The repeal of the Glass-Steagall Act and its contribution to the 2007-08 Global Financial Crisis

(1)

UNIVERSITY OF AMSTERDAM AMSTERDAM BUSINESS SCHOOL

The repeal of the Glass-Steagall Act and

its contribution to the 2007-08 Global

Financial Crisis

Master Thesis

MSc Business Economics: Finance

Joanna Ropska

07/2015

Thesis Supervisor: Professor Torsten Jochem

Abstract

The 2007-08 Global Financial Crisis has spurred a fierce debate regarding the role that the repeal of the Glass-Steagall Act has played in destabilizing the financial sector. Arguments have been presented claiming that the repeal has led to commercial banks entering securities activities and this way pursuing excessive risk. This study addresses this issue by performing an empirical analysis investigating the relationship between investment banking risk, estimated over the period 2000-2006, and Crisis performance, for a sample of Bank Holding Companies and Financial Holding Companies. The results indicate that investment banking risk does not explain more detrimental Crisis performance. These findings imply that the contribution, of the Glass-Steagall Act’s repeal, to the Crisis, cannot be attributed to the commonly discussed risk-taking mechanism.

(2)

Statement of Originality

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

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

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

(3)

In the light of the 2007-08 Global Financial Crisis, substantial discussion has been extended regarding the role that large and complex financial institutions [LCFIs] played in destabilizing the financial system. The central argument in this debate has been the, allegedly, excessively risky and hazardous behaviour by the LCFIs.

Those discussions have ultimately led to the topic of the Glass-Steagall Act‟s repeal (dictated by the Gramm-Leach-Bliley Act). Established after the Great Depression, in 1933, the Glass-Steagall Act provided for a separation of commercial and investment banking. Such distinction was, at the time, believed to enhance the stability of the financial system, and provide a solution to the agency problems (Crawford, 2011). The Act‟s repeal, in 1999, allowed for the creation of universal (non-traditional) banks [Financial Holding Companies]. This implied that commercial banks could enter the investment banking business and investment banks could affiliate with commercial banks, this way, gaining access to deposits. While the first mechanism could have led to improperly managed risk, the second one could have meant a gamble with depositors‟ money. Both could imply an increase in the systematic as well as specific risk.

This paper will make an attempt at answering the question of whether the Glass-Steagall Act‟s repeal contributed to the 2007-08 Global Financial Crisis. The analysis will treat the side of commercial banks entering investment banking. To the author‟s best knowledge, no such empirical analysis has been done which explains bank performance, during the Crisis, with the investment banking risk estimated over the period of 2000-2006.

The issue investigated is particularly relevant, as several contradicting opinions have been expressed by prominent figures. Some argue that the Act‟s repeal is to be blamed for the Crisis (Stiglitz, 2009) while others claim that the repeal prevented an even deeper economic downturn (Blinder, 2010). Moves such as the establishment of the Dodd–Frank Wall Street Reform and Consumer Protection Act also seem to indicate that the spirit of Glass-Steagall is followed. An empirical analysis of this issue would, thus, undoubtedly, be of great relevance for both policy makers as well the public. For the former, it would provide a quantitative basis for the proposed regulations. For the latter, it would serve as guidance for the governmental actions as well as the behaviour of the financial sector.

In order to answer the research question, two hypotheses require testing. First, one needs to investigate whether entry into investment banking, by commercial banks, has significantly and detrimentally affected the banks‟ risk profile. Had this been the case, the first condition (risk creation) for the Act repeal‟s contribution to the Crisis would likely be fulfilled. The second hypothesis treats the risk and its reflection during the Crisis – one needs to analyse if

(5)

the investment banking risk, of commercial banks, was an explanatory element of the banks‟ worse Crisis performance. A confirmation would imply that, indeed, the granting of investment banking powers, to commercial banks, by the Gramm-Leach-Bliley Act, significantly influenced their performance during the downturn. This is not to say, however, that Gramm-Leach-Bliley caused the Crisis – for such an assertion one would need to look at the magnitude of the studied effect.

The method for testing those two hypotheses is the following. The starting sample is composed of 251 holding companies of banks – Bank Holding Companies [non-FHCs] and Bank Holding Companies, which, in 2000, formed Financial Holding Companies [FHCs]. Their financial data over the years of 2000-2006 and during the Crisis (December 3rd, 2007 to June 30th, 2009) are used.1 The first hypothesis is tested with a panel regression. Namely, a measure of risk profile (ROA volatility, Z-score) is regressed on the investment banking activity (share of non-interest income to net interest income) and controls. From this regression, investment banking risk is predicted, which is then used in a cross-section of Crisis returns, as the main regressor.

The findings suggest that investment banking activity has been influencing commercial bank risk profile significantly and negatively – more investment banking is associated with more risk. The association holds when using two additional transformations of non-interest income. The investment banking risk, however, is not significant when explaining more detrimental Crisis performance. The result is robust for a different Crisis period specification and an alternative measure of investment banking (trading income). This is line with the raw data where one can observe FHCs having lower risk and greater returns than non-FHCs. This pattern seems to suggest that the argument of commercial banks engaging in investment banking and, this way, pursuing excessive risk does not hold. Alternatively, the results could be driven by the lack of informativeness, of the estimated relationship, (risk-investment banking) in terms of predicting the detrimental performance during the Crisis.

In the context of the current literature, the conclusion of investment banking‟s detrimental effect on risk is in line with numerous other findings. The second part of the empirical analysis, on the other hand, is rather novel. Hence, the scope for reference is limited. The most related study is the one by Brunnermeier et al. (2012) who find a negative effect of non-interest income on recession returns one year later. The critical, risk mechanism, however, is not treated in this analysis. This element is, thus, the primary contribution of the current study

1_{The chosen period is the 18-month contraction period as reported by the NBER's Business Cycle Dating} Committee.

(6)

– by investigating the effect of investment banking risk on performance it directly responds to the commonly extended criticism of Gramm-Leach-Bliley leading to excessive risk-taking, via securities activities, of commercial banks, which, in turn, resulted in a market crash. No empirical support is found for this claim.

The paper is structured as follows. First, the Glass-Steagall Act, the Gramm-Leach-Bliley Act and their theoretical relation to the Crisis are introduced. Second, an overview of the relevant empirical studies is provided. Third, the research approach is explained. Fourth, descriptive statistics is analysed. Fifth, empirical results are presented. Sixth, the results‟ robustness is assessed. Seventh, conclusions are drawn.

1. Introduction

1.1 The Glass-Steagall Act

The Glass-Steagall Act was a part of the Banking Act of 1933, which was established as a response to the Great Depression and applied to all national and state-chartered banks belonging to the Federal Reserve System (Crawford, 2011 and Litan, 1988). The Glass-Steagall Act referred to four sections (16, 20, 21, 32) of the Banking Act. Namely, those centred on the separation of commercial and investment banking. Sections 16 and 20 prevented commercial banks from underwriting corporate debt and equity securities. Section 21 forbid an entity, performing underwriting, from taking deposits. Section 32 made any kind of employee/direct interlocking between a bank and an underwriter illegal (Litan, 1988). One of the primary reasons for establishing a wall between commercial and investment banking was conflict of interests (Brown, 1995). It originated from the fact that commercial banks perform loan-granting activities. Those, in turn, provide the bank with valuable information, which is unknown to external investors or investment banks. The possession of the private information can either facilitate a certification role or give rise to conflict of interests – banks can protect their own interest, at the expense of investors, by underwriting risky securities, when the proceeds are used to service a loan (Puri, 1996).

Another reason for the separation were concerns of the instability of the financial system, which resulted in a market crash in 1920s. It was argued that the use of deposits for speculative and reckless purposes, within the securities business, was contributing to the equity markets‟ volatility, which, eventually, resulted in a market crash (Securities Industry Association as cited in Markham, 2009). Hence, it seems that a belief was held that a wall

(7)

between commercial and securities business would contribute to a more stable financial system.

While the Act stayed valid until 1999, in 1987 a significant relaxation of its provisions was introduced. Namely, the Federal Reserve Board allowed for establishment of investment banking subsidiaries, of commercial banks, under Section 20. Such operation was deemed legal as long as the subsidiary did not generate more than 5% of the revenues (in 1989 this was extended to 10% and in 1996 to 25%). This freedom was further enhanced in 1997 when the Federal Reserve allowed for a direct acquisition of an investment bank as opposed to establishing a new subsidiary (Cornett et al., 2002). Hence, one can observe that by the time the Glass-Steagall Act was repealed, a significant expansion of commercial banks‟ investment activities freedom was already introduced.

1.2 The Repeal and the 2007-08 Global Financial Crisis – theoretical framework

In 1999, the Glass-Steagall Act was repealed with the establishment of the Gramm-Leach-Bliley Act [GLBA], (or: The Financial Services Modernization Act). With the repeal, banks, insurance companies and securities firms were allowed to merge under the umbrella of a holding company (Financial Holding Company [FHC]) (Geyfman and Yeager, 2009 and Carnell et al. 2009 as cited in Wilmarth, 2009). An establishment of a FHC implied that an entity would become a universal bank and, hence, engage in commercial as well as investment banking. As demonstrated in figure 1, investment banking activities (as measured with non-interest income), performed by commercial banks, increased significantly around this period. Similarly, the share of non-interest income to net interest income experienced a drastic increase (figure 2).

Figure 1. Development of non-interest income [ths], for the sample of FHCs and non-FHCs, over the years 1990-2006 [Source: Compustat].

0 20000 40000 60000 80000 100000 120000 140000 160000 fhc non-fhc

(8)

Figure 2. Share of non-interest income to interest income, for the sample of FHCs and non-FHCs, over the years 1990-2006 [source: Compustat]

There is some speculation that the repeal came about due to the pressure exercised by the financial industry. Some argue that by the time it was introduced, Glass-Steagall was virtually dead as a result of the judicial interpretation (Cornett et al., 2002). An example for this claim is the allowed merger of the Citicorp and Travelers Insurance Company in 1998 (Crawford, 2011).

The repeal was a result of a substantial discussion in which the following for and against arguments were laid down, in a Congressional Research Service Report (Jackson, 1987). The arguments against the repeal included the previously mentioned conflict of interests inherent in the granting and use of credit. Furthermore, it was argued that due to the substantial power that depository institutions have, when it comes to managing other peoples‟ money, guaranteeing their soundness and competition is critical. This argument was related to the presumption that securities activities are by nature risky and have the potential of provoking significant losses. Since this threatens the deposits, which the government insures, it could lead to a stretching of the safety net. Finally, a concern was presented that risk management might not be exercised properly in the speculative securities activities. This was supported with the example of a collapse of the real estate trust funds, which were sponsored by bank holding companies a decade before. Thus, one can observe that risk shifting was a dominant concept behind the considered arguments. Moreover, it seems that investment banking, at the time of the Glass-Steagall‟s repeal, was understood as narrowly as the securities activity. This definition will also be applied in this study.

The arguments against the repeal were weighted with those for it. To start with, it was believed that the conflict of interests could be managed with proper legislation and with the

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 fhc non-fhc

(9)

establishment of separate securities subsidiaries. Furthermore, there was a discussion regarding the competitiveness of depository institutions against securities firms and foreign institutions – an observation was made that depository institutions were losing market share. This argument was further substantiated with a claim that the line between loans, securities and deposits was already becoming unclear. It was also assumed that securities activities, which depository institutions were interested in, were low in risk and could produce diversification benefits, this way reducing the overall risk. Finally, it was claimed that, in other countries, depository institutions were successfully engaging in investment banking. From the analysed arguments, for and against the Glass-Steagall Act‟s repeal, we are mostly interested in the arguments against. One can narrow those down to three following concepts – conflict of interests, risk-taking (shifting), rise of too-big-to-fail [TBTFS] institutions. Those

are precisely the concepts which were at the centre of the public debate revolving around the 2007-08 Global Financial Crisis. While the rise of the TBTFS is possible to be identified,

quantifying conflict of interests (and excessive risk-taking) is more difficult. In the following discussion, the repeal of the Glass-Steagall Act will be referred to as GLBA.

Having described the Glass-Steagall Act‟s provisions, and concerns around its repeal, one could ask how the GLBA could be related to the Crisis. With this respect, the omnipresent concept is the excessive risk-taking, by commercial bank, after they started performing investment banking activities. While investment banking involves multiples activities, the focus, in this discussion, lies with the securities business – this is the most risky element of investment banking and also the factor, which lied at the heart of the Crisis. The primary mechanism, during the Crisis, was securitization. Combined with excessive expansion in the subprime mortgage market, it resulted in massive losses when defaults started (Markham, 2009). The banks, which were exposed to the securitized instruments the most, were the primary losers (White, 2009). Through the network of spillovers, the losses were transmitted to other institutions and the real economy (Bordo, 2008).

Before the GLBA, banks could sell their loans to entities, which would then securitize and sell them further (Markham, 2009). The banks could not, however, trade and securitize themselves. After the GLBA‟s introduction, commercial banks, which formed FHCs, could freely engage in the securities business. This implied buying assets, often with borrowed money, and, through securitization, selling them at the times of high prices. In situations of bad times, a bank would need to liquidate its portfolio and sell at prices lower than the fundamentals – such mechanism destabilized prices. Banks performing such operations, naturally, accepted the risk the less at stake they had (Shleifer and Vishny, 2010). The

(10)

volatility of those instruments provided for greater risk. Since banks kept some of the residual exposure on their balance sheets (e.g. some tranches for signaling effects), the risk was not entirely shifted (Markham, 2009). Due to the existence of the “too big to fail subsidy”, bank managers could pursue higher returns and disregard this implied risk (Stiroh and Rumble, p.29, 2006). In addition, other investment banking activities, such as advisory, which required an even greater operational and financial leverage, exposed the banks to business cyclicality, which generated an additional volatile position on the balance sheet (Lepetit et al., 2008). The described mechanism was largely at work, during the Crisis, and has resulted in substantial losses, as depicted in figure 3. Income diversification was unlikely to provide diversification benefits –interest and non-interest income were highly correlated.2

Figure 3. Development of trading income [ths], for a sample of FHCs and non-FHCs, over the years 2000-2014 [Source: Compustat]

The publicly expressed criticism, arguments at the time of the GLBA‟s introduction and the securitization mechanism explained above lead to a specification of the hypothesis, which this study will investigate. Namely, the proposition is that commercial banks, which entered investment banking, have experienced a risk creation that has led to their detrimental performance, during the Crisis, and, perhaps, has substantially destabilized the entire financial system – hypothesis A.

2_{According to the author‟s own calculations, in the period 2007-2009, the correlation coefficient between net} interest income and non-interest income amounted to 0.94. The calculation is based on Compustat data.

0 10 20 30 40 50 60 70 80 90 100 -20000 -15000 -10000 -5000 0 5000 10000 15000 20000 25000 30000 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 fhc [left scale] non-fhc [right scale]

(11)

2. Literature review

The literature relevant in this paper can be classified under two headings – the risk implications of GLBA, and the effects of bank income diversification on financial performance. While the first one conveys information about the change in bank risk profile, specifically after the GLBA, the second one studies the concept of the risk effects of bank expansion into non-traditional banking. Thus, its relevance lies in explicitly addressing the relationship between investment banking and risk. This is contrary to the first group of papers where the mechanism, through which the GLBA is argued to have influenced the risk profile, is not treated. Table 1 contains a summary of the empirical studies referred to, in this paper.

(12)

Table 1 – Summary of Empirical Research. Risk and performance effects of the Gramm-Leach-Bliley Act

Cornett et al. (2002) 40 publicly traded commercial banks

04/1987 – 12/1997 Ratio of pre-tax cash flows ROA to year-end book value of assets

Overall improvement in performance and no effect on risk is found.

Mamun et al. (2005) 343 banks 01/1998 – 12/2000 Stock returns, systematic risk (market, foreign exchange, interest rate)

Most welfare effects were accrued by Money Centers followed by Super Regional Banks. In an alternative banking system categorization, banks with Section 20 subsidiary gained most. In case of both groups, significant reduction in systemic risk is found. Boyd et al. (1993) 1000 synthetic

banks

1971 – 1987 Z-score Commercial banks, entering non-banking business, can experience a risk reduction when merging with insurance firms. The effect is opposite in case of mergers with securities and real estate firms. Allen and Jagtiani

(2000)

729 synthetic banks

1986 – 1994 Monthly return volatility Entry into non-banking activities yields a reduction in the specific risk and an increase in the market risk.

Narayanan et al. (2002)

253 institutions 01/02/ 1996 – 31/12/1997 Stock price, change in daily return volatility

Welfare effects are positive for BHCs for two events. Risk effects (unsystematic), on the other hand, show a detrimental influence for all groups (banks with and without Section 20 subsidiary, their competitors, their customers).

Akhigbe and Whyte (2004)

406 institutions 2/07/1999-15/11/1999 300-day event window

Market-based total, systematic, unsystematic risk

Following the GLBA, an increase in total and unsystematic risk for banks and insurance firms is found. The opposite is found for securities firms. Systematic risk declines across all types of banks.

Geyfman and Yeager (2009)

777 banks 1990 – 2007 Weekly return volatility Engagement in investment banking is positively correlated to total and unsystematic risk.

(13)

Risk effects of non-traditional banking

Stiroh (2004) >14,000 banks 1970s – 2002 Z-score, Sharpe ratio Banks with more non-interest income experience more volatility and less risk-adjusted returns.

DeYoung and Rice (2004)

4, 712 Commercial banks

1998 – 2002 Return on equity, return on equity volatility, Sharpe ratio

Non-interest income increases with bank size, stress on customer relationship, some technological advances. It decreases for well-managed banks.

Non-interest income is positively correlated with return on equity and return on equity‟s volatility and negatively with Sharpe Ratio. Stiroh and Rumble

(2006)

1, 816 FHCs 1997 – 2002 Risk-adjusted ROE/ROA, Z-score

Across banks, income diversification produces benefits, which are offset by greater earnings volatility.

Within banks, no diversification benefits exist. More non-interest income is correlated with worse risk-adjusted performance.

Brunnermeier et al. (2012)

538 Commercial banks

1986 – 2008 CoVar, SES Banks with more non-interest income contribute more to systemic risk.

Banks with more trading income, one year prior to recession, exhibit worse performance during the recession.

Lepetit et al. (2008) 734 European banks

1996 – 2002 ROA, ROE, weekly return volatility, beta, specific risk, loan loss provisions to net loans, Z-score, ZP-score, distance to default DD

More non-interest income is positively correlated with risk and insolvency. For small listed banks, more trading income can lead to less risk.

Köhler (2014) 16 760 German banks

2002 – 2010 Z-score, risk-adjusted ROA/ROE, leverage risk

For retail-oriented banks, more non-interest income implies less risk. The opposite is true for investment-oriented banks. Only fees and commissions part, of non-interest income, has a significant effect.

(14)

2.1 Gramm-Leach Bliley Act and bank risk profile

The question of the GLBA‟s consequences for banks has been given some attention in empirical research.3 Papers investigate the effects of the repeal both with simulations as well as with actual data. The authors have looked at consequences in terms of risk-taking, and performance and wealth effects. Moreover, some studies have focused purely on a limited degree of the Glass-Steagall Act provisions‟ relaxation (establishment of a Section 20 investment banking subsidiary). This literature overview will start by discussing the research into the effects of establishing a Section 20 subsidiary. The subsequent parts will be organized along the lines of the research methodologies used.

2.1.1 Section 20 subsidiary

Cornett et al. (2002) focus on the effects of establishing a Section 20 subsidiary, by BHCs, on performance. The interest lies in the change, of the ratio of return on assets to year-end book value of assets, from pre [-3, -1] to post [+1, +3] subsidiary establishment. The change in the performance indicator is compared to a matched sample. Overall, an improvement in performance is found. For the risk effect, the volatility of daily return on equity is used [-1,+1 year]. No significant change in total and systematic risk is found. Since the paper considers risk effects of engaging, by commercial banks, in investment banking on a small scale (up to 10% of revenues), it can only provide an intuition of what could be expected from a full relaxation of investment banking constraints. It is not certain whether, on a full scale, the effect would translate in the exact same way.

Similarly to Cornett et al. (2002), Mamun et al. (2005) investigate the welfare and risk effects of the establishment of the GLBA. The authors compare two portfolios – Money Centers, Super Regional Banks and other banks versus a portfolio with banks with a Section 20 subsidiary, prior-GLBA, and those that had one post-GLBA but not prior. In order to calculate the portfolio returns, a market model with a dummy variable for the event date and lagged values of the market index is used. Risk is classified into market, foreign exchange and interest rate. In case of the first portfolio specification, the results indicate that most welfare effects were accrued by Money Centers followed by Super Regional Banks. In terms of systematic risk, a reduction is found. In case of the second portfolio specification, the greatest welfare effects are found for banks with a Section 20 subsidiary prior-GLBA. Similarly to the previous group, a reduction in systematic risk is found.

(15)

2.1.2 Simulations/ Synthetic Banks

Boyd et al. (1993) estimate the risk effects by simulating a merger between a BHC and a firm active in the non-banking business. The authors then calculate the risk measures of those merged firms and compare them with those BHCs, which did not merge (actual).4 The

employed risk measure is the Z-score (calculated with both market and accounting data). The findings are that mergers with insurance companies can decrease risk, which is not the case for mergers with securities firms or real estate firms.

Similarly, Allen & Jagtiani (2000) create synthetic banks in order to analyse the GLBA risk effects. In order to create a synthetic bank, the authors produce a portfolio with one depository institution, one securities firm and one insurance firm. Risk is defined as the volatility of monthly returns – value-weighted average monthly return of one institution from each of the company types (as defined above), in a given portfolio. The authors find that non-banking activity produces a reduction in the overall bank risk but an increase in the overall market risk.

2.1.3 Effects using an event study

Narayanan et al. (2002) analyse the welfare effects for banks (with and without Section 20 subsidiary), their customers and competitors, from expanding the activities of BHCs to the securities business. The authors perform an event study on three occurrences5 in the period preceding the GLBA. The interest lies with the price behaviour and risk. The risk effects are computed as a change in variance of the daily return, of a portfolio, between pre- and post-event periods. The results indicate that all portfolios experience a small but statistically significant increase in unsystematic risk. The welfare effects show a positive effect for the BHCs, at the expense of the customers and the competitors, in case of two out of three events.

2.1.4 Risk effects

Akhigbe & Whyte (2004) analyse the risk effects for commercial, insurance and securities firms. Within banks, a distinction is made between those with Section 20 subsidiary and those already with a status of a FHC. The authors assess the changes in total, systematic and

4

The authors use historical data where they randomly combine one BHC with one non-banking firm. In order to achieve a desired ratio, for the first post-merger year, of nonbanking to consolidated assets, the data of the nonbanking firm is scaled. The scaling factor is calculated based on a variable N (0≤N˂1), which implies the portfolio weight. The resulting scaling factor is where Ab are the assets of the BHC and An are assets of the nonbanking firm. The ratio is determined for the first post-merger year and can subsequently vary. This way a predetermined initial portfolio weight is produced. Subsequently, the data of the BHC and the scaled nonbanking firm are summed – this way data for the hypothetical firm is obtained.

5

Expansion of Section 20 subsidiary‟s revenue limits (from 10% to 25%), relaxation of restrictions (information, resource and funding interchange) between a bank and its security affiliate, announcement of a willingness for a future acquisition of Alex Brown & Sons by Bankers Trust Corporation.

(16)

unsystematic risk by investigating the change in volatility of returns for the pre- and post-event 300-day window. The pre-post-event period is the period before the first announcement indicating that new regulation will be introduced. The post-event period follows the signing of the GLBA into law. The results imply an increase in total risk for all banks (regardless of the presence of a Section 20 subsidiary or a conversion to a FHC) and insurance firms. The opposite is true for securities firms. This pattern is followed in the case of unsystematic risk. Systematic risk, on the other hand, exhibits a decline across all types of institutions.

Geyfman & Yeager (2009) also use the period around the GLBA‟s enactment in order to study the market-based risk differences between universal (commercial & investment banking) as opposed to traditional banks. Applying the Heckman two-stage regression model, the authors investigate the effect of being a universal bank (transformation into a FHC, ratio of Section 20 subsidiary‟s assets to BHC assets, ratio of investment banking income to net operating revenue) on risk. The authors find that more engagement in investment banking activities leads to more total and unsystematic risk. No significant change in systematic risk is found. Importantly, in the post repeal period, the authors find a reduction in systematic risk. They explain this result, however, with a sample period rather than fundamentals. The current literature has found diverging results regarding the component of total risk which has been subject to change after the GLBA. It does, however, agree on the existence of a pattern, whichever component changes, the risk increases. In the context of this paper, the most informative finding is the one by Geyfman and Yeager (2009) where an account, of the change in the bank risk profile, is made specifically for the post-GLBA period. The authors‟ finding of a risk increase serves as a first step in the current paper‟s framework regarding the Glass-Steagall Act‟s contribution to the 2007-08 Crisis. Since the risk is documented to have increased, the question, which should and will be addressed, is whether this risk has contributed significantly to a detrimental performance during the Crisis.

The following section will discuss the second strand of literature – the effect of investment (non-traditional) banking on bank risk profile.

2.2 Non-interest income on performance

Stiroh (2004) first takes up the issue of the existence of diversification benefits in expanding the range of bank services. He investigates the relationship between non-interest income and risk and risk-adjusted profits. First, he reports high and upwards trending correlation between interest and non-interest income. This way, following the portfolio theory, he reasons that interest income is unlikely to produce diversification benefits. Second, in a similar vein, he

(17)

documents a strong correlation (when removing trading income) between non-interest income and GDP growth, showing that the expansion into non-traditional banking is unlikely to remove cyclicality in earnings. Finally, the author performs an empirical analysis of the relationship between non-interest income and risk and return. He finds that more non-interest income implies more volatility and less risk-adjusted return.

DeYoung & Rice (2004) analyse the concept of non-interest income both in terms of its prevalence as well as its association with the bank financial performance. First, the authors look at the characteristics explaining the reliance on non-interest income. With a panel data of commercial banks, they find that well-managed banks tend to rely on non-interest income less, whereas bank size, stress on customer relationships and some financial technology improvements all reveal a positive relationship with non-interest income. Second, an analysis of non-interest income and financial performance is presented. The authors investigate the effect on return on equity, its volatility and Sharpe ratio. The findings indicate a positive relationship between non-interest income and return on equity and return on equity‟s volatility, and a negative one with Sharpe ratio. The results are robust when performing an instrumental variable analysis and for most subsample periods. Interestingly, the authors find that in the first subsample, the non-interest income improves the risk-return tradeoff whereas it is detrimental in the second part. They use this finding and argue that the long-term expansion into non-traditional banking has peaked and that traditional banking activities will likely remain the focus of US commercial banks.

Stiroh and Rumble (2006) elaborate on the earlier paper (Stiroh, 2004) and investigate the relationship between bank revenue diversification and risk-adjusted performance. A sample of Financial Holding Companies is used in order to answer the question of whether expanding the range of services implies performance benefits. The methodology combines two analyses – cross-section, for effects across banks, and panel data for effects within banks. The association of interest is that between risk-adjusted return on equity, return on assets and Z-score, and a diversification measure [DIV]6. Moreover, the model introduces a second element – the share of non-interest income itself. Such an approach allows the authors to identify the effect of the “indirect diversification” (revenue composition) and “direct exposure” (introduction of new activities) (Stiroh and Rumble, p.2, 2006). The findings imply that, across banks, diversification benefits exist but are offset by the greater earnings volatility. Within banks, on the other hand, no diversification benefit is revealed but,

6

Composition of revenue as broken down into net interest income and non-interest income. A higher DIV implies a greater degree of diversification.

(18)

similarly to the cross-section, greater non-interest income is found to be negatively and significantly correlated with risk-adjusted performance. Hence, a robust result of detrimental effect of non-interest income, on performance, is found.

Brunnermeier et al. (2012) investigate the question of non-interest income‟s contribution to systemic risk. With a sample of commercial banks the analysis involves studying the effect of non-traditional banking income (trading income, investment banking/venture capital income) on the stability of the financial sector. The latter is measured with two variables – CoVar, SES. While CoVar quantifies a bank‟s externality on the system, SES quantifies the opposite. The findings imply a greater contribution, to systemic risk, of banks with greater non-interest income. This result is robust for decomposing non-interest income and checking for endogeneity (a diff-diff analysis with the Lehman Brothers bankruptcy as an exogenous shock). Further, the authors relate the amount of non-traditional banking income to the recession performance. They find that more trading income, one year prior to recession, implies a significantly worse performance during the recession. The relationship does not stand when using investment banking/venture capital income.

Finally, some papers study the risk effects of non-traditional banking in other countries. Lepetit et al. (2008) analyse product diversification of the European banks and its effect on risk. Proxing product diversification with non-interest income, further decomposed to trading income and commission and fees, the authors investigate the banks‟ risk exposure and insolvency risk. On an aggregate level, they find that a higher share of non-interest income (to net operating income) is significantly correlated with a higher risk and insolvency. They find that small banks and commissions and fees income drive this result. Moreover, for the small listed banks, trading income can even lead to a reduction in the risk exposure and default risk.

Köhler (2014) investigates the effect that an increasing share of non-interest income has on bank risk. Using a sample of German banks, the author finds that the effect depends upon the business strategy of a bank – for retail-oriented banks (savings and cooperative banks), an increase in the share of non-interest income implies a reduction in riskiness (i.e. diversification benefits). The opposite is true for investment-oriented banks. The analysis of the effect, with respect to different business strategies, is enhanced with the use of quintile regressions. Interestingly, the author finds a significant impact of only certain components of non-interest income (fees and commission) while trading income yields no significance. The author also observes that the composition of non-interest income is significantly different for the retail and investment-oriented banks.

(19)

Overall, one can observe a consensus, in the literature, regarding the effects of an increasing share of non-interest income on bank financial performance. Referring both to the American as well as the European banking systems, one can conclude that an expansion into non-traditional banking involves an increase in riskiness. Interestingly, the authors find diverging results as to which components of non-interest income drive the increase in risk. While some „blame‟ the fees and commissions, others claim that trading income is the source of risk. Since this paper investigates the effects of commercial banks entering investment banking, by engaging in securities activities, the second proposition is considered more relevant. In line with the presented empirical findings, a hypothesis will be derived that, after the GLBA, an increase in non-traditional (investment) banking has been positively correlated with an increase in bank riskiness – hypothesis B.

The contribution, and an expansion of the existing studies, will lie with the second part of the empirical analysis. Namely, an account will be made of how bank riskiness, originating from investment banking, has been related to the performance during the Crisis. This analysis will be novel, yet related to the paper by Brunnermeier et al. (2012). The current paper is a direct response to the criticism expanded regarding the Glass-Steagall‟s repeal – that it has provoked more risk-taking, which, in turn, influenced the financial sector to the detriment. Hence, the focus explicitly lies on the risk profile of commercial banks entering the securities business. The paper by Brunnermeier et al. (2012) analyses the issue more generally where the interest lies in the relationship between non-traditional banking and performance, during times of market distress – the risk mechanism is not treated. The current paper, thus, directly addresses the main point of discussion – the risk-taking behavior. Therefore, the findings of the two papers are likely to complement each other forming a coherent and an informative framework. Furthermore, to the author‟s knowledge, the method used for predicting risk, from investment banking, used in this paper, has not been formulated before.

3. Methodology

The primary interest of this research is to study whether the GLBA contributed to the Crisis. For this purpose, the empirical analysis will be composed of investigating two elements – the effect that engaging in investment banking had on a commercial bank‟s risk profile, and the effect that investment banking risk exercised on banks‟ performance during the Crisis. While the first effect relates to hypothesis B, the second one will address hypothesis A.

(20)

3.1 Sample selection

The starting point for the sample collection is the CRSP dataset. Institutions with SIC codes between 6000 to 6300 and 6710, 6711, 6712 with data available from the year of 2000 are used. Foreign incorporated companies are removed from the sample. Next, the National Information Center – a repository of financial data and institution characteristics collected

by the Federal Reserve System, is referred to. Here, each institution is evaluated according to

its structure before the GLBA (prior to 2000) and subsequent changes. Institutions, which, prior to 2000, were Bank Holding Companies [later referred to as non-FHCs] and stayed as such and those which were Bank Holding Companies and, in 2000, changed to Financial Holding Companies are kept in the sample. This way only holding companies of banks are used. Such classification allows the dataset to have less heterogeneity, which could introduce significant noise. Entities, which underwent structural changes, such as renaming, are identified based on permco or a similar company name7. Banks, which underwent a merger are either kept in the sample (if acquired by another bank in the sample) or removed as a result of a lack of subsequent data availability. The resulting dataset is then merged with the Compustat data. Citigroup is removed from the sample because of its known investment banking activities before the GLBA. This leaves the final sample at 251 banks.

The sample construction is the critical element of this study, which directly relates the empirical analysis to the question. Namely, the period studied starts at the time of the repeal and ends right before the Crisis. This way the mechanisms at work can be attributed to the GLBA itself.8 Similarly, institutions which could, some of which did, convert to FHCs are used – such transformation being the fundamental freedom granted by the GLBA (i.e. an implication of a bank entering the investment banking business).

3.2 The model

The primary interest is the influence, which risk originating from commercial banks‟ securities activity, exercised on the Crisis performance. One, thus, needs to identify this risk for the entities in question. This identification will be performed with a panel data. The time period covered will be 2000-2006. The analysis will be performed in two stages.

First, an investigation of the investment banking‟s effect on the risk profile will be performed. While the risk profile will be measured with return on assets [ROA] volatility and

7_{Company name changes are confirmed with the National Information Center – a repository of financial data}

and institution characteristics collected by the Federal Reserve System

8_{This is a technical assumption. The attribution occurs based on the design of the empirics where the time frame} for the studied effects starts at the time of the GLBA. This is not to say, however, that it can be said, with certainty, that these effects only started occuring after the GLBA. It is the implied assumption but it has not been proved in this paper.

(21)

the Z-score, investment banking activity will be proxied with the share of non-interest income to net interest income.9 Loans to Assets [LA] and Equity to Assets [EA] ratios will be included as control variables in order to account for other sources of a commercial bank‟s risk – credit and liquidity risk. The direction of those effects could be two-fold. In case of loans to assets, a higher ratio could imply greater risk due to higher credit and liquidity risk. It could, on the other hand, indicate a bank‟s greater engagement in traditional banking. In case of equity to assets, on the other hand, a higher ratio could point to greater conservatism and, hence, lower risk. Yet, risky entities might hold more equity for turbulent times (Geyfman & Yeager, 2009). Total assets [TA] control for firm size. Year dummies are also included. The endogeneity (reverse causality) of the model is limited with the inclusion of the Mills ratio.

From this regression, investment banking risk will be predicted and averaged over the years 2000-2006.10 Such procedure directly relates the studied risk mechanism to the GLBA – we are interested in the risk, which was created right after the repeal and which was at work until the Crisis began.

The following model, estimated with standard errors clustered at bank level, and winsorized at 1%, will form the basis for this stage‟s analysis:

In the introduction to this paper, arguments were presented which claim that the GLBA has led to excessive risk-taking, by commercial banks, which started engaging in investment banking. From these arguments, as well as other studies, a hypothesis [B] was formed that more investment banking implies greater risk. In line with this framework, an expectation is formed that the coefficient will be positive – more investment banking, higher risk profile. Moreover, with the inclusion of Mills Ratio and other control variables, this relationship does not seem to suffer from obvious endogeneity.

9_{Z-score is a measure of bank insolvency. A higher value implies a lower insolvency risk (De Haan and} Poghosyan, 2012). As the two risk measures differ in construction, the risk from ROA volatility is higher when the coefficient is positive whereas the risk from the Z-score is higher when the coefficient is negative. This difference is the reason why the regressions on the buy-and-hold return will have opposite signs when using the risk predicted from ROA volatility and Z-score. Different signs will, thus, imply that the results are consistent. 10_{Risk is predicted by multiplying the investment banking coefficient with the investment banking measure.}

(22)

In the second stage, the predicted investment banking risk will be used as the main independent variable explaining bank performance during the Crisis. Bank performance will be measured with the buy-and-hold return during the Crisis (December 3rd, 2007 to June 30th, 2009). The model will control for, book-to-market value [BM], firm size (market value, MV), leverage [L] and Tier 1 capital ratio [T1]. A dummy variable for FHC is also included.

MV+

The X variable is the predicted investment banking risk, which is the primary variable of interest. Similarly to the first regression, the expectation of its sign is implied in the hypothesis [A]. The line of argumentation was that the excessively risky behaviour of banks is to be blamed for the occurrence of the Crisis. This assertion, combined with the repeal of the Glass-Steagall, would imply that the repeal has facilitated the engagement in investment banking, which has led to excessive risk-taking and, in turn, provoked the Crisis. Such framework would point to an expectation of a negative coefficient, in the second

regression. Similarly, is likely to be negative – the debate revolving around the Crisis blames the large financial conglomerates for the downturn. Tier 1 capital ratio is expected to unravel a positive sign. Market value, book-to-market value and leverage are expected to be negative.

Findings of a significantly negative coefficient would speak in favour of the claim that GLBA contributed to the detrimental performance of banks, during the Crisis. The magnitude of the coefficient will form basis for the discussion regarding the causal relationship between the GLBA and the Crisis‟s occurrence.

3.2.1 Endogeneity issue

The entry into investment banking business was enabled with the GLBA where Bank Holding Companies could choose to engage in securities activities by establishing a Financial Holding Company. Hence, the critical element of any empirical analysis, on the topic, is the endogenous choice, of a bank, to become a FHC. Not accounting for this element introduces endogeneity problem as reflected in reverse causality – e.g. a more risky commercial bank engaging in investment banking in order to diversify some of the risk away.

A partial solution to this issue, applied in this paper, is the Heckman Two-Stage Selection Model. This procedure allows for limiting selection bias where the sample used is not random (Winship and Mare, 1992).

(23)

The mechanics of the method will involve two steps. The first step is a probit regression of a variable indicating entering into investment banking, i.e. becoming a FHC (0 or 1), on a variable “large” which is equal to 1 if a bank was among the 10% largest banks, in terms of total assets, in 2000. The reason for such choice is that the largest banks were the ones, which were most likely to enter investment banking. The second step involves calculation of the Mills Ratio, for every bank, which is equal to probability density function over cumulative density function. This Mills ratio is then included in the final regression as a control variable (Geyfman & Yeager, 2009).

The second regression is unlikely to suffer from reversed causality – the risk is predicted as the average of years 2000-2006 and this is related to returns between 2007-2008. The omitted variable bias is limited, yet not removed, with the inclusion of standard controls.

4. Data and Descriptive Statistics

Tables 2 and 3 present the sample summary statistics for regressions 1 and 2, respectively. The numbers are reported separately for FHCs and non-FHCs. This allows for a comparison of raw statistics between the commercial banks, involved in investment banking and those, which did not start engaging in the securities business. Appendix C contains summary statistics for the whole sample (FHCs and non-FHCs combined). All financial data comes from Compustat and CRSP databases.

(24)

Table 2. Sample Summary Statistics – investment banking on risk

This table presents summary statistics [$mln] for the sample of FHCs and non-FHCs in the period 2000-2006. FHCs are the banks, which, after the repeal, converted to universal banks (i.e. entered investment banking). FHCs have higher measures of investment banking activity (non-interest income, trading income and their shares).

Non-FHCs FHCs

N Mean Standard Deviation

Min Max Median N Mean Standard

Deviation

Min Max Median

Total Assets 1163 4,791.47 13,911.53 43.55 203,638 1,178.39 392 68,454.18 196,394.40 224.55 1,459,737.00 3,207.35

Non-interest income 1161 126.67 667.38 -44.22 9,024 12.14 392 1,753.35 5,037.65 0.87 40,195 36.41

Net interest Income 1161 157.64 465.83 1.24 7,397 41.48 392 1,801.40 4,633.80 8.23 34,591 112.20

Non-interest income (loss)/total assets (%)

1161 1.32 1.48 -0.27 15.27 1.00 392 1.72 1.05 0.28 6.69 1.46

Net interest income/total assets (%)

1161 3.53 0.71 1.12 6.43 3.50 392 3.38 0.66 0.91 6.30 3.38

Non-interest income/net interest income (%)

1161 40.36 65.56 -10.16 1,047.95 28.48 392 58.37 61.32 6.88 503.42 41.60

Trading income/net interest income (%)

1161 0.10 1.01 0 21.73 0 392 1.71 8.14 0 75.27 0

Equity to Assets ratio 1163 0.09 0.03 0.03 0.47 0.09 392 0.09 0.02 0.05 0.19 0.09

Loans to Assets ratio 1163 0.64 0.13 0 0.94 0.66 392 0.63 0.13 0.04 0.85 0.65

ROA volatility (%) 1160 0.26 0.62 0.00 8.39 0.11 391 0.19 0.27 0.01 2.32 0.1

(25)

Table 3. Sample Summary Statistics – investment banking risk on Crisis performance

The table presents summary statistics [$mln] for the sample of FHCs and non-FHCs at the fiscal year-end 2006 and 2007-2009 (buy-and-hold return). FHCs are the banks, which, after the repeal, converted to universal banks (i.e. entered investment banking).

Non-FHCs FHCs

N Mean Standard deviation

Min Max Median N Mean Standard

deviation

Min Max Median

Buy-and-hold return 111 -0.42 0.36 -1.00 0.68 -0.47 53 -0.37 0.35 -0.96 0.47 -0.41

Tier 1 capital ratio (%) 111 11.70 2.73 6.30 23.20 11.30 53 10.65 2.30 7.47 16.77 9.90

Book-to-market Value 111 0.58 0.29 0.01 3.00 0.53 53 0.54 0.15 0.29 1.10 0.50

Market Value 111 2,944.71 22,715.07 8.84 239,757.00 295.14 53 13,013.36 32,427.67 41.85 167,550.70 1,105.71

(26)

For the first regression, the total sample is composed of 251 banks out of which a majority are non-FHCs. This implies that converting into a FHC was not an obvious choice for the banks. Regarding their size, one can observe a sizeable difference in total assets – FHCs are, on average, fourteen times as big as non-FHCs. The median value, however, is three times greater (1,178.39 versus 3,207.35) which implies that there are significant outliers – the standard deviation, in case of FHCs, is 196,394.40. This observation is in line with the assumption made for the calculation of the Mills ratio – the biggest banks were the ones, which were most likely to engage in investment banking. A natural consequence of this is the value of non-interest income – FHCs have an average value of 1,753.35, which is much larger than that of non-FHCs (126.67). Scaled by total assets, however, the difference is no longer that substantial – the average values are 1.32% and 1.72% for non-FCHs and FHCs, respectively. Moreover, the maximum of the entire sample is among non-FHCs. The occurrence of positive values, for non-interest income, in case of non-FHCs, is a critical observation. Since non-interest income is meant to proxy for investment banking, one would expect to observe values very close to 0, for non-FHCs. This element might introduce noise to the analysis and should, thus, be acknowledged. It either means that the measure has several components, not all of them being related to investment banking, or that non-FHCs perform investment banking activities without having the FHC status. Nonetheless, the significant difference in the mean and median values would seem to point to some degree of the measure‟s relevance. This assumption will be subject to close investigation in the empirical analysis. In case the measure proves to be too noisy, an alternative of trading income will be used – as documented in table 2, this form of investment banking is substantially larger for FHCs (average of 1,71% as opposed to 0,10%). Scaling non-interest income by net interest income (as opposed to total assets) presents a similar result. The mean values are 58.37% and 40.36% for FHCs and non-FHCs, respectively. While the median is also larger for FHCs, the maximum value is again recorded among non-FHCs. In the literature, a range of values has been reported. While Brunnermeier et al. (2012) document a value of 23%, in 2000, Stiroh and Rumble (2006) find 33.1% and 47.8% for small and large banks, respectively.11 Interest income (scaled by total assets), on the other hand, is slightly larger for non-FHCs, which is consistent which the structure of the banks‟ activities. Nonetheless, the difference is smaller than one would expect – average of 3.53 as opposed to

11

Scaling non-interest income by total assets implies how much of the business comes from non-traditional banking whereas scaling it by net interest income suggests how much profit comes from non-traditional banking activities. The literature, referred to in this paper, has primarily used the latter.

(27)

3.38 of FHCs. This means that either non-FHCs are not focused purely on traditional banking, which is in line with the larger values of non-interest income, or that non-FHCs pursue a significant amount of both traditional as well as non-traditional banking. The loans to asset ratio suggests the latter – we observe average values of 0.64 and 0.63 for non-FHCs and FHCs, respectively. This indicates that FHCs did not sacrifice the commercial banking activities in the name of investment banking. Instead, they started pursuing it in addition to the traditional banking business. This statistic is in line with the numbers reported by Geyfman and Yeager (2009) – in the years 2001-2007, the authors document 0.67 for non-FHCs and 0.58 and 0.66 for non-FHCs, with and without section 20 subsidiary, prior to 2000, respectively. Stiroh and Rumble (2006) report a similar value of 0.64. The equity to assets ratio does not differ – both groups record 0.09. This indicates that banks in both subsamples are either equally conservative or that FHCs do not recognize a need to keep a greater buffer for turbulent times – this would mean that they should not have a greater risk profile, which is contrary to the expectations. Geyfman and Yeager (2009) find the same pattern. Moreover, for FHCs, Stiroh and Rumble (2006) report equity to assets of 0.09. In this context, of relevance are the values for banks‟ riskiness. With both measures, FHCs show less risk – ROA volatility of 0.26 and 0.19, and Z-score of 5.66 and 5.81 for non-FHCs and FHCs, respectively. This is a critical finding, which is likely to be reflected in the empirical analysis – based on summary statistics, FHCs appear less risky. The median values suggest the same conclusion.

In case of the second regression, the total sample is smaller as some banks no longer were active during the analysed period – 164 banks. The observation of primary interest is the Crisis performance – the buy-and-hold return. The mean (-0.42 vs. -0.37), minimum (-1.00 vs. -0.96) and median (-0.47 vs. -0.41) values are lower for non-FHCs. Yet, the maximum is greater in the non-FHC group (0.68 vs. 0.47). The returns are almost equally volatile for the two groups. The range of returns is largely in line with the findings by Fahlenbrach et al. (2012) – the authors find a mean Crisis return of -0.31, min of -1 and max of 0.47.12 The Crisis return statistics further add to the conclusion drawn from table 2 – FHCs have less risk and experienced greater returns during the Crisis. Similarly to the Total Assets variable in table 2, market value points to FHCs being bigger than non-FHCs – 13,013.36 as opposed to 2,944.71. The median FHC market value is almost four times the median of the non-FHC group. As reported in table 13 (appendix C), across the two groups, the mean market

(28)

capitalization is 6,198.61, which is similar to 5,439.75 reported by Fahlenbrach et al. (2012). Additionally, we observe a larger Tier 1 capital ratio for non-FHCs – 11.70% as opposed to 10.65%. Both groups, however, on average, meet the minimum requirement of 10%. Fahlenbrach et al. (2012) document an average of 10.86%. The standard deviation from those values is similar for the two groups – 2.73 and 2.30 for non-FHCs and FHCs, respectively. Further, non-FHCs seem to be slightly more overvalued with the average book-to-market value of 0.58, which is 0.04 higher than that of FHCs – a value of 0.60 is to be found in the paper by Fahlenbrach et al. (2012). Non-FHCs also appear to have a greater leverage. The summary statistics have reasonable values and are consistent with those reported in other studies.

5. Empirical results

In order to answer the research question, two empirical analyses must be performed. The first one investigates the relationship between investment banking and risk. If investment banking activity had been significant in explaining bank risk profile then it already constitutes a first step towards arguing some intuitive association between GLBA and bank performance. The second one looks at whether investment banking risk significantly explains the Crisis returns. A finding of investment banking risk‟s significance in explaining bank returns, during the Crisis, would further substantiate the discussion. This is particularly so due to the construction of the sample and the time period used – both directly address the GLBA itself.

5. 1 Investment banking on risk

The first part of the empirical analysis aims at quantifying the effect of investment banking on risk. The hypothesis is that expanding the range of bank services has been detrimental in terms of bank risk. Table 4 presents support for this claim.

(29)

Table 4. Effect of investment banking on risk 2000-2006

The table reports the full-sample correlation between measures of bank risk and investment banking activity. Columns (1) and (2) present the results from panel data regressions of annualized return on asset volatility (%) on one-year lag of non-interest income to net interest income ratio (%) and bank characteristics. Columns (3) and (4) present the results from panel data regressions of annualized log of Z-score on one-year lag of non-interest income to net non-interest income ratio (%) and bank characteristics. The covariates include the mills ratio, log of total assets, loans to total assets ratio and equity to total assets ratio. Year dummies are also included. Appendix A provides variable definitions. All variables are winsorized at 1%. Standard errors are clustered at bank level. The results indicate that investment banking activity has been significant in explaining bank risk profile. Numbers in parenthesis are the standard errors and the stars indicate significance level - ***1%, **5% *10%.

Dependent variable ROA volatility Log (Z-score)

(1) (2) (3) (4)

Non-interest Income to Net Interest Income t-1 0.002*** 0.002*** -0.005*** -0.006*** (0.000) (0.000) (0.000) (0.000)

Log (Total Assets) t-1 -0.017* 0.111***

(0.01) (0.032)

Equity to Assets ratio t-1 -0.246 7.848***

(0.553) (1.552)

Loans to Assets ratio t-1 -0.062 0.177

(0.134) (0.343) FHC -0.059*** 0.153* (0.021) (0.086) Mills Ratio -0.052 0.235 (0.060) (0.222) D (2001) -0.018 0.054 (0.029) (0.302) D (2002) 0.064** -0.190** (0.028) (0.083) D (2003) -0.013 0.093 (0.022) (0.072) D (2004) -0.009 -0.064 (0.021) (0.068) D (2005) -0.005 0.007 (0.023) (0.072) Constant 0.144*** 0.394*** 5.936*** 4.059*** (0.015) (0.152) (0.050) (0.536) Observations 1,301 1,301 1,301 1,301 Number of banks 251 251 251 251 R-squared 0.060 0.082 0.056 0.115

As shown, the significance of non-interest income, in explaining bank risk, is already demonstrated in the univariate regression. A positive effect is found where a one standard deviation increase in non-interest income share yields an effect of 0.24 standard deviation

(30)

increase in ROA volatility and 0.35 standard deviation decrease in Z-score.13 Hence, the direction of the effect, when using two different risk measures, is the same. While the magnitude of the effects is not substantial, they are sizeable.

Adding controls, which take out the noise from the regression, does not change the univariate regression‟s results. In case of ROA volatility, as the risk measure, the coefficient remains positive. In case of the Z-score, the magnitude of the coefficient experiences a slight increase. The insignificance of the Mills ratio, in both regressions, suggests that there is no selection bias. The size control (log of total assets) is significant in both regressions and implies that larger banks have lower risk (a one standard deviation increase in size implies a 0.05 standard deviation lower ROA volatility and 0.18 standard deviation higher Z-score). Previous studies have found the same direction of the effect (Stiroh (2004), Lepetit et al. (2008), DeYoung and Rice14 (2004), Köhler 15 (2014). This result is consistent with the summary statistics – FHCs, which are, on average, bigger than non-FHCs, have lower risk. The FHC also supports this conclusion. FHCs have lower volatility which, controlling for size, investment banking activity, credit and liquidity risk, is statistically significant. If a bank is a FHC, it has a 0.059% lower ROA volatility and 0.153% higher Z-score than a non-FHC. Credit and liquidity risk controls (equity to assets, loans to assets) provide consistent results – a positive effect on risk (lower ROA volatility, higher Z-score) but only in one case is the coefficient significant (Z-score regressed on equity to assets ratio). In the context of the relevant empirical studies, authors have found a variety of results regarding the direction of the equity and loans to assets ratios‟ effect. Equity to assets ratio‟s coefficient was found positive when explaining both the Z-score and ROA volatility (Stiroh (2004), Stiroh and Rumble (2006). Loans to assets coefficient has been reported negative in terms of Z-score, ROA volatility, Sharpe ratio and ROE volatility by Stiroh (2004), DeYoung and Rice (2004) and Stiroh and Rumble (2006). Finally, using market-based measures of risk, Geyfman and Yeager (2009) documented negative signs on total assets and equity to assets, and positive one on loans to assets. One can, thus, observe certain similarities and differences between the different studies and the current analysis. As explained, equity and loans to assets ratios can both be expected to demonstrate either a positive or a negative sign. Hence, the diverging results are not largely surprising. The significant element is the consistency, of the effect, sustained across the two regressions.

13

Standardized coeffiecient = (coefficient * s.d.X)/s.dY

14_{Measuring the effect of size on the volatility of return on equity.} 15_{An effect only on ROA volatility}

(31)

As reported in Table 5 the results are robust – using two additional transformations of non-interest income (log of non-non-interest income, ratio of non-non-interest income to total assets) a detrimental effect on bank risk profile is found. This implies that expressing the investment banking activity as a portion of income or business does not vary the results. In case of the first specification, the magnitude of the effect is larger with one standard deviation increase in the log of non-interest income being associated with a 0.45 and 0.69 standard deviations increase and decrease in ROA volatility and Z-score, respectively. The second transformation, on the other hand, exhibits a weaker effect – one standard deviation increase in the non-interest income to total assets implies a 0.20 standard deviation increase and 0.30 standard deviation decrease for ROA volatility and Z-score, respectively. The control variables demonstrate a consistent picture. The effect of total assets maintains its direction and significance gaining magnitude. Similarly, equity to assets ratio keeps its sign and significance. The magnitude is slightly gained. Further, loans to assets ratio gains in magnitude and significance (in case of the log of Z-score regressed on the log of non-interest income). The FHC dummy again demonstrates a positive effect on risk with significance maintained and magnitude slightly gained. The Mills ratio maintains its sign and gains significance when using the log of non-interest income as the investment banking measure.

The repeal of the Glass-Steagall Act and its contribution to the 2007-08 Global Financial Crisis