1
Abstract
This paper investigates the relationship between bank capital levels prior to the crisis and subsequent stock price performance during the financial crisis that started in 2007. My paper follows a similar analysis as Demirguc-Kunt et al (2010) and Akhigbe et al (2012). Both papers analyze the relationship between bank capital and subsequent stock returns while also controlling for other bank specific characteristics such as size and liquidity. However, unlike these papers, my sample consists only of banks from the European Union countries. My sample of 118 banks is further divided in three different groups depending on total assets size. Another grouping of banks is based on whether or not they are members in the Eurozone. My null-hypothesis (H0)is
that capital levels are not related to subsequent stock price performance. The alternative hypothesis (H1) is that capital is negatively related to subsequent stock price performance i.e.
banks had poor quality assets. To test my hypothesis, first I run an OLS regression including only bank capital (defined as prior crisis Tier 1 ratio). The results on the full sample and the three different groups of banks based on size are insignificant. Only for the banks from non-Eurozone countries Tier 1 has a marginally significant effect. Furthermore, to deepen my analysis, I also include six other control variables such as size, reliance on non-interest income, and net-interest margin. I find that bank size is negatively related to subsequent stock returns at 1% significance level. Other variables that are found to be significant in explaining the stock performance during the crisis include assets growth in the year prior to the crisis, the ratio of liquid assets to total deposits and borrowings, and net-interest margin.
2
Introduction
The financial crisis which started in 2007 injected a large degree of uncertainty into financial markets all over the world. Financial institutions suffered large losses related to subprime mortgages, mortgage-backed securities (MBS), collateralized debt obligations (CDO) and other assets that were wrongly assumed to be safe and of excellent quality. The stock prices of many banks plummeted and are yet to recover. Other banks were acquired at distressed prices and some were even forced to declare bankruptcy. Some financial institutions suffered large losses and ended up being acquired at a very low price by more stable banks. The financial crisis caused a recession in many developed economies around the world and had damaging longer-term effects on worldwide economic growth. Much has been published in recent years about the causes of the financial crisis, about banks’ business models, and many other topics.
My paper aims to explore the relationship between the bank capital of European banks prior to the crisis and the subsequent stock price declines for the same banks during the financial crisis. Spence (1974) is one of the first authors who explore signaling theory and its use in the presence of asymmetric information. Since then this theory has been the foundation for many theoretical models and served as the basis for testing a wide range of hypotheses. Akhigbe et al (2012) mention that one of the main assumptions in the capital signaling theory is that there is information asymmetry between bank managers and investors regarding the quality of the assets held by the bank. The information available to the public about a bank’s assets is limited, which in turn leads to investors not being able to fully evaluate the real quality of those assets. Because of the existing requirements about risk-based capital, a certain bank’s capital levels may be perceived as a signal about the quality of its assets. Therefore, many investors may rely largely on the capital ratios as a signaling device of the bank’s risk exposure.
3
for different financial instruments and also underestimated interconnectedness and the related systemic risk. Choudhry (2012) implies that an unplanned result of the original Basel accord was that securitization became very popular. Basel II was supposed to improve the Basel I rules and introduce an even safer framework, however, it failed to properly address securitization and structured products, which were two of the main reasons for the crisis.
This study tests a capital signaling hypothesis where capital is indicative of asset quality. My analysis assumes that there are two main reasons why banks held large amounts of capital. On the one hand, some banks engage in more risk averse practices, thus holding more capital would hedge against some risk. These banks would have performed better during the financial crisis. On the other hand, banks may hold higher capital but the quality of their assets might turn out to be poor. The higher capital would imply a worse performance in the financial crisis because in a situation like this asset prices decline severely and capital turns out to be insufficient. This would lead to worse stock price performance for banks which were originally holding larger amounts of capital. My paper follows a similar analysis done by Akhibe et al (2012) and Demirguc-Kunt et al (2010).
Nevertheless, there is some disagreement in the literature that explores the effects of bank capital over the stock performance. On the one hand, Akhibe et al (2012) find that better capitalized banks were the ones that experienced sharper declines in their stock prices. On the other hand, Demirguc-Kunt et al (2010) find some evidence that banks with higher capital (measured as total regulatory capital scaled by total un-weighted assets) prior to the crisis also had better performance during the crisis. The coefficient for their result, however, is small and only marginally significant. A main reason for the different results that these authors acquire is that they use different samples of banks for their empirical analysis. Akhibe et al (2012) use only U.S. banks, while Demirguc-Kunt et al (2010) use banks from all over the world. Therefore, these findings suggests that if some evidence works for the U.S., this does not mean that it will work for samples that consist of banks from other countries or regions. However, I was not able to find any paper that focuses on European banks. That is why, my analysis which uses sample of 118 European Union banks, will try to shed some light over this question – whether or not prior bank capital levels of European Union banks are related to subsequent stock returns during the financial crisis.
4
of the specific nature of the financial crisis in Europe. The European Union crisis is different than the one in the U.S. – it lasts longer and the existence of the euro addresses a different set of challenges than sovereign currencies (i.e. there is a much stronger separation between fiscal and monetary policy in Europe than in most other countries). Speed of response to shore up capital and assist banks is also slower in Europe than in the US because the mechanism through which they decide is more cumbersome. By breaking the sample into Eurozone and non-Eurozone banks I also want to check if the regulatory framework in the Eurozone worked differently than outside of it.
This topic of what might affect stock price performance during crisis periods is particularly important both for regulatory authorities and for investors. For regulation purposes, it is essential to see what the relationship, if any, is between various bank characteristic and bank performance during the crisis because that way the regulator might intervene with emergency liquidity injections, special assistance and even funds to recapitalize. Also, such an analysis might prove to be helpful in identifying which banks are more likely to be affected – that is why I also break up my full sample in three different groups of banks depending on their total assets size. For these reasons, an empirical research on this topic could prove to be useful for regulators – determining the factors that lead to better or worse performance during the crisis can be used to improve current regulation frameworks
From an investment perspective, such analysis is relevant because it can give insights into which characteristics of a bank can be predictors of better performance in the event of a crisis. There are different types of investors depending on their risk appetite. Therefore, risk-averse investors for example, will be able to make a better choice between different bank investments when looking at their specific characteristics.
5
Section II. Literature Review
Akhibe et al (2012) investigate the relationship between the bank capital level prior to the 2007–2009 financial crisis and the exposure of bank stock prices and stock volatility during the crisis. They use weighted least squares to perform a multivariate analysis applied to the pre-crisis and crisis period for 288 U.S. banks. Besides bank capital, the authors also use some additional bank characteristics such as bank size, concentration in real estate loans, bank stock liquidity etc., measured prior to the crisis. The results support the capital signaling hypothesis where bank capital levels prior to the crisis serves as an indicator of the quality of the assets during the crisis. Moreover, the authors find that larger banks actually experienced larger stock price decline during the crisis. The paper also shows that banks’ stock returns during the crisis were positively related with pre-crisis levels of non-interest income and return on average assets.
In the same vein, Demirguc-Kunt et al (2010) investigate whether better capitalized banks experienced smaller declines in their stock prices during the financial crisis. The authors perform an OLS regression and besides using different capital ratios (Tier 1, Tier 2, Leverage Ratio), they also control for other independent variables such as liquidity, net loans, deposits and others. Their analysis of 381 banks from all over the world, shows that differences in capital levels prior to the crisis did not really affect subsequent stock price performance. They only find that the pre crisis leverage ratio, measured as total capital ratio divided by total assets, is marginally significant and with a small positive coefficient. However, they find strong evidence that, during the crisis period, banks with higher capital levels performed better (i.e. they experienced smaller subsequent decline in their stock prices). Still, this effect is large only for the sample that consists of large banks. For this group of banks, this relationship is particularly strong when capital ratio is measured as the regulatory capital divided by total assets. The authors mention that this particular finding indicates that market participants did not see the risk-adjustment under Basel to be informative and capable in capturing the actual risk for banks during the crisis.
6
that, all else equal, an increase in the capital ratio requirements reduces the risk taking incentives. Another result from their analysis is that in order to meet higher capital standards, banks that focus on value maximization choose to raise additional capital rather than just sell assets or retire deposits.
Altunbas and Manganelli (2011) analyze how bank risk relates to their business models during the financial crisis. They show that institutions with higher risk exposure were large in size, undercapitalized and relied heavily on short-term funding. On the other hand, business models geared towards reduced bank risk are characterized by a solid deposit base and diversified income sources. The authors consider that, due to structural changes brought by deregulation and innovations, transparency in the banking sector has decreased and made the institutions more dependent on financial markets. The paper also shows that banks that possessed strong deposit base performed better than those that were more dependent on short-term market funding.
Fahlenbrach et al (2011) investigate whether there is a relationship between banks’ performance in the 1998 crisis and their performance in the recent financial crisis. They mainly test for two hypotheses- the learning hypothesis and the business model hypothesis. The learning hypothesis suggests that banks that performed poorly in the 1998 crisis learned from their mistakes and took actions to make sure that they would not face the same experience again. The business model hypothesis, on the other hand, suggests that banks’ sensitivity to crises depends on their business model and banks do not tend to change business models, usually because it would not be profitable for them to make this change. As an example that business models do matter, Loutskina and Strahan (2011) find that mortgage lenders that are more geographically concentrated invest more in information collection and therefore performed better than diversified lenders during the crisis. The results from Fahlenbrach et al (2011) support the business model hypothesis. The authors find that banks that were affected in a crisis, for some reason (possible reduction of future profitability, for example), do not appear to change their business models and become more risk-averse. Their results show that a bank’s stock performance during the 1998 crisis is a good predictor for a bank’s performance during the 2007 financial crisis.
7
governance models, country regulations and balance sheet indicators prior to the crisis affected bank’s subsequent performance. A main finding of Beltratti and Stulz (2009) is that banks from countries with stricter restrictions on bank activities and banks with more Tier 1 capital performed better. Better performance was also related to less leverage and lower returns right before the crisis. The paper also gives evidence that banks with the largest stock increases prior to the crisis were in fact the ones that experienced greater stock price decline during the financial crisis.
Gorton and Metrick (2010) address repo transactions that are often collateralized with securitized bonds. The authors argue that securitization in combination with repo finance had a strong negative impact on banks and deepened the financial crisis. They use data that includes credit spreads for a large number of securitized bonds to investigate how the crisis spread from subprime assets related to housing into other markets that had no relation to housing. The authors also mention that at one point the market became aware of the large subprime-related risks and that in turn led to doubts about bank solvency ant the quality of the collateral used in repo transactions.
Amel-Zadeh and Meeks (2011) mention that even though a bank possesses enough capital according to Basel regulations, it might still fail because of a sudden lack of liquidity. Allen and Wood (2005) also discuss the capital adequacy rules and raise some reasonable doubts that Pillar 1 of the Basel 2 regulatory standards, which consists of minimum capital requirements, might actually do more harm than good. One of the reasons they give for this assessment is that Pillar 1 actually represents an “officially designed framework for risk management”, which after a certain period of time will undoubtedly become outdated. However, because of its mandatory character and slow implementation of new improvements, banks will have fewer incentives to develop their own risk management.
8
market participants were actually well aware of the changes in the risk exposure of the bank’s portfolio and these changes were subsequently reflected in the bank’s stock performance.
Cetorelli and Goldberg (2011) find that globalization is an important factor that has a deep impact on the consequences of the international liquidity shocks. In particular, the authors observe that global banks that have active internal capital markets contribute to international shocks because a liquidity squeeze is now transferred to those banks’ foreign branches or subsidiaries. Goldberg (2009) also provides arguments that globalization, through internal capital markets and bank’s foreign subsidiaries, is a main factor in the transmission of shocks. This article also implicates that, initially, host countries’ regulators and supervisors might not be fully prepared to assess the new and more complex products that are introduced by foreign banks. This suggests that globalization is also related to host country regulatory and supervisory authorities and these countries should make sure that their evaluation capabilities of new products and services are kept up to date.
Section III. Methodology
9
My analysis will test whether banks held higher levels of capital because they wanted to be "safer", which would imply a better performance in the financial crisis, or, whether banks held higher levels of capital and the quality of their assets was poor, which would imply a worse performance in the financial crisis. The null-hypothesis (H0)is that capital levels are not related
to subsequent stock price performance. The alternative hypothesis (H1) is that capital is
negatively related to subsequent stock price performance (i.e. banks had poor quality assets). Demirguc-Kunt et al (2010) provide strong evidence that supports the hypothesis that banks with higher levels of capital experienced a smaller decline in their stock prices during the crisis. However, this holds when Tier 1 is measured when the crisis already started, not prior to the crisis. Besides the full sample of banks in their analysis, the authors also have a sample consisting only of large banks.
To test my hypothesis, first I collected financial data for 118 banks using Bankscope. The sample consists only of European banks, unlike Demirguc-Kunt et al (2010) who use banks from all over the world. My sample also excludes any foreign subsidiaries that might operate in the European Union. I will also study the relationship between capital and subsequent stock price return when I control for bank characteristics that can be expected to impact stock price performance during the financial crisis. Those additional characteristics include various assets and income ratios. As opposed to Demirguc-Kunt et al (2010), who only control for large banks, I will also divide the full sample of banks in different groups – depending on total assets size I differentiate between large, medium-sized and small banks. Furthermore, I also provide another division of the full sample based on membership in the Eurozone.
Models employed
To estimate my model, I will first run the following OLS regression using only the Tier 1 capital ratio as the independent variable. :
10
where is the bank’s stock return for the period between April 2007 and February 2009, is the Tier 1 capital ratio measured as of December 31, 2006, and is the error term.
I will run another regression in which I also control for other variables such as bank size, dependence on non-interest income, net-interest margin etc. The equation for this second equation will be:
( )
where SIZE is the bank size measured as the natural logarithm of the bank’s total assets, NONINT is the non-interest income divided by gross revenues, LIQASD is the amount of available liquid assets divided by the amount of total deposits and borrowings, ROAA is the return of average assets, GROW is the percentage growth of the total assets in the year prior to the crisis and NIM is the net-interest margin of the bank. All of these additional independent variables are measured as of December 31, 2006. My equation is similar to the model in Demirguc-Kunt et al (2010), with a difference in the time period over which bank returns during the financial crisis are measured.
Heteroskedasticity and outliers
The existence of heteroskedasticity is a major concern when performing OLS regressions because it can cause the standard errors of the estimates of the coefficients to be biased – these values will be lower or higher than their true value. Hence, I will test for heteroskedasticity using White’s test. If heteroskedasticity is detected, in order to correct the standard errors, I will use the tools presented in White (1980) i.e. heteroskedasticity -consistent covariance matrix, which calculates the standard errors using White’s correction for heteroskedasticity.
11
values of the bottom 5% and the upper 5% are replaced with values corresponding to the 5th percentile and 95th percentile, respectively.
As an alternative solution to the outlier problem, and for further robustness, I implement a method proposed by Iglewicz and Hoaglin (1993) and their modified Z-score method. This method overcomes the problem when the sample mean and standard deviation are affected substantially by a small number of extreme values. Besides that, the modified Z-score is also better suited when handling small samples. Hence, I can also use it to remove outliers from the different groups of banks based on assets size or Eurozone membership. The results from this modified Z-score technique are showed in Appendix 5.
Section IV. Data Variables
Next, I present a short description of each variable that I use and its importance to the analysis.
Stock Price Performance
Stock price performance is the dependent variable in my analysis. Banks’ stock performance is measured for the period between 30 April 2007 and 27 February 2009, using closing monthly prices. This period is based on the performance of the SX7L index, which reflects the performance of European companies in the banking sector. In April 2007 the SX7L index reached its peak and then started declining until the beginning of March 2009 when it reached its lowest point. Chart 1 below shows the movement of the SX7L index between April 2007 and March 2009. Because the SX7L index is based on the performance of European banks, it summarizes the overall performance of companies operating in the banking sector. Therefore, the period between its peak and its lowest point can be viewed as the period when the financial crisis severely hit European banks. Banks’ stock price performance (STKPERF) is measured as the change in stock price between the two dates above.
12
robustness test. This is the period when the subprime crisis really unfolded but it is also the period before the Troubled Assets Relief Program (TARP)1 was implemented.
Chart 1
SX7L Index
As we can see from the chart above, the index was close to its peak in early summer 2007, During the summer of 2007, as the subprime crisis started to unfold, the SX7L index started slowly declining. Brunnermeier (2009) highlights some of the events that occurred in the period 2007 – 2009. The first important event around that time (May 2007) was the closure of UBS hedge fund due to large subprime-related losses. Later in the summer, credit rating agencies started downgrading many subprime tranches. In June 2007, Bear Stearns hedge funds experienced serious problems when facing margin calls and the bank had to provide them with billions in liquidity in an effort to preserve its reputation. Another major event from the summer of 2007 was the freezing of three investment funds by BNP Paribas because the bank was not
1
13
capable to value their holdings of structured products. Sharper declines in the index, however, started in January 2008. In mid January, the downgrade of Ambac Financial Group Inc., which was one of the biggest bond insurers, negatively impacted share prices of banks around the world. March 2008 marked the acquisition of Bear Stearns by JP Morgan at a remarkably low price per share. In summer 2008, Fannie Mae and Freddie Mac which were publicly traded government-sponsored enterprises and securitized a large portion of the US mortgage market, started experiencing troubles as mortgage default rates continued to increase. Market worries about the ability of Freddie and Fannie to absorb losses on their large mortgage portfolios eventually forced the government to put them in federal conservatorship. Other events that happened in September 2008 helped transform the U.S. financial crisis into a global one at faster rates. The bankruptcy of Lehman Brothers, the acquisition of Merrill Lynch by Bank of America, and the bailout of the international insurance company, AIG, had severe worldwide consequences on stock prices.
Tier 1 Capital Ratio
To test my main hypothesis, I obtained the Tier 1 capital ratio for each bank in the sample as of December 31, 2006. The Tier 1 ratio is the ratio of the core equity bank capital divided by the bank’s risk-weighted assets, which include all assets that the bank holds and that are regularly weighted for risk. As a general rule, the higher this ratio is, the more stable the bank is. Banks with higher ratios are expected to have better protection against losses in case of a financial distress. Depending on the level of Tier 1 capital ratio, regulators can assess a bank’s capital adequacy. I plan to test whether this was actually the case during the financial crisis – whether banks with higher capital levels were the ones that performed better compared to the banks with lower capital levels. The Tier 1 ratio (TIER1) is measured as of 31 December, 2006.
Bank size
14
support the so called “too big to fail” banks with various programs. Brunnermeier (2009) argues that the rise in popularity of structured and securitized products is the main determinant that led to cheap credit and a decline in lending standards. This resulted in rapid growth of the housing prices and laid the foundations for the crisis. Since large banks were among the most active buyers of structured and securitized products, it is expected that they were also the ones that experienced a rather sharp decline in their stock prices. In my analysis, SIZE is measured as the natural logarithm of the bank’s assets as of December 31, 2006.
Available liquid assets
Liquidity is one of the main indicators that bank managers closely observe. Especially during crisis a lot of banks experience liquidity shocks because of the insufficient possession of liquid assets. Banks are reluctant to lend to one another and an already troubled bank might not be able to issue new equity. Even if a bank tried to issue new equity, that would likely cause an even larger depreciation of its stock price. The new shares would compete with investors trying to dispose of that bank’s shares, further diluting the price. At the same time the new emission of shares would also signal potential problems with raising funds. Thus, it is important for banks to have quick and easy access to liquid funds during crisis times. My liquidity variable, LIQASD, is measured by the ratio of available liquid assets divided by the amount of total deposits and borrowings as of December 31, 2006.
Assets growth
Similarly to bank size, over the long term, banks with stable and increasing assets growth are expected to be safer and less risky institutions. However, aggressive growth prior to the crisis was, for many banks, the result of increased exposure to structured finance products. These products led to the largest losses in the crisis. Therefore, a bank with high asset growth during the years before the crisis might actually perform worse than one with moderate growth. Assets growth (GROW) is measured as the percentage growth of the assets of the bank over the period January 1, 2006 - December 31, 2006.
Non-interest income
15
crisis because non-interest income is not related to the interest rate of their lending activities. Thus, banks with higher proportion of non-interest income might be expected to perform better during the crisis. However, Brunnermeier, Dong and Palia (2011) find that banks with higher levels of non-interest income to interest income ratios have contributed to the rise of systemic risk in the industry and, thus, bank’s operations related to non-core activities have actually contributed to the adverse outcomes during the financial crisis. The reliance on non-interest income (NONINT) is measured for the year ending December 31, 2006 as the non-interest income divided by gross revenues.
Return on average assets
The return on average assets (ROAA) shows how profitable a company is in relation to its total average assets. ROAA is also an indicator of the efficiency of the management in using the company’s assets to generate earnings. During a crisis, banks with higher return on assets prior to the crisis are expected to have a relative advantage compared to those with low ratios, because higher ROAA would have reduced the reliance on an external funding (stock and bond issues). ROAA is measured for the year ending December 31, 2006 as net income divided by total average assets.
Net interest margin
The net interest margin (NIM) is measured as the difference between the interest income that a bank received and the amount of interest that was paid to its lenders, divided by the amount of average earning assets for the year ending December 31, 2006. NIM essentially examines how efficient a bank’s investment policy is compared to its debt funding costs. If NIM is negative this implies that the bank did not conduct a successful strategy because the interest expenses were higher than the interest income. The IMF Global Financial Stability Report (2012) mentions that banks with lower net interest margins were subject to larger intervention by the government.
Summary statistics
16
December 31, 2006. Thus, I have 118 observations to use in my analysis. This sample includes three banks that filed for bankruptcy during the crisis. The complete list of banks is included in Appendix 1. In addition to analyzing the full sample, I divided my sample in three groups: large, medium-sized and small banks and conducted the same analysis as on the complete sample. The group of the large banks includes banks with total assets above 50 billion euro; the medium-sized banks group consists of banks with total assets between 10 and 50 billion euro; and small banks are defined as those with total assets below 10 billion euro. Because some of the countries in the European Union do not use or have not yet converted to euro, I used the corresponding exchange rate as of December 31, 2006 to adjust all values to euros. Table 1 below provides information on the banks’ country of origin as well as a breakdown by total assets.
Table 1
Number of banks from each country grouped depending on total asset size
Country Banks Banks with Total
Assets > 50 billion
Banks with Total Assets between 10 and 50 billion
17
As I pointed out earlier, it is important to break down the full sample of banks into different groups depending on their total assets size because, as the IMF Global Financial Stability Report (2012) and Brunnermeier (2009) find, large banks suffered larger shocks because of their interconnectedness and large investments in structured products. Tarashev et al (2009) show that higher exposures to a common factor, i.e. structured products in the recent financial crisis, increase the probability of joint failures and also lead to higher systemic risk. The authors also show that in terms of bank size, larger banks additionally strenghten the possible negative effects of the common factor exposure. This implies that, from a regulatory point of view, steps must be taken to solve the “too big to fail” problem – because it seems that some large banks take excessive risks on purpose knowing that they would get bailed out eventually. As governments seem to tailor different policies by bank size, the effects of the crisis and the dynamics of the model may respond differently to different groupings by size. By separating the banks into groups by asset size, my model adds to the literature by allowing for different effects of the independent variables on stock performance within each group.
18 Descriptive statistics
Table 2 below provides descriptive statistics of the different variables that will be used in the model.
Table 2
The table presents descriptive statistics for the different variables of 118 banks from the European Union that will be used in the model. Abbreviations correspond to the following variables: ASSETS = bank total assets (€million); NONINT = the ratio of total non-interest income to gross revenue; GROW = asset growth in the year prior to the crisis; LIQASD = the ratio of liquid assets to total deposits and borrowings; NIM = net interest margin; ROAA = return on average assets; TIER1 = the Tier 1 capital ratio. All independent variables are measured as of 31 December, 2006. STKPERF = stock returns for the period April 2007 – February 2009; STKPERFR = stock returns for the period July 2007 – September 2008.
Variable Mean Median Maximum Minimum Std. Dev.
ASSETS 147,937.70 15,145.20 1,571,768 73.6 339,448.60 NONINT 0.4233 0.4027 0.8716 0.0429 0.1541 GROW 0.2076 0.1371 4.8349 -0.2469 0.4521 LIQASD 0.2622 0.2256 1.5186 0.0477 0.1955 NIM 0.0259 0.0239 0.1184 0.0025 0.0152 ROAA 0.0135 0.0107 0.0789 0.0003 0.0101 TIER1 0.1033 0.0893 0.549 0.056 0.0546 STKPERF -0.6956 -0.7018 -0.0246 -1 0.2029 STKPERFR -0.4486 -0.4561 0.0744 -0.9122 0.205 Observations 118 118 118 118 118
19
Table 3 below presents the descriptive statistics of the variables after winsorising the data in order to reduce the effects of possible outliers.
Table 3
The table presents descriptive statistics for the different variables of 118 banks from the European Union that will be used in the model, after winsorising the independent variables. Abbreviations correspond to the following variables: ASSETS = bank total assets (€million); NONINT = the ratio of total non-interest income to gross revenue; GROW = asset growth in the year prior to the crisis; LIQASD = the ratio of liquid assets to total deposits and borrowings; NIM = net interest margin; ROAA = return on average assets; TIER1 = the Tier 1 capital ratio. All independent variables are measured as of 31 December, 2006. STKPERF = stock returns for the period April 2007 – February 2009. STKPERFR = stock returns for the period July 2007 – September 2008.
Variable Mean Median Maximum Minimum Std. Dev.
ASSETS 134,195.3 15,145.2 1,139,056 209.4 288,495 NONINT 0.4249 0.4027 0.7459 0.2187 0.1358 GROW 0.1716 0.1371 0.4795 0.0152 0.1198 LIQASD 0.2456 0.2256 0.5580 0.0678 0.1272 NIM 0.0252 0.0239 0.0487 0.0067 0.0118 ROAA 0.0130 0.0107 0.0305 0.0038 0.0078 TIER1 0.0988 0.0893 0.1760 0.0634 0.0314 STKPERF -0.6946 -0.7018 -0.0246 -1 0.2017 STKPERFR -0.4486 -0.4561 0.0744 -0.9122 0.2050 Observations 118 118 118 118 118
20
and -62% for the small banks. Descriptive statistics based on Eurozone breakdowns are also added in Appendix 3.
In Appendix 2 I provide pairwise correlation coefficients for all variables after winsorising the data. An interesting observation is the strong negative correlation between the Tier 1 capital ratio and bank size. Demirguc-Kunt et al (2010) also find particularly strong negative correlation between Tier 1 ratio and bank size.Although the correlation between some of the variables is relatively high, it is not high enough to suggest multicollinearity.
Section V. Results
Following the methodology outlined above, I estimate equations (1) and (2) for the overall sample as well as by groups, separated first by size and then by membership in the Eurozone. I begin by analyzing the full sample along with the samples grouped by asset size. First, I estimate equation (1), a simple linear regression, followed by equation (2), which expands the analysis to multiple explanatory variables. Then, I present the same type of analysis for the groups defined by membership in the Eurozone. All analysis below is conducted on the winsorised data. The results from the regressions without winsorising the data and using the modified Z-score method for excluding outliers can be found in the Appendix 4 and Appendix 5, respectively.
The results from equation (1) based on the full sample and different samples based on size are presented in Table 4 below
Table 4
The estimated coefficients are presented in (1) with the standard errors in parentheses. The
regression tests only the relationship between Tier 1 capital and subsequent stock price performance. “Large” banks are those with total assets > 50 billion euro, “Medium” are banks with total assets between 10 and 50 billion euro, and “Small” banks have total assets < 10 billion euro.
Variable Full Sample Large Medium Small
Tier 1 Capital Ratio 0.0046
[0.0059] 0.0068 [0.0220] -0.0304 [0.0188] -0.0044 [0.0070] F-statistic 0.6220 0.0958 2.6029 0.3959 Prob.(F-statistic) 0.4319 0.7587 0.1192 0.5321 Number of Observations 118 39 27 52 R-squared 0.01 0.01 0.09 0.01
21
None of the coefficients presented above is significant. The R-squared value is also very low, except for the group of medium-sized banks. Because the p-value is greater than the significance level of 10%, the test fails to reject the null hypothesis that levels of capital prior to the crisis are not related with stock performance during the crisis. The results indicate that there is not any particular relationship between prior Tier 1 capital levels and subsequent stock price performance for any of the different samples of banks.
Next, I add the control variables as in equation (2) to determine if any of them helps explain the stock price decline in the period April 2007 – February 2009. Table 5 below presents the results.
Table 5
OLS regression as it is presented in (2)
. The estimated coefficients are presented with the standard errors in parentheses. Besides
Tier 1 Capital Ratio, the regression also includes other control variables. “Large” banks are those with total assets >
50 billion euro, “Medium” are banks with total assets between 10 and 50 billion euro, and “Small” banks have total assets < 10 billion euro.
Variable Full Sample Large Medium Small
Tier 1 Capital Ratio -0.0070
[0.0084] -0.0128 [0.0266] -0.0222 [0.0235] -0.0011 [0.0108] Bank Size -0.0388*** [0.0131] -0.0119 [0.0318] -0.1862** [0.0814] -0.0236 [0.0261] Non-interest income 0.0021 [0.0017] 0.0041 [0.0026] -0.0037 [0.0069] 0.0004 [0.0021] Liquid Assets to Total
Deposits and Borrowings
-0.00245 [0.0015] 0.0007 [0.0029] -0.0050 [0.0057] -0.0036** [0.0016]
Return on average assets -0.0114
[0.0375] 0.1163 [0.1209] 0.0306 [0.1347] -0.0115 [0.0478] Assets Growth -0.0055*** [0.0014] -0.0011 [0.0029] -0.0032 [0.0031] -0.0072*** [0.0025] Net-interest margin -0.0171 [0.0328] 0.0401 [0.0611] -0.0460 [0.1024] -0.0790** [0.0391] F-statistic 4.7302 1.1469 1.7795 4.0353 Prob. (F-statistic) 0.0001 0.3606 0.1503 0.0017 R-squared 0.23 0.21 0.40 0.39 Adjusted R-squared 0.18 0.03 0.17 0.29 Number of Observations 118 39 27 52
22
The Adjusted R-squared value of the regression for the full sample indicates that 18% of the variation in stock price changes is explained using these variables. The Adjusted R-squared value for the sample of large banks is significantly lower compared to the others at only 0.03.The Adjusted R-squared value is 0.17 for the medium-sized banks, while for the small banks it is 0.29. Of note is the somewhat unusual result from the model specification over the medium size banks. The F-statistic is insignificant at the 10% level (suggesting litlle to no evidence that at least one coefficient is different from zero), even though the coefficient on bank size is sigificant at the 5% level. This result is likely driven by the small number of observations in the sample compared to the number of regressors.
When performing an OLS regression, however, heteroskedasticity might become an issue. It occurs when the standard errors do not have a constant variance. Therefore, to account for possible heteroskedasticity in my regressions, I perform a White’s test, for heteroskedasticity. This test is performed as in Brooks (2008). The results of the White’s test are summarized in Table 6 below:
Table 6
White’s heteroskedasticity test over the OLS regression as it is presented in (2)
based on the different samples of banks
depending on total assets size. “Large” banks are those with total assets > 50 billion euro, “Medium” are banks with
total assets between 10 and 50 billion euro, and “Small” banks have total assets < 10 billion euro.
Sample F-statistic Probability LM Probability Scaled explained SS Probability
All 2.2353 0.0366 14.6947 0.0401 14.9395 0.0368
Large 0.2100 0.9806 1.7656 0.9717 0.8809 0.9965
Medium 1.5388 0.2140 9.7689 0.2021 3.4620 0.8392
Small 1.6823 0.1383 10.9787 0.1395 6.1169 0.5262
where LM is the lagrange multiplier test statistic, calculated as the product of the sample size and R-squared:
(3) LM = n.R2
23
10% significance level. The third indicator, “Scaled explained SS”, is based on a normalized version of the explained sum of squares of the auxiliary regression and also suggests that there are no signs for heteroskedasticity for the different groups of banks based on total assets size. However, for the full sample of banks the F-statistic, the LM and the Scaled Explained SS are all significant. Their values are 2.24, 14.69 and 14.94, respectively. This suggests strong evidence of heteroskedasticity and thus, the null-hypothesis of the test, which suggests homoskedasticity, is rejected. Therefore, the standard errors in Table 5 are corrected to count for heteroskedasticity, but only for the full sample of banks.
As we can see from Table 5, the Tier 1 capital ratio is still insignificant. However, some of the other variables that were included show interesting results. The size of bank is statistically significant at 1% for the full sample consisting of 118 banks, and at 5% for the group of banks with total assets between 10 billion and 50 billion euro. The coefficient for both results is negative implying that larger banks actually performed worse than their smaller counterparts in terms of stock returns during the crisis. Akhigbe et al (2012) also find an inverse relationship between bank size and subsequent stock return. The liquid assets to total deposits and borrowings ratio is significant for the small group of banks at the 5% level. Again, the coefficient suggests that this ratio is negatively related with subsequent stock returns. Specifically, this implies that for each percentile increase in the ratio of liquid assets to total deposits and borrowings, the stock price decreased by 0.0036% on average. This might seem a bit surprising, however, Demirguc-Kunt et al (2010) also find a negative relationship between their liquidity ratio and subsequent stock price performance. The authors suggest that a possible explanation for this is that higher level of liquid assets might have been related to larger possession of mortgage-backed securities, which were one of the main reasons for the crisis and turned out to be highly illiquid and of poor quality.
24
total assets growth decreases stock returns by 0.0055% on average for the full sample and by 0.0072% on average for the small banks. A possible explanation for this result is that banks increased their total assets by making investments in assets which quickly decreased in quality during the crisis. Prior to the crisis, structured products were extremely popular and, as Brunnermeier (2009) states, banks were actually among the most active buyers of those products. After winsorising the data, the net-interest margin also becomes significant at 5% but only for the sample of small banks. There is a negative relationship between net-interest margin and performance of smaller banks, which provides evidence to contradict the IMF Global Financial Stability Report (2012) estimate, which suggests that banks with lower net-interest margins were subjects to higher interventions. A possible explanation might be the fact that from the group of small banks, those who had higher net-interest margin also had higher reliance on interest income. Thus, during the crisis if the net-interest margin decreases, this would lead to a loss of income for those banks. Another explanation may be that all the banks in that sample are small. Banks that were more prone to be targeted for intervention were larger banks, therefore observing a negative relationship over small banks (which were not necessarily the target of strong intervention) does not contradict the International Monetary Fund finding, especially if there was no statistical effect over the entire sample.
Next, I reestimate equations (1) and (2) over the subsamples of banks defined based on whether or not they are members of the Eurozone area. Table 7 and Table 8 below present the results.
Table 7
OLS regression as it is presented in (1) . The estimated coefficients are presented with the
standard errors in parentheses. The estimated coefficients are presented with the standard errors in parentheses. The regression tests only the relationship between Tier 1 capital and subsequent stock price performance The column “Eurozone” presents results from banks that are from Eurozone countries, while “Non-Eurozone” provides results for the sample of banks from countries that are not in the Eurozone. The Full Sample column is added for comparison
Variable Full Sample Eurozone Non-Eurozone
Tier 1 Capital Ratio 0.0046
[0.0059] -0.0024 [0.0103] 0.0128* [0.0066] F-statistic 0.6220 0.0532 3.7772 Prob.(F-statistic) 0.4319 0.8182 0.0584 Number of Observations 118 72 46 R-squared 0.01 0.01 0.08
25
The R-squared value for the sample of banks in the Eurozone is very low (0.01). For those that are not in the Eurozone it is a little higher (0.08), which means that around 8% of the variation in stock price returns is explained by the model. An interesting finding here is that the Tier 1 capital ratio is significant at 10% for the non-Eurozone countries. The coefficient is also positive, which suggests a positive relationship between Tier 1 capital and subsequent stock price performance. This means that banks with higher Tier 1 ratios prior to the crisis, also performed better during the crisis i.e. they experienced lower stock price decline.
Table 8
OLS regression as it is presented in (2)
The estimated coefficients are presented with the standard errors in parentheses. Besides
Tier 1 Capital Ratio, the regression also includes other control variables. The column “Eurozone” presents results from banks that are from Eurozone countries, while “Non-Eurozone” provides results for the sample of banks from countries that are not in the Eurozone. The Full Sample column is added for comparison.
Variable Full Sample Eurozone Non-Eurozone
Tier 1 Capital Ratio -0.0070
[0.0084] -0.0204 [0.0128] 0.0058 [0.0113] Bank Size -0.0388*** [0.0131] -0.0695*** [0.0159] -0.0052 [0.0158] Non-interest income 0.0021 [0.0017] 0.0012 [0.0022] 0.0034 [0.0027] Liquid Assets to Total
Deposits and Borrowings
-0.00245 [0.0015] -0.0022 [0.0023] -0.0027 [0.0024]
Return on average assets -0.0114
[0.0375] 0.0572 [0.0647] -0.0159 [0.0609] Assets Growth -0.0055*** [0.0014] -0.0064*** [0.0018] -0.0033 [0.0027] Net-interest margin -0.0171 [0.0328] -0.0458 [0.0436] 0.0257 [0.0377] F-statistic 4.7302 5.2101 1.2282 Prob(F-statistic) 0.0001 0.0001 0.3117 R-squared 0.23 0.36 0.19 Adjusted R-squared 0.18 0.29 0.03 Number of Observations 118 72 46
26
The Adjusted R-squared value for the Eurozone sample is 0.29, which means that 29% of the variation in stock price returns is explained using this model. The Adjusted R-squared for the non-Eurozone countries is only 0.03. The F-statistic of the non-Eurozone banks is not significant and indicates that even when all the coefficients are taken together, they are still insignificant.
When I include all the other control variables, once again I test for heteroskedasticity using White’s test. The results are presented below.
Table 9
White’s heteroskedasticity test over the OLS regression as it is presented in (2)
. The column “Eurozone” presents results from
banks that are from Eurozone countries, while “Non-Eurozone” provides results for the sample of banks from countries that are not in the Eurozone.
Sample F-statistic Probability LM Probability Scaled explained SS Probability
Eurozone 2.5362 0.0228 15.6354 0.0287 10.9333 0.1416
Non-Eurozone 1.2497 0.3009 8.6079 0.2820 5.4602 0.6040
The F-statistic for the Eurozone banks is 2.5, while for the non-Eurozone banks it is 1.3. The Lagrange multiplier statistic (LM) and the Scaled Explained SS version of the test have values of 15.6 and 10.9, respectively, for the Eurozone sample; and 8.6 and 5.5 for the non-Eurozone sample. For the non-Eurozone, the F-statistic and LM version are significant at 5%, while Scaled Explained SS is not significant even at 10%. Therefore, the conclusion of this test is somewhat ambiguous. However, because both F-Statistic and LM versions are significant at 5%, I would err on the side of caution and assume that heteroskedasticity appears in this group. None of the statistics for the non-Eurozone countries is significant, therefore there are no signs of heteroskedasticity in this group. Thus, the standard errors of the OLS regression over the Eurozone banks in Table 8 are corrected using White’s heteroskedasticity -consistent standard errors.
Looking at the banks from the Eurozone, the results resemble those of the full sample. Bank size is again significant at 1% and with negative relationship with subsequent stock performance, again consistent with Akhigbe et al (2012). Assets growth in 2006 is also significant at 1% and with a negative sign of the coefficient.
27
is not significant anymore. The Tier 1 ratio was also significant at 10% when it was the only independent variable in the model, but again, this effect disappeared when other variables were included. A possible explanation of this might be the high negative correlation that is found between Tier 1 and bank size. This would lead to a reduced Tier 1 effect when bank size is included in the OLS regression.
Besides winsorising the data and correcting the standard errors where necessary, for further robustness I also use a different period to estimate stock price performance during the crisis. Instead of the current April 2007 to February 2009, I use July 31, 2007 to September 30, 2008. The July 31, 2007 starting date is just before the subprime crisis really started with the redemption freeze of the BNP Paribas funds on August 9, 2007. The September 30, 2008 date is after the bankruptcy of Lehman Brothers and the associated drop in the stock market, but before the US Treasury announced the Capital Purchase Program through which it purchased capital in the biggest US banks on October 14, 2008 (the Troubled Asset Relief Program – TARP). I will still use the independent variables as of December 31, 2006, so that, similarly to the estimations above I see how the pre-crisis capital and ratios are related to subsequent stock performance during the crisis. That way I verify whether the results presented above also apply to an alternative specification of the financial crisis. Table 10 and Table 11 below present the results over full sample and the different groups of banks based on total assets size.
Table 10
OLS regression as it is presented in (1) . The estimated coefficients are presented with the
standard errors in parentheses. The regression tests only the relationship between Tier 1 capital and subsequent stock price performance. “Large” banks are those with total assets > 50 billion euro, “Medium” are banks with total assets between 10 and 50 billion euro, and “Small” banks have total assets < 10 billion euro.
Variable Full Sample Large Medium Small
Tier 1 Capital Ratio 0.0029
[0.0061] 0.0134 [0.0301] -0.0159 [0.0173] -0.0028 [0.0070] F-statistic 0.2267 0.1965 0.8424 0.1538 Prob.(F-statistic) 0.6349 0.6602 0.3675 0.6966 Number of Observations 118 39 27 52 R-squared 0.002 0.005 0.030 0.003
28
Changing the period over which the stock returns are measured does not yield different results when only Tier 1 ratio is included. It is still insignificant in every group and the R-squared values are even lower when compared to Table 4.
Table 11
OLS regression as it is presented in (2)
The estimated coefficients are presented with the standard errors in parentheses. Besides
Tier 1 Capital Ratio, the regression also includes other control variables. “Large” banks are those with total assets >
50 billion euro, “Medium” are banks with total assets between 10 and 50 billion euro, and “Small” banks have total assets < 10 billion euro.
Variable Full Sample Large Medium Small
Tier 1 Capital Ratio -0.0102
[0.0097] -0.0091 [0.0327] -0.0386 [0.0229] -0.0025 [0.0114] Bank Size -0.0030 [0.0123] 0.0377 [0.0389] -0.1392* [0.0793] -0.0046 [0.0275] Non-interest income 0.0039** [0.0019] 0.0071** [0.0032] -0.0034 [0.0068] 0.0010 [0.0022] Liquid Assets to Total
Deposits and Borrowings
-0.0017 [0.0018] -0.0003 [0.0036] 0.0019 [0.0055] -0.0039** [0.0017]
Return on average assets 0.0152
[0.0455] 0.0787 [0.1476] 0.1389 [0.1312] 0.0210 [0.0503] Assets Growth -0.0043** [0.0017] 0.0010 [0.0035] -0.0041 [0.0030] -0.0063** [0.0026] Net-interest margin 0.0346 [0.0315] 0.1415* [0.0746] -0.0122 [0.0997] -0.0726** [0.0412] F-statistic 1.7621 2.5769 1.0355 2.9232 Prob. (F-statistic) 0.1021 0.0324 0.4395 0.0133 R-squared 0.1 0.37 0.28 0.32 Adjusted R-squared 0.04 0.23 0.01 0.21 Number of Observations 118 39 27 52
Note: (*), (**) and (***) stand for statistically significant at the 10%, 5% and 1% level, respectively.
29
intervention was more pronounced, and also targeted more towards larger banks. This in turn could have affected their stock price performance. Therefore, the original result of significance in the bank size might have been driven by bank size acting also as a proxy for the effects of government intervention.
Furthermore, reliance on non-interest income is now significant at 5% for the full sample and the large banks. The coefficient is positive meaning that banks with higher revenues from non-interest related activities performed better. Akhigbe et al (2012) also find positive relationship between reliance on non-interest income and stock returns during the crisis. Liquid assets to total deposits and borrowings has the same sign and significance level (5%) for the small banks, consistent with the results from Table 5. Assets growth in 2006 is again significant and with negative sign for the full sample and the small sample of banks. However, its significance level has changed from 1% to 5% for both groups. Once again, net-interest margin has the same significance level (5%) and sign for the sample consisting of small banks. However, net-interest margin is now also significant at 10% for the group of large banks. The coefficient for this group is positive, meaning that large banks with higher levels of net-interest margin experienced lower stock price decline.
Although some of the variables are significant and with the same sign in both Table 5 and Table 11, all the other changes infer that the results are not completely robust. Table 12 and Table 13 below present the results for the Eurozone and non-Eurozone banks when I use the 31 July 2007 – 30 September 2008 returns.
Table 12
OLS regression as it is presented in (1) . The estimated coefficients are presented with the
standard errors in parentheses. The estimated coefficients are presented with the standard errors in parentheses. The regression tests only the relationship between Tier 1 capital and subsequent stock price performance The column “Eurozone” presents results from banks that are from Eurozone countries, while “Non-Eurozone” provides results for the sample of banks from countries that are not in the Eurozone. The Full Sample column is added for comparison
Variable Full Sample Eurozone Non-Eurozone
Tier 1 Capital Ratio 0.0029
[0.0061] 0.0021 [0.0099] 0.0051 [0.0079] F-statistic 0.2267 0.0450 0.4074 Prob.(F-statistic) 0.6349 0.8326 0.5266 Number of Observations 118 72 46 R-squared 0.002 0.001 0.001
30
When using the different period for stock returns, the Tier 1 ratio is not significant anymore for the non-Eurozone banks. The R-square values are also lower compared to those of Table 7 meaning that this specific model does not explain the variation of stock returns.
Table 13
OLS regression as it is presented in (2)
The estimated coefficients are presented with the standard errors in parentheses. Besides
Tier 1 Capital Ratio, the regression also includes other control variables. The column “Eurozone” presents results from banks that are from Eurozone countries, while “Non-Eurozone” provides results for the sample of banks from countries that are not in the Eurozone. The Full Sample column is added for comparison
Variable Full Sample Eurozone Non-Eurozone
Tier 1 Capital Ratio -0.0102
[0.0097] -0.0229* [0.0126] -0.0004 [0.0138] Bank Size -0.0030 [0.0123] -0.0299* [0.0160] 0.0274 [0.0193] Non-interest income 0.0039** [0.0019] 0.0051** [0.0023] 0.0018 [0.0033] Liquid Assets to Total
Deposits and Borrowings
-0.0017 [0.0018] -0.0015 [0.0023] -0.0019 [0.0029]
Return on average assets 0.0152
[0.0455] 0.0827 [0.0746] 0.0331 [0.0745] Assets Growth -0.0043** [0.0017] -0.0048** [0.0019] -0.0032 [0.0034] Net-interest margin 0.0346 [0.0315] 0.0413 [0.0423] 0.0469 [0.0462] F-statistic 1.7621 3.2048 0.5251 Prob. (F-statistic) 0.1021 0.0057 0.8099 R-squared 0.1 0.26 0.09 Adjusted R-squared 0.04 0.18 -0.08 Number of Observations 118 72 46
Note: (*), (**) and (***) stand for statistically significant at the 10%, 5% and 1% level, respectively.
For the group of non-Eurozone banks, the different period for the stock returns does not change the outcomes for any of the independent variables – all of them are still insignificant, consistent with the results from Table 8.
31
30 September 2008. This result is consistent with Akhigbe et al (2012) but disagrees with Demirguc-Kunt (2010). This is also the only model specification that supports the capital signaling hypothesis for this sample of European banks. For this period, non-interest income is also significant at 5% and is with a positive coefficient meaning that higher prior levels of non-interest income to gross revenues led to better performance during the crisis. The coefficient of assets growth in 2006 once again has a negative sign, however the significance level is reduced from 1% to 5%. As with the different groups of banks based on total assets size, there are certain changes that occur after the period for the stock returns is changed indicating that the results from the model are not fully robust.
Conclusions
The financial crisis that started in 2007 showed the urgent need for new reforms in the area of bank supervision and regulations. There are already discussions about the implementation of Basel III, which would increase the capital that is required for holding different instruments and would also require that banks have enough liquid assets to reduce the likelihood of a liquidity crisis. Another regulatory improvement through Basel III would be the consideration of the “too big to fail” problem and possible solutions for it.
32
33
References:
Akhigbe, A., Madura, J., Marciniak, M., 2012. Bank capital and exposure to the financial crisis. Journal of Economics and Business, Volume 64, Issue 5, 377-392.
Allen, W.A. and Wood, G. (2006). Defining and achieving financial stability. Journal of Financial Stability 2 (2006) 152–172
Altunbas, Y., Manganelli, S., Marques-Ibanez, D., 2011. Bank risk during the financial crisis. Do business models matter? Unpublished working paper.
Amel-Zadeh, A., Meeks, G., 2011. Bank failure, mark-to-market and the financial crisis. Unpublished working paper. Judge Business School, University of Cambridge.
Beltratti, A., Stulz, R., 2011. Why did some banks perform better during the credit crisis? A cross-country study of the impact of governance and regulation. Journal of Financial Economics, forthcoming.
Bessler, W., Kurmann, P. 2012. Bank risk factors and changing risk exposures of banks. Capital market evidence before and during the financial crisis. Unpublished working paper. Justus-Liebig University Giessen.
Brooks, C., 2008. Introductory Econometrics for Finance 2nd Edition. Cambridge University Press, New York.
Brunnermeier, M., 2009. Deciphering the Liquidity and Credit Crunch 2007-08, Journal of Economic Perspectives 23(1), 77-10.
Brunnermeier, M., Dong, G., Palia, D., 2011. Banks’ Non-Interest Income and Systemic Risk. AFA 2012 Chicago Meetings Paper. Princeton University, Columbia University, Rutgers Business School.
Cetorelli, N., Goldberg, L. (2011). Banking globalization and monetary transmission. Journal of Finance, forthcoming.
34
Demirguc-Kunt, A., Detragiache, E.,Merrouche, O., 2010. “Bank capital: Lessons from the Financial Crisis”. Journal of Money, Credit and Banking, forthcoming.
Fahlenbrach, R., Prilmeier, R., Stulz, R. (2011). This Time Is the Same: Using Bank Performance in 1998 to Explain Bank Performance During the Recent Financial Crisis. Journal of Finance, 2012, v.67(6), 2139-2185.
Furlong, F. T., and M. C. Keeley, 1989, Capital Regulation and Bank Risk-Taking: A Note, Journal of Banking and Finance, 13, 883-891.
Goldberg, L.S. (2009). Understanding Banking Sector Globalization, IMF Economic Review 56(7), 171-197
Gorton, G., Metrick, A., 2010. Securitized banking and the run on repo. Journal of Financial Economics, forthcoming.
International Monetary Fund, October 2012. Global Financial Stability Report: Restoring Confidence and Progressing on Reforms.
Iglewicz, B., Hoaglin, D. How to detect and handle outliers. ASQC Quality Press, 1993
Loutskina, Elena, and Philip E. Strahan, 2011, Informed and uninformed investment in housing: The downside of diversification, Working Paper, Boston College.
Spence, A. M. (1974). Market signaling: Information transfer in hiring and related processes. Cambridge, MA: Harvard University Press
Tarashev, N., C. Borio, and K. Tsatsaronis (2009). The systemic importance of financial institutions. BIS Quarterly Review. September, Basel
U.S. Department of the Treasury Press Center. "Treasury Announces TARP Capital Purchase Program Description". Available at: http://www.treasury.gov/press-center/press-releases/Pages/default.aspx
35
Appendix 1
The appendix list all sample banks used in the analysis. Shown is the name, the country of origin and whether the bank is large, medium or small depending on total assets size.
№ Bank Country Large Medium Small
1 Aareal Bank AG Germany
2 Alliance & Leicester Plc UK
3 Allied Irish Banks Plc Ireland
4 Alm. Brand A_S Denmark
5 Alpha Bank AE Greece
6 Amagerbanken, Aktieselskab Denmark
7 Arbuthnot Banking Group Plc UK
8 Attica Bank SA Greece
9 Banca Carige SpA Italy
10 Banca Finnat Euramerica SpA Italy
11 Banca Generali SpA Italy
12 Banca Ifis SpA Italy
13 Banca Monte dei Paschi di Siena Italy
14 Banca popolare dell_Emilia Italy
15 Banca Popolare di Milano Italy
16 Banca Popolare di Sondrio Italy
17 Banca Popolare di Spoleto Italy
18 Banca Profilo SpA Italy
19 Banco Bilbao Vizcaya Spain
20 Banco BPI SA Portugal
21 Banco de Andalucia SA Spain
22 Banco de Sabadell SA Spain
23 Banco de Valencia SA Spain
24 Banco Desio Italy
25 Banco di Sardegna SpA Italy
26 Banco Espirito Santo SA Portugal
27 Banco Guipuzcoano SA Spain
28 Banco Popular Espanol SA Spain
29 Banco Santander SA Spain
30 Bank für Tirol Austria
31 Bank Handlowy w Warszawie Poland
32 Bank Millennium Poland
33 Bank Ochrony Srodowiska SA Poland
34 Bank of Cyprus Public Comp Cyprus
35 Bank of Greenland-Gronland Denmark
36
37 Bank of Valletta Plc Malta
38 Bank Polska Kasa Opieki SA Poland
39 Bankinter SA Spain
40 Barclays Plc UK
41 BKS Bank AG Austria
42 BNP Paribas France
43 Bradford & Bingley Plc UK
44 BRE Bank SA Poland
45 Bulgarian-American Credit Bulgaria
46 Close Brothers Group Plc UK
47 Commerzbank AG Germany
48 Crédit Agricole S.A. France
49 Credito Bergamasco Italy
50 Credito Emiliano SpA-CREDEM Italy
51 Credito Valtellinese Soc Italy
52 Cyprus Popular Bank Cyprus
53 Danske Bank A_S Denmark
54 Deutsche Bank Germany
55 Dexia Belgium
56 DiBa Bank A_S Denmark
57 Djurslands Bank A_S Denmark
58 DVB Bank SE Germany
59 Erste Group Bank AG Austria
60 Espirito Santo Financial G Luxembourg
61 Eurobank Ergasias SA Greece
62 FHB Mortgage Bank Plc-FHB Hungary
63 Finibanco Holding SGPS SA Portugal
64 Fionia Holding A_S Denmark
65 First Investment Bank AD Bulgaria
66 Hellenic Bank Public Cyprus
67 HSBC UK
68 Hvidbjerg Bank Aktieselskab Denmark
69 Hypo Real Estate Holding AG Germany
70 IKB Deutsche Industriebank Germany
71 Investec Plc UK
72 Irish Bank Resolution Corp Ireland
73 Jyske Bank A_S Denmark
74 Kas Bank NV The Netherlands
75 Komercni Banka Czech Republic
76 Kreditbanken A_S Denmark
77 Laan & Spar Bank A_S Denmark
78 Landesbank Berlin Holding Germany
37
80 Lollands Bank Denmark
81 Mediobanca SpA Italy
82 Moens Bank A_S Denmark
83 National Bank of Greece SA Greece
84 Noerresundby Bank A_S Denmark
85 Nordea Bank AB Sweden
86 Nordfyns Bank Denmark
87 Nova Kreditna Banka Maribo Slovenia
88 Oesterreichische Volksbank Austria
89 Oestjydsk Bank A_S Denmark
90 OTP Bank Plc Hungary
91 Permanent TSB Plc Ireland
92 Piraeus Bank SA Greece
93 Probanka d.d. Maribor Slovenia
94 Proton Bank S.A. Greece
95 Ringkjoebing Landbobank Denmark
96 Royal Bank of Scotland UK
97 Salling Bank A_S Denmark
98 Skandinaviska Enskilda Banken Sweden
99 Skjern Bank Denmark
100 Société Générale France
101 Spar Nord Bank Denmark
102 Sparekassen Faaborg A_S Denmark
103 Sparekassen Himmerland Denmark
104 Standard Chartered Plc UK
105 Svendborg Sparekassen A_S Denmark
106 Svenska Handelsbanken Sweden
107 Swedbank AB Sweden
108 Sydbank A_S Denmark
109 Tatra Banka a.s. Slovakia
110 Totalbanken A_S Denmark
111 TT Hellenic Postbank S.A Greece
112 UniCredit SpA Italy
113 Unione di Banche Italiane Italy
114 USB Bank Plc Cyprus
115 Van Lanschot NV The Netherlands
116 Vestfyns Bank A_S Denmark
117 Vordingborg Bank A_S Denmark
