The effect of subordinated debt holdings on bank risk taking and the functioning of market discipline: A financial stability perspective Master Thesis MSc International Business & Management

(1)

RIJKSUNIVERSITEIT GRONINGEN

FACULTY OF ECONOMICS AND

BUSINESS

The effect of subordinated debt holdings on bank

risk taking and the functioning of market discipline:

A financial stability perspective

Master Thesis

MSc International Business & Management

Miroslava Karadzhova Supervisor: prof. dr. Hans van Ees

S1939890@student.rug.nl Co-assessor: drs. Henk Ritsema

(2)

2

Abstract

This paper focuses on the impact of subordinated debt holdings on bank

risk taking applying an unbalanced panel data set of 262 commercial,

savings and cooperative European banks in the period 2005-2009. After

controlling for time-variant factors, specific bank characteristics and

country level differences, results indicate that banks with higher

subordinated debt ratio (measured as subordinated debt holdings to total

assets) are more risky. No support is found for the hypothesis that this

relationship is non-linear, which questions the benefits of using

subordinated debt as a facilitator for market discipline. Furthermore,

savings, cooperative and banks with dispersed ownership are found to be

less risky, while dispersed ownership also has a moderating effect on the

subordinated debt holdings – risk relationship.

(3)

3

The 2007-2009 financial crisis has been undoubtedly a result of excessive bank leverage, inadequate and low-quality capital and insufficient liquidity buffers. Repeatedly, regulators across the world have raised their concerns that monitoring and controlling the increasingly complex financial institutions with traditional tools is no longer effective (Bliss & Flannery, 2002). Researchers, regulators and analysts have been investigating in depth the reasons for excessive risk taking by banks’ management with the ultimate goal to construct more sophisticated measures for prevention of similar crises in the future. There has been a wide debate on whether stricter regulations, converging international standards or more efficient mechanisms for monitoring and disciplining should be applied by regulators with the goal to achieve a sustainable risk taking level of banks.

(5)

5 debt bonds can serve as a signal for the stability of a bank provided the market’s perception for the bank’s risk profile reflects accurately risk taking and financial stability.

Those characteristics have made subordinated debt increasingly attractive and already early research in the 1990’s (e. g. Flannery & Sorescu, 1996) has proposed that banks should be required to issue subordinated debt and maintain sufficient subordinated debt holdings in order to constrain risk taking. However, it is questionable whether the effect of subordinated debt is entirely positive. According to opposing arguments subordinated debt has an ambiguous effect on risk taking. It can reduce risk only in case banks are credibly willing to commit to a sustainable level of risk (Blum, 2002). Furthermore, replacing equity holdings with subordinated debt holdings in the bank capital structure may destabilize the bank, because it increases leverage and consequently the possibility of insolvency (Levonian, 2001). Apparently the potential risk-subordinated debt relationship is complex and possibly twofold, especially considering the effect of subordinated debt holdings.

Financing through subordinated debt instead of equity has also been preferred by bank management in the resent years, since equity is viewed as a more expensive source of capital. Accordingly the increased capital requirements from Basel III have prompted a wide discussion and management of large international banks has expressed views that those capital requirements could not be as beneficial for the financial system, because they may restrict lending activities and impede economic growth.

(6)

6 On the other hand, explicit deposit insurance guarantees and implicit bailout policies by monetary authorities diminish the possible losses of bank creditors (Kato & Hagendorff, 2010). This gives rise to a potential moral hazard problem and lack of incentives for creditors to monitor banks. Those factors, together with the increased trend in institutional ownership in banks in the latest years suggests a new unhealthy situation that banks may face in the near future, because of the unaligned incentives of large institutional shareholders and creditors. Consequently, this may mean that banks would also accept higher risk in order to minimize the effects of the higher costs of financing.

Following the above discussion, this paper focuses on finding out what is the real effect of subordinated debt holdings on bank risk taking. As indicated, subordinated debt is an issue of current importance, with implications for the overall stability of the financial system and the development of future regulatory policies. So far, empirical literature has mostly focused on the relationship between price of subordinated debt and market valuation of banks in order to infer the existence of market discipline. However, the balance sheet effect of subordinated debt holdings on risk taking has been largely neglected. The paper aims to close this gap and investigates to what extent subordinated debt holdings can explain risk taking by testing how they affect the financial stability of banks interacting with other relevant bank characteristics. This relationship has important implications for the understanding of the functioning of the market discipline. Furthermore, it can serve as a basis for regulators to set standards for the continuously growing market of subordinated debt. Due to the fact that there is no converging view on this relationship, the study takes on an exploratory character. It aims to find the shape of this relationship, including additional factors discussed in the literature, which are considered and found to be of significant influence on bank risk taking.

(7)

7 is found that the negative relationship between subordinated debt and financial stability is mitigated by the presence of a dispersed ownership structure at the bank.

The paper is structured as follows. Section 2 provides an overview on the empirical literature on subordinated debt and develops the research question, hypotheses and economic model applied. Section 3 explains the choice of data, methodology and the specification of the econometric model. The results and robustness checks are presented in section 4. Finally, concluding remarks, limitations and implications for further research are provided in section 5.

2. Literature review

Before addressing in depth empirical findings, it is essential to provide meaningful definitions of the ground concepts, which are used throughout this research, and the relationships between them.

2.1 Theoretical background

As already stated subordinated debt accrues in the form of fixed income instruments that are unsecured and junior to all other obligations of the bank, in the event of bankruptcy (Caldwell, 2005). Unsecured means that in the case of bankruptcy no underlying bank assets can be claimed by bondholders. Junior implies that its holders receive the lowest priority. Subordinated debt covenants also state that it is not redeemable before maturity without regulators’ approval and include a lock-in clause prohibiting payment of either interest or principal (even at maturity), if the payment would cause the issuing bank’s risk-based capital ratio to fall or remain below the minimum required ratio (Kaplanski, 2008). Subordinated debt has been associated with bank risk mainly in two ways, i.e. through direct and indirect market discipline. Market discipline is defined as the effects of publicly available market signals from bank-issued securities that lead to less risk taking by the issuing bank, relative to otherwise similar banks that do not issue publicly traded securities (Kwan, 2004). In the case of subordinated debt, the logic is that holders of such debt are very sensitive about default risk and have an incentive structure similar to that of regulators (Evanoff et al., 2011), because their claims have no upside potential and at the same time are to be repaid last in case of default.

Direct market discipline refers to new debt and originates from the risk premium

(8)

8 but also uninsured subordinated debt. In the case where the bank defaults, the insured depositors receive compensation from the government. On the contrary, subordinated creditors face the possibility to lose their investment in the bank, thus they would care more about the banks’ asset choices. Simultaneously, because the market for subordinated debt is competitive, the creditors require different interest rates depending on their anticipation of the banks’ risk-orientation (its future asset choices), which is done by looking at the riskiness of the current assets a bank holds. Direct market discipline puts emphasis mostly on short term subordinated debt, because it could be easily withdrawn and has to be renewed more often. This can induce managers to act more in line with the preferences of creditors in order to secure future debt. Early research on subordinated debt has largely emphasized this function. One important issue within the debate about market discipline is the extent to which private investors can observe and price the risk taken by banks and affect bank’s management decisions.

The second function that focuses on existing debt is indirect market discipline, which is also referred to as signaling. It is linked to monitoring by investors on the secondary market, supervisors and central banks. In essence, the yields of a bank’s risk-sensitive source of funds are applied as a signal for bank supervisors to improve their risk monitoring and controlling tasks (Kwast et al., 1999). Subsequently, this implies that banks would pay additional attention to their risk taking activities in order to maintain lower price of their debt. Evanoff and Wall (2000) stress especially on the importance of indirect discipline in relation to subordinated debt prices as a market mechanism. Due to the junior status of subordinated debt, its price should be very sensitive and will reflect more accurately how market participants1 evaluate banks’ risk. However, this would only be possible in case of a well-functioning market, with no asymmetrical information. Studies on the subordinated debt market present rather diverging results. While Flannery and Sorescu (1996) find support for the risk-spread relationship, Jagtiani, Kaufman and Lemieux (1999), as well as Morgan and Stiroh (2000), present conflicting results and a weak relationship, if any. Consequently, the disciplining effect of the subordinated debt market on bank risk is not unambiguously proven.

The Basel Committee on Banking Supervision is actively promoting the concept of market discipline as an effective tool for prudential regulation in banking. It became a key pillar in the second Basel Accord. According to the Basel II capital requirements long-term subordinated debt is part of Tier 2 capital allocated for market risk and short-term – part of

(9)

9 Tier 3 capital. The newly introduced Basel III framework proposed that short-term subordinated debt should be excluded from bank’s capital ratio calculation, implying that it doesn’t serve as effectively as equity in buffering against bank losses and protection for the claims of depositors. Furthermore, the subordinated debt eligible for Tier 2 capital should have maturity of at least 5 years and it should include covenants for mandatory write-down or conversion to common equity (Basel Committee on Banking Supervision, 2010). Neither Basel II nor Basel III entail mandatory subordinated debt issuance, which is also an indication that regulators do not consider subordinated debt as a perfect substitute for equity in the financing mix.

It has been argued that issuance of subordinated debt serves for the enhancement of market discipline as long as the yields of the debts are correlated with the incentives for risk taking (Evanoff & Wall, 2000; Goyal, 2005). There has been a wide debate on whether the latter actually holds in reality. In theory, a bank would be incentivized to take less risk in order to be able to achieve a lower cost of debt. However, once the interest rate on debt is already contracted, the bank could again invest in riskier assets, especially if debt holders are not willing or able to exert influence on the bank’s risk appetite. Furthermore, the incentives of subordinated debt holders change, if the bank is not financially stable. Their claims resemble much more those of equity holders, if the bank is near insolvency, because they will accept higher risk in order to increase the possibility of full repayment (Levonian, 2001).

Proponents of the introduction of mandatory subordinated debt claim that it is one of the most effective channels for market discipline (Basel Committee on Banking Supervision, 2003; Evanoff, and Wall, 2000; Lang and Robertson, 2002). They consider that requiring banks to issue some amount of subordinated debt repetitively can constrain their risk taking. The explanation for this mechanism is that creditors will charge riskier banks higher interest rate, thus banks would think twice before taking excessive risk, because they would continuously have to “pass the test of the market” (Evanoff et al., 2011). Some proposals (Evanoff, and Wall, 2000) have suggested that subordinated issuance has to be made compulsory, but only for large banks.2 In order for the market to function efficiently banks would have to issue debt once or twice a year. The most popular market discipline proposals suggest that banks should be required to hold a minimum amount of subordinated, uninsured debt (Lang and Robertson, 2002). This amount ranges from 2% of total liabilities to 3-5% of deposits, 1-4% of risk weighted assets (Basel Committee on Banking Supervision, 2003).

(10)

10 According to this view, banks that issue debt on a regular basis and hold subordinated debt will be sensitive to market prices of debt and will constantly readjust their risk level, in order to send positive signals to the market. In essence, this implies that if a bank does not hold (or hold an amount below the minimum) subordinated securities, secondary market discipline will not be present and the bank will not be constrained to take excessive risk. However, no expectations are made about banks that have higher subordinated debt holdings and how this affects their risk taking incentives.

Empirical research examining mostly cross-sectional data provides evidence that creditors indeed charge riskier banks a higher interest rate on subordinated debt (i.e. Covitz, Hancock & Kwast, 2004; Fan et al., 2003; Sironi, 2003), which is an indication for the desired monitoring effect, but this result does not imply that banks are ultimately disciplined by the market. There are opposing views (Bliss & Flannery 2002; Evanoff & Wall, 2000), which argue that the prices of subordinated debt may not reflect the risk level of banks accurately due to information asymmetry. Potential (and current) investors may lack the ability (and/or willingness) to observe and measure risk. Furthermore, the expectation of (explicit and implicit) government guarantees for all obligations undertaken by banks weakens monitoring incentives. It also has to be taken into account is that effective monitoring does not necessarily mean influencing bank behavior (Ashcraft, 2008; Bliss & Flannery, 2002). Effective monitoring is present, when investors manage to evaluate banks’ financial stability accurately and price banks’ securities accordingly. While influencing (or as also used interchangeably in the literature - disciplining) implies that changes in securities prices prompt bank management to corrective action. Bliss and Flannery (2002) test for the hypothesis that bank management responds to changes in subordinated debt and equity prices and find no support for it.

(11)

11 that the probability of subordinated debt claims being repaid in full rises with an increase in asset volatility. Consequently such incentives arising from subordinated debt combined with high subordinated debt holdings on the bank balance sheet may have an adverse effect on stability, offsetting the potentially positive effects of market discipline.

2.2 Problem indication, originality and relevance for the Management Theory Earlier proposals for mandatory subordinated debt issuance of banks provide rather general explanations about its effects on risk taking. Mostly, they focus on the premium that prospective investors would require on debt for banks with riskier assets before issuance and only stress on the monitoring and influencing effects. The other wide stream of literature on subordinated debt aims at assessing its effectiveness as a direct or indirect market discipline tool through investigation of bond spreads. The implications from the possible subordinated debt holdings – risk taking relationship are rather underestimated.

As hinted by Blum (2002), commitment to a certain risk level (or the inability to do so) should be considered as an underlying factor in determining the effect of subordinated debt on risk. Commitment to a certain risk level is done at the level of bank’s management, which in turn is influenced by shareholders. Bank’s specialization, though arguably an exogenous factor, also influences the type of assets banks invest in. Thus, ownership concentration and type of bank are factors that predetermine the bank’s risk aversion to a certain extent.

(12)

12 considering subordinated debt holdings in the context of agency problems, the amount of subordinated debt, issued and held by banks in order to boost capital ratios, could have a negative effect and lead to increased risk taking, especially for banks that are not able to commit to a certain level of risk ex ante. High subordinated debt holdings in the bank capital structure increase leverage and may aggravate the risk taking incentives for equity holders significantly, thus it is of great interest to focus on how this affects financial stability. Simultaneously, risk taking incentives and the choice of assets, are also determined by the ownership structure and specialization of banks. If a bank is able to commit to investing in safe assets disregarding its capital structure, then the leverage effect of subordinated debt holdings may not be as severe.

In order to achieve a certain perspective, the basic problem of evaluating the effect of subordinated debt holdings on bank risk is subdivided into three more specific ones, which point out the benefits of the proposed research. The first and most important problem that arises is that currently it is not known how the already issued subordinated debt has affected the risk taking of banks. There have been some theoretical propositions about what should be expected, but no empirical analysis has been done on this phenomenon. This research aims to achieve originality and innovativeness (Thomas, 2006) by investigating the empirical reality in that matter and in this sense it is rather exploratory. Results in that respect will be beneficial to policymakers for gauging the real effect of subordinated debt on banks, which will allow for drawing conclusions about the extent to which its use has to be promoted, required or constrained by regulators.

Secondly, the researched problem is considered of extremely contemporary importance, due to the newly arisen situation, in which banks were almost “forced” to increase the amount of their subordinated debt financing due to the contracted market for deposits and the expensive equity financing. This current state has to be analyzed with caution since the extraordinary measures taken during the crisis may not reflect reasonable thinking or theoretical assumptions of how the banking sector operates. Given the changes in the empirical reality, it is plausible to question existing theory in the field.

(13)

13 2.3 Main research question and sub-questions

In order to narrow the scope of the proposed research it is considered appropriate to focus the investigation on the banking sector within the EU. The main reason for this choice is the fact that there has been a lack of research on this entity. Most research on subordinated debt has focused on the US. Furthermore, there have been multiple papers focused on a single market such as Canada (Caldwell, 2005), Japan, Taiwan (Chih-Yung, Yu-Fen & Gu-Shin, 2010). The European banking industry has been largely neglected in this regard. To the best of my knowledge there are only three studies on it (Bruni & Paternó, 1995; Groop, Vesala & Vulpes, 2002; Sironi, 2003). However, they do not directly address the relationship between risk taking and subordinated debt holdings. Following from the theoretical discussion above it is evident that subordinated debt has been regarded only as a facilitator for market discipline, while the effects of subordinated debt holdings on the bank balance sheet have been neglected. Considering that the benefits from market discipline may not outweigh the costs of excessive leverage and the chosen scope of the paper, the main research question is posed as:

What is the effect of subordinated debt holdings on bank risk taking in the EU banking industry?

(14)

14 leveraged capital structure due to high subordinated debt holdings, it may further aggravate risk. Thus, ownership concentration is added to the theoretical model to increase its explanatory power.

Sub-question 1: How do subordinated debt holdings affect risk taking in the EU banking industry when ownership concentration is considered?

Furthermore, since asset preferences differ for banks with different specialization, it has to be taken into consideration that a bank’s risk-orientation could also be constrained by the specialization of the bank. Banking specialization determines the objectives of a bank to a great extent. For example, a commercial bank has the objective of shareholder value maximization, while savings and cooperative banks may focus on more socially responsible lending and providing value for their customer owners (Westman, 2009). Traditionally, savings banks have been operating with the goal to enable savings opportunities for the general public and to support regional development projects (Gardener et al., 1997). On the other hand, the main objective of cooperative banks has been to offer favorable deposits and loan prices to representatives of a particular profession or other group of retail customers (O'Hara, 1981). Those different objectives provide a basis for different risk-orientation in itself (Westman, 2009), which implies that being a cooperative or a savings bank in itself may be a risk-constraining factor mitigating the leverage effects of subordinated debt on the bank balance sheet.

Sub-question 2: How do subordinated debt holding affect risk taking in the EU banking industry when bank specialization is considered?

(15)

15 to the US, where the FDICIA3 was applied after 1991 and bank failures occurred through the recent financial crisis. As a result the risk taking and ability to commit to a sustainable level of risk in the EU should be even more dependent on bank’s specialization and ownership, which in turn may mitigate the effect from excessive leverage due to high subordinated debt holdings.

2.4 Expectations and the economic model

2.4.1 Expectations about the significant negative relationship between subordinated debt holdings and bank risk taking

The right hand side of the bank balance sheet is a mix of equity, insured deposits and possibly uninsured subordinated debt. In order for the bank to provide loans to finance its lending activities, it has to make decisions about its financing mix. The higher the debt to equity ratio at the bank, the more leveraged it is. Leverage leads to distortion in the lending decisions of the bank and preference for more risky assets, in order to lower the costs of funding. Furthermore, return on equity increases with leverage, because of the higher probability of financial distress. Since higher leverage induces higher return on equity, it creates frictions and incentive for excessive risk taking. Subsequently it is wrongly considered that the increase in equity requirements increases bank’s funding costs, because more equity on the balance sheet decreases leverage, and accordingly reduces the riskiness and the required return on equity. Another advantage of more equity financing is that it leads to better capitalization and less need for external financing needed to fund growth. Results from recent studies confirm that bank value is positively related to capitalization (Mehran and Thakor, 2010) and that better capitalized banks are more able to survive financial discress (Berger and Bouwman, 2010).

In order for banks to respond to the higher capital requirements of regulators, they may be incentivized to substitute equity financing for subordinated debt, since it could increase their capital ratio and also the interest payments on debt are tax deductible. However, as noted by Levonian (2001), this substitution increases leverage, leads to a more fragile capital structure and increases the possible social costs associated with implicit government guarantees. Due to such guarantees, even though the claims of subordinated debt holders are not insured, they have fewer incentives to monitor and influence banks’ asset choices and risk taking. If it is expected that the government will bail out the bank in the case of default, the

(16)

16 creditor will not be willing to incur extra costs in monitoring. Subsequently, it follows that substituting equity for subordinated debt actually exacerbates governance and type III agency problems and does not serve as a solution to the moral hazard problem, present from the convex claims of shareholders. To the contrary, the presence of subordinated debt on the bank balance sheet may aggravate the incentives for management to choose for assets with higher return yields, but which increase the probability of distress.

Another important factor is that debt holders are in general more dispersed, and may not only be unwilling but also unable to engage in monitoring the bank. Moreover, their claims are pre-specified and as long as the bank is able to satisfy them, they wouldn’t be concerned for the stability of the bank. This is true especially for large banks with systemic importance, which are deemed “too big to fail”. As a result, those banks could achieve lower interest rates on debt, substitute more expensive equity for debt and increase leverage and the probability of distress. The reasoning behind promotion of subordinated debt as a facilitator for market discipline comes from the assumption that subordinated debt holders face only downside risk and have no upside potential gains, thus their risk preferences should converge with those of the deposit insurer. This is not necessarily the case because subordinated debt holders also benefit from asset volatility and are willing to impose more risk, if they believe that the bank is unstable and can face default. Those incentives combined with high leverage due to increase in subordinated debt holdings, are expected to lead to higher risk taking. Accordingly the first hypothesis is posed as:

H1: Subordinated debt holdings have a significant negative effect on bank risk taking.

2.4.2 Expectations about the non-linearity of the relationship between subordinated debt holdings and bank risk taking

(17)

17 Niu (2008) gives a more detailed explanation of how exactly disciplining would work by presenting the bank’s asset choice in the presence or absence of subordinated debt holdings on the bank balance sheet. If a bank is financed only with insured deposits and equity (i.e. no subordinated debt on the balance sheet), it will always choose for risky assets after the interest rate on deposits is contracted, because the risky asset provides a higher return when successful. In this case the bank will not care for the downside risk, because of its limited liability and the fact that all its depositors are insured by the government in case of default. On the other hand if a bank chooses for a financing mix which also includes subordinated debt, its will be more sensitive about risk and asset choices, because subordinated debt creditors are not insured and may impose pressure on the bank to take on a more risk averse strategy through charging higher interest rates and further influencing the bank’s decisions.

This positive effect however, is not expected to hold when subordinated debt holdings are excessively large, because then bank incentives change again. With high subordinated debt holdings and excessive leverage the bank’s incentives to invest in risky assets would prevail, since this will lower the bank’s expected cost of debt. Having in mind the possible changes in asset preferences due to different subordinated debt holdings, it is possible that a U-shaped relationship exists between risk and subordinated debt holdings, in case banks asset choices are influenced by how investors price subordinated debt securities and the corrective actions regulators may impose due to the market signals. This implies that banks with low or none subordinated debt holdings on the balance sheet would take higher risk than banks that hold a sufficient amount of subordinated debt, because they will not face evaluation by the market. A research on the Canadian banking sector provides some evidence for this possibility by showing that riskier banks are less likely to resort to the subordinated debt market for financing in order to avoid market evaluation (Caldwell, 2005). On the other end, where subordinated debt holdings are very high, the effect of market discipline is likely to be offset by the leverage effect discussed in the previous subchapter.

As already mentioned, empirical evidence for the presence influencing effect of the subordinated debt market is rather mixed. But considering that there could be a disciplining effect that may offset the negative leverage effect from subordinated debt holdings, it is of interest to test for the possible non-linearity of the subordinated debt – risk taking relationship. Consequently, the second hypothesis is formulated as:

(18)

18 2.4.3. Expectations about the moderating effects of bank specialization and ownership concentration on the relationship between subordinated debt holdings and bank risk taking

The main argument behind inclusion of specialization and ownership concentration as possible moderators of the subordinated debt – risk taking relationship, is the importance of the commitment capabilities of a bank when assessing the effects of subordinated debt as identified by Bloom (2002). If a bank is able to commit ex ante to a given level of risk, then this may decrease the negative leverage effect of subordinated debt holdings and exert a moderating effect. Having in mind that there is already empirical evidence suggesting that cooperative, savings banks and banks with dispersed ownership are less risky, it is also considered meaningful to check whether the leverage effect would be less severe for those banks.

In general, it is assumed that cooperative banks as well as savings banks are far more conservative in their asset choices compared to commercial banks due to different objectives. When the objectives of a bank have a more socially responsible character and do not only promote shareholder value maximization at any cost, excessive risk taking due to leverage from subordinated debt holdings may be less severe. Commercial banks are mainly active in a combination of retail banking (individuals, SMEs), wholesale banking (large corporations) and private banking. Commercial banks’ main objective is shareholder value maximization, while this is not the case for savings and cooperative banks. Furthermore, the inclusion of such classification is justified due to the substantial presence of savings and cooperative banks in the European banking system. Savings banks are mainly active in retail banking and usually belong to a group of savings banks. They account for about 25% of nonbank customer deposits in the EU, 50% in Spain and 40% in Germany (Williams, Peypoch & Barros, 2011).4 They have a decentralized distribution network and provide local outreach. Traditionally, their objective has been to enable savings opportunities to the general public, to carry out social or beneficiary work and to support regional development projects (Bergendahl & Lindblom, 2008).

Cooperative banks have a cooperative ownership structure and are mainly active in retail banking. They have emerged as institutions providing favorable deposits and loans to representatives of a particular profession (Altunbas, Evans & Molyneux, 2001). The ownership rights of both cooperative and savings banks are often ambiguous, thus the disciplining effect of the market for corporate control is absent (Williams et al, 2011). Thus, it

(19)

19 has been largely considered that they are less profitable than commercial banks. High profitability in banking is achieved through investing in riskier assets with higher returns and findings from recent research document that lower profitability is often associated with lower risk (Garcia-Marco & Robles-Fernandez, 2008; Westman, 2009). Lower risk also originates from the agency relationships within such organizations. Cooperative banks, for example, are owned by their customers, which implies that there is no agency conflict between shareholders and depositors. Since bank specialization in itself may be an explanation for risk differences and specific asset choices, this has to be accounted for when evaluating the effect of subordinated debt holdings. Accordingly, the third hypothesis is formulated below:

H3: The relationship between subordinated debt holdings and risk taking is moderated by bank specialization.

Ownership concentration in banking has also been extensively researched. Several papers (Saunders, Strock, and Travlos, 1990; Gorton and Rosen, 1995; Demsetz, Saidenberg

(20)

20

H4: The relationship between subordinated debt holdings and risk taking is moderated by less concentrated ownership at the bank.

2.4.4. Control variables and the economic model

In order to measure the net effect of subordinated debt holdings, a set of other relevant bank characteristics is introduced as well. First of all, it is controlled for size, since in general public firms tend to be larger than private firms and commercial banks tend to be larger than savings and cooperative ones. Moreover, smaller banks have more difficult access to subordinated debt markets due to high costs of issuance. Also large banks tend to be more efficient than small banks, but at the same time they may be more risky and opaque. Another issue is that when a bank is too large, there is a threat of systemic impact in case of failure. The implicit government guarantees for a bail-out in such cases induce management to increase risk. Size matters especially in the case of assessing the effects of leverage from subordinated debt holdings, because large banks that are more likely to be bailed out from the government, thus subordinated debt creditors will have even lower incentives to monitor and influence them. In relation to this argument, it is also noted by Poole (2009) that debt capital flows more readily to larger banks, even inefficient ones.

Second, market discipline may have different effect on risk taking for public and private banks. Essentially, the effect of market exposure on public banks could be two-fold. Publicly-owned banks can gain access to additional equity at a lower cost than privately-owned banks. Public banks face strong incentives to conduct their business in a sound manner and maintain a high level of equity capital to face potential losses. Thus, public ownership facilitates the market for corporate control. Previous research indicates that publicly held banks tend to hold more capital than privately held ones (Kwan, 2004). As capital measures the ability of the bank to absorb losses, it is considered that banks holding more capital are less risky. On the other hand, public banks have the ability to expand and become more complex. Consequently, this leads to a higher degree of freedom to manage their equity and invest in risky projects with a higher expected return.

(21)

21 shareholders were achieved through the investment in very risky assets and the underestimation of the consequences of those actions. Thus, in the long run this cannot be considered as good performance and in order to achieve sustainably good results a bank should be able to commit to a given strategy and level of risk. Consequently, bank performance and capitalization are controlled for.

Fourth, although the EU banking system is becoming more and more integrated through converging international standards and the efforts of the Basel Committee on Banking Supervision, differences in legislation, market characteristics and competition on country level still exist. Accordingly, it is also controlled for the country of origin of the banks included in the research.

Thus, a basis model including subordinated debt holdings, bank specialization, ownership concentration and other relevant controls (size, listing status, profitability, capitalization and country of origin) is deemed appropriate to explain risk taking. Additionally, the possible presence of moderation effects of ownership concentration and bank specialization is explored. The economic model is developed as:

RISK = f [subordinated debt holdings, bank specialization, ownership concentration, moderating effects of bank specialization and ownership concentration, controls (size, listing status, profitability, capitalization and country of origin)]

3. Data and methodology 3.1 Data selection

(22)

22 BankScope lists 1623 commercial, savings and cooperative banks that are active and provide financial statements for the years considered. A few restrictions are imposed on this initial sample. First, only consolidated statements5 are used for all banks. This restriction helps to focus on the ultimate ownership as much as possible. Furthermore, strategic decisions as those for debt issuance are made at the top of the organization with all subsidiaries and operations taken into account. Another expected benefit noted by Westman (2009) is that the use of consolidated data, instead of data for individual savings and cooperative entities for example, increases the comparability of those banks with commercial ones. When imposing the consolidation restriction the sample reduced to 1635 year-observations for 331 banks. However there were missing values for the subordinated debt holdings for some years.6 After deletion of those, as well as banks with only one year-observation the sample reduced further to 1390 year-observations from 306 banks. Observations with missing values for ownership concentration and negative values for Z-score7 were also deleted. This resulted in the final panel of 1180 observations for 262 banks where 68% of all banks have observations for all 5 years, 20% for 4 years and 12% for 3 or 2 years, out of which 122 observations are for savings and 203 for cooperative banks. Table 1 in Appendix B displays summary statistics for all measures used for the further calculation of variables.

3.2 Relation between concepts and measures

For the reader’s convenience and further reference, each concept discussed below, the variable through which it is operationalized, the abbreviation of each variable and a thorough variable description can also be found in Appendix A.

The concept of risk taking is rather difficult to operationalize and this has been problematic for researchers in the field of banking as well as regulators. This research follows Laeven and Levine (2009) and takes on a financial stability perspective in measuring risk taking. More specifically, a bank’s Z-score, which indicates the distance from insolvency is used as a proxy for risk taking. It was first proposed by Boyd and Graham (1986) and later applied in multiple studies (Hannan & Hanweck, 1988; De Nicolò, 2001; Laeven & Levine, 2009 among others). It is an accounting based measure which is constructed based on data available and reported on an annual basis. It is the inverse of the probability of failure for a

5_{A consolidated statement integrates the statements of all subsidiaries.}

6_{Due to the fact that subordinated debt holdings in 2005 were not widely reported.}

7_{12 bank-year observations had non-positive Z-score values, thus they had to be deleted, since it is not possible}

(23)

23 given bank, with high scores implying low probability. It is considered very useful for comparison of different types of financial institutions and it furthermore accounts for overall bank risk (Berger et al, 2008) since it captures profitability, leverage and volatility.

It is calculated as Z-score = (ROA+CAR)/σROA, where ROA = Net Income/Total Assets, CAR = Equity8/Total Assets expressed in percentage and σ(ROA) is the standard deviation of return on assets, a measure of the returns volatility. σ(ROA) is calculated over the researched period separately for each bank, taking all year-observations into account.9 The interpretation of the Z-score is that it indicates the number of standard deviations that the bank’s return on assets has to drop below before it becomes insolvent. Following Westman (2009), Iannotta, Nocera and Sironi (2007) the natural logarithm of the Z-score is used (referred to as LNZ further on), since this measure is found to be highly skewed (results from a Shapiro-Wilk test reject the hypothesis that Z-score is normally distributed at 1%).

The concept of subordinated debt holdings (SND) is easily related to the empirical reality. Banks provide information on the amount of debt outstanding on an annual basis. The SND data are also highly skewed, however a logarithmic transformation is not deemed appropriate, since the observations with zero debt for a particular year will be lost, while they still provide meaningful information. Using the total amount of SND as an independent variable itself is not deemed appropriate because of the differences in bank size. Researchers relate SND to different other measures such as bank liabilities, deposits or risk-weighted assets (Sironi, 2001). Due to the fact that risk-weighted assets are subject to differences in reporting, SND is related to total assets as applied by the U.S. Shadow Financial Regulatory Committee (2000), where SND_Ratio = (SND/TA)*100.10 For the purpose of testing H2, a square of the SND_Ratio variable is also introduced, where SND_Ratio_sq = (SND_Ratio)².

Bank specialization is divided in three groups: commercial, savings and cooperative. Out of the 18 available options on the BankScope database only those are chosen, as these constitute the largest groups and it is considered that for those groups it is possible to derive meaningful conclusions. Dummy variables for savings and cooperative are included (referred to as SAVE and COOP, respectively), which indicate whether bank i is a savings, cooperative, or otherwise a commercial bank. In general it is expected that there is a mitigating effect of

8_{As per BankScope’s definition equity consists of common shares and premium, non-controlling interest,}

revaluation reserves and reserves for general banking risk

9_{Some studies calculate it on a three year-rolling time window, however this is not possible here, due to limited}

number of year-observations.

10_{The U.S. Shadow Financial Regulatory Committee (2000) argues that a ratio calculated with risk weighted}

(24)

24 those types on risk. Additionally, interaction variables between SND_Ratio and type are included to test the possible moderating effect of bank specialization on the subordinated debt holdings - risk relationship as proposed in H3.

In order to construct variables representing ownership concentration BankScope database’s Independence Indicator is used. This indicator considers direct and total ownership of any bank with known recorded shareholders. The sample banks are divided in three groups as per the following criteria. According to BankScope’s classification a bank is considered ‘independent’ (INDEP), i.e. having dispersed shareholder base, if no shareholder (excluding public and aggregated shareholders, which are considered as unable to exert controlling power over the bank) owns more than 25% of the bank directly or in total. In another group fall banks, which have one or more small block holders with an ownership percentage of 25 or more - SMBLOCK, but at the same time no shareholder owns more than 50% of the bank. The third group consists of those banks, which have a large block holder who owns 50% or more of the bank. This group is omitted in the model to prevent perfect collinearity. It is expected that banks with dispersed ownership or small block holders will be less risky. Again interaction variables are added to test H4.

The following control variables are introduced:

Size is measured through the natural logarithm of the book value of total assets (LNTA) as proposed in multiple studies (for example Chen, Steiner & Whyte, 1998; Anderson & Fraser, 2000; Sullivan & Spong, 2007). Total assets measure bank size, which is entered with a natural logarithm transformation to account for its skewness (results from a Shapiro-Wilk test reject the hypothesis that Z-score is normally distributed at 1%). Ceteris paribus, larger banks tend to hold more diversified asset portfolios, resulting in a lower level of risk (Chen et al, 1998). However, this is the case only if their portfolios are replications of smaller banks’ portfolios. Larger banks are inherently more opaque and also hold more off-balance sheet risk, which implies higher risk (Anderson & Fraser, 2000). Morgan and Stiroh (2001) also present evidence that the spread - risk relationship on bank bonds is weaker for larger and less transparent banks. Due to the conflicting results in previous studies, no particular expectations for size are made.

(25)

25 This is deemed appropriate since such companies no longer have access to cheaper capital, thus market discipline for them should have the same implications as for unlisted ones. Changes in the listing status of sample banks are also accounted for. To account for the differences in listing status a dummy variable LISTED is included in the model, which takes the value of one, if the bank is listed and zero otherwise. Banking is a rather homogenous industry, with presumably similar production functions. Thus it seems plausible to view listing status as an exogenous variable, especially after controlling for firm size. Although it has been suggested that effects from listing may be two-fold, results from a recent paper on the European banking industry (Barry, Lepetit, & Tarazi, 2011) suggest that market forces align the risk taking behavior of publicly held banks. Thus, the expected sign associated to the variable LISTED is positive.

Profitability is measured through Net Interest Margin (NETIRM), which represents the difference of interest income and interest paid out to depositors, divided by the total interest earning assets for each year-observation. Flannery and Sorescu (1996) argue that there is a positive relationship between profitability and default likelihood, due to the fact that higher profits may be achieved at the cost of higher risk. However, if profitability is achieved through a diversified portfolio, it may not necessarily be positively related to risk and banks with more diversified sources of income are found to be less risky (Ellul and Yerramilli, 2010). No expectations are posed about this control variable.

Capitalization is measured by the ratio book value of equity/total assets (CAP). Well capitalized banks are less vulnerable to economic shocks (Kato & Hagendorff, 2010), since equity acts as a cushion against asset malfunction. Thus, the higher this ratio is the more protected the bank is. Consequently, it is expected that capitalization is positively related to Z-score.

Differences on country level are also considered and the country of origin of each bank is also added as a control variable. It is deemed appropriate since, different investor protection laws across countries may have influence on the agency relationship within a bank and consecutively on the risk appetite. Furthermore, differences between regulations and market specificities across the European Union still exist and should be taken into account. 3.3 Econometric model and specification

(26)

26 Griffiths & Lim, 2008). The following section discusses the specification of the models used for testing the researched hypotheses.

A non-probability purposive sampling selection is used (Thomas, 2006; Teddlie & Tashakkori, 2009). Still, the cross section observations can be viewed as independent, identically distributed draws from the population since the number of observations is much larger compared to the number of periods (Wooldridge, 2010). The use of panel data is preferred, because it provides the possibility to measure more precisely the relationship between the introduced variables, since it controls for heterogeneity, i.e. factors that are not measurable or directly observable (Baltagi, 2008). In this case, such factors are bank specific characteristics with affect the risk taking orientation of the bank and cannot be caught through accounting data. Furthermore, panel data gives a possibility to account for time-variant factors, which change in time, but stay constant across entities (Baltagi, 2008), such as changes in market conditions or regulatory policies at the European level.

When considering the possible estimation techniques, it has to be taken into account that using Pooled Ordinary Least Squares (POLS) is possible, but it is very likely to produce coefficients, which suffer from an omitted variable bias. There are two different estimators associated with panel data - fixed effects (FE) and random effects (RE), which control for different types of omitted variables by observing changes in the dependent variable over time. The use of fixed effects gives the opportunity to analyze the impact of variance in subordinated debt holdings on bank risk within each bank entity. It is reasonable to assume that each bank has its own unobservable specific characteristics which may influence the independent variables in the model. Compared to applying POLS, fixed effects provide a more accurate picture on the net effect of SND_Ratio on LNZ, taking out those bank-specific characteristics, which are constant over time.

(27)

27 fixed and between effects. An assumption of the RE estimator is that the unobserved omitted variables (differences in country regulations, legislation and competitiveness in the local banking industry, for example) should not be correlated with the explanatory variables applied in the model, otherwise the estimated coefficients from the RE can be inconsistent. This assumption is checked for with a Hausman test (Hill et al., 2008). If it is violated the FE estimator is still unbiased. In case the assumption holds, this implies that the coefficients estimated for time-variant variables with FE and RE do not exhibit systematic differences, but the RE estimator is more efficient.

Results from a Hausman test11 provide support for the preference of random over fixed effects estimation (see Appendix D, Table 1). The χ² statistic for the test is 13.34, with a p-value of 0.04, thus the hypothesis that the difference in coefficients estimated with FE and RE is not systematic can’t be rejected at 1% significance level. But this is not the case at a 5% level. Since it is likely that the rather high χ² is due to the loss of statistical power when applying the smaller sample, the same test is performed on the initial sample with 1390 observations as a robustness check. The value of χ² is 0.95, which is can’t be rejected at 5%, confirming the conclusion that the difference in coefficients is not systematic. Moreover, the applicability of the RE estimator is checked with a Breusch-Pagan/Lagrange-Multiplier test. It rejects the Ho, that Pooled Ordinary Least Squares (POLS) should be applied, at 1% significance (see Appendix D, Table 1), which implies that differences exist across entities.

Additionally, the period under investigation matches with the recent financial crisis. It is plausible that the economic conditions in each of those years will have a systematic effect on the banks’ Z-scores. In order to account for this and not distort the results for the relationship between risk and subordinated debt holdings, time-fixed effects are added. The need for inclusion of those effects is also tested and found significant (see Appendix D, Table 1).

If H1 holds a certain subordinated debt ratio at a particular moment within a given bank would correspond to a certain level of risk. This is considered valid since bank deposits have absolute priority over other financial liabilities, which should increase the urgency with which subordinated bondholders feel the results of adverse changes in asset value (Bliss & Flannery, 2002). Thus, the effect of subordinated debt holdings on risk should be visible shortly after changes on the right hand side of the balance sheet of the bank. This justifies the

11_{Provides a comparison between the estimated coefficients with FE and RE and checks whether the unique}

(28)

28 use of accounting-based measures of risk at year-ends and suggests that there should be no need for any time lags to be applied. Such line of reasoning also helps to disregard reverse causality and possible feedback relationships as a potential problem. In order to test H1, the following model is specified:

(1): LNZit = β1*SND_Ratioit + β2*LNTAit + β3*CAPit + β4*NETIRMit +

β5*LISTEDit +Yt + COUNTRYi + α + uit + εit

where α is the intercept parameter, uit accounts for between-entity error and εit for within-entity error. Yt represent dummy variables for the years 2006-2009 and COUNTRYi represents the country of origin of the bank.

Furthermore, in case of presence of a significant relationship between SND_Ratio and LNZ, it is hypothesized (H2) that this relationship is possibly non-linear and could be graphed as an inverted U-shaped curve (equivalent to a U-shaped curve for the subordinated debt holdings – risk taking relationship accordingly). This implies that very low and very high subordinated debt holdings at the bank correspond to a low Z-score, less stability and aggravated risk taking. Simultaneously, the positive effects of market discipline offset the negative leverage effects when subordinated debt holdings are sufficient for market discipline to occur. When residuals appear to trend down at both high and low predicted values curvilinear regression12 can be an appropriate fit, thus the variable SND_Ratio_sq is added to model (1). The inverted U-shaped relationship will be present, if SND_Ratio yields a significant positive coefficient, while simultaneously SND_Ratio_sq a significant negative one. In order to test H2, a second extended model is specified as:

(2): LNZit = β1*SND_Ratioit + β2*SND_Ratio_sqit + β3*LNTAit + β4*CAPit +

β5*NETIRMit + β6*LISTEDit + Yt+ COUNTRYi + α + uit + εit

A third extended model is built to test H3 and H4. A favorable assumption of random effects is that entity errors (ui) are not correlated with the regression errors (εit) in the model. Thus, it is possible to include time-invariant variables as explanatory variables in the model. Since it is expected that bank specialization and ownership concentration affect risk taking,

12_{It refers to linear OLS regressions that include non-linear transformations of the original y or x variables. It fits}

(29)

29 the dummy variables SAVE, COOP, INDEP and SMBLOCK are added to measure their direct effect on LNZ. In order to test their moderating effect on the SND_Ratio – LNZ relationship, interactions are added. The aim of introducing interaction terms is to test whether there is any combined effect of two predictors on the dependent variable. If an interaction effect exists, the impact of one of the predictors depends on the level (state, in the case of categorical variables) of the other one. In this case not only the intercept, but also the slope of the relationship differs per subgroup. Consequently, the following model is specified:

(3): LNZit = β1*SND_Ratioit + β2*LNTAit + β3*CAPit + β4*NETIRMit + β5*LISTEDit +

β6*Range1it + β6*COOPi + β7*SAVEi + β8*INDEPi + β9*SMBLOCKi + β10*SND_Ratioit* COOPi + β11*SND_Ratioit* SAVEi + β12*SND_Ratioit* INDEPi +

β13*SND_Ratioit* SMBLOCKi + Yt + + COUNTRYi + α + uit + εit

4. Results

4.1 Descriptive statistics

Table 2 in Appendix B provides summary statistics for all variables. The measure of bank risk is score, which is inversely related to the probability of default. The calculated Z-score variable has a mean of 35.57, and a quite high standard deviation of 43.05. The lowest Z-score value is 0.19 and the highest 445.02, which signals for the heterogeneous risk profiles of the banks included in the sample. Other researchers report similar average Z-score values. Berger et al (2008) estimated a mean Z-score of 56.76, Turk Ariss (2010) – 30.15, Barry et al (2009) – 46.7. Differences in the averages are mainly due to sampling banks from different countries or regions. As discussed previously, LNZ is applied in regressions in order to account for the skewness of Z.

Furthermore, summary statistics divided by specialization, ownership concentration and country are provided in the Appendix B (see Tables 3, 4 and 6). Cooperative bank have the highest mean Z-score of 72.01 and are on average better capitalized with a mean capital ratio of 9.22, while savings banks have only a slightly higher mean Z-score (36.02) and lower capital ratio (7.02) as compared to the entire sample. Banks with dispersed ownership exert a mean Z-score of 52.57 and those with small block holders only 32.53. When looking at the sample divided by country, it is interesting to note that the mean Z-scores for Ireland, Greece, Belgium and Denmark are the lowest.

(30)

30 in the Z-score can be detected for 2008 and 2009, as well as a higher standard deviation from the mean. Simultaneously, in 2008 and 2009 a higher SND_Ratio corresponds to a lower Z-score compared to 2007. Another interesting observation is that the SND_Ratio decreased during the years of the financial crisis and increased again afterwards.

4.2 The relationship between subordinates debt and risk

As it was mentioned before the nature of this study is rather exploratory, since no previous research provides indications for the relationship between SND holdings and risk. Thus, stepwise results for model (1) are reported in Table 1 of Appendix C. As a first step LNZ is regressed solely against SND_Ratio (POLS1) using Ordinary Least Squares estimator. In the next regressions, the bank-specific control variables (POLS2 in Table 1, Appendix C) country and year dummies (POLS3 in Table 1, Appendix C are added.13 Further, both fixed and random effects estimators are used (FE1, RE1, FE2, RE2 and RE3). FE2 and RE2 include year dummies, while RE3 also includes country dummies. Due to the presence of heteroskedasticity robust standard errors are reported in all estimations.

There is a significant negative relationship between the subordinated debt ratio and bank’s Z-score, which holds across all eight regressions. The coefficient of SND_Ratio is negative and significant at either 1% or 5% in all regressions, implying that the relationship is robust to adding controls, time and country effects. When adding the bank-specific controls, country and year dummies (POLS2 and POLS3 in Table 1, Appendix C), the coefficient diminishes slightly (from -0.149 to -0.140 to -0.115) but it stays negative and significant at 1%. Further, the negative relationship persists across FE and RE regressions at 5% significance. Although both FE and RE yield consistent results, in the specification section it has been concluded that RE is the more efficient estimator. Thus, looking at the estimates from the RE regression which includes all relevant variables and accounts for country and year effects (RE3 in Table 1) it can be inferred that 1 unit change in SND_Ratio is associated with 4.1% decrease in Z-score, all other factors held constant. The results from applying model (1) prove a significant negative relationship between LNZ and SND_Ratio at 1%. This implies that H1 is confirmed.

Furthermore, when adding the controls R-square increases in the POLS regressions (from 0.03 to 0.12 to 0.32), which means that the overall fit of the model increases. However,

13_{Other control variables were also considered (such as income diversity and asset quality), but they did not}

(31)

31 the coefficients estimated using POLS are overstated since POLS estimators partially attribute unobserved effects to the SND_Ratio coefficient. Next, looking at both FE1 and RE1 regressions the significant negative relationship persists, although with a lower coefficient. This is due to the fact that those estimators account for changes in LNZ within banks (for FE) and also across banks (for RE), which allows for the net effect of SND_Ratio to be captured. Moreover, R² increases further in the FE1 and RE1 regressions, which implies an improved model fit. For the FE regressions the relevant R² is the one that reports the variance within each bank (0.44), while for RE both within (0.44) and between (0.09) R² are meaningful measures of fit. When adding time effects, the R²Within improves from 0.44 to 0.51 and the overall R² improves from 0.9 to 0.10. Finally, with introduction of country dummies in RE3, the model fit improves further and standard errors decrease for all estimated coefficients.

Turning to the control variables, LNTA, CAP and NETIRM have positive and significant coefficients in RE3. This implies that larger banks, more profitable banks and especially better capitalized banks are less risky. The relationship between capitalization and Z-score is quite robust among all regressions (positive sign and significant at 1%), which is in line with the expectations. However, LNTA has a significant coefficient only in RE2 and RE3, which is in line with results from previous studies that find ambiguous effects of size on bank risk. Berger et al (2008) and Turk Ariss (2010) also find a positive relationship, while Boyd and De Nicolò (2005) and Anderson and Fraser (2000) find a negative one. The coefficient of NETIRM is positive in all regressions, but it is significant only at 10% in RE1 and at 5% in FE1, FE2, thus no firm conclusions about it are made. The variable LISTED is not significant and its coefficient changes from positive to negative when year dummy variables are added in FE2 and RE2, which is not in line with the expectations posed previously. This can possibly be explained by its twofold effect on risk.

(32)

32 was hypothesized that if subordinated debt holdings at the bank are sufficient14 for market discipline to be present, an inverted U-shaped curve would fit the data better than a straight line with a negative slope, since it will outweigh the effects of leverage due to subordinated debt. 15 As per the stated expectations, when adding the SND_Ratio_sq variable, it should enter the regression with a negative coefficient and the SND_Ratio coefficient should change to positive (as in an inverted quadratic function).

When looking at the results from RE4 presented in Table 2, it is evident that both SND_Ratio and SND_Ratio_sq enter the regression with a negative sign. The coefficient for SND_Ratio is insignificant, while the coefficient for SND_Ratio_sq is significant only at 10%. When tested for the joint significance of the SND_Ratio and SND_Ratio_sq coefficients an F-statistic of 5.77 is obtained, which is insignificant at 5%. Furthermore, there is no change in R² when SND_Ratio_sq is added, implying that the model does not fit the data better. It can be concluded that there is no evidence for non-linearity of relationship between SND_Ratio and LNZ and the inclusion of the quadratic term is redundant. Accordingly, H2 cannot be confirmed.

4.4 Effects of bank specialization and ownership concentration

Model (3) is used to test H3 and H4. First, results for the direct effect of specialization and ownership concentration on risk are presented in Table 3. Results for the moderating effects are presented in Table 4. Since RE estimator has been identified as consistent and the most efficient one, again only results using RE are presented here. Here RE3 also serves as a means for comparison. In RE5 only the dummy variables for ownership concentration – INDEP and SMBLOCK are added. RE6 reports the results when only bank specialization is considered and RE7 includes dummies for both specialization and ownership concentration. Again, due to the presence of heteroskedasticity robust standard errors are reported in all regressions.

By looking at the R² first, it can be concluded that adding the two categories of dummies slightly increases the explanatory power of the model. Bank ownership concentration and specialization does not change over time for the sample banks, which obviously results in a constant R²Within. When adding INDEP and SMBLOCK R²Between

14_{The mandatory subordinated debt proposals opt for different optimal subordinated debt holdings and range}

from 2-4% of total deposits or liabilities, 1-4% of risk-weighted assets (Sironi, 2001), 1-4% of total assets (U.S. Shadow Financial Regulatory Committee, 2000).

15_{Other possible non-linear functions were also considered (such as 1/SND_Ratio, square root of SND_Ratio,}