The Effect of Capital on the Euro Area Banks’ Performance during
Systemic Crises
Friso Steinebach1,2 University of Groningen Faculty of Economics and Business
f.a.steinebach@student.rug.nl Supervisor Sebastiaan Pool
7 June 2018
Abstract
This paper examines how bank capital influences banks’ performance during systemic crises in terms of survival and market share and how this outcome diverges with Greece, Ireland, Portugal, Spain, and Cyprus (GIPSC) and individual countries in the Euro area. The results suggest that bank capital does not affect both measures of bank performance. However, significant results emerge when performing additional analyses. Capital positively influences medium banks’ survival probability for the GIPSC countries. In addition, when examining individual countries, support for the positive relationship is found. Regarding the market share, the influence of bank capital for individual systemic crises shows conflicting results.
Keywords: Bank capital, survival, market share, systemic crises JEL classifications: G01, G21, G28
2 1. Introduction
The financial and Eurozone crisis show that highly leveraged banks create negative externalities. A small decrease in asset value can result in financial distress and ultimately cause insolvency for banks. The inter-connectiveness of the financial sector can cause the system to freeze and severely damage the rest of the economy. To minimize social costs, governments spent large amounts on bailouts and recovery resolutions. Insolvency for the bank is not the largest problem, but highly leveraged banks are compelled to sell sizable assets in order to reduce their leverage. These sales put large pressure on the asset market, thereby, reducing prices and affecting other banks. Banks act in their own interest and pay limited attention to systemic risk. Financial regulation tries to avoid such systemic risk and associated social cost to safeguard the soundness of the financial system (Thakor, 2014). Due to the recent systemic crises, it is natural to consider deleveraging banks. As response, the Basel committee has increased the equity requirements of banks (BIS, 2010).
However, the banking sector has a strong view about the related cost to increased equity requirements, while understating the significant benefits related to the decline of systemic risk. The view of banks is inconceivable, given the costs of the recent crises. Analytical attempts of researchers show that increased capital requirements lowers lending growth and consequently economic activity (Cline, 2015). The literature suggests that banks should have a more nuanced view (e.g. Berger and Bouwman, 2013; Mehran and Thakor, 2014). Given the different views between banks and regulators, the goal of this paper is to analyze the effect of bank capital on the performance of banks during systemic crises in the Euro area. I use two key measures to track the performance of banks, namely survival and market share of banks, similar to Berger and Bouwman (2013).
Literature shows that bank capital increases the probability of survival during crises. The survival is attributed to two main effects. First, bank capital acts as a buffer for banks during crisis periods (Repullo, 2004; Mehran and Thakor, 2011; Garel and Petit-Romec, 2017). This is the mechanical effect of bank capital and increases the odds of survival for banks. Second, bank capital positively influences the management incentives. These risk-management incentives include screening, monitoring, and asset-substitution. Screening incentives increase the efficiency of asset allocation (Coval and Thakor, 2005). The
3 The theories on the other key bank performance measure, namely market share, show contradicting results. Holstrom and Tirole (1997), Allen and Gale (2004), Allen et al. (2011), and Thakor (2014) suggest that higher bank capital results in a competitive advantage for banks, which implies that higher-capitalized banks are able to obtain more market share. Conversely, Brander and Lewis (1986) and Lyandres (2006) suggest that firms with high leverage have more aggressive market expansion strategies and are able to gain larger market share.
This study uses the ECB/ESRB EU crisis database to identify systemic crises in the Euro area during the financial and Eurozone crisis, because the start and the end of the systemic crises differed between countries in the Euro area. The database covers all the EU countries and Norway from 1970-2016. The research builds on the empirical model of Berger and Bouwman (2013) and requires that the Euro must be in circulation two years before the start of the systemic crisis to calculate the pre-crisis capital and the control variables. After applying these restrictions, the following countries emerge: Austria, Belgium, Cyprus, France, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Slovenia and Spain (Appendix A). The banks are divided into three separate classes, small banks (total assets up to €1
billion), medium banks (€1 billion till €3 billion), and large banks (€3 billion and larger), because the effect of bank capital depends on the bank’s size (Vallascas and Keasey, 2012; Berger and Bouwman, 2013; Black, Correa, Huang, and Zhou, 2016).
The effect of capital on survival of banks is researched by using logistic regressions. The survival dummy is regressed with the pre-crisis capital ratio. The outcome of the
regressions shows the odds ratios. Based on Allison (1999), Mood (2010) shows that the odds ratio cannot clarify the true effects neither can the ratio be compared within models and across other estimations. Several authors argue that the odds ratio is affected by additional variables even though these variables are independent of the other variables (Gail, Wieand, and Piantadosi, 1984; Allison, 1999; Mood, 2010). To understand the accurate effect of bank capital on the survival of banks, I use marginal effects for the logistic regressions. A set of control variables is included to diminish the effect of the omitted variables problem, including risk, ownership, and profitability. The second part of the research focuses on the effect of capital on market share by using an OLS regression. The change in market share between the crisis and pre-crisis period is regressed on the bank’s average pre-crisis capital ratio. The same control variables are included to diminish the potential effect of the omitted variables problem.
4 countries in the Euro area to prevent bankruptcy of their country and their banks. I include an intercept and slope dummy to investigate whether the effect of capital on the banks’
performance is amplified in the GIPSC countries. In addition, the systemic crises are examined individually because it allows a more specific view on how capital is affected by the specific circumstances of the systemic crisis in each country.
A variety of robustness checks is performed to test the reliability of this research. First, banks considered too-big-to-fail (TBTF) are dropped from the sample. Second, the bank cutoff of medium and large banks is altered to test the sensitivity of bank size classification. Lastly, the sample mainly consist of banks from Italy. These banks are dropped to check if it influences the results.
The contribution of this study is threefold. First, this research is one of the first using the ECB/ESRB EU crisis database to identify systemic crises during the financial and
Eurozone crisis in the Euro area. This database presents a distinctive and detailed overview of systemic crises in the Euro area and ensures a more accurate definition of crises than most researchers do. Most researchers set a time window for the financial and Eurozone crisis for all the countries in the Euro area. However, the financial and Eurozone crisis started and ended at different points in time for countries in the Euro area. Second, this paper examines the effect of capital on bank performance for 12 countries in the Euro area, which allows checking between groups of countries and individual countries. Last, this paper is one of the few papers that separates between banks sizes while empirical evidence shows that bank’s size affects the influences the effect of capital (e.g. Berger and Bouman, 2013; Vazquez and Federico, 2015; Black et al., 2016)
The first part of the research shows that bank capital does not affect the survival probability of banks. However, bank capital positively affects the survival probability of the banks in the GIPSCs countries. In the second part of the research, it is shown that bank capital does not affect the market share of the banks. However, capital does affect market share during the individual systemic crises but the effect of capital on the market share shows contradicting results.
The remainder of the paper is organized as follows. Section two discusses the existing literature and develops hypotheses. Section three describes the methodology. Section four examines the data and discusses the limitations of the data. Section five analyses the results of the general effect of capital, the effect on the GIPSC countries and, the effect on the
5 2. Literature review
This section first elaborates on how bank capital, from the Modigliani and Miller perspective, influences banks. Then the focus will be on how capital affects the performance of banks by examining the relationship between bank capital and survival. Subsequently, the effect of capital on market share will be discussed.
Theoretical foundation: Capital and banks
In any research, focusing on the impact of financing structures on banks, appropriate attention has to be paid to the Modigliani and Miller theorem (or capital structure irrelevance theorem). Modigliani and Miller (1958)find that the financing structure of a firm does not affect the market value of a company and the share of equity and debt to the total financing mix is irrelevant. This theory has received substantial criticism because it is based on strong assumptions. The Modigliani and Miller theorem is considered the foundation of modern corporate finance and the starting point for further research on its applicability to
non-financial firms and the banking sector. The theory serves as the basis for this research to show how equity (bank capital) affects the performance of banks, in the form of survival and
market share. Regulators view the Modigliani and Miller theorem fundamental in the banking regulation (Aboura and Lépinette, 2015). Bank capital is the focal point in regulations from national regulators in both micro- and macro prudential regulation. The global regulatory framework including the Bank for International Settlements focuses on the guidelines of capital set by national regulators (Thakor, 2014).
The banking sector disagrees with the view of the regulators, stating that capital is expensive and equity requirements, having substantial benefits in preventing crises, also impose costs on the financial system and possibly on the economy. In return, banks are less profitable, reduce lending and as consequence weaken economic activity (Cline, 2015). In addition, they argue that holding more bank capital is costly for banks because debt reduces agency cost and asymmetric information. Debt provides discipline on management to minimize depletion or diversion of funds, which reduces the marginal cost of debt (Gertler, Kiyotaki, and Queralto, 2010). Furthermore, asymmetric information can increase the cost of raising external equity compared to raising debt (Myers and Majluf, 1984). However,
according to Myerson (2013), the problem of asymmetric information is avoided when the regulator requires the banks to sell a new equity share under a transparent set of requirements instead of the discretionary decisions of the management.
6 perform their function, including loaning, deposit taking and issuing bonds. Banks with more capital have fewer distortions in lending decisions and function better, which indicates that highly levered banks are not optimal from a social perspective. The social optimal levels of capital exceed the private levels because banks themselves do not account for systemic bank failures. Increasing current capital requirements increases social benefits while minimizing social costs(Admati, DeMarzo, Hellwig, and Pfleiderer, 2013). These findings are in line with empirical evidence. Based on the principals of the Modigliani and Miller theorem, Kashyap, Stein, and Hansen (2010) show that higher capital has a small effect on the funding rate in the long run. Mehran and Thakor (2014) present evidence that higher capital links to higher bank values while the lending declines modestly.
However, public policy based on the Modigliani and Miller theorem might not hold because several distortions and frictions in the economy influence banks (Admati et al., 2013). Governments create most of these distortions and frictions. For example, tax benefits favor debt over equity, because debt is tax deductible while equity is not. The value of the tax deductions diminishes when a bank replaces debt with equity. These differences become most apparent during the raising of funds. Another reason why the Modigliani and Miller theorem does not hold is the safety net in the form of deposit insurance and implicit government guarantee (Cline, 2015). First, deposits can be raised close to the risk-free rate due to the implicit deposit insurance made by the government. This incentivizes banks to attract more debt instead of equity. Second, the deposit insurance scheme is seen as a subsidy for banks. Empirical evidence of Tsesmelidkis and Merton (2012) finds that large banks take advantage of low funding cost due to the guarantees. The implicit support for Too-Big-To-Fail (TBTF) banks create substantial wealth transfer to the large financial institutions. The insurance can be seen as free put options issued by the government on the bank’s debt and owned by the bank’s bondholders, whereby shareholders benefit from higher leverage (Tsesmelidkis and Merton, 2012). This does not create social benefits for the whole society, but only for shareholders. Vickers (2012) recalls that prospective bailouts cheapen the private but not social cost of debt relative to equity, which leads to a reason why Modigliani-Miller theorem is not fully applicable to banks.
Survival of banks
7 or even negative, banks are able to lower its payout to equity holders without any notion of default.
The other effect of capital on the survival of banks focusses on how capital effects the risk-management incentives of banks. These risk-management incentives include screening, monitoring, and asset-substitution. Based on the screening incentives of Coval and Thakor (2005), a minimum amount of bank capital is crucial to perform efficient asset allocation (Berger and Bouwman, 2013). The increased efficient allocation of assets maximizes the expected profitability of the assets, and indirectly increasing the survival odds of banks. The monitoring effect shows that banks with higher bank capital induce higher levels of borrower monitoring by the bank (Holstrom and Tirole, 1997; Carletti and Marquez, 2011; Mehran and Thakor, 2011). The monitoring reduces the probability of default of loans and increases the banks’ odds of survival. The theory on asset-substitution states that bank capital reduces the excessive risk-taking incentives caused by limited liability and government protection (Acharya, Mehran and Thakor, 2011). This increases the odds of survival due to the reduced riskiness of the bank’s portfolio. The theories on risk management incentives do essentially not focus on crises, but they do focus on the survival probability. Bank capital can enhance the probability of survival during crises because the survival probability is lower during crises.
However, some theories suggest that capital does not increase the survival of banks during crises. Koehn and Santomero (1980) argue that banks increase their portfolio risk when capital is optimal. These riskier portfolios increase the risk of bankruptcy and correspondingly decrease the odds of survival. Moreover, Besanko and Kanatas (1996) suggest that higher capital negatively affects the bank safety. This arises when issuing equity dilutes the ownership of bank insiders, which reduces their incentives to exert effort on the behalf of the bank’s equity holders. The dilution effect can eliminate the major benefit of capital to reduce to substitution moral hazard problem. The literature concludes that most papers, and especially most recent papers, show a positive relationship between bank capital and survival of banks.
Empirical evidence on the effect of capital on survival shows a positive relationship. Mehran and Thakor (2014) show that by using cross-section banks, more capital is associated with higher bank values and higher probability of survival. The increased capital causes modest declines in lending. The overarching message is that low bank capital increases the systemic risk of the financial sector and increases the probability of bailouts by the
government. In addition, by using a novel empirical approach, Vallascas and Keasey (2012) show that the reduction of the leverage ratio used as in Basel-III improves the bank’s
8 The current adequacy ratio and implicit government support actually lead to increases in systemic risk, which may indicate that micro prudential regulation gives banks incentives to take on more systemic risk. They find that bank size and leverage ratio increase the systemic risk of the financial sector. This decreases the survival of banks during systemic crises. Moreover, Vazquez and Federico (2015) find that banks with weaker structural liquidity and higher leverage before the financial crisis are more likely to fail after the financial crisis. They also point out that Basel-III regulations should particularly focus on leverage for the systemically important institutions. Furthermore, Berger and Bouwman (2013) find that capital increases the survival probability of banks during crises. Especially the size of banks affects the probability of survival for banks. Larger banks benefit from capital during banking crises while smaller banks take advantage from capital during crises and normal times.
Deposit insurance can also influence the effect of capital on the survival of banks. It enables banks to raise funds close to the risk-free rate, increasing the profitability of banks. The increased profitability increases the odds of survival for banks. However, Merton (1977) argues that deposit insurance corresponds to a common stock put option and its value declines when capital increases. The impact of the put option is only affects banks with low levels of capital (Ronn and Verma, 1986). Thakor (2014) suggests that both deposit insurance and other forms of protection do play a role in leverage choice of banks. Therefore, the effect of capital is larger in the absence of deposit insurance and other forms of protection. In addition, Fahri and Tirole (2012) argue that safety nets induce coordinated behavior that increases systemic risk of the banking sector. Empirical evidence of Nier and Bouwmann (2006) show that deposit insurance indeed results in lower capital buffers for individual banks across numerous countries. The influence of bank capital on the survival of banks is still present with deposit insurance and decreases the capital of banks, which decreases the probability of survival of banks.
The debt overhang problem as described by Myers (1977) resists calls for higher bank capital and leaves banks with high leverage. The equity holders view issuing additional bank capital during crises as a negative NPV project (Thakor, 2014), since equity holders share the gains from the project with the bank’s creditors even when the total value of the banks increases. Admati, DeMarzo, Hellwig, and Pfleiderer (2014) state that the equity-debt holder conflict does not only lead to debt-overhang but also leads to “leverage ratchet effect”. This effect causes, when debt amounts are set, equity holders to resist any forms of debt reduction, even when the total value of the firm increases. Their empirical evidence shows that debt buyback reduces the probability of bankruptcy and therefore increases the probability of survival. However, the equity holders are unable to gain from the reduced probability of bankruptcy and the debt holders accrue all the value gains. In terms of the banking sector, the debt-overhang problem and refusal to reduce debt are even more interesting due to the
9 connected banks will benefit. The benefits for other banks create lower probability of default and therefore, increase the odds of survival. This gives another perspective on the effect of capital on the survival of banks.
Market share
Research about the relationship between capital and market share shows mixed results. On the one hand, literature shows that banks derive a competitive advantage of holding more capital. This implies that better-capitalized banks are able to obtain more market share (Holstrom and Tirole, 1997; Allen and Gale, 2004; Allen et al., 2011; Thakor, 2014).
Empirical evidence is also present. Allen and Gale (2004) state that banks with higher capital have the ability to gain market share in a duopolistic market when the other bank is affected by a solvency crisis. Moreover, Berger and Bouwman (2013) investigate how pre-crisis capital affects the market share of banks. They find that higher capital helps small banks to increase their market share during all crises while medium and large banks are only able to increase their market share during banking crises.
On the other hand, Lyandres (2006) argues that the performance of companies is positively correlated with leverage. Higher levered companies have more aggressive growth strategies and are able to obtain more market share. Most of these theories do not focus on the effect of capital during crisis periods, which can have different implications during crisis periods. González (2013)shows that, on average, the performance of highly leveraged firms decreases largely during crises. Therefore, the enhanced performance of highly levered banks is only during normal times and diminishes during crisis periods. Additionally, Berger and Bouwman (2013) point out that the competitive advantage of capital is more pronounced during crisis periods, for several reasons. First, customers are more concerned about the bank’s capital during crisis periods and better-capitalized banks are able to attract customers from less capitalized banks. Second, better-capitalized banks have higher flexibility to make certain loans decisions due to regulatory-and market constraints during crisis periods. Third, banks that experience near bankruptcy tend to be acquired by other banks with higher capital levels. These M&As need to be approved by regulators, which prefer banks with higher capital (Berger and Bouwman, 2013).
Hypotheses
The insights from the existing literature presents a direction for further analysis. This research paper tries to answer the following two questions and shows the hypotheses:
10 2. Does capital affect the market share of banks during crises?
Hypothesis: Capital increases the bank’s market share during systemic crises.
3. Methodology
This paper builds upon the empirical models of Berger and Bouwman (2013) to investigate how capital affects the performance of banks during systemic crises. First, the methodology to study the effect of capital on the survival of banks is explained. Second, the methodology that describes the influence of bank capital on the market share of banks is demonstrated. In addition, additional analyses used in this paper are explained.
Survival
The effect of capital on the survival of the banks is examined by using the estimation:
𝑆𝑈𝑅𝑉𝐼𝑉𝐴𝐿( = 𝛼+ + 𝛼-𝐸𝑄𝑅𝐴𝑇(,23456+ 𝛼7𝑋(,23456 + 𝜖( (1)
𝑆𝑈𝑅𝑉𝐼𝑉𝐴𝐿( is a dummy variable that equals one if the bank is in the sample one quarter
before the start of the systemic crisis and still is in the sample one quarter after the systemic crisis has ended. The dummy variable has a value of zero when the bank is dissolved during the systemic crisis or when the bank has merged during the systemic crisis. Mergers are also classified as non-surviving because the underlying reason of the merger is not clear.
𝐸𝑄𝑅𝐴𝑇(,23456 is the bank’s capital ratio, which is calculated as the average equity-to-assets
ratio over 2 years before the start of the systemic crisis. 𝑋(,23456 is a set of control variables included to reduce the omitted variable problem. The control variables included are based on risk, ownership, and profitability.
11 Additional analyses
The severity of the crises between countries in the Euro area differed during the financial and especially during the Eurozone crisis. The imbalances can be noticed at the balance sheets of the National Central banks (NCBs) in the Euro area. The current account deficits of Greece, Ireland, Portugal and Spain increased from the introduction of the Euro in 2002, to 929 billion euros in 2010, or 7% of their aggregated GDP over the period (Sinn, 2011). The European Central Bank (ECB) allowed and encouraged money creation and lending by the NCBs of the periphery at the expense of the core countries. The bubble busted in the GIPS countries during the financial crisis and the equity of Europe’s banks in the periphery was deprived. Private investors started to doubt if the GIPS current account deficits were sustainable. Therefore, the financing of these countries ended and the investors started to retreat from these countries to protect their wealth (Sinn and Wollmershäuser, 2012).
In effect, Greece, Ireland, Portugal, Spain and Cyprus (GIPSC) lost market access and needed support from other countries in the Euro area to prevent bankruptcy of their country and their banks. The EFSF and ESM were created to lend to these countries at very favorable conditions compared to the market rates to ensure that the countries themselves and their banks did not go bankrupt. The impact of the systemic crises in the GIPSC countries during the financial and Eurozone crisis is larger than in other countries. The following model is estimated to test whether the effect of capital on the survival of banks is amplified for these countries:
𝑆𝑈𝑅𝑉𝐼𝑉𝐴𝐿( = 𝛼+ + 𝛼-𝐸𝑄𝑅𝐴𝑇(,23456+ 𝛼7𝐺𝐼𝑆𝑃𝐶( + 𝛼=𝐺𝐼𝑃𝑆𝐶(𝐸𝑄𝑅𝐴𝑇(,23456+ (2) 𝛼>𝑋(,23456+ 𝜖(
The observant readers notice that model (2) is similar to equation (1), except an intercept dummy variable for the GIPSC countries and the interaction term between bank capital and GIPSC countries are added to equation (2).
12 Market share
The second part of the research focuses on the effect of bank capital on the market share of banks during systemic crises compared to the pre-crisis period. The estimation strategy is identical to the strategy in the first section. First, the general affect of capital on the market share of banks is examined. Second, using additional analysis to focus on the effect of capital in the GIPSC countries and lastly focus on the effect of capital during the individual crises.
An Ordinary Least Squares (OLS) regression is used to determine how capital influences the market share of banks. The estimation reads:
∆𝑀𝐾𝑇𝑆𝐻𝐴𝑅𝐸( = 𝛽++ 𝛽-𝐸𝑄𝑅𝐴𝑇(,23456+ 𝛽7𝑌(,23456+ 𝜖( (3)
∆𝑀𝐾𝑇𝑆𝐻𝐴𝑅𝐸( is the change in bank i’s market share during the crisis compared to the pre-crisis period. The market share is calculated as the bank’s total assets divided by the
industries’ total assets. The change in market share is calculated as the bank’s average market share during the systemic crisis minus the average market share over the two years before the crisis. The market share is not adjusted for mergers and acquisitions, because it is an approach for banks to extend their market share in the industry. 𝐸𝑄𝑅𝐴𝑇(,23456 is the bank’s capital ratio, which is calculated as the average equity-to-assets ratio over 2 years before the start of the systemic crisis. 𝑌(,23456 are the control variables also used in (1).
Additional analyses
Second, the effect of bank capital on the market share for the GIPSC countries is investigated by the following estimation:
∆𝑀𝐾𝑇𝑆𝐻𝐴𝑅𝐸( = 𝛽++ 𝛽-𝐸𝑄𝑅𝐴𝑇(,23456+ 𝛽7𝐺𝐼𝑃𝑆𝐶( + 𝛽=𝐸𝑄𝑅𝐴𝑇(,23456𝐺𝐼𝑃𝑆𝐶( (4) +𝛽>𝑌(,23456 + 𝜖(
Model (4) is similar to equation (2), except the dependent variable is changed from
𝑆𝑈𝑅𝑉𝐼𝑉𝐴𝐿( to ∆𝑀𝐾𝑇𝑆𝐻𝐴𝑅𝐸( to have the ability to investigate the effect of capital on the
change in market change during the crisis compared to the pre-crisis period.
13 Control variables
This study includes control variables related to risk, ownership, and profitability to mitigate the omitted variable problem and to capture solely the effect of capital on the performance of banks. Appendix A2 explains the definition of all the control variables used in the regressions. This section describes the reasons for including the control variables. All control variables are averaged over the 2 years before the years, unless stated otherwise.
The first factor included is the risk factor. Banks with riskier and less-transparent portfolios are less likely to survive (e.g. Ng and Rusticus, 2011) and might find it hard to improve market share. Three variables are included to control for the difference in risk.
First, a measure of credit risk is included to control for the riskiness of the assets of the banks. This is calculated as the bank’s risk-weighted assets divided by total assets. A high ratio of risk-weighted assets to total assets suggests that the bank holds a portfolio of risky assets. The riskier portfolio increases the possibility of default during a crisis, which decreases the odds of survival for banks during systemic crises.
Second, a measure of liquidity is included because banks with more liquid assets are able to reduce their probability of insolvency, which is a large concern during a systemic crisis. Banks with more liquid assets have a higher probability of survival. This is calculated as liquid assets as percentage of deposits and liabilities.
Last, trading assets are included to capture the effect of what literature calls the “dark side” of liquidity (Morgan, 2002).Trading assets are transparent and liquid, but change easily, whichmakes them hard to monitor. This factor is calculated by dividing trading assets with total assets.
The inclusion of risk as control variable is important because it is a primary reason why banks hold capital, because it helps to isolate the effect of capital on the performance of banks. Literature normally includes only one measure of risk (e.g. Vallascas and Keasey, 2012). Nonetheless, I try to include all three measures to capture all the information of the three factors of risk into a single specification of risk. However, when potential
multicollinearity arises, the variable causing problems is dropped.
The second factor focuses on the ownership status of the banks by concentrating on the bank holding company (BHC) status of the bank. The BHC is a dummy variable that has a value of one when the bank is part of a BHC and zero otherwise. The holding company acts as a strength for the banks that are part of a BHC because it provides internal capital markets in times of distress. This increases the probability of survival for banks during crises and helps strengthening the competitive position of the bank. For example, Houston, James, and Marcus (1997) find that BHC membership affects the loan growth.
14 ROAE is applied because it captures both on- and off-balance sheet activities and banks have to allocate capital against every off-balance-sheet activity. ROAE is the average net income divided by the stockholders’ equity.
Endogeneity
When examining the effect of capital on the performance of banks two potential issues of endogeneity appear. These are reversed causality and the omitted variable problem.
Reversed causality occurs because bank capital effects the survival and market share, but survival and market share also affect bank capital. I use pre-crisis capital to minimize the reversed causality. The pre-crisis capital is lagged over the two years before the start of the systemic crisis, which decreases the causality.
The omitted variable problem arises because variables that are both correlated with the explanatory and dependent variable are excluded from the regression and will create biased coefficients. The following factors are included to reduce the omitted variable problem: risk, ownership, and profitability.
Logistic regression
The logistic regression used in equation (1) and (2) creates problems when interpreting the results. The outcomes of the regressions show the odds ratios. Based on Allison (1999), Mood (2010)shows that the odds ratio cannot clarify the absolute effects neither can the ratio be compared within models and across other estimations. To understand the true effect of bank capital on the survival of banks, I use marginal effects for the logistic regressions. The problems arise due to the specification of the logistic models.
These models do not estimate b but they estimate 𝛽 𝜎F , which increases when omitted independent variables are added to the model (Norton, 2012). However, the effect diminishes when marginal effects are used instead of odds ratios when problems of omitted variables exist (Lee, 1982; Yatchew and Griliches, 1985; Wooldridge, 2010).
15 heteroscedasticity, because the models are extremely sensitive. Therefore, I do not adjust for heteroscedasticity.
4. Data Sample
Before constructing the sample, the systemic crises during the financial and Eurozone crisis in the Euro area need to be defined. The ECB/ESRB EU crisis database is used to identify systemic crises in the Euro area. Both methodologies use pre-crisis capital of banks two years before the start of the systemic crises. This study only focuses on the Euro area, which implies that the Euro must be in circulation during the pre-crisis periods. After applying these restrictions, the following countries emerge: Austria, Belgium, Cyprus, France, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Slovenia and Spain (Appendix A).
The end date of the systemic crises is the end of the management date in the
ECB/ESRB EU crisis database because the end of the management date can be seen as the end of the direct period of the crisis (Bussière and Fratzscher, 2008). The end date of the systemic crisis in Greece is missing since the systemic crisis was still active after publishing the ECB/ESRB EU crisis database in 2017. I assume that the crisis has ended after 2017 to include Greece in the sample.
To construct the sample, the banks include listed and unlisted banks, which are
classified as commercial, savings or cooperation banks. Investment banks are excluded in this paper due to their different risk characteristics (DeYoung and Torna, 2013). Investment banks do not take deposits and are more exposed to market price fluctuations. The total dataset consists of 1554 banks after dropping the banks with no observations. This sample of banks is divided into small, medium and large banks as discussed before.
16 Variable construction
Dependent variables
Survival and market share are included as measures of bank performance during systemic crises because bank managers are most concerned about these performance measures. The survival of banks is not only strategic for banks, but also influences the decisions made by regulators, which are interesting in banking stability. Gaining market share is an important objective for most companies (e.g. Aghion and Stein, 2008) and banks often asses their performance relative to their peer banks (Berger and Bouman, 2013).
For regulators, who are weighing the level and other specifics of capital requirements to achieve banking stability, it is interesting to find out whether bank capital has significant effect of the bank’s survival probability and how this effect changes depending on size and country. Even though the market share of banks is a zero-sum game, it is still important for regulators, because it affects the modus operandi of banks.
Appendix A2 shows the definitions of the dependent variables used in this study. Presented in Appendix A3, panel A, B, and Cshow the descriptive statistics per country. Table 1 shows all the descriptive statistics for small, medium, and large banks. First, survival as dependent variable will be reviewed. The table shows that the percentage survival of banks differs among bank sizes. The banks with the highest percentage of survival are the medium banks with 0.902 or 90.2%, adjoining a standard deviation of 0.298. They are followed up by small banks, which have a percentage of survival of 0.891 or 89.1%, adjoining a standard deviation 0.311. The banks with the lowest survival probability are the largest banks with 0.795 or 79.5%, adjoining a standard deviation of 0.404.
The other dependent variable is market share, which is the other measure of bank performance. The descriptive statistics show that large banks are able to obtain market share during the crisis compared to the pre-crisis with 0.003 or 0.3%, adjoining a standard deviation of 0.025. Small banks do not gain or lose market share on average, adjoining a standard deviation of 0.006. The medium banks experience a negative change in market share during the crisis compared to the pre-crisis with -0.001 or 0.1%, adjoining a standard deviation of 0.006.
Bank capital
17 Appendix A2 shows the definition of EQRAT. Presented in Appendix A3, panel A, B, and Care the descriptive statistics per country. Table 1 shows the descriptive statistics for small, medium, and large banks. Small banks have the largest amount of bank capital with 0.141 or 14.1%, adjoining a standard deviation of 0.148. Medium banks hold less capital than small banks, namely 0.091 or 9.1%, adjoining a standard deviation of 0.298. Largest banks hold the lowest amount of capital with 0.073 or 7.3%, adjoining a standard deviation of 0.066. It shows that the amount of bank capital (EQRAT) decreases with bank size; small banks have most bank capital while the largest have the lowest amount of bank capital.
Control variables
The descriptive statistics in table 3 show that data on the control variable is nearly absent for the small and medium banks. Consequently, the control variables for small and medium banks cannot be used in the regressions. Appendix A2 shows the definition of the control variables. Presented in Appendix A3, panel A, B, and Cshow the descriptive per country. Table 1 show the descriptive statistics for the three separate bank sizes.
The table shows that credit risk for large banks is 0.418, adjoining a standard deviation of 0.151. This means that, on average, large banks hold 41% of their assets are risk-weighted to the total amount of assets. The second factor of risk has a coefficient 34.114, adjoining a standard deviation of 29.447. This means that, on average, large banks hold 34% liquid assets as percentage of deposits and liabilities. The last factor of risk shows that on average 0.040 or 4%, adjoining a standard deviation of 0.081, of their assets are trading assets.
The second factor included concerning ownership shows that almost half of the large banks are member of a bank holding company at 0.458 or 45.8%, adjoining a standard deviation of 0.499. The banks that are part of a bank holding company decreases with size. For medium banks, 0.174 or 17.4% of the banks is part of a bank holding company and only 0.037 or 3.7% of the small banks.
18 Table 1: Descriptive statistics for the 12 European countries during systemic crises, divided by small, medium, and large banks
This table reports the descriptive statistics. The mean value is the average value of the variable. The highest value of the value is presented by the maximum (Max), while minimum (Min) represents the lowest value of the variable. The number of observations included is presented by Obs.
Variables Mean Standard Deviation Min Max Obs
19 Multicollinearity
Table 2 shows the pair wise correlation between all variables in this study. Kennedy (2008) states that correlation coefficients above 0.7 may indicate multicollinearity problems for the regression analysis. The coefficient TRADING_ASSETS is strongly correlated with
CREDIT_RISK. It shows a correlation coefficient of 0.934. TRADING_ASSETS are
removed to deal with the multicollinearity problem and consequently, this paper only includes two measures of risk in the control variables.
Table 2: Pair wise Correlation Matrix
This table reports the correlation coefficient for the main dependent and explanatory variables.
[1] [2] [3] [4] [5] [6] [7] [1] SURVIVAL - [2] EQRAT 0.0713 - [3] ΔMKTSHARE 0.0407 -0.0798 - [4] CREDIT_RISK 0.0196 0.1491 -0.041 - [5] LIQUID_ASSETS 0.1821 -0.2475 0.0348 -0.005 - [6] TRADING_ASSETS 0.0299 0.0934 -0.0358 0.934 0.028 - [7] BHC_MEMBER 0.1664 -0.0935 0.1293 -0.135 0.066 -0.150 - Potential problems
The dataset used in this study causes several limitations. The Bankscope database
discontinued in 2017. The new database, namely BankFocus does not contain data more than three years ago. I obtained data from Bankscope until 2013.3 The data from 2014 onwards is obtained from BankFocus. The index number from Datascope is used to match with the banks in BankFocus, since both databases are from Bureau van Dijk and the index numbers
correspond to the same banks in both databases. However, the index numbers of the banks in Bankscope did not equal all the banks in BankFocus. Consequently, banks are dropped from the sample that can be found in BankFocus but not in Bankscope. This reduces the number of banks in the sample, which can potentially influence the results.
A similar issue arises during the selection of companies for control variables. For the control variables I use the SNL financial database because BankFocus does not provide data for the control variables, because the data is absent for more than three years ago. The SNL financial database mostly includes data on the largest banks as seen in the descriptive
20 statistics. This influences the research since I am not able to include the control variables for medium and small banks and reduce the omitted variable problem to capture only the effect of bank capital on bank performance.
In equation (3) and (4), the methodology of calculating ∆𝑀𝐾𝑇𝑆𝐻𝐴𝑅𝐸( causes
problems, because the ∆𝑀𝐾𝑇𝑆𝐻𝐴𝑅𝐸( cannot be calculated when data is missing on the total
assets during the pre-and during the crisis period. I use interpolation between observations to decrease the amount of missing observations.
Table 3: Sample divided by size and country
This table describes the amount of banks by size and country.
Country Small banks Medium banks Large banks Total
Austria (AT) 207 31 27 265 Belgium (BE) 17 10 17 44 Cyprus (CY) 5 2 7 14 France (FR) 49 37 140 226 Greece (GR) 2 3 14 19 Ireland (IE) 0 0 14 14 Italy (IT) 404 112 112 628 Luxembourg (LU) 15 20 47 82 Netherlands (NL) 5 8 19 32 Portugal (PT) 12 4 13 29 Slovenia (SI) 3 8 8 19 Spain (ES) 63 30 89 182 Total 782 265 507 1554
Table 3 shows the composition of banks by bank size and country. The total banks amount to 1554 banks, divided into small, medium, and large banks. It shows that 32.6% of the banks can be identified as large banks. The amount of small and medium banks compared to the total amount of banks is remarkable low. This suggests that inclusion of smaller and
especially medium banks in this sample is missing; this can influence the inferences about the effect of capital on the bank performance. As robustness, the TBTF banks are dropped to check whether the largest banks influence the results and the cutoff size between medium and large banks is altered.
21 between bank capital and survival or between bank capital and market share might represent the behavior of the Italian banks. As robustness check, Italy is dropped from the sample to check if the effect of capital remains when the Italian banks are absent.
5. Results
This section is divided into two parts. First, the results of the effect of capital on the survival of banks are described. Second, the influence of bank capital on the market share of banks is explained. Both parts of this section first describe the main outcome. Subsequently, the outcome of the GIPSC countries is reported and lastly the influence on the individual countries are reported.
Regarding the main results and the additional regression of the GIPSC countries small, medium, and large banks are first regressed to study the effect of bank capital on the survival of banks without control variables. Second, the regressions are run with control variables for large banks. This strategy is applied because it gives the possibility to compare the
regressions of the large banks with and without control variables to check whether the control variables reduce the omitted variable problem and therefore, are able to capture the true effect capital on the bank performance.
Survival
Before discussing the results, the outcome of the logistic regressions ((1) and (2)) showing the odds ratios are in Appendix A4 and A5. In addition, the odds ratios of the individual crisis are displayed in Appendix A6. I report the marginal effects for logistic regressions to have the correct effect of bank capital on the survival of banks and have the ability to between small, medium and large banks. The effects are proven effectively when models have problems of omitted variables (Lee, 1982; Yatchew and Griliches, 1985).
Table 4 reports the marginal effects for the influence of bank capital on the survival of banks. The results display that bank capital does not affect the survival of banks during systemic crises. This suggest that bank capital does not influence the survival of banks, contrary to the empirical evidence (e.g. Berger and Bouwman, 2013; Vazquez and Frederico, 2015). These researchers find that bank capital positively influences the bank’s probability of survival.
22 However, they also investigate the effect of capital on the survival of banks only during the financial crisis. It shows that only small banks are significantly influenced by bank capital, arguing that government interventions during the financial crisis were largely focused on the larger banks. The insignificant results of the larger banks arise because government assistance partly substituted for pre-crisis bank capital.
This gives support for the insignificant results. Government assistance was both needed for the financial and Eurozone crisis. However, the insignificant results for small banks remain unsolved. The number of small banks compared to the total sample might create insignificant the results, because BankFocus mostly contains larger banks.
Table 4: The effect of the bank’s pre-crisis bank capital on its ability to survive systemic crises
This table shows the marginal effects of the logistic regressions that investigate how pre-crisis capital affects the banks’ ability to survive systemic crises. The pre-crisis periods of the 12 countries is pooled together to investigate the influence of capital on the survival of banks. The descriptive statistics show that data on the control variables for small and medium banks is absent. First, the logistic regressions without control variables are run for all bank sizes (panel A). Second, the logistic regression is run for large banks with control variables (panel B). The results are divided into small banks (total assets up to €1 billion), medium banks (total assets exceeding €1 billion and up to €3 billion), and large banks (total assets exceeding €3 billion).
The dependent variable is SURVIVAL, which is a dummy variable that corresponds to one when the bank is in the sample one quarter before the systemic crisis and is still active one quarter after the systemic crisis has ended. EQRAT is averaged over two years before the start of the systemic crisis. EQRAT is defined as equity to total assets. The parentheses represent the z-statistics. *, **, and *** denote significance at 10%, 5%, and 1% level.
Panel A: SURVIVAL: Marginal effects without control variables
Variables Small banks Medium banks Large banks
EQRAT 0.173 0.003 0.421
(1.56) (0.01) (1.10)
Number of
23 Panel B: SURVIVAL: Marginal effects with control variables
Variables Large banks
EQRAT 1.636 (1.40) CREDIT_RISK -0.270 (-1.15) LIQUID_ASSETS 0.002 (1.14) BHC_MEMBER 0.245*** (2.58) ROAE 0.001 (0.6) Number of observations 109
The control variables in panel B do not show significant results except for BHC_MEMBER, which is positive and significant. Indicating that being part of a bank holding company increases the probability of survival. Banks that are part of bank holding company increase their odds of survival by 24.5%. This is in line with Berger and Bouwman (2013) that show a positive and significant result.
Additional analyses
24 Table 5: The effect of capital on survival for the GIPSC countries
This table shows how pre-crisis capital affects the probability of survival for banks in the GIPSC countries during systemic crises.
The descriptive statistics show that data on control variables for small and medium banks is absent. First, the logistic regressions without control variables are run for all bank sizes (panel A). Second, the logistic regression is run for large banks with control variables (panel B).
The banks are divided into small banks (total assets up to €1 billion), medium banks (total assets exceeding €1 billion and up to €3 billion), and large banks (total assets exceeding €3 billion). The dependent variable is SURVIVAL, which is a dummy variable that corresponds to one when the bank is in the sample one quarter before the systemic crisis and is still active one quarter after the systemic crisis. EQRAT is averaged over two years before the start of the systemic crisis. EQRAT is defined as equity to total assets. GIPSC is an intercept dummy that has the value one if it is one of the GIPSC countries, namely Greece, Ireland, Portugal, Spain, and Cyprus. GIPSC*EQRAT is an interaction dummy that is defined as EQRAT when it represents one of the GIPSC countries. The parentheses represent the z-statistics. *, **, and *** denote significance at 10%, 5%, and 1% level.
Panel A: SURVIVAL: Marginal affects without control variables
Variables Small banks Medium banks Large banks
25 Panel B: SURVIVAL: Marginal effects with control variables
Variables Large banks
EQRAT -0.427 (-0.83) GIPSC -0.229 (-1.51) EQRAT*GRIPSC 1.856 (1.3) CREDIT_RISK 0.030 (0.19) LIQUID_ASSETS 0.001 (0.64) BHC_MEMBER 0.156* (1.85) ROAE 0.007 (1.18) Number of observations 109
Table 5 shows the results of the influence of capital on the survival of the banks in the GIPSC countries. The results show that bank capital positively influences the probability of survival for medium banks. This suggests that bank capital increases the odds of survival for banks in the GIPSC countries. A 1% increase of bank capital increases the probability of survival with 204.6%. This is in line with the empirical evidence (e.g. Berger and Bouwman, 2013;
Vazquez and Frederico, 2015), stating that bank capital increases the probability of survival for banks.
This seems an incorrect interpretation, because the value of the dependent variable is between zero and one. However, it is important to remember that the slope of a function (marginal effect) can be greater than one, even when all the values of the function are all between zero and one. The precise influence of capital on the survival of banks cannot be explained, but the outcome shows a significant and a strong positive relationship between capital and survival of banks.
26 part of a bank holding company. This is in line with the findings of Berger and Bouwman (2013).
Last, the other additional analysis to find the influence of capital on the survival during the individual systemic crises is examined to investigate the particular characteristics surrounding the individual crises and countries. Before discussing the outcomes, the N/A denote the regressions that cannot be regressed, because the insufficient number of non-survivors in the country’s sample.
The results in table 6 show positive and significant results for France (large banks) and Spain (small and medium banks). This suggests that bank capital positively influences the probability of survival for France and Spain. This is in line with the existing empirical literature (e.g. Berger and Bouwman, 2013; Vazquez and Frederico, 2015).
However, a large number of insignificant results appear with contradicting coefficients for all bank sizes. Especially, the substantial number of negative coefficients is surprising, because both literature and empirical evidence suggest a positive relationship between bank capital and survival of banks. Two reasons are observed from table 6 to explain the negative coefficients. First, the number of banks in Belgium, the Netherlands, and Portugal are low, which means that mismeasurement due to the low sample size can influence the coefficient in these countries. Second, most of the results with negative coefficients emerge for large banks, which gives support to Berger and Bouman (2013), who state that government intervention during the financial crisis and for this research the Eurozone crisis was largely focused on the larger banks. The government assistance partly replaced the pre-crisis bank capital,
decreasing the value of holding capital and leading to insignificant and negative coefficients.
Table 6: The effect of capital on survival during individual systemic crises
This table shows how pre-crisis capital affects the survival of banks for the 12 individual systemic crises.
The banks are divided into small banks (total assets up to €1 billion), medium banks (total assets exceeding €1 billion and up to €3 billion), and large banks (total assets exceeding €3 billion). The control variables are not included because the data on the small and medium banks is missing. The dependent variable is SURVIVAL, which is a dummy variable that corresponds to one when the bank is in the sample one quarter before the systemic crisis and is still active one quarter after the systemic crisis has ended. EQRAT is averaged over two years before the start of the systemic crisis. EQRAT is defined as equity to total assets.
The parentheses represent the z-statistics. *, **, and *** denote significance at 10%, 5%, and 1% level.
Variables Small banks Medium banks Large banks
EQRAT
Austria -0.022 0.014 -0.453
(-0.12) (0.05) (-1.06)
27
Belgium -0.452 N/A N/A
(-0.35)
Cyprus N/A N/A N/A
France 0.030 -0.334 1.365*
(0.28) (-0.59) (1.92)
Greece N/A 7.444 9.456
(0.33) (1.44)
Ireland N/A N/A 12.021
(1.27)
Italy -0.005 0.400 -0.111
(-0.04) (0.54) (-0.68)
Luxembourg N/A N/A 0.464
(0.29)
Netherlands N/A -4.828 -1.736
(-1.39) (-0.62)
Portugal 0.142 N/A -0.223
(0.29) (-0.22)
Slovenia N/A N/A N/A
Spain 0.649** 4.600* 1.942
28 Market share
The first part of the research shows that bank capital does not affect the survival of banks. The second part of this research focuses on another measure of bank performance, namely the change in market share, to find if capital affects the market share of banks. In all the regressions, robust standard errors are used to deal with heteroscedasticity since all the Breusch-Pagan / Cook-Weisberg tests for heteroscedasticity are rejected (Appendix A7).
Table 7 shows the results of the OLS estimation to investigate whether bank capital affects the market share of banks. Panel A shows that bank capital does not affect the market share of banks during the crisis compared to the pre-crisis period. This is in contradiction with my hypothesis and the empirical literature (Allen and Gale, 2004; Berger and Bouwman, 2013; González, 2013). These papers find a positive relationship between capital and survival of banks.
This research mostly relates to the study of Berger and Bouman (2013) because a similar methodology is used to empirically examine the effect of capital on the market share of banks. The differences in results can be partly explained due to the fact that they use US data to study the effect of capital on the market share of banks. The geographical differences might explain the different results. In addition, Berger and Bouwman (2013) group crises as banking and market crises over the period 1980 until 2010 while this study focuses on the banks in the Euro area and only focuses on the more recent financial and Eurozone crisis.
Moreover, they also investigate the effect of capital on the survival of banks only during the financial crisis. It shows that only small banks are significantly influenced by bank capital, arguing that government interventions during the financial crisis were largely focused on the larger banks. The insignificant results of the larger banks arise because government assistance partly substituted for pre-crisis bank capital.
However, this does not explain why the small banks do not show significant results. The descriptive statistics show that the number of small banks is very low compared to the total number of banks in the sample. This might influence the results and generate
29
Table 7: The effect of the bank’ pre-crisis bank capital on its market share during systemic crises
This table shows the outcome of the OLS estimation that investigates how pre-crisis capital affects the banks’ ability to survive systemic crises. The pre-crisis periods of the 12 countries is pooled together to investigate the impact of capital on the survival of banks. The descriptive statistics show that data on the control variables for small and medium banks is absent. First, the logistic regressions without control variables are run for all bank sizes (panel A). Second, the logistic regression is run for large banks with control variables (panel B).
The results are divided into small banks (total assets up to €1 billion), medium banks (total assets exceeding €1 billion and up to €3 billion), and large banks (total assets exceeding €3 billion). The dependent variable is DMKTSHARE that is the change in the market share of the bank during the crisis minus the market share pre-crisis period. EQRAT is averaged over two years before the start of the systemic crisis. EQRAT is defined as equity to total assets.
The parentheses represent the t-statistics in panel A and B. The Breusch-Pagan / Cook-Weisberg test for heteroscedasticity is used to determine if robust standard errors are needed (Appendix A7). *, **, and *** denote significance at 10%, 5%, and 1% level.
Panel A: DMKTSHARE: Regression without control variables
Variables Small banks Medium banks Large banks
30 Panel B: DMKTSHARE: Regression with control variables for large banks
Variables Large banks
EQRAT -0.092 (-1.10) CREDIT_RISK 0.092 (0.96) LIQUID_ASSETS 0.000 (0.73) BHC_MEMBER 0.017*** (2.78) ROAE 0.000 (1.49) Constant -0.040 (-0.78) Number of observations 79 R2 0.115
Panel B includes control variables for large banks, because data for the control variables on small and medium banks is missing. It shows that the effect of capital on the market share is insignificant. This indicates that bank capital does not affect market share of large banks when including control variables.
31 Additional analyses
The main results show that bank capital does not affect the market share during the crisis compared to the pre-crisis period of banks. The first part of the result section shows similar outcomes, but additional analysis shows that the effect of capital changes for the GIPSC countries. The severity of the financial and Eurozone crisis varied between countries in the Euro area. The GIPSC countries required bailouts during the Eurozone crisis. An interaction term is included for the GISPC countries to study the effect of capital on the survival of their banks.
Table 8 shows the results of the effect of capital on the market share of banks in the GIPSC countries. Panel A shows that bank capital does not affect the market share of banks during the crisis compared to the pre-crisis period for the GIPSC banks. This suggests that the effect of capital on the market share is not different for GIPSC countries and capital does not influence the market share of banks during the crisis compared to the pre-crisis period. In addition, the results still are in contradiction with the empirical literature (Allen and Gale, 2004; Berger and Bouwman, 2013; González, 2013). These papers show a positive relationship between capital and market share of banks.
32 Table 8: The effect of the bank’ pre-crisis bank capital on its market share during
systemic crises for GIPSC countries
Panel A describes how bank capital effects the market share of the GIPSC countries during the systemic crises by using an interaction term on capital for the GIPSC countries. Panel A describes the regressions without control variables because the data is nearly absent for small and medium banks. Panel B describes the effect of bank capital on large banks in the GIPSC countries by including control variables.
The results are divided into small banks (total assets up to €1 billion), medium banks (total assets exceeding €1 billion and up to €3 billion), and large banks (total assets exceeding €3 billion). The dependent variable is DMKTSHARE that is the change in the market share of the bank during the crisis minus the market share of the banks pre-crisis. EQRAT is averaged over two years before the start of the systemic crisis. EQRAT is defined as equity to total assets. GIPSC is an intercept dummy that has the value one if it is one of the GIPSC countries, namely Greece, Ireland, Portugal, Spain, and Cyprus. GIPSC*EQRAT is an interaction dummy that is defined as EQRAT when it is one of the GIPSC countries.
The parentheses represent the t-statistics in panel A and B. The Breusch-Pagan / Cook-Weisberg test for heteroscedasticity is used to determine if robust standard errors are needed (Appendix A8). Respectively, *, **, and *** denote significance at 10%, 5%, and 1% level.
Panel A: DMKTSHARE: Regression without control variables
Variables Small banks Medium banks Large banks
33 Panel B:DMKTSHARE: Regression large banks with control variables
Variables Large banks
EQRAT -0.358 (-1.39) GIPSC -0.026 (-1.07) EQRAT*GIPSC 0.445 (1.24) CREDIT_RISK 0.099 (1.02) TRADING_ASSETS 0.000 (0.62) BHC_MEMBER 0.012 (2.01)** ROAE 0.000 (1.79)* Constant -0.021 (-0.53) Number of observations 79 R2 0.150
The previous outcomes show that bank capital does not affect market share for the main regression of the 12 systemic crises and the additional analysis on the GIPSC countries. Table 9 reports the results on the effect of capital on the market share for the individual countries. The results show that the influence of capital on the market share displays contradicting outcomes for both bank sizes and countries.
One the one hand, table 9 shows that Austria (medium banks), Cyprus (large banks), Luxembourg (small banks), and Spain (small banks) exhibit positive and significant results. This suggest that bank capital positively influences the market share of banks during the crisis compared to the pre-crisis period. This is line with the hypothesis and the empirical evidence (e.g. Allen and Gale, 2004; Berger and Bouwman, 2013; González, 2013), suggesting a positive relationship between bank capital and market share.
34 have negative significant results. This suggest that bank capital negatively affects the market share of banks during the crisis compared to the pre-crisis period. This is in contradiction with the existing empirical literature (Allen and Gale, 2004; Berger and Bouwman, 2013;
González, 2013).
The results are mixed. The empirical evidence can explain the positive outcomes, while the negative results cannot be clarified by the existing empirical literature. However, the results of the negative relationship between capital and market share can be interpreted based on the reviewed literature. Lyandres (2006) suggests that highly levered companies have more aggressive growth strategies, which enhances the performance of firms. This indicates that banks with more bank capital have less aggressive growth strategies, which results in a lower market share during the crisis compared to the pre-crisis period. The paper of Lyandres (2006) is based non-crisis periods and as pointed out by Berger and Bouman (2013), the effect of capital on market share is more pronounced during crisis periods, which means that the effect of the aggressive growth strategies should be diminished during crisis periods. Thus, the negative influence of capital on the market cannot be explained by reviewing the existing literature.
35 Table 9: The effect of capital on the market share during individual systemic crises
This table shows the effect of the bank’ pre-crisis bank capital on its market share during each systemic crisis per country.
The results are divided into small banks (total assets up to €1 billion), medium banks (total assets exceeding €1 billion and up to €3 billion), and large banks (total assets exceeding €3 billion). All the regressions do not include control variables to mitigate the omitted variable problem because the data on small and medium banks is absent. The constant is not displayed for brevity.
The dependent variable is DMKTSHARE that is the change in the market share of the bank during the crisis minus the market share pre-crisis. EQRAT is averaged over two years before the start of the systemic crisis. EQRAT is defined as equity to total assets.
The parentheses represent the t-statistics. *, **, and *** denote significance at 10%, 5%, and 1% level, respectively.
Variables Small banks Medium banks Large banks
EQRAT Austria 0.000 0.047*** 0.242 (-1.350) (19.010) (1.160) Belgium -0.109*** -0.098*** -0.457 (-45.170) (-18.970) (-1.490) Cyprus -0.013 N/A 3.452* (5.710) (-1.820) France 0.000 0.000 -0.012* (-1.530) (0.150) (-1.920)
Greece N/A N/A -0.081
(-0.090)
Ireland N/A N/A 0.427
36 Portugal 0.000 -0.021 0.079 (-0.440) (-2.030) (0.620) Spain 0.003** -0.001*** 0.000 (2.430) (-4.850) (0.010) Robustness checks
In this section, the robustness of the estimation results for the main regression and the additional analysis on the GIPSC countries (equations (1), (2), (3), and (4)) are tested. First, TBTF banks are excluded from the sample, because bank capital might not provide the same benefits as to other large banks. Second, the chosen cut-off size between medium and large banks might influence the results. The cut-off size between medium and large banks is changed from €3 to €5 billion. Last, Italy is dropped from the sample since it makes up more than 50% of the small banks and more than 40% of the medium banks, which might influence the results of this research. The same control variables for large banks are included in the regressions, but to preserve space, the control variables are not presented in the results. Exclude too-big-to fail banks
The largest banks in the Euro area can be considered TBTF. Capital might not provide the same benefits to TBTF banks, because TBTF banks receive government support when
needed. The TBTF banks are dropped from the sample to diminish the influence of the TBTF banks.
37 Table 10: Robustness: Exclude too-big-to fail banks
This table describes the results by excluding too-big-to fail banks. Two measures of too-big-to fail banks are included. First, the large banks with more than €50 billion are excluded. Second, the systemic banks are excluded to investigate the relationship between capital and survival or change in market share.
Panel A does not include control variables to mitigate the omitted variable problem because the data on small and medium banks is absent. Panel B includes control variables for large banks.
The dependent variables are DMKTSHARE that is the change in the market share of the bank during the crisis minus the market share pre-crisis and SURVIVAL is a dummy variable that corresponds to one when the bank is in the sample one quarter before the systemic crisis and is still active one quarter after the systemic crisis. EQRAT is averaged over two years before the start of the systemic crisis. EQRAT is defined as equity to total assets. GIPSC is an intercept dummy that has the value one if it is one of the GIPSC countries, namely Greece, Ireland, Portugal, Spain, and Cyprus. GIPSC*EQRAT is an interaction dummy that is defined as EQRAT when it is one of the GIPSC countries.
The parentheses represent the z-statistics for SURVIVAL regressions and t-statistics for DMKTSHARE regressions in panel A and B. The Breusch-Pagan / Cook-Weisberg test for heteroscedasticity is used to determine if robust standard errors are needed for the DMKTSHARE regressions (Appendix A8). *, **, and *** denote significance at 10%, 5%, and 1% level,
respectively.
Panel A: Exclude too-big-to fail banks without control variables
Large banks (drop banks with total assets > €50 billion)
Large banks (drop systemic banks)
Key variables SURVIVAL DMKTSHARE SURVIVAL DMKTSHARE
38 Panel B: Exclude too-big-to fail banks with control variables
Large banks (drop banks with total assets > €50 billion)
Large banks (drop systemic banks)
Key variables SURVIVAL DMKTSHARE SURVIVAL DMKTSHARE
EQRAT 0.000 -0.098 1.952 -0.055 (0.1) (-1.03) (1.48) (-0.75) GIPSC countries EQRAT -1.672 0.194 -0.535 0.189 (-1.28) (-0.89) (-0.9) (-1.08) GIPSC -0.638** 0.008 -0.292* 0.014 (-2.06) (-0.38) (-1.72) (-0.65) GIPSC*EQRAT 5.631* 0.260 2.286 0.317 (1.72) (0.82) (1.44) (0.96) Number of observations 69 46 103 73 R2 (Equation 1) 0.202 0.130 R2 (Equation 2) 0.111 0.174
Table 10 show the results. Panel A shows the results of the main regression and the additional analysis on the GIPSC countries without control variables. While the marginal effects on the survival of large banks are similar to the results presented in table 4, the sign of the effect of capital on the market share of banks changes compared to the results in table 7 when both dropping systemic banks and banks with more than €50 billions of total assets. However, these results remain insignificant.
Panel B shows the results including control variables. The results of the change in market share during the crisis compared to the pre-crisis period are consistent with the results in table 7. In addition, similar results arise when the systemic banks are dropped from the sample compared to table 4. However, when dropping the banks with more than €50 billion in total assets the results change for the additional analysis on the GIPSC countries compared to table 5. The marginal effects of the interaction term GIPSC*EQRAT increase in magnitude and become significant. A 1% increase in capital increases the probability of survival with 563.1%.
