggests that net income7_{can be an adequate approximation of current management performance. Mansur et al.}

(1993) argues that the sign of the coefficient can not be postulated a priori. When net income increases because of aggressive risk taking by the management, the relationship between management quality risk and market beta will be a positive one. However, if net income increases because of acquisition of better quality assets and market shares, this relationship will be negative. Results indicate that Mansur et al. (1993) fails to prove that management quality risk affects market betas of banks.

The second method is based on Angbazo (1997) which measures management quality risk by dividing earning assets (EA) by total assets (TA). Since, management decisions influence the composition of assets, higher (lower) earning assets may be the result of increased (decreased) management quality risk. Therefore Angbazo (1997) argues that, the earning assets to total assets ratio should be positively related to net interest margins. In addition, Angbazo (1997) shows some evidence that net interest margins are positively related to management quality risk indeed. However, it is also logically that earning assets are inversely related to net income in the short run. When investments are made in earning assets, net income may be decreased, where net income may increase when earning assets are divested. Although investing in earning assets may generate more net income in the long run, it may decrease net income in the short run due to costs of these investments. Indirectly based on Mansur et al. (1993), this relation between earning assets and net

income suggests that the relationship between the earning assets to total assets ratio and market betas will be inverse.

2.3.3. LIQUIDITY RISK

Liquidity risk is the risk of not having enough cash or borrowing capacity to meet new loan demand or deposit withdrawals, and thereby forced to borrow funds at excessive cost. According to Pástor & Stambaugh (2003), these excessive costs are unwelcome to an investor whose wealth will drop and therefore will expect a higher return on equity which increases market betas. Liquidity risk is measured through several different methods. For instance, Mansur et al. (1993) uses three different approaches for measuring liquidity risk, (1) total loans to total deposits (TD) ratio, (2) total loans to total assets and (3) cash and due from banks (CDFB) to total assets ratio. Since loans are non liquid assets, an increase in loans should harm the liquidity of a bank. Therefore, ratios (1) and (2) are positively related to market beta. Cash and due from banks are liquid assets, so an increase of ratio (3) will decrease market beta. However, none of these methods results in a statistically significant role for liquidity risk. Angbazo (1997) measures liquidity risk by dividing liquid assets (LQA) by total liabilities (TLB). Since an increase in liquid assets decreases liquidity risk, the relationship between the liquid assets by total liabilities ratio and market beta is inverse. In addition, Angbazo (1997) proves that net interest margins are negatively related to liquidity risk. Beaver et al. (1970) uses the current ratio for measuring liquidity risk and their results point out that there is no reason to see liquidity risk as an factor that explains market betas. Beaver et al. (1970) suggests in advance that there will hardly be any effect of the current ratio on market betas at all, so their method will not be included in the model.

2.3.4. LEVERAGE RISK

Angbazo (1997) divides core capital by total assets to calculate leverage risk. This measure for risk of insolvency is expected to be negatively related to market beta, since substituting equity for debt reduces the risk of insolvency and therefore also the cost of borrowed funds. Moreover, since an increased capital ratios reduce tax shields, earnings are expected to decrease. Consequently, expected return on equity will decrease, which indicates lower market betas. In addition, Modigliani & Miller (1958) shows that when debt is introduced, the earnings stream of the common shareholders becomes more volatile.

Mansur et al. (1993) uses an indirect method to measure leverage risk, the shareholders’ equity (SE) to total deposits ratio. An increase in this ratio due to more equity or less deposits, signifies less financial risk. Employing leverage amplifies potential gain from an investment or project, but also increases potential loss, since interest and principal payments may be higher than the investment returns. The coefficient of this ratio is therefore expected to be negative. The Ridge regression technique used by Mansur et al. (1993) indicates that leverage significantly influences market beta. Thus, leverage and market betas tend to be negatively related. Another article that also confirms this relationship is that of Karels et al. (1989), which examines the relation between capital adequacy and market betas using data of 24 U.S. banks over the period 1977 - 1984. The empirical results provided by Karels et al. (1989) indicate that higher levels of capital adequacy are related to lower market betas.

However, Angbazo (1997) provides evidence that net interest margins are positively related to core capital. Since equity is a more expensive source than debt, an increase in equity may increase the average cost of capital. Moreover, Berger (1995) provides evidence that capital ratios are positively related to return on equity. When investors expect more in return, market betas tend to increase. Differently, Beaver et al. (1970) divides total senior securities, including current liabilities, by total assets to calculate leverage risk. That study shows that leverage has also a positive significant influence on market beta. In fact, the methods of Angbazo (1997), Mansur et al. (1993) and Beaver et al. (1970) are practically the same, so solely one method will be incorporated in the model. Since the accounting ratio used by Mansur et al. (1993) is the one that is readily observable from balance sheets, that is taken to measure leverage risk.

Although there are plausible reasons to suggest a negative relationship between market beta and underlying risks, this study is mainly focused on the downside market risk aspect. For this reason, the expected signs of variables are unknown when upside market beta is used as dependent variable.

TABLE 1 :EXPECTED SIGNS OF VARIABLES BY BETA TYPE *

βT _βAG _β– _β+ _σ

Credit Risk NCO / TL + + + ? +

LLR / TL + + + ? +

Management Quality Risk NI / TA ? ? ? ? ?

EA / TA - - - ? -Liquidity Risk TL / TD + + + ? + TL / TA + + + ? + CDFB / TA - - - ? -LQA / TLB - - - ? -Leverage Risk SE / TD ? ? ? ? ?

3.

M

ETHODOLOGY

3.1 DERIVING MARKET BETAS

Since bank stock return volatility is time-varying, it has to be modeled accordingly. Using the Autoregressive Conditional Heteroskedastic (ARCH) model8 developed by Engle (1982) and the Generalized Autoregressive Conditional Heteroskedastic (GARCH) model introduced by Bollerslev (1986) will avoid this problem. Similar to Elyasiani & Mansur (2005), there will also be analyzed with a normal OLS model for robustness purposes. Thus, traditional market beta βT will be a dependent variable as well as market beta adjusted for ARCH and GARCH effects, βAG_{. Market betas are derived annually, since the time span must be}

corresponding to that of financial statements, which are traditionally given by annual reports. The model that derives market betas is the following

8_{ARCH and GARCH betas are traditional market betas adjusted for ARCH and GARCH effects. ARCH and}

ikt o ijk mkt ikt R C 



R 



(6) 2 0 1 , 1 2 , 1 ikt ik t ik t

h







 

_





h

_ (7)

)

,

0

(

~

1 ikt t ikt





N

h



(8),

where Rikt stands for the return on the stock of bank i at day t in year k. Rmktis representing the

market return index at day t in year k , while C0is a constant and εikt is the error term of bank i at

time t in year k. Equation (7) and (8) are added for ARCH and GARCH purposes, where hiktstands

for the conditional variance of bank returns. The error term



_ikt is dependent on information set

1 t

 . Market beta, given by βijk , represents the sensitivity of stock returns to changes in market

returns of bank i at year k, measured by method j. As abovementioned these methods j are traditional market betas βT_{, market betas adjusted for ARCH and GARCH effects β}AG_{, downside}

market betas β–_{, upside market betas β}+_{, and total risk σ}2_.

3.2 MODEL SPECIFICATION

Once all these betas are derived, the influence of accounting ratios on these betas will be investigated using a pooled OLS model. This model will be of the following form



        n k ik k k ikm ikm ikm ikm ijk C C CR C MQR C LQR C LVR D 1 5 4 3 2 1







(9),

where βijk represents market beta of bank i derived by method j in year

k

. Besides the constant C1

and the error term εik, only underlying risks with their accessory betas are presented in the model.

C2 , C3 , C4, and C5 measure the influence of credit risk (CRikm), management quality risk

(MQRikm), liquidity risk (LQRikm) and leverage risk (LVRikm) on the market beta of bank i at year k

respectively. Dummy variable9_D

k is included to account for year effects, where k = 200410–

2007. As suggested by the literature section underlying risk can be approximated by several measurements. Credit risk is approximated by the net charge offs to total loans ratio and by the

9_{In appendices the coefficients of dummy variables are given by C}

loan loss reserve to total loans ratio. Management quality risk is measured by the net income to total assets ratio and by the earning assets to total assets ratio. Liquidity risk is measured by four ratios, the total loans to total deposits ratio, the total loans to total assets ratio, the cash and due from banks to total assets ratio and by the liquid assets to total liabilities ratio. Leverage risk is approximated by the shareholders’ equity to total deposits ratio. Concerning the balance sheet items, accounting ratios are based on the average value of begin and end year values. However concerning the income statement items the values are derived the way they are given.

Since most underlying risks are calculated by more than one method, several combinations of these methods can be made. There can not be derived from existing literature which accounting ratios are the best approximates, separately as well as simultaneously when they are combined in the model, so for completeness reasons all possible combinations11_{will be}

examined.

4.

D

ATA

The dataset is composed by the 89 largest listed12_European13_{banks and the 44 largest}

listed U.S. banks, of which all suggested underlying risks can be derived. However, unfortunately not all accessory market betas can be derived due to lack of market data or due to inappropriate market data. After removing these observations the data set contains 17714_{observations of 74}

European banks and 170 observations of 38 U.S. banks. Concerning European banks, only banks that are located in the European Union (EU) are taken into account. Countries from which banks are included in the data set, trading with other currencies than the euro are the United Kingdom, Denmark and Sweden. The remaining members of the European Union, which are incorporated by this study, are Austria, Belgium, Cyprus, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal and Spain. As can be derived from Appendix 2a, there can be stated that each country is more or less equally represented in the European data set, especially when country size is kept in mind.

Daily stock returns of banks from 2003 - 2007 are derived from Datastream. These stock returns are adjusted for stock splits and dividends. For the daily market index returns, also

11_{An overview of all these combinations is provided by Appendix 1.} 12_{The largest banks at 31}st_{of December 2008 according to Bankscope.}

13_{An overview of which European and U.S. banks are incorporated by the dataset is provided by} Appendices 2a and 2b.

derived from Datastream, the country’s leading market index return is taken for European countries, and for the U.S. the S & P 500 is taken as the appropriate market index.

All types of proposed risks are derived from the available annual reports which are provided by Bankscope. These reports are published more than once a year, however for simultaneity reasons, solely end of accounting year reports are used. Whether these accounting ratios are correlated is presented by Table 2 for European data and by Table 3 for U.S. data. From Table 2 there can be derived that there are barely noteworthy correlations. The only15

correlation coefficient that is substantial is that between the liquid assets to total liabilities ratio and the shareholders’ equity to total deposits ratio, which is 0,64. However, this is not surprising, since both ratios incorporate total deposits as the aliquot.

TABLE 2 : CONTROLLING FOR MULTICOLLINEARITY IN EUROPEAN DATA

NCO / TL LLR / TL NI / TA EA / TA TL / TD TL / TA CDFB / TA LQA / TLB SE / TD NCO / TL 0,09 -0,06 0,05 -0,17 -0,13 -0,01 0,19 -0,04 LLR / TL -0,13 -0,17 -0,13 -0,01 0,36 0,30 -0,12 NI / TA -0,20 -0,08 0,20 0,22 -0,05 0,25 EA / TA 0,04 0,15 -0,36 -0,05 -0,15 TL / TD 0,36 0,15 -0,11 0,64 TL / TA 0,17 -0,20 0,35 CDFB / TA 0,54 0,14 LQA / TLB -0,11 SE / TD

Moreover, this correlation coefficient will not create a problem, since there are more approximations of liquidity risk. Taking a closer look at Table 3, concerning the correlations between U.S. accounting ratios, indicates that the shareholders’ equity to total deposits and the total loans to total deposits are highly correlated with a correlation of 0,82. The correlations between the loan loss reserve to total loans ratio and the net income to total loans ratio, and that between the liquid assets to total liabilities ratio and the total loans to total deposits are highly correlated as well, with correlations of 0,85 and -0,59 respectively. However, these concern accounting ratios of the same underlying risk, which will not be incorporated simultaneously.

TABLE 3 : CONTROLLING FOR MULTICOLLINEARITY IN U.S. DATA NCO / TL LLR / TL NI / TA EA / TA TL / TD TL / TA CDFB / TA LQA / TLB SE / TD NCO / TL 0,85 0,13 -0,33 -0,02 -0,07 0,02 0,22 0,26 LLR / TL 0,16 -0,39 -0,07 0,16 0,14 0,12 0,17 NI / TA 0,03 -0,03 0,10 0,00 -0,11 -0,14 EA / TA 0,02 0,32 -0,22 -0,25 -0,17 TL / TD 0,27 0,22 -0,09 0,82 TL / TA 0,17 -0,59 0,17 CDFB / TA 0,21 0,21 LQA / TLB 0,02 SE / TD

5.

E

MPIRICAL

E

VIDENCE

5.1 RESULTS OF A POOLED OLS MODELANALYSIS USINGβAGANDβTAS DEPENDENT VARIABLES

Analyzing data for 16 different models, while distinguishing between traditional market betas and market betas adjusted for ARCH and GARCH effects provides a vast amount of results. A complete overview of the results of all the models for Europe and the U.S. is provided by Appendices 3a – 3d. The significance of the underlying risks is presented by Table 4, which is a short summarization of Appendices 3a – 3d. Concerning the normality of the distribution of European data there can be stated that these are normally distributed for all 16 sixteen models. All these models have proper Jarque Bera values for traditional market betas as well as for market betas adjusted for ARCH and GARCH effects. Nearly all models are significant at an 1 percent level, so these European data can be classified as trustworthy. Unfortunately, U.S. data do not show comparable characteristics. Concerning the statistical significance of the models, there can be stated that they are all significant at 1 percent level. However, Jarque Bera values of these models are dramatically high. The fact that the data are not normally distributed is corresponding with theory supplied by Roy (1952), Kahneman & Tversky (1979) and Gul (1991), that investors differently value downside losses and upside gains. Although the non normal distribution of U.S. data can be explained by economic rationale, the results must be interpreted with care.

concerning European data as is displayed by Table 4, while U.S. data only shows marginal influence of the net charge offs to total loans ratio on market betas. Concerning the sign of the coefficient there can be derived from Table 4 that models using European data fail to generate a clear sign, while all models suggest a positive relationship between the net charge offs to total loans ratio and market betas when U.S. data are used.

TABLE 4 : SIGNS OF VARIABLESDERIVED FROM A POOLED OLS MODEL ANALYSIS USING

βAG_ANDβT_{AS DEPENDENT VARIABLES BY WORLD REGION AND SIGNIFICANCE.*}

Variable Measure Maximum

- + - + - + - + CR NCO / TL 8 4 4 4 4 0 8 0 8 [ 4 ] [ 1 ] LLR / TL 8 0 8 1 7 0 8 0 8 [ 3 ] [ 2 ] [ 2 ] MQR NI / TA 8 0 8 0 8 8 0 8 0 [ 3 ] [ 3 ] [ 4 ] [ 4 ] EA / TA 8 8 0 8 0 2 6 1 7 [ 7 ] [ 8 ] LQR TL / TD 4 2 2 2 2 4 0 4 0 [ 1 ] [ 2 ] [ 2 ] [ 2 ] TL / TA 4 3 1 3 1 4 0 4 0 [ 2 ] [ 1 ] [ 4 ] [ 3 ] CDFB / TA 4 0 4 1 3 0 4 0 4 [ 3 ] [ 2 ] LQA / TLB 4 1 3 1 3 0 4 0 4 [ 4 ] [ 4 ] LVR SE / TD 16 14 2 15 1 12 4 11 5 [ 7 ] [ 10 ] [ 7 ] Dummies 2007 16 8 8 1 15 0 16 0 16 [ 8 ] [ 8 ] [ 16 ] [ 16 ] 2006 16 8 8 8 8 9 7 7 9 [ 8 ] [ 8 ] 2005 16 8 8 8 8 0 16 0 16 [ 8 ] [ 6 ] 2004 16 - - - - 15 1 16 0 World Region Europe U.S. βT βT βAG _βAG

* Amount of models16in which the concerned variable is significant at a 5 percent level are in parentheses. When none of the models generates a significant value of the concerned variable, a blank space is given.

This positive relation in the U.S. is in line with existing literature, since Angbazo (1997) and Ramcharran & Kim (2003) suggest this relationship as well. The loan loss reserve to total loans ratio is not very convincing as well. Derived from Table 4, European and U.S. data show that there is ample reason to think that the loan loss reserve to total loans ratio does influence market betas. Despite the fact that the ARCH and GARCH models perform better than normal OLS models, they fail to prove a significant positive relationship between credit risk and market betas as suggested by Mansur et al. (1993). Although nearly all models indicate a positive relationship between the loan loss reserve to total loans ratio and market betas, only a few signalize a statistical significant positive influence of credit risk. Thus, convincing relations between credit risk and traditional market beta and between credit risk and market beta adjusted for ARCH and GARCH effects can not be found.

Liquidity risk is measured through the total loans to total deposits ratio, the total loans to total assets ratio, the cash and due from banks to total assets ratio and the liquid assets to total liabilities ratio. Concerning the total loans to total deposits ratio there can be stated that it’s influence on both market betas is inconclusive using European data and inverse when data of U.S. banks are used. Table 4 indicates that there is a significant influence of the total loans to total deposits ratio on both market betas in the United States, however not all models support this. The inverse relationship is contradicting to existing literature, since Mansur et al. (1993) argues that total loans are non liquid assets and therefore an increase in total loans leads to higher liquidity risk. Concerning the total loans to total assets ratio there can be stated that there exists a mainly inverse relationship between this accounting ratio and both market betas, although this relationship is stronger for U.S. data. This relationship is also not in line with related literature, comparable to the previously discussed accounting ratio that does approximate for liquidity risk. The third liquidity ratio proposed by Mansur et al. (1993), the cash and due from banks to total assets ratio, is positively related to both market betas. However, Mansur et al. (1993) argues that an increase in liquid assets will have a negative influence on market beta, thus this result is contradicting to existing literature as well. Concerning the liquidity ratio proposed by Angbazo (1997), there can be stated that it has a positive influence on both market betas, although this effect is not statistically significant for European banks. Again, this contradicts related literature, since an increase in liquid assets is supposed to reduce liquidity risk and should therefore lower market betas. All accounting ratios that approximate for liquidity risk are as such related to market betas that they are unanimously contradicting existing literature. An adequate explanation is hardly to find for this contradiction, although the economic rationale that excessive cash may signalize bad investment opportunities makes any sense. Furthermore, based on the liquidity theory of Edgeworth (1888), increased liquid assets may indicate an expected bank run, since that theory suggests that excess reserves will be held to meet the liquidity requirements of an uncertain future.

effects are more convincingly present, especially in the year 2007. Overall, there can be stated that the differences between traditional market betas and market betas adjusted for ARCH and GARCH effects as being market risk measurements are minimal.

5.2 RESULTS OF A POOLEDOLS MODELANALYSIS USINGDOWNSIDE MARKET BETAS AND UPSIDE MARKET BETAS AS DEPENDENT VARIABLES

A complete overview of all the models for Europe and the U.S. concerning downside market beta and upside market betas as the dependent variable is provided by Appendices 3e – 3h. The significance of the underlying risks is presented by Table 5, which is a short summary of Appendices 3e – 3h. Concerning the normality of the distribution of European data there can be stated that these are normally distributed for models which have downside market beta as dependent variable. All these models have proper Jarque Bera values, however these values are higher for the models which have upside market beta as dependent variable. Since the data used for upside market beta models are slightly positively skewed and leptokurtic, there can be stated that these follow a non normal distribution. Nearly all models are statistically significant, thus these European data can be classified as trustworthy. Concerning the distribution of data of U.S. banks there can be stated that the normality is improved compared to the traditional market beta analysis. Jarque Bera values of the traditional market beta analysis and the market beta adjusted for ARCH and GARCH effects analysis are on average above 200 while these values for the downside market beta analysis are around 50 and for the upside market beta analysis around 20. Although the data is not adequately normally distributed, there can be derived that separating downside market betas from upside market betas does explain the data more effectively than traditional market beta. Concerning the skewness of European as well as U.S. data there can be stated that they differ between downside market beta analysis and upside market beta analysis. The fact that these differ is corresponding with theory supplied by Roy (1952), Kahneman & Tversky (1979) and Gul (1991), that investors differently value downside losses and upside gains.

TABLE 5 : SIGNS OF VARIABLES DERIVED FROM A POOLED OLS MODEL ANALYSIS USINGβ–

ANDβ+AS DEPENDENT VARIABLES BY WORLD REGION AND SIGNIFICANCE*

* Amount of models in which the concerned variable is significant at a 5 percent level are in parentheses. When none of the models generates a significant value of the concerned variable, a blank space is given.

Upside market betas are also influenced by credit risk concerning U.S. data, however the effect is not that strong as on downside market betas. European data for the upside market beta analysis is classified as inconclusive.

Concerning management quality risk there can be derived from Table 5 that management quality risk has a reversed influence17_{on European data compared to U.S. data, which is also}

shown by the traditional market beta analysis. Also comparable to the traditional market beta analysis is the reversed influence on both market betas of the net income to total assets ratio

17_{Since the results shown by Table 3 are similar to results provided by Table 1 concerning the sign of} management quality risk’s coefficients, comparing results to existing literature will be identical to the traditional market beta analysis and therefore will be superfluous.

Variable Measure Maximum

- + - + - + - + CR NCO / TL 8 0 8 5 3 0 8 1 7 [ 3 ] LLR / TL 8 0 8 3 5 1 7 3 5 [ 2 ] [ 1 ] MQR NI / TA 8 0 8 0 8 8 0 8 0 [ 4 ] [ 3 ] [ 8 ] EA / TA 8 8 0 8 0 2 6 3 5 [ 1 ] [ 5 ] LQR TL / TD 4 2 2 2 2 4 0 4 0 [ 2 ] [ 2 ] [ 4 ] [ 4 ] TL / TA 4 4 0 3 1 4 0 4 0 [ 2 ] [ 2 ] [ 4 ] [ 3 ] CDFB / TA 4 1 3 0 4 2 2 0 4 [ 1 ] LQA / TLB 4 0 4 2 2 0 4 0 4 [ 4 ] [ 3 ] LVR SE / TD 16 10 6 14 2 11 5 12 4 [ 6 ] [ 6 ] [ 3 ] [ 12 ] Dummies 2007 16 0 16 6 8 0 16 0 16 [ 8 ] [ 4 ] [ 16 ] [ 16 ] 2006 16 0 16 11 5 16 0 0 16 [ 8 ] 2005 16 4 12 11 5 16 0 0 16 [ 8 ] [ 3 ] [ 16 ] 2004 16 0 16 16 0 0 6 0 16 [ 8 ] [ 3 ] World Region Europe U.S.

compared to the influence of the earning assets to total assets ratio. Apparently, management quality risk has the same influence on downside market beta as well as on upside market beta. Regarding the statistical significance there can be derived from Table 5 that both ratios have a substantial significant influence on both market betas concerning European data whereas the convincing statistical significant influence concerning U.S. data solely belongs to the influence of the net income to total assets ratio on downside market beta. Therefore, for U.S. data, there can be assumed that the affect of the net income to total assets ratio is stronger on downside market beta than on upside market beta.

The first ratio18_{that approximates liquidity risk, the total loans to total deposits ratio has a}

inconclusive influence on both market betas using European data, however has a convincing negative influence on downside market beta as well as on upside market beta when data of U.S. banks are used. As argued in the discussion about the traditional market beta analysis this negative sign is contradicting to existing literature. An explanation for the inverse influence of liquidity risk on downside market beta as well as on upside market beta is also given in that discussion. A possible explanation for the negative influence of liquidity risk on upside market beta can be that excessive cash may signalize bad investment opportunities. On the other hand, a feasible clarification for the inverse relationship between liquidity risk and downside market beta can be that increased liquid assets may indicate an expected bank run as suggested by Edgeworth (1888).

Concerning leverage risk there can be stated based on Table 5 that it has mainly a negative affect on both market betas. This is in line with existing literature since Mansur et al. (1993) points out that increased shareholders’ equity or decreased total deposits will lower market beta. For European data, it does apparently not make any difference to this influence whether there is a bullish economy or a depressed economy, since the relationship between leverage risk and market beta is as even strong for downside market beta as for upside market beta. This relationship is somewhat confusing when U.S. data is used especially concerning downside market beta since they are mainly negatively influenced by leverage risk, however the thrifty significant results surprisingly have a positive sign.

Without being very explicit about the year dummies, there can be derived from Table 5 that their signs are more convincing for downside and upside market betas than for traditional market betas. Furthermore, the influence of unknown factors in a particular year on market beta

do apparently differ between downside and upside market betas. For U.S. data, they have a similar affect, concerning the sign, on both market betas in 2004 and in 2007, however this affect is dissimilar in the years in between. Despite the fact that there are some doubtful results, the results show predominantly that unknown factors in any particular year are increasing prevalently downside market beta. This increase pertains as well Europe as U.S. and is very convincing for the year 2007, however it is questionable whether this can directly be related to the recent financial crisis. Financial crises tend to affect the entire market, which can be measured by market beta since it is a systematic risk. However, bank runs are a form of unsystematic risk as argued in the literature section and may therefore not captured by market beta.

Overall, there can be stated that the differences between downside market betas and upside market betas being market risk measurements are limited. Credit risk as well as dummy variables have a different influence on downside market beta than on upside market beta. Remaining variables do mainly have an identical influence on both market risk measurements. As can be derived from Table 1, the expected signs of the variables influencing upside market beta were unknown. However, results point out that the influence of underlying risks on market beta does not heavily change with the state of economy.

5.3 RESULTS OF A POOLED OLS MODEL ANALYSIS USING TOTAL RISK AS DEPENDENT VARIABLE

Concerning total risk19 _{as dependent variable, the model changes slightly, since a}

semilogarithmic regression method is used, which changes the model into a nonlinear model. Applying a loglin form of semilogarithmics transforms the left hand side of (9) into ln(βijk),

leaving the right hand side unchanged. Using semilogarithmic regressions, nonlinear relationships are approximated by linear relationships, which causes errors in the coefficients of the parameters as suggested by Halvorsen & Palmquist (1980) and Thornton & Innes (1989). Comparing these coefficients to coefficients derived from linear models may therefore lead to wrong conclusions. Thornton & Innes (1989) argues also that when the sum of a coefficient times a unit change20in the accessory independent variable is close to zero, the coefficient can be seen as close the true coefficient. Since only the earning assets to total assets ratio provokes this sum to be obviously

19_{Semilogarithmics are also incidentally used concerning downside and upside market betas as dependent} variables for U.S. data.

larger than zero, as can be derived from Table 6, the coefficients derived from models using semilogarithmics are supposed to be close to the true coefficients21_.

TABLE 6 : SUM OF COEFFICIENTS TIMES UNIT CHANGE IN VARIABLE

Variable

Europe U.S. Europe U.S. Europe U.S. Europe U.S. Europe U.S. NCO/TL 0,00 0,05 0,00 0,05 0,05 0,03 -0,01 0,02 0,03 0,03 LLR/TL 0,04 0,09 0,02 0,07 0,05 0,03 0,02 0,02 0,14 0,01 NI/TA 0,02 -0,10 0,03 -0,08 0,03 -0,06 0,02 -0,05 -0,03 -0,04 EA/TA -2,34 0,10 -2,80 0,22 -1,22 0,11 -3,17 0,08 -0,38 0,95 TL/TD -0,05 -0,03 0,00 -0,04 -0,01 -0,06 -0,06 -0,06 -0,06 -0,05 TL/TA -0,16 -0,22 -0,15 -0,22 -0,16 -0,24 -0,12 -0,19 -0,46 -0,39 CDFB/TA 0,03 0,15 0,04 0,14 0,01 0,00 0,08 0,09 0,11 0,15 LQA/TLB 0,01 0,09 0,02 0,07 0,03 0,07 0,01 0,06 0,10 0,03 SE/TD -0,15 -0,02 -0,21 0,00 -0,13 0,02 -0,18 -0,06 0,01 0,11 2007 0,06 0,04 0,07 0,06 0,20 0,04 0,02 0,11 -0,07 0,09 2006 0,03 0,00 0,04 0,00 0,17 -0,01 -0,04 0,01 -0,16 -0,10 2005 0,01 0,01 0,02 0,01 0,10 -0,02 -0,03 0,04 -0,26 -0,07 2004 0,00 0,00 0,00 -0,01 0,02 0,01 -0,01 0,01 -0,02 -0,07

ARCH/GARCH Beta Traditional Beta Downside Beta Upside Beta Total Risk

Taking total risk as dependent variable the market risk measure incorporates systematic as well as unsystematic risk. Therefore, compared to market beta, total risk should be able to capture the effect of sector specific events like bank runs, which are an unsystematic risk. Results of analyzing with a pooled OLS model with total risk as dependent variable are given by Appendices 3i and 3j, however a short summarization of these appendices is provided by Table 7. Concerning the statistical trustworthy of this result there can be stated that the European data follow a normal distribution. Their Jarque Bera values are all from an acceptable size and the models do all generate significant F-statistics. On the other hand, Jarque Bera values for U.S. data are not satisfying, with values that are fluctuating around 40. Again, this is confirming literature that suggests that stock return data are not normally distributed as among others argued by Roy (1952), Kahneman & Tversky (1979) and Gul (1991).

Concerning credit risk there can be stated for both accounting ratios that approximate for this risk, that they are statistical convincingly affecting European market betas in a positive way. This confirms and extends previous evidence about the relation between credit risk and market betas. Regarding U.S. data the convincing influence of credit risk dilutes strongly compared to the influence of credit risk on market betas. The coefficients of credit risk show even inverse results compared to the previous analyses incorporating market betas. As explained in the theoretical section, defaulting debtors can be seen as a macro economical problem, which makes

credit risk a systematic risk. Therefore, it is no curious apparition that total risk is a less suitable risk measurement for determining a strong influence of credit risk.

Regarding management quality risk there can be derived from Table 7 that the results are quite similar to that of Table 4. Only for European data there is observed a difference, since the earning assets to total assets ratio does apparently influence total risk in a negative way. However, none of the models qualifies this effect as statistically significant so no reliable conclusions can be based on this.

TABLE 7 : SIGNS OF VARIABLES DERIVED FROM A POOLED OLS MODEL ANALYSIS USING

TOTAL RISK AS DEPENDENT VARIABLE BY WORLD REGION AND SIGNIFICANCE.*

Variable Measure Maximum

- + - + CR NCO / TL 8 0 8 8 0 [ 8 ] [ 1 ] LLR / TL 8 0 8 3 5 [ 7 ] MQR NI / TA 8 8 0 8 0 [ 4 ] [ 8 ] EA / TA 8 7 1 0 8 [ 2 ] LQR TL / TD 4 2 2 4 0 [ 2 ] TL / TA 4 4 0 4 0 [ 3 ] CDFB / TA 4 0 4 0 4 LQA / TLB 4 0 4 1 3 LVR SE / TD 16 11 5 0 16 [ 2 ] [ 12 ] Dummies 2007 16 8 8 0 16 [ 8 ] [ 16 ] 2006 16 16 0 16 0 [ 8 ] [ 16 ] 2005 16 16 0 16 0 [ 16 ] [ 16 ] 2004 16 16 0 16 0 [ 16 ] Europe U.S. World Region

Concerning liquidity risk there can be stated that the signs of the coefficients are predominantly comparable to that of the traditional market beta analysis. However, European data do not signalize any significant variable, while only the total loans to total deposits ratio and the total loans to total assets ratio appear to have a significant influence on total risk of U.S. banks. Although the variables are not statistically significant for European data their sign is once more contradicting to theory suggested in the literature section.

Since total risk incorporates unsystematic risk as well, this should logically be a better way to determine the sensitivity of stock returns to for example bank runs as suggested by the literature section. Of course, there are more and perhaps even better examples of unsystematic risks, however bank runs are a major threat to banking companies and are therefore supposed to illustrate adequately the importance of incorporating unsystematic risk. In case of a bank run people are en masse withdrawing their money, which leads to dramatically decreased deposit accounts of banks. Therefore, it is argued that the threat of a bank run may provoke increased leverage risk. As can be derived from Table 7, only concerning U.S. banks, there appears to be an extremely strong relation between leverage risk and total risk. The pooled OLS model generates solely positive relationships between leverage risk and total risk, of which almost all are statistically significant. Since pooled OLS models that use other market risk measurements indicate neither or opposite results, there can be argued that total risk is an appropriate risk measurement to determine the influence of increased leverage risk, which is plausibly caused by an imminent bank run. Furthermore, the significant positive influence of leverage risk and the contradicting influence of liquidity risk on market betas are supporting each other. When banks signalize declining deposit accounts they may be threatened by a bank run and will therefore hold more liquid assets. Although there may be other plausible explanations for the positive relation between leverage risk and total risk, the bank run example makes certainly sense.

5.4 APPLYING INTERRELATED INDEPENDENT VARIABLES

As abovementioned, there are several accounting ratios that have a significant influence on traditional market betas as well as on the other risk measures. These influences are measured over a period of 5 years, however it is questionable whether these influences should not change with annually changing economic conditions. Since there is reason to think that economic conditions do change frequently over time, the influence of each underlying risk will be measured yearly, using interrelated variables. Although these economic conditions are not specified, and they are therefore representing unknown factors, it is expected that the influence of a certain underlying risk will change in subsequent years. The annual influence of each underlying risk on market betas will be examined, however not all accounting ratios are taken into account. For simplicity reasons, solely the most convincing representing accounting ratios will be used for this interrelated variable analysis. A complete overview of the gathered empirical evidence is given by Table 8. As can be derived from this table, there is little difference between the convincingness of both accounting ratios that approximate credit risk.

TABLE 8 : EMPIRICALLY DERIVED SIGNS OF VARIABLES BY BETA TYPE*

Variable Ratio

Europe U.S. Europe U.S. Europe U.S. Europe U.S. Europe U.S. Credit Risk NCO / TL ? + ? ++ + ++ ? + +++

-LLR / TL + + ++ + + + ? ? +++ ?

Man. Quality Risk NI / TA ++ - - ++ - - ++ - - - ++ - - -

-EA / TA - - - + - - - ? - ? - - ? - + Liquidity Risk TL / TD ? - - ? - - ? - - - ? - - - ? -TL / TA - - - -CDFB / TA ? ++ + +++ ? ? + + + + LQA / TLB ? +++ ? +++ + +++ ? +++ + ? Leverage Risk SE / TD - - ? - - - ? - - - +++ Dummies 2007 ++ +++ ++ +++ ++ +++ + +++ - - +++ 2006 ++ ? ++ ? ++ - ? + - - -2005 ++ + ++ + ++ - ? +++ - - - -2004 ? - ? - ++ + - + - -βAG β– β+ σ2 βT

+++ = In at least 75 % of all models, the variable has a positive significant influence. ++ = In at least 50% of all models, the variable has a positive significant influence. + = In less than 50% of all models, the variable has a significant positive influence. ? = There is no convincing sign of the variable.

However, since the loan loss reserve to total loans ratio gives a unanimous sign for all market risk measures, this ratio is chosen to represent credit risk. Concerning management quality risk there can be stated that the net income to total assets ratio is the best representing ratio. This ratio gives for Europe as well as for the United States a convincing sign. The most convincing accounting ratio that approximates liquidity risk is the total loans to total assets ratio. Although there are no contradicting signs between the four liquidity risk approximations, the total loans to total assets ratio is the most consistent one for Europe as well as for the United States. Leverage risk is solely measured by the shareholders’ equity to total deposits ratio, which will logically be used. Summarizations22_{of the results of these analyses for European and U.S. data are given by}

Table 9 and Table 10 respectively. Although the non-significant signs of coefficients might provide useful insights, only the significant signs of coefficients are being discussed.

Concerning credit risk there can be stated based on Table 9 that the risk that European banks bear changes annually, when total risk is taken as the dependent variable. Results point out that the lowest amount of credit risk is found in the year 2004. The highest level23_{of credit risk is}

found in the year 2005, whereupon it slightly diminishes in the following years. Thus, there is convincing24_{evidence that the relationship between credit risk and total risk changes over time}

and can therefore not be judged as being constant. Management quality risk, measured by the net income to total assets ratio, influences all betas except βAG _{significantly, when interrelated}

variables are used. The highest levels of management quality risk are prevalently found in the year 2005, however this is only the case when market risk measurements are used, which only measure systematic risk. Results indicate that, when total risk is used as the dependent variable, the least lowest level of management quality risk is found in 2006. Although the signs and annual size of coefficients differ between systematic risk and total risk measurements, there can be stated that the influence of management quality risk measured by the net income to total assets ratio changes annually. Concerning the total loans to total assets ratio, which approximates liquidity risk, there can be stated that analyzing with total risk as dependent variable leads to inconclusive results. On the contrary, analyzing with traditional market beta, market beta adjusted for ARCH and GARCH effects and with downside market beta indicates that the coefficients of the relation between liquidity risk and market beta is prevalently the largest in the year 2007.

22_{Results are not provided by appendices, however they are demandable through mailing to the email} address mentioned on the front page.

23_{The highest level is resembled by that coefficient that is positively influencing market beta and is larger} than coefficients of other years.

TABLE 9 : SIGNS25OF VARIABLES DERIVED FROM A POOLED OLS MODEL ANALYSIS USING

INTERRELATED VARIABLES BY SIGNIFICANCE USING EUROPEAN DATA.*

Variable Measure Max.

- + - + - + - + - + CR LLR / TL 8 1 7 0 8 0 8 8 0 8 0 [ 8 ] LLR / TL 8 8 0 8 0 8 0 5 3 0 8 2007 _{[ 8 ]} LLR / TL 8 8 0 8 0 8 0 0 8 0 8 2006 _{[ 8 ]} LLR / TL 8 5 3 7 1 8 0 0 8 0 8 2005 _{[ 8 ]} LLR / TL 8 0 8 1 7 8 0 0 8 0 8 2004 _{[ 2 ]} MQR NI / TA 8 8 0 7 1 0 8 0 8 8 0 [ 4 ] [ 4 ] [ 3 ] [ 8 ] NI / TA 8 1 7 3 5 8 0 8 0 0 8 2007 _{[ 4 ] [ 2 ] [ 2 ] [ 6 ]} NI / TA 8 1 7 1 7 8 0 4 4 0 8 2006 _{[ 4 ] [ 4 ] [ 2 ] [ 8 ]} NI / TA 8 0 8 1 7 8 0 2 6 0 8 2005 _{[ 4 ] [ 4 ]} LQR TL / TA 4 0 4 4 0 4 0 4 0 2 2 [ 1 ] [ 2 ] TL / TA 4 3 1 0 4 0 4 0 4 2 2 2007 _{[ 1 ] [ 1 ]} TL / TA 4 4 0 1 3 1 3 0 4 2 2 2006 _{[ 1 ] [ 2 [} TL / TA 4 4 0 0 4 0 4 0 4 2 2 2005 _{[ 2 ]} LVR SE / TD 16 7 9 15 1 16 0 16 0 8 8 [ 4 ] [ 1 ] [ 9 ] SE / TD 16 9 7 1 15 0 16 0 16 8 8 2007 _{[ 9 ]} SE / TD 16 10 6 2 14 1 15 0 16 8 8 2006 _{[ 6 ]} SE / TD 16 13 3 6 10 6 10 0 16 8 8 2005 _{[ 3 ]} σ2 βT _βAG _β– _β+ Dependent Variable

* Amount of models in which the concerned variable is significant at a 5 percent level are in parentheses. When none of the models generates a significant value of the concerned variable, a blank space is given.

Although the coefficients are not very significant, there can be stated that there is reason to argue that the influence of liquidity risk, measured by the total loans to total assets ratio, differs by year. The shareholders’ equity to total deposits ratio, which approximates leverage risk, influences market beta in a slightly convincing way, when the market beta adjusted for ARCH and GARCH effects, downside market beta or upside market beta is used as dependent variable.

However, since significant results are mainly found when upside market beta is used as dependent variable, there is no reason to think that the influence of leverage risk on market beta changes annually.

TABLE 10 : SIGNS OF VARIABLES DERIVED FROM A POOLED OLS MODEL ANALYSIS USING

INTERRELATED VARIABLES BY SIGNIFICANCE USING U.S. DATA.*

Variable Measure Max.

- + - + - + - + - + CR LLR / TL 8 0 8 0 8 1 7 0 8 0 8 [ 5 ] [ 4 ] [ 8 ] [ 1 ] LLR / TL 8 8 0 8 0 0 8 8 0 8 0 2007 _{[ 2 ]} LLR / TL 8 8 0 8 0 8 0 8 0 8 0 2006 _{[ 1 ]} LLR / TL 8 8 0 8 0 3 5 8 0 8 0 2005 _{[ 5 ] [ 8 ] [ 1 ]} LLR / TL 8 8 0 8 0 0 8 8 0 8 0 2004 MQR NI / TA 8 4 4 5 3 8 0 0 8 3 5 [ 2 ] NI / TA 8 8 0 7 1 0 8 8 0 8 0 2007 _{[ 6 ]} NI / TA 8 8 0 8 0 8 0 7 1 8 0 2006 NI / TA 8 7 1 2 6 0 8 8 0 8 0 2005 NI / TA 8 7 1 1 7 0 8 8 0 8 0 2004 LQR TL / TA 4 4 0 4 0 4 0 4 0 4 0 [ 1 ] [ 2 ] [ 4 ] [ 2 ] TL / TA 4 0 4 0 4 0 4 2 2 0 4 2007 TL / TA 4 4 0 2 2 1 3 2 2 0 4 2006 TL / TA 4 0 4 0 4 0 4 2 2 0 4 2005 TL / TA 4 3 1 0 4 0 4 4 0 0 4 2004 LVR SE / TD 16 14 2 16 0 0 16 16 0 0 16 [ 13 ] SE / TD 16 0 16 0 16 0 16 0 16 16 0 2007 _{[ 1 ]} SE / TD 16 12 4 1 15 11 5 16 0 16 0 2006 _{[ 2 ]} SE / TD 16 13 3 12 4 12 4 16 0 16 0 2005 _{[ 9 ] [ 6 ]} SE / TD 16 0 16 0 16 0 16 0 16 16 0 2004 σ2 Dependent Variable βT _βAG _β– _β+

Although the analyses of European data yield several interesting annual changes, analyzing with U.S. data yields less significant result. Concerning credit risk of U.S. banks, there can be stated that significant results are prevalently been found in the year 2005, however this can not be claimed for all market risk measurements. Results do also indicate that the coefficients of the loan loss reserve to total loans ratio in the year 2005 are larger than in other years.

Especially when the upside market beta is used as dependent variable these annually changing size of coefficients appears to be significant. When the influence of management quality risk is measured using U.S. data, there can be derived from Table 10 that this influence is significantly present, when total risk is used as dependent variable. This influence is only significant in the year 2007. Additionally, results point out that the coefficients of these significant variables are lower than in other years. Thus, there can be stated that U.S. banks bear less management quality risk in 2007, compared to preceding years. Of course, recalling from Table 1, the expected sign is inconclusive, so a rising coefficient may indicate that banking management does more aggressively takes risk or may indicate that banking management does acquire worse quality assets and market shares. As presented by Table 10, liquidity risk seems apparently not to have any annual changing influence on market beta and will therefore not be discussed furthermore. Concerning leverage risk there can be stated that the most convincing results are found in the year 2005. In addition, results do also suggest that the influence of leverage risk on market beta is significantly lower in 2005, compared to other years. Although the coefficients are not statistically significant, there can be stated that leverage risk in the U.S. is at it’s highest level in the year 2007.

6.

C

ONCLUSION

6.1 IMPORTANT EVIDENCE