artikel smk (bank)

Journal of Banking & Finance 34 (2010) 1590–1606

Contents lists available at ScienceDirect

Journal of Banking & Finance

journal homepage: www.elsevier .com/locate / jbf

Endogenously structured boards of directors in banks

Shams Pathan a,*, Michael Skully b

a UQ Business School, The University of Queensland, Brisbane, Queensland 4072, Australiab Department of Accounting and Finance, Monash University, Melbourne, Victoria 3145, Australia

a r t i c l e i n f o a b s t r a c t

Article history:Received 4 September 2009Accepted 7 March 2010Available online 10 March 2010

JEL classification:G21G28G30G32G34L22K22

Keywords:Board of directorsIndependent directorsCEOCEO dualityEndogenousBank holding companies

0378-4266/$ - see front matter � 2010 Elsevier B.V. Adoi:10.1016/j.jbankfin.2010.03.006

* Corresponding author. Tel.: +61 7 3346 8075; faxE-mail addresses: [email protected] (

buseco.monash.edu.au (M. Skully).

This paper examines the trends and endogenous determinants of boards of directors (board size, compo-sition, and CEO duality) for a sample of 212 US bank holding companies, from 1997 to 2004. Overall, theresults show that the costs and benefits of boards’ monitoring and advising roles could explain bankboard structures with caveats. For example, due to the regulatory nature and comparatively intensivescrutiny of bank officers and directors, it is argued that bank managers have less control over the direc-tors’ selection processes. Thus, bank board independence should not be the outcome of negotiation withCEOs. Consistent with this view, bank CEOs are found not to affect bank board independence. The trendanalysis also provides some important results. In contrast to non-bank evidence, for instance, board sizewas discovered to decrease over the sample period for large and medium-sized banks, while board sizeremained relatively stable for small banks. These results are robust with respect to different estimationspecifications. Furthermore, the study’s findings have important policy implications for bank regulatorsand investors.

� 2010 Elsevier B.V. All rights reserved.

1. Introduction associated with boards’ monitoring and advising functions (Her-

A wide range of accounting, finance and management literaturehas determined that a certain type of board structure is preferredto monitor managers. For instance, a small number of board direc-tors and more independent directors are considered to be impor-tant elements of an effective board (e.g., Yermack, 1996; Famaand Jensen, 1983). This issue was further emphasized by the intro-duction of the Sarbanes–Oxley Act of 2002 and the associated list-ing rules by NYSE, NASDAQ, and AMEX as they require a majorityof independent board directors and a completely independentaudit committee. Hence, these developments are in favor of a uni-form board structure, irrespective of the industry in question.However, if we believe Alchian’s (1950) economic theory of ‘Dar-winism,’ it is important to understand why some firms still main-tain large boards, while others have majorities of non-independentor executive directors. To answer this question, several studies at-tempt to explain this observation by relating the costs and benefits

ll rights reserved.

: +61 7 3346 8166.S. Pathan), michael.skully@

malin and Weisbach, 1998; Raheja, 2005; Adams and Ferreira,2007; Harris and Raviv, 2008). Based on these theoretical works,among others, Baker and Gompers (2003), Boone et al. (2007),Coles et al. (2008), Linck et al. (2008), and Lehn et al. (2009) empir-ically find evidence in support of the endogenous formation ofboards of non-financial firms.

While the same theoretical underpinnings relating to boardstructure are valid to both banks and non-bank firms, the existingempirical studies exclude banks from their sample, and several fac-tors (such as regulation, high leverage) could limit generalizingnon-financial board structure findings to banks. This study aimsto fill this knowledge gap by investigating whether the costs andbenefits of the boards’ monitoring and advising functions couldalso explain board structure (board size, composition, and CEOduality1) in a regulated industry, like banks.

The global financial crisis also highlights the importance ofimproving understanding of bank governance. Indeed, the studyon bank board structure deserves special attention for several rea-sons. Perhaps the bank board of directors is even more important

1 ‘CEO duality’ refers to a situation in which the CEO is also the board chair.

http://dx.doi.org/10.1016/j.jbankfin.2010.03.006

mailto:[email protected]



http://www.sciencedirect.com/science/journal/03784266

http://www.elsevier.com/locate/jbf

2 Macey and Miller (1993, pp. 401–407) define fiduciary duties as ‘‘. . . themechanism invented by the legal system for filling in the unspecified terms ofshareholders’ contingent [contracts].” In addition, see Macey and O’Hara (2003, pp.93–95) for a good discussions of bank directors’ fiduciary duties, or ‘duty of care’ and‘duty of loyalty’ to shareholders, depositors and regulators.

3 Source: The OCC Web site http://apps.occ.gov/EnforcementActions/ (viewed onAugust 27, 2009) for a search engine for enforcement actions.

S. Pathan, M. Skully / Journal of Banking & Finance 34 (2010) 1590–1606 1591

as a governance mechanism than its non-bank counterparts be-cause banks have become larger, complex and more diversified,following the deregulation with the Riegle–Neal Interstate Bankingand Branching Efficiency Act of 1994, as well as the Gramm–Leach–Bliley Act of 1999. In addition, the presence of regulationcould have different implications for bank board structure deter-minants. For example, as discussed later in Section 2, bank direc-tors and managers are subject to stringent regulatory scrutiny,compared to non-bank board directors. This regulation is com-monly justified for three reasons. First, there are costly conse-quences in the case of bank failure (Flannery, 1998). Second,bank shareholders have distorted objective of excessive risk-takingin the presence of deposit insurance (Galai and Masulis, 1976). Fi-nally, bank debtors do not have the incentive to monitor bankmanagers due to high information asymmetry (Demirgüç-Kuntand Detragiache, 2002). This constant regulatory monitoring couldlimit bank managers’ self-serving behavior (such as perks). Hence,bank managers, including CEOs, cannot influence the directorselection process. As a result, in contrast to non-bank studies,CEO power (i.e., CEO’s ability to influence board decisions) shouldnot be an important determinant of bank board independence.Thus, it is important to examine, even in the presence of such reg-ulation, whether bank board structure can still be explained by thecosts and benefits associated with boards’ monitoring and advisingfunctions, given bank characteristics and other governance mecha-nisms. It is also important to determine whether board structurehas changed significantly for regulated banks due to the enactmentof the Sarbanes–Oxley Act (SOX) of 2002 and associated listingrules changes. The study of the banking industry also provides aunique setting in which to enhance our understanding of boardstructure determinants.

Using a sample of 212 US bank holding companies monitoredbetween 1997 and 2004, this study finds some evidence in favorof endogenously chosen boards of directors. This supports theargument that banks structure their boards consistently with thecosts and benefits associated with boards’ monitoring and advisingfunctions. More specifically, the results show that: (i) larger andmore diversified banks have larger and more independent boards,and also combine both CEO and board chair titles; (ii) bank boardindependence is not the outcome of negotiations with the CEO; (iii)banks in which managers’ opportunities to consume private bene-fits are high have larger boards, while banks in which the cost ofmonitoring managers is low have more independent boards; (iv)banks in which managers have substantial influence and the con-straints on managerial influence are weak combine both CEO andboard chair roles; and (v) banks in which insiders’ shareholdingis high and the outsiders’ shareholding is low have smaller boards.

The trends in bank board structure over the sample period alsoprovide some significant insights. For example, bank board size de-clines over the sample period, particularly for large and mediumsize banks, which is in contrast with non-bank firm evidence. How-ever, the percentage of independent directors increases substan-tially, especially during the post-SOX period.

This study contributes to the existing literature on board struc-ture determinants in several important ways. This is the first studyto demonstrate that even in a regulated industry like banking, thecosts and benefits of monitoring and advising functions of boardscould explain their structure. This paper complements and extendsAdams and Mehran’s (2009) study, which investigates bank boardgovernance for a sample of 35 BHCs from 1959 to 1999. They illus-trate that bank board size relates to M&A activity and organiza-tional structure. However, they have not shown the determinantsof other important board features, such as board compositionand leadership structure. Likewise, they have not exclusivelyexamined the determinants of bank board structure (such as nego-tiations with the CEO, ownership incentive structure), in view of

existing non-bank evidence by Lehn et al. (2009), Linck et al.(2008), and Boone et al. (2007), among others. This study alsobroadens our knowledge by showing that bank CEOs do not influ-ence the board selection process due to fear of regulatory action.This result challenges the existing non-bank evidence and thushas important policy implications, while designing appropriategovernance system for banks. In terms of methodology, a broadset of diagnostic and statistical consistency tests were conductedto confirm the robustness of the results, including several ap-proaches that account for unobserved heterogeneity and simulta-neity. For example, a system generalized method of moments(GMM) estimation technique was used to directly control for any‘dynamic endogeneity’ problem. Finally, this is perhaps the firststudy to provide some evidence that bank board structure has sig-nificantly changed, following SOX and the associated changes man-dated by the stock exchanges. Such findings are vital to evaluatingthe possible impact of SOX on regulated banks’ board structure.

The rest of the paper is structured as follows. Section 2 furtherdrives the study of bank board structure determinants by discuss-ing the regulatory oversight of boards of directors in banks. Section3 reviews the literature on board structure determinants and for-mulates the relevant hypotheses for banks. Then, Section 4 de-scribes the data and methodology. Section 5 provides theempirical results, while Section 6 demonstrates the robustness ofthe results, using different estimation techniques. Section 7 reportssome results with regard to the impact of SOX and associated list-ing rules’ changes on bank board structure determinants. Finally,Section 8 concludes the paper.

2. Regulatory oversight of boards of directors in banks

Banks’ boards of directors historically have not been legallybound to solely serve the shareholders, as is typically the case fornon-bank firms. The ‘fiduciary’2 responsibility (i.e., duty of loyaltyand care) of the bank directors and managers extends beyond share-holders to depositors and bank regulators (for more details, see Ma-cey and O’Hara, 2003; Fanto, 2006). Bank regulators set detailedstandards of conduct for directors and managers and monitor indi-vidual conformity with these standards to ensure ‘safe and sound’bank system. The regulators have considerable disciplinary powersavailable, if they discover bank directors and managers in any viola-tion of the standards. The disciplinary actions include suspensionand removal from the bank, and even a life-long ban from the indus-try; regulators can also refer the matter to federal prosecutors. Withthe passing of the Financial Institutions Reforms, Recovery andEnforcement Act (FIRREA) of 1989, and the FDIC Improvement Actof 1991, Congress further empowered bank regulators in taking‘prompt corrective actions’ against bank directors and officers fortheir decisive roles (see Shepherd, 1992 for details). For example,Section 1821(k) of FIRREA 1989 stated that directors and officersof insured institutions would be held personally liable for any mis-conduct of bank business (Shepherd, 1992, p. 1122).

The available data indicate that bank regulators frequently usethese disciplinary powers against bank directors and managers.For example, in 2005, of the 32 consensual removal orders by theOffice of the Comptroller of the Currency (OCC), 12 involved seniorbank officers, including CEO and directors.3 Thus, bank directorsand managers have a legal duty to recognize their obligation to the

http://apps.occ.gov/EnforcementActions/

1592 S. Pathan, M. Skully / Journal of Banking & Finance 34 (2010) 1590–1606

debt claimants, especially when a bank is in a weakened financialcondition. Fanto (2006, p. 6) uses the metaphor that ‘‘bank regulatorsstay with bank directors and officers from the cradle to the grave” toillustrate how intensively bank regulators monitor and disciplinebank management. In this regard, Pathan (2009) shows that banksin which CEOs have more power to influence board decision take lessrisk.

It should also be noted that certain regulations at the bank level,as opposed to the BHC’s level, can constrain bank board size andcomposition. For example, the board of a national bank (regulatedand supervised by the OCC) must consist of 5–25 directors. How-ever, the comptroller can hold the national bank exempt from sucha limit. The majority of directors of national banks must also be se-lected from a certain proximity to the bank’s head office, unless theresidency requirement is waived by the comptroller. The bankdirectors are required to have invested funds in their banks. Simi-larly, boards of state member banks are subject to specific stategovernment directives. For example, New York State requires itsmember banks to maintain boards of seven to thirty directors(when their capital stock, surplus, and divided profits are in excessof $50 million), while two-thirds of these directors should be non-executives, i.e., outsiders.4 During this study’s sample period, theNew York State member banks were required to hold a minimumof ten meetings per year (two conference call meetings were al-lowed). State regulations on the number of meetings could influencethe bank’s choice of directors, as those living within proximity mightbe more likely to be selected.

Although boards of national and state banks are required tomeet such regulatory standards, BHCs’ boards – the focus of thisstudy – are exempt from such state requirements. The Federal Re-serve System acts as the ‘umbrella supervisor’ for BHCs and do notimpose any such specific requirements on BHCs’ boards. Therefore,the regulatory environment alone cannot fully explain BHC boardstructure.

3. Literature review and hypotheses’ development

Prior studies argue that optimal board structure is based on thecosts and benefits of the board monitoring and advising roles,along with other firm and governance characteristics (Linck et al.,2008, p. 311). The two most important roles of a board of directorsare monitoring and advising (e.g., Raheja, 2005; Adams and Ferre-ira, 2007; Linck et al., 2008). As a monitor of managers, the boardsupervises the management so as to refrain them from any self-serving behaviors, such as shirking and perquisites. In its advisingrole, the board provides opinions and directions to managers forkey strategic business decisions. Typically, in explaining boards,previous studies model two specific elements of the board: boardsize and board composition (i.e., independent directors) as pointsof reference. Indeed, the theoretical arguments on board structuredeterminants can be extended to explain other board structurevariables. Accordingly, the following Sections, 3.1–3.4, elaborateon the ‘scope of operations’, board monitoring requirements, CEOs‘negotiations’ power and their succession process, and ownershipincentives structure as possible determinants of board structure(board size, independence, and CEO duality).5

6 The term ‘monitoring costs’ is used to describe the costs related to informationacquisition and processing in transforming directors’ expertise to the specific firmsfor which they serve as directors (Linck et al., 2008, p. 311).

7 The term ‘private benefits’ is defined as the insiders’ efforts-aversion, perks from

3.1. Scope of operations

The term ‘scope of operations’ refers to the nature, diversity andcomplexity of the business production process (Boone et al., 2007;

4 Source: www.banking.state.ny.us.5 It is important to note that the terms used to explain the different determinants

of board structure are from the relevant literature, to avoid any confusion.

Linck et al., 2008). In comparison to small firms, large and diversi-fied firms require additional board members to support their com-plex and diversified activities (Agrawal and Knoeber, 1996), andalso to monitor management performance (Coles et al., 2008; Lehnet al., 2009). Large firms could also require more directors to serveon their board’s sub-committees, handling the nomination ofboard members, compensation, and auditing (Boone et al., 2007).Similarly, the information requirements of larger and more com-plex firms generally result in the need for larger boards. In this re-gard, prior studies have established a positive relationshipbetween board size and the firm’s ‘scope of operations’ (e.g., Booneet al., 2007; Coles et al., 2008; Linck et al., 2008; Lehn et al. 2009).

In addition to board size, the firm’s ‘scope of operations’ couldalso affect the board composition, i.e., board independence (Booneet al., 2007; Coles et al., 2008; Linck et al., 2008; Lehn et al., 2009).Since outside independent directors are better monitors, large andcomplex firms could require more of them so as to reduce the aug-mented agency problems of being large (Lehn et al., 2009). In addi-tion, Fama and Jensen (1983) and Linck et al. (2008) consideroutside directors to be of high importance to large and complexfirms, since they bring valuable expertise and potential networksthat could be beneficial to the firm. They further contend that eventhough ‘monitoring costs’6 increase with a firm’s ‘scope of opera-tions’, the benefits from effective monitoring offset the costs ‘on bal-ance’. These arguments are consistent with those of Boone et al.(2007), Coles et al. (2008) and Linck et al. (2008), all of whom holdthat board independence is positively related to the scope of opera-tions. To capture the different aspects of the so-called scope of oper-ations, prior studies have used multiple proxies for it, such as firmsize (i.e., total assets), age, leverage and the number of business seg-ments involved (Boone et al., 2007; Linck et al., 2008).

Thus, for large and diversified banks, additional board membersand possibly more independent directors are required to monitormanagement (Boone et al., 2007, p. 70) and to advise on new prod-uct markets, technology, regulations, M&A and so forth (Lehn et al.,2009, p. 4). Based on these arguments, the first Hypothesis H1 re-lated to the ‘scope of operations’ is as follows:

Hypothesis H1. Bank board size and the percentage of inde-pendent directors are positively related to the bank’s scope ofoperations.

3.2. Board monitoring requirements

The term ‘board monitoring requirements’ is used to expressthat board structure is also affected by the net benefits of monitor-ing managers’ ‘private benefits,’7 as well as the ‘monitoring costs’to directors (Raheja, 2005; Adams and Ferreira, 2007; Harris andRaviv, 2008). With regard to ‘private benefits’, the benefit gainedfrom monitoring managers increases if managers have opportunityto extract more ‘private benefits’ from their firms (Boone et al.,2007; Chi and Lee, 2010). Generally, for managers, such opportuni-ties arise when firms have free cash flows, and managers areimmune to any shareholders’ activism, i.e., M&A activities (Booneet al., 2007). For instance, if boards are ‘staggered,’8 then sharehold-ers or potential acquirer could not be able to discipline managers, asit will be difficult, if not impossible, to remove all the directors at the

inferior projects and opposition to acts against the CEO (Raheja, 2005, p. 298).8 A board is staggered when its directors are divided into classes, generally three

classes, with only one class of directors standing for re-election each year. Thus,shareholders cannot replace the majority of directors in any given year, even thoughthere could be widespread shareholder support for such a change.

http://www.banking.state.ny.us


same time. In the presence of such opportunity for greater ‘privatebenefits’ to insiders, boards will hire more independent directorsand so increase in overall size (Boone et al., 2007, p. 71).

With regard to ‘monitoring costs,’ these are greater for firmswith high information asymmetry (Fama and Jensen, 1983). Priorstudies suggest that firms with greater ‘monitoring costs’ shouldrely less on outside directors (Fama and Jensen, 1983; Lehn et al.,2009). It is costly to transfer firm-specific information to outsidersbecause they are less informed about the firm’s projects (Linck etal., 2008, p. 311). In addition, large boards could have less motiva-tion (to incur additional costs or efforts) to acquire information dueto ‘free-riding’ problems, as well as higher coordination and directcosts, such as remuneration (Linck et al., 2008). In contrast, insidedirectors have access to firm-specific information as part of theirday-to-day activities. Thus, theoretical models of Raheja (2005),Adams and Ferreira (2007) and Harris and Raviv (2008) on boardstructure predict that the number of outsiders decreases with‘monitoring costs.’ Therefore, firms with high information asym-metry could benefit from smaller board size and a greater repre-sentation of inside directors because of high monitoring costs.Generally, information asymmetry is high for firms with high stockreturn volatility (Fama and Jensen, 1983), high growth potential(Boone et al., 2007; Linck et al., 2008), and high R&D expenditures(Boone et al., 2007; Coles et al., 2008).

Thus, an optimal board will have more outside independentdirectors and be larger in overall size, when management ‘privatebenefits’ are high and the cost of monitoring is low. Accordingly,Boone et al. (2007) find a statistically significant positive (negative)relation between monitoring ‘private benefits’ (‘monitoring costs’)and board size, but not the same for board independence. Mean-while, Linck et al. (2008) and Lehn et al. (2009) support the nega-tive impact of ‘monitoring costs’ on both board size and boardindependence. Linck et al. (2008) do not explore the effect of mon-itoring ‘private benefits’ on board size, but instead find a statisti-cally significant positive relation between monitoring ‘privatebenefits’ and board independence. Thus, based on this discussion,the second hypothesis related to bank board monitoring require-ments is as follows:

Hypothesis H2. Bank board size and the percentage of inde-pendent directors are positively related to management privatebenefits and negatively related to directors’ monitoring costs.

3.3. CEOs negotiations power and their succession process

The ‘negotiation’ theory illustrates that board structure, partic-ularly board independence, is the outcome of negotiations betweenthe board and CEOs (Hermalin and Weisbach, 1998). That is, boardindependence decreases with CEO’s (negotiation) power and in-creases with constraints on such CEO power (Hermalin and Weis-bach, 1998; Baker and Gompers, 2003; Boone et al., 2007; Linck etal., 2008). CEO (negotiation) power generally derives from theCEO’s perceived ability (relative to a replacement) to influenceboard decisions, as can be proxied by firm performance, CEO ten-ure or CEO ownership. Likewise, CEO’s (negotiation) power canbe constrained by the presence of shareholdings of non-affiliatedblock-holders or outside directors on the board (Boone et al., 2007).

Board independence could also be influenced by the CEO suc-cession process (Hermalin and Weisbach, 1998; Linck et al.,2008). The two most common types of CEO-succession processesare ‘horse races’ and ‘passing the baton’ (Vancil, 1987; Brickleyet al., 1997). Using the ‘horse races’ argument, the firm conductsa tournament among eligible candidates for the CEO position(Brickley et al., 1997). Under the ‘passing the baton’ argument,the board chooses a designated successor for the CEO in advance(Vancil, 1987). This succession process indicates that board inde-

pendence decreases as the CEO approaches retirement (Vancil,1987), as is proxied by CEO age (Linck et al., 2008).

However, for banks – due to statutory and regulatory consider-ations, as mentioned earlier in Section 2 – bank management,including CEOs, neither can influence the directors’ selection pro-cesses nor they have any succession planning in fear of severe pen-alties for any misconduct. For example, on November 24, 2003,Richard M. Thomas, former CEO and President of First NationalBank, was disciplined by OCC with an industry ban, $50,000 in res-titution, transfer of shares, and a $1000 civil money penalty for hismisconduct (OCC No. 2003-106). Likewise, on December 22, 2003,Eduardo Masferrer, former CEO and board chair of Hamilton Bank,was disciplined with an industry ban, $960,000 in restitution, and a$40,000 civil penalty (OCC No 2003-150). Therefore, in contrast tonon-bank firm evidence, this study reasonably predicts that bankboard independence is neither negatively related to CEO negotia-tion power and closeness of CEO retirement, nor positively relatedto constraints on such CEO power. Thus, the third hypothesis re-lated to the CEO negotiation theory and succession process is sta-ted in an alternative format as:

Hypothesis H3. The percentage of independent directors isnegatively related to CEO power and closeness of CEO retire-ment, and positively related to constraints on such CEO power.

3.4. Ownership incentives structure

The board could also reflect the firm’s ‘ownership incentivesstructure’ (Rediker and Seth, 1995; Raheja, 2005; Linck et al.,2008). This ‘ownership incentives structure’ explains that varia-tions in the firm’s ownership structure can also be important inaligning both managers’ and shareholders’ interests (McConnelland Servaes, 1990; Tong, 2008). These incentives may substituteor complement the board as internal governance mechanismsand hence can be considered as a relevant determinant of boardstructure. Raheja (2005) contends that when both shareholderand management incentives are aligned, boards will be smaller.This is because there is a low demand for outside monitors, wheninsiders are less likely to select inferior projects. Linck et al. (2008)use this notion to support the negative relationship between theCEO shareholdings (as proxy for insider incentives) and board size.

Raheja (2005) further argues that greater outside directors’shareholdings raises their incentives to monitor management, aswell as to verify the projects more vigorously because of their in-creased shares of the firm’s profit. These shareholdings reduce pro-ject verification costs, and potentially communication andcoordination costs as well. Raheja’s (2005, Proposition 6, p. 296)model, therefore, specifies that an optimal board is larger andhas a majority of independent directors when verification costsare low. However, Linck et al. (2008) find that outside directors’shareholdings decrease both board size and board independence.They reason that fewer outside directors may be required wheneach director has stronger incentives to monitor management.Likewise, for banks, Rediker and Seth (1995) find that theproportion of outside directors is negatively related to both thenon-affiliated block-holders’ shareholdings and the managerialshareholdings. However, consistent with Raheja’s (2005) proposi-tion for banks, Whidbee (1997) shows that the proportion of out-side directors is negatively related to CEO ownership, whilepositively related to non-affiliated block ownership. Based on thisdiscussion, the fourth hypothesis with regard to ownership incen-tive alignment is as follows:

Hypothesis H4. Bank board size is negatively related to insid-ers’ incentives alignment and positively related to outsiders’incentives alignment.


3.5. CEO duality determinants

The existing theoretical studies do not explicitly address thedeterminants of CEO duality. However, their implications – alongwith existing empirical findings – can nevertheless help in formu-lating some specific expectations regarding CEO duality (e.g., Her-malin and Weisbach, 1998; Linck et al., 2008). As an insider, theCEO possesses firm-specific knowledge that is important for large,complex and diversified banks. Likewise, banks with high monitor-ing costs, i.e., with high information asymmetry, could benefit fromCEO duality. It could also be argued that in the presence of oppor-tunities for insiders to extract private benefits, the CEO and boardchair roles should be separated to achieve a balance between boardindependence and such opportunities. Based on these arguments,the following two Hypotheses H5A and H5B, for bank CEO dualitydeterminants relate to scope of operation and board monitoringrequirements respectively:

Hypothesis H5A. CEO duality is positively related to the bank’sscope of operations.

Hypothesis H5B. CEO duality is negatively related to manage-ment private benefits and positively related to directors’monitoring costs.

Consistent with Hermalin and Weisbach’s (1998) ‘negotiation’theory, as explained in Section 3.3, it can be argued that a CEO withgreater power will also chair the board, so as to gain more power.Similarly, constraints on CEO power will favor separating the CEOand board chair positions. Compatible with Brickley et al. (1997)succession planning theory, CEOs are honored with board chair ti-tle as they approach retirement. In support of these notions, Lincket al. (2008) find that the probability of CEO duality increases withCEO power and closeness of CEO retirement, and decreases withconstraints on CEO power. Thus, Hypothesis H5C for bank CEOduality determinants related to the CEO negotiation theory andsuccession process can be specified as follows:

Hypothesis H5C. CEO duality is positively related to CEO powerand closeness of CEO retirement, and negatively related toconstraints on such CEO power.

9 Any loans should comply with the applicable law, including Regulation ‘‘O” ofBoard of Governors of Federal Reserve and Section 13(k) of the Securities ExchangeAct of 1934.

10 Bebchuk et al. (2009) put forth the entrenchment index as the composite of sixdummy variables: staggered boards, limits to shareholder by-law amendments,supermajority requirements for mergers, supermajority requirements for charteramendments, poison pills and golden parachutes. Definitions of these variables are inthe study by Bebchuk et al. (2009, p. 824). In an unreported correlation between aBebchuk entrenchment index and EINDEX is 0.67 (p-value < 0.01) for a sample of 49banks.

4. Data and empirical method

4.1. Data and sample procedure

The initial sample consists of the largest BHCs headquartered inthe United States with standard industrial classification (SIC) of6021 and 6022 for respective national and state commercial banks,over the period from 1997 to 2004. The data are sourced from DEF14A proxy statements, BANKSCOPE, FR Y-9C, DATASTREAM, andSDC Platinum.

Detailed information on bank board structures is hand collectedfrom DEF 14A proxy statements of annual meetings found in theSEC’s EDGAR filings. Following Adams and Mehran (2009), the gov-ernance data are measured on the date of the proxy statement, i.e.,at the beginning of the respective fiscal year. The data collectionprocedure is then adjusted to account for when the proxies dis-close some governance information for the previous fiscal year(e.g., the percentage of CEO shareholding) and others for the fol-lowing fiscal year (e.g., the number of directors). The financialinformation on BHCs are mostly obtained from the BANKSCOPEdatabase and complemented by the fourth quarter ConsolidatedFinancial Statements for BHCs, i.e., Form FR Y-9C, from the FederalReserve Board. The market information on BHCs is collected fromthe DATASTREAM database. Similarly, the US three-month Trea-sury bill rate in the two-index market model for bank risk compu-

tations is obtained from the Federal Reserve Bank of St. Louis. Theinformation on M&A activities of the sample BHCs over the sampleperiod are obtained from Thomson Financial’s SDC Platinum data-base. The initial sample begins with the three hundred largestBHCs, as ranked by the 2004 year-end book values of total assets.The final sample, an intersection of the data on BHCs with SIC6021 and 6022 in DEF 14A proxy statements, BANKSCOPE, DATA-STREAM, and with a minimum two consecutive years’ data be-tween 1997 and 2004, consists of 1,534 observations on 212 BHCs.

4.2. Measures of variables

The three left-hand side variables, i.e., the board structure vari-ables, explained in this paper are board size (BS), independentdirector (INDIR), and CEO duality (DUAL). BS is the total numberof directors serving on the bank board at the end of each fiscal year.INDIR is the number of independent directors, as a percentage ofthe total number of board directors. Following prior studies (e.g.,Hermalin and Weisbach, 2003; Adams and Mehran, 2009), thisstudy defines independent directors as those whose only businessrelationship with the bank is their directorship. When evaluatingdirector’s independence, following Adams and Mehran (2009),the borrowing and depositing by directors with their BHCs or theirsubsidiaries are also considered.9 DUAL is a dummy variable thatequals one if the CEO also chairs the board, but is otherwise zero.

The measurements of the four explanatory variables (i.e., deter-minants) –related to the testing hypotheses about scope of opera-tions, monitoring requirements, negotiations with CEOs andownership incentives structures – are as follows. Bank ‘scope ofoperations’ is proxied by three variables: bank size, bank age,and revenue diversification index. Bank size (TA) is the bank’s totalassets, and bank age (AGE) is the number of years since a bank wasfirst listed on DATASTREAM. Following Stiroh and Rumble (2006, p.127), bank revenue diversification index (DIVER) is computed asone minus the sum of the squared fraction of operating incomefrom interest and the squared fraction of net operating incomefrom non-interest sources. Prior studies, including Lehn et al.(2009), Linck et al. (2008), and Boone et al. (2007), used thenumber of business segments as a measure of diversification fornon-bank firms. However, for banks, Stiroh and Rumble’s (2006)revenue diversification approach seemed more appropriate be-cause it captures the complexity and the level of diversificationof banks through their income sources. DIVER measures the degreeof diversification in a BHC’s net operating revenue. Prior studies,including Lehn et al. (2009), Linck et al. (2008), and Boone et al.(2007), used the number of business segments as a measure ofdiversification for non-bank firms. However, for banks, Stiroh andRumble’s (2006) revenue diversification approach seemed moreappropriate because it captures the complexity and the level ofdiversification of banks through their income sources.

A shareholder’s restrictive rights index (EINDEX), an approxi-mation of Bebchuk, Cohen and Ferrell’s (2009) entrenchment in-dex,10 is used as a proxy for managers’ opportunities for privatebenefits. EINDEX is computed as the sum of two dummy variables:staggered board (STAGG) and poison pill (POISON). The dummyvariable, STAGG, equals one if the board is staggered; otherwise, it


is zero. The dummy variable, POISON, equals one if the bank boardhas the provision for poison pill, but is otherwise zero. The two prox-ies for directors’ monitoring costs are bank charter value (CV) andbank risk (RISK). CV is calculated following Keeley’s Q (1990) asthe ratio of the market value of total assets to the book value of totalassets, while the market value of total assets is the sum of marketvalue of equity plus book value of liabilities. All these values are mea-sured at the end of fiscal year. It is important to note that CV itself isconsidered to be an important controlling device for banks, as it isaffected by regulation (Jonghe and Vennet, 2008) and hence altersthe need for monitoring by independent directors. High charter va-lue could reduce moral hazard problems in banks because share-holders in banks with high charter values have more to lose if arisky project becomes insolvent. Thus, charter value reduces theneed for independent directors as a monitoring device (Marcus,1984). RISK is calculated as the standard deviation of the bank’s dailystock return for each year.

CEO power is measured by three proxies: the prior period re-turn on average (ROA) which is the net income after tax as a per-centage of average total assets; CEO tenure (CEOT), which is thenumber of years spent as the bank CEO; and the percentage ofthe bank’s total outstanding shares owned by the bank CEO(CEOWN). CEO succession planning is proxied by the CEO’s age inyears (CEOAGE). Likewise, following Boone et al. (2007) and Lincket al. (2008), constraints on CEO power are measured by two vari-ables: outside ownership (OUTSIDEOWN) and non-affiliated blockownership (NBLOCKOWN). OUTSIDEOWN is calculated as thenumber of shares held by the bank officers and directors, excludingthe CEO, as reported in DEF 14A proxy statements as a percentageof the total number of outstanding shares.11 Finally, NBLOCKOWNis calculated as the number of shares owned by non-affiliated blockholders who own 5% or more shares as a percentage of total numberof outstanding shares.12 Non-affiliated block holders exclude insidedirectors and any trust company holding stocks on behalf of thebank’s employee stock ownership plan because the bank managersmay have influence over those trust companies through the con-tractual nature of their relationships.

Several additional variables are included as control variables toreduce biases in the coefficient estimates due to omitted variables.These additional bank-specific control variables include: bank cap-ital ratio, dummies for merger, post-SOX period, and lag of BS andINDIR. The bank capital ratio (CAPITAL) is measured as bank totalequity as a percentage of the bank’s total assets. CAPITAL is ex-pected to positively affect both BS and INDIR because a high capitalratio means a lower level of debt. Debt, such as subordinated debt,is considered to be an important market monitoring mechanism indisciplining bank managers (Flannery, 1998). Hence, given the ab-sence of such monitoring mechanisms, other internal governancetechniques, such as independent directors, may become moreimportant. The dummy for M&A (MERGER) equals one for a bankthat completed any M&A activity during that year, but is otherwisezero. MERGER is included to control for any prior period M&Aactivity (if any) by a bank because any recent M&A activity couldaffect bank board structure (Boone et al., 2007; Adams and Meh-ran, 2009). For instance, the positive association between bankboard size and prior period M&A activity might reflect the addition

11 Boone et al. (2007) use the percentage of total outstanding shares owned byoutside directors, rather than by officers and directors, as a proxy for outsiders’incentive alignment. Board ownership can be a reasonable proxy for outside directorownership because the correlation between these two variables is 0.74 andstatistically significant at 1% level for a random sample of 40 banks.

12 SEC requires banks to disclose those shareholders owning at least five percent ofthe firm’s total outstanding shares in the DEF 14A proxy statement. The shareholdingsof others with less than five percent are also sometimes reported. However, in thedeterminant analysis for BS and BM, NBLOCKOWN is omitted, as it proved to behighly correlated with OUTSIDEOWN (�.1015 with p-value <0.01).

of directors from an acquired bank’s board into the merged/acquir-er bank’s board. The dummy for the post-SOX period (DSOX)equals one if the period is either 2003 or 2004; otherwise, it equalszero. It is included to control for the post-SOX environment andalso to capture the impact of SOX on the board variables. The coef-ficient on DSOX also complements the univariate test results inSection 6, showing if bank board variables changes were statisti-cally significant in the post-SOX period. Finally, following priorstudies (e.g., Boone et al., 2007; Linck et al., 2008), lags of BS andINDIR are included in the respective regression equations to cap-ture the interaction between different board structure variables,i.e., to reduce endogeneity of the variable of interest.

4.3. Empirical models and estimation methods

4.3.1. Empirical modelsThe following three regression equations, Eqs. (1)–(3), are spec-

ified to test formally the determinants hypotheses, respectively, forBS, INDIR, and DUAL, given the theoretical and empirical discussionin Section 3. The hypothesized signs of the equations are shownbeneath the respective variables in the following equations.

lnðBSÞi;t ¼

aþ b1 lnðTAþÞi;t þ b2 lnðAGE

þÞi;t þ b3ðDIVER

þÞi;t

þd1ðEINDEXþ

Þi;t þ d2ðCV�Þi;t þ d3ðRISK

�Þi;t

þc1ðCEOWN�

Þi;t þ c2ðOUTSIDEOWNþ

Þi;t þ f1ðINDIRþÞi;t�1

þf2ðCAPITALþ

Þi;t þ f3 ðMERGERÞi;t�1þ

þu ðDSOXÞi;tþ

þei;t

8>>>>>>><>>>>>>>:

ð1Þ

ðINDIRÞi;t ¼



þÞi;t

þd1ðEINDEXþ

Þi;t þ d2ðCV�Þi;t þ d3ðRISK

�Þi;t

þ/1ðROA�Þi;t�1 þ /2 ðCEOT

�Þi;t þ /3ðCEOWN

�Þi;t

þ/4 lnðCEOAGE�

Þi;t þ k1ðOUTSIDEOWNþ

Þi;tþk2ðNBLOCKOWN

þÞi;t þ f1 lnðBS

þÞi;t�1 þ f2 lnðCAPITAL

þÞi;t

þf3 ðMERGERÞi;t�1þ

þu ðDSOXÞi;tþ

þei;t

8>>>>>>>>>>>>>><>>>>>>>>>>>>>>:

ð2Þ

ðDUALÞi;t ¼



þÞi;t

þd1ðEINDEX�

Þi;t þ d2ðCVþÞi;t þ d3ðRISK

þÞi;t

þ/1ðROAþÞi;t�1 þ /2 ðCEOT

þÞi;t þ /3ðCEOWN

þÞi;t

þ/4 lnðCEOAGEþ

Þi;t þ k1ðOUTSIDEOWN�

Þi;tþk2ðNBLOCKOWN

�Þi;t þ f1ðINDIR

�Þi;t�1

þf2 lnðCAPITALþ

Þi;t þ f3 ðMERGERÞi;t�1�

þu ðDSOXÞi;t�

þei;t

8>>>>>>>>>>>>>><>>>>>>>>>>>>>>:

ð3Þ

where subscripts i denotes individual BHC (i = 1, 2, . . . , 212), t refersto time period (t = 1997, 1998, . . ., 2004), and ln is the natural log-arithm. b, d, /, k, c, f, u, and W are the parameters to be estimated. eis the idiosyncratic error term. The definitions of the variables andthe relevant hypothesized (or expected) signs in regression Eqs.(1)–(3) are as already discussed in Sub-Section 4.2 and summarizedin Table 1.

4.3.2. Estimation methodsFollowing prior studies, including Boone et al. (2007), Coles

et al. (2008), and Linck et al. (2008), the primary estimation meth-od of regression Eqs. (1) and (2) for board size (BS) and indepen-dence (INDIR), respectively, is pooled ordinary least squares(OLS). Due to the binary nature of the variable, regression Eq. (3)for CEO duality (DUAL) is estimated with the maximum-likelihoodLOGIT model. A priori, the variance-covariance matrix in thepooled-OLS estimates will be adjusted with Huber (1964) orWhite’s (1980) estimators, which are robust with respect to heter-oskedasticity. Adopting Petersen (2009) procedure, observationsare also clustered by both panels (i.e., by banks) and time period

Table 1Summary of definitions and predicted signs of variables for Eqs. (1)–(3).

Variables Definitions BS INDIR DUALEquation (1) (2) (3)

Panel A: Dependent variablesBS The number of directors in the BHC’s boardINDIR The percentage of independent directors in the BHC’s boardDUAL A dummy variable that equals 1 if the CEO also chairs the BHC’s board,

otherwise zero

Panel B: Explanatory variablesTA The total assets of the BHC as at the end of each fiscal year b1 (+) b1 (+)AGE The number of years since the BHC was listed in the DATASTREAM database b2 (+) b2 (+)DIVER Stiroh and Rumble’s (2006) ‘revenue diversification index’ which is

calculated as 1 – (squared of fraction of operating income from interest plussquared of fraction of net operating income from non-interest sources)

b3 (+) b3 (+)

EINDEX The sum of two dummy variables: staggered boards, and poison pills. Thedummy for staggered boards equals 1 if the BHC’s board is classified,otherwise zero. The dummy for poison pills equals one if the BHC’s boardhas the poison pill provision, otherwise zero

d1 (+) d1 (+) d1 (�)

CV Keeley’s Q (Keeley, 1990) which is calculated as the sum of the marketvalue of equity plus the book value of liabilities divided by the book value oftotal assets

d2 (�) d2 (�) d2 (+)

RISK The standard deviation of daily BHC’s stock returns in a year d3 (�) d3 (�) d3 (+)ROA The return on average total assets which is calculated as net income after

tax as a percentage of the BHC’s average total assets. Average total assets issimply the average of beginning and end of year BHC’s total assets

/1 (�) /1 (+)

CEOT The number of years the BHC CEO has served in this position /2(�) /2 (+)CEOAGE The BHC CEO’s age in years /3(�) /3 (+)CEOWN The percentage of total outstanding shares owned by the BHC CEO c1 (�) /4 (�) /4(+)OUTSIDEOWN The percentage of total outstanding shares owned by the BHC officers and

directors excluding those of the CEOc2 (+) k1 (+) k1 (�)

NBLOCKOWN The percentage of total outstanding shares owned by non-affiliated block-holders who hold at least 5% of outstanding shares

k2 (+) k2 (�)

Panel C: Other control variablesCAPITAL The BHC’s total equity as a percentage of total assets f2 (+) f2 (+) f2 (�)MERGER A dummy for any M&A, i.e. a dummy variable which equals one for BHC

that made an acquisition in a year, otherwise zerof3 (+) f3 (+) f3 (�)

DSOX A dummy for post-SOX periods, 2003 and 2004, i.e. a dummy variablewhich equals one if the period is either 2003 or 2004, otherwise zero

u (+/�) u (+/�) u (+/�)

This table summarizes the definitions and predicted signs of the variables in regression Eqs. (1)–(3) related to the testing of Hypotheses H11, H2, H3, H4, H5A and H5B. Column1 shows the list of variables and their definitions are in column 2. Columns 3–6 show coefficients along with their predicted signs in regression Eqs. (1)–(3) for thedeterminants of respective BHC board size (BS), percentage of independent directors (INDIR), and CEO duality (DUAL).


to address random unobserved serial and cross-sectional correla-tion respectively (if any) in residuals.

This study applies several measures to reduce endogeneity inthe right-hand side variables. For example, lagged values of bothBS and INDIR are included as instrumental variables in the respec-tive board structure determinants regression equations. As an addi-tional robustness check, simultaneous equation model (three-stageleast square (3SLS)) is also estimated and reported in Section 6.1, inwhich structural models are specified as endogenizing board struc-ture variables. Particularly, Eqs. (1)–(3) for BS, INDIR, and DUALrespectively are all estimated in a simultaneous system. SystemGMM estimates in Section 6.2 are also robust to unobserved heter-ogeneity, simultaneity and dynamic endogeneity (if any).

4.4. Descriptive statistics and correlation analysis

The descriptive statistics for the various board structures, CEOcharacteristics, ownership, and bank-characteristics variables arepresented in Table 2. The board structure variables in Panel A of Ta-ble 2 show that the mean (median) number of bank board direc-tors, BS, is 12.92 (12.00), with a minimum of 5 and a maximumof 31. The mean (median) percentage of independent directors, IN-DIR, of 64.52 (66.67%) is similar to the 66.52% reported by Cornettet al. (2009) for banks. Seventy-four percent of the sample bankshave staggered boards, and thirty-four percent have a poison pillprovision. The mean value of shareholders’ restrictive right index(EINDEX) is 1.08.

In Panel B of Table 2, the descriptive statistics of CEO character-istics indicate that 58% of the sample banks combined both CEOand board chair titles (DUAL). The mean (median) tenure of theCEO, CEOT, is 8.85 (7.00) years. The mean (median) age of theCEO, CEOAGE, is 56.09 (56.00) with consistent with that reportedby Cornett et al. (2009). The mean (median) CEOWN is 4.41%(1.30%). In Panel C of Table 3, within the two ownership variables,the mean (median) OUTSIDEOWN is 10.25% (7.24%), which is com-parable to the 9.63% reported by Anderson and Fraser (2000). Themean (median) NBLOCKOWN is 3.67% (7.24%), which is lower thanthat of 6.64%, as reported by Anderson and Fraser (2000). For brev-ity, the descriptive statistics of ownership and bank-specific vari-ables (in Panels C and D of Table 2) are not described further.

Table 3 presents the Pearson product-moment correlation coef-ficients among the governance and bank-specific variables. It ad-vances some initial guesses about the determinants of boardstructure. For example, the highly positive correlation betweenbank board size (BS) and bank size (TA) indicate that board size in-creases with bank size. It also indicates that the multicollinearitycould be a concern in the multivariate analysis. For instance, TA,DIVER and AGE are highly positively correlated.

4.5. Trends in bank board structure

To place the study in comparison to non-bank studies, Fig. 1 be-low shows the time trends of bank board structure (i.e., board size,percentage of board independence, and CEO duality) from 1997 to

Table 2Descriptive statistics.

Variables Mean SD Min. 1st Quartile Median 2nd Quartile Max. Skew. Kurt.

Panel A: Board structure variablesBS (No.) 12.92 4.54 5 10 12 15 31 0.96 3.83OUTDIR (%) 84.62 8.73 37.5 80 86.96 90.91 100 �1.49 5.94INDIR (%) 64.52 15.72 10 55.56 66.67 75 96.55 �0.58 3.1STAGG 0.74 0.44 0 0 1 1 1 �1.11 2.22POISON 0.34 0.47 0 0 0 1 1 0.7 1.49EINDEX 1.08 0.72 0 1 1 2 2 �0.12 1.91

Panel B: CEO characteristicsDUAL 0.58 0.49 0 0 1 1 1 �0.31 1.1CEOT (Years) 8.85 7.74 0 3 7 14 46 1.18 4.52CEOAGE (Years) 56.09 7.32 34 51 56 60 85 0.4 3.84CEOWN (%) 4.41 8.8 0 0.55 1.3 3.46 65.19 3.72 18.72

Panel C: Ownership structureOUTSIDEOWN (%) 10.25 9.96 0.19 4.04 7.24 13.36 83.32 2.7 14.46NBLOCKOWN (%) 3.67 8.54 0 0 0 5.7 98.5 4.9 38.07

Panel D: Bank-specific variablesTA (in bil.) 23.66 105.78 0.16 1.02 2.07 7.66 1484.10 8.23 81.31CV 1.1 0.07 0.94 1.05 1.09 1.13 1.64 1.82 10.54CAPITAL (%) 9.26 1.9 3 7.99 9.09 10.16 21.59 1.34 7.93DIVER 0.36 0.09 0.06 0.3 0.36 0.43 0.5 �0.53 2.71AGE (Years) 13.73 9.01 1 6 12 22 31 0.46 1.9MERGER 0.11 0.32 0 0 0 0 1 2.45 6.99RISK (%) 2.26 1.2 0.65 1.65 2.02 2.53 17.32 4.81 42.75ROA (%) 1.24 0.51 �6.24 1 1.22 1.47 5.59 �1.01 41.62

This table presents the distribution of variables by showing mean, standard deviation (SD), minimum (Min.), first quartile (1st Quartile), median (Median), second quartile(2nd Quartile), skewness (Skew.), and kurtosis (Kurt.). See Table 1 for variable definitions.

Table 3Correlation matrix.

Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

1 BS 1.00 0.16 0.01 0.08 0.00 �0.01 �0.15 0.08 �0.04 0.38 0.02 �0.02 0.28 0.18 0.05 �0.1 �0.052 INDIR 1.00 0.05 0.14 �0.05 0.01 �0.12 �0.36 0.04 0.33 0.05 0.01 0.17 0.32 0.00 �0.18 0.023 EINDEX 1.00 �0.04 �0.18 �0.17 �0.14 �0.25 �0.05 �0.01 0.03 �0.07 �0.02 0.05 �0.02 �0.12 �0.034 DUALCEO 1.00 0.25 0.22 0.11 �0.19 �0.03 0.24 0.09 �0.07 0.20 0.22 �0.07 �0.04 0.055 CEOT 1.00 0.42 0.18 0.08 �0.06 �0.04 �0.03 0.04 �0.01 0.05 �0.05 �0.02 0.006 CEOAGE 1.00 0.03 0.02 �0.04 0.01 0.01 0.08 �0.05 0.08 �0.02 0.04 0.057 CEOWN 1.00 0.06 0.04 �0.15 �0.13 0.02 0.00 �0.10 �0.07 0.09 0.008 OUTSIDEOWN 1.00 �0.10 �0.34 �0.18 �0.13 �0.19 �0.37 �0.02 0.08 �0.149 NBLOCKOWN 1.00 0.07 0.01 �0.01 0.02 0.09 �0.04 �0.03 0.0410 LNTA 1.00 0.31 �0.07 0.51 0.68 0.12 �0.24 0.2011 CV 1.00 0.17 0.08 0.29 �0.02 �0.13 0.5912 CAPITAL 1.00 �0.07 0.08 0.08 0.01 0.4013 DIVER 1.00 0.39 0.01 �0.18 0.0614 AGE 1.00 0.00 �0.17 0.2315 MERGER 1.00 �0.03 �0.0416 RISK 1.00 �0.1917 ROA 1.00

This table shows Pearson pair-wise correlation matrix. Bold texts indicate statistically significant at 1% level or better. See Table 1 for variables definitions.


2004. The three size groupings are developed by ranking the banksinto quartiles based on their total assets in each year. Then, the firstquartile banks are grouped as small, the second and third quartilesas medium, and the fourth quartile large.

Panel A shows the trend in the mean bank board size. Consis-tent with the non-bank evidence, larger banks have larger boards.Bank boards generally decrease over the sample period, except forduring 1997 and 1998. This decrease is most remarkable for largeand medium banks. Panel B reports the time trend in the mean per-centage of independent directors. The larger banks appear to havemore independent directors than medium and small banks. Thepercentage of independent directors remains relatively flat for bothlarge and medium banks until 2001, and then increases slightly in2004. However, small banks exhibit the most dramatic swing inthe percentage of independent directors over the sample period.

For small banks, it declines from 64.18% in 1997 to 55.66% in2000, and then increases to 61.98% in 2004. Finally, Panel C showsthat as with board size and independence, bank size is an impor-tant determinant of CEO duality, i.e., whether banks combine bothCEO and board chair positions. Similar to non-bank evidence, wecan see that there is no strong time trend in terms of bank CEOduality.

5. Empirical results

Table 4 below reports the results of regression Eqs. (1)–(3). Therelevant diagnostic tests in Panel C of Table 4 are based on pooled-OLS without any robust adjustment for residuals. The average var-iance inflation factors (AVIF) across all the columns indicate that

Panel A: Board Size

10.00

11.00

12.00

13.00

14.00

15.00

16.00

17.00Small Medium Large

Panel B: % Independent Directors

50.00

55.00

60.00

65.00

70.00

75.00Small Medium Large

Panel C: % of Banks with CEO is also Board Chair

0.400.450.500.550.600.650.700.750.800.850.90

1997 1998 1999 2000 2001 2002 2003 2004

1997 1998 1999 2000 2001 2002 2003 2004 1997 1998 1999 2000 2001 2002 2003 2004

Small Medium Large

Fig. 1. Bank board structure trends: 1997–2004. The sample includes 212 BHCs over 1997–2004, i.e. a total bank observations of 1534. The size groups are formed by rankingthe banks into quartiles based on their total assets each year. We label the first quartile banks ‘‘small”, quartiles second and third ‘‘medium” and the fourth quartile ‘‘large”.Panel A, B, and C report the trends in the mean board size, the percentage of independent directors, and the percentage of banks with CEO is also board chair, respectively forbanks.


the multicollinearity among the regressors should not be a concernin estimating the regression equations.13 The White (1980) alterna-tive test for heteroskedasticity (P1) shows statistically significantLM-statistics for each regression, which confirms the presence ofheteroskedasticity with normal OLS estimates. Likewise, thepooled-OLS estimates appear to suffer from first-order serial correla-tion, as indicated by statistically significant F-statistics across allregressions with Wooldridge’s (2006) test for first-order serial corre-lation (P2). The presence of first-order serial correlation in the paneldata also indicates the presence of an ‘unobserved firm-fixed effect’(Wooldridge, 2002, p. 176). These justify the pooled-OLS estimates ofEqs. (1) and (2) and LOGIT estimates of Eq. (3), with Huber (1964) orWhite (1980) heteroskedasticity robust standard errors. The obser-vations are also clustered by both banks and times to control for un-known fixed- and time-effects in the estimates. In Panel B of Table 4,the regression equations are well fitted with adjusted/pseudo R-squared of 27.55%, 21.25%, and 16.74%, respectively, for BS, INDIR,and DUAL regressions, with statistically significant F-statistics.

With regard to Hypothesis 1, as mentioned earlier in Section 4.2,three variables are used to proxy for the bank’s ‘scope of opera-tions’ TA, AGE, and DIVER. Even with the criticism of ‘attenuationbias’ due to the inclusion of multiple proxies for one underlyingvariable, the coefficients on TA and DIVER remain statistically sig-nificant in the BS regression. The use of multiple proxies for anunderlying variable in one regression equation could bias the coef-ficient toward zero, which is commonly known as ‘attenuationbias’ (Lubotsky and Wittenberg, 2006). Similarly, the coefficientson TA and AGE are still positive and statistically significant forthe INDIR regression. More to the point, the Wald (1943) tests –for the joint significance of the coefficients of all the three mea-sures of scope of operations (TA, AGE and DIVER) for BS and INDIRregression (in Panel D of Table 4) – indicate statistically significantF-statistics. Thus, Hypothesis 1 is well-supported and confirms that

13 According to Chatterjee et al. (2000), the guidelines for detecting multicollinearityare: (i) the largest VIF is greater than 10, and (ii) the mean VIF is larger than 1.

bank board size and the percentage of independent directors arepositively related to the bank’s scope of operations.

With regard to the monitoring hypothesis, i.e., Hypothesis 2, inthe BS regression, the coefficient on the EINDEX is positive and sta-tistically significant at the 5% level, and the coefficients on the CVand RISK are both negative but statistically significant for CV at10%. In the INDIR regression, while the coefficient on EINDEX isnot statistically significant, the coefficients on CV and RISK are bothnegative and statistically significant at 10% or better.14 The Wald(1943) tests for the joint significance of these two coefficients (CVand RISK) of the monitoring costs (in Panel D of Table 4) produce sta-tistically significant F-statistics for INDIR regression, but not for BSregression. Thus, the overall monitoring hypothesis (Hypothesis 2)is partially supported, as board size is positively related to the mon-itoring management’s private benefits, while the percentage of inde-pendent directors negatively relate to directors’ monitoring costs.

With regard to the hypothesis related to CEO negotiation andsuccession planning (i.e., Hypothesis 3), as expected, none of thenegative coefficients on all of the three measures of CEO power(lag ROA, CEOT, and CEOWN) and a positive coefficient on theproxy for CEO succession planning (CEOAGE) is statistically signif-icant in the INDIR regression. Similarly, the Wald (1943) test statis-tic for the joint significance of these four variables (lag ROA, CEOT,CEOWN and CEOAGE) is not statistically significant (F-statis-tic = 0.40, p-value = 0.81). Contrary to the prediction, the coeffi-cients on the two proxies for constraints on CEO power,OUTSIDEOWN and NBLOCKOWN, are negative and statistically sig-nificant for the former at the 1% level. The results of OUTSIDEOWNand NBLOCKOWN, however, are consistent with findings by Lincket al. (2008) for non-bank firms and Rediker and Seth (1995) andWhidbee (1997) for banks. They interpret these results as the sub-stitution effects of governance mechanisms, i.e., fewer outsidemonitors are required when each director and non-blockholder

14 In an unreported regression where RISK is the only proxy for monitoring costs,along with other relevant variables, a statistically significant negative coefficient onRISK is indicated.

Table 4Regression results for the determinants of the bank board structure.

Variables Pred. BS (1) Pred. INDIR (2) Pred. DUAL (3)

Panel A: Explanatory variablesTA + 0.087***(6.74) + 0.021*(1.72) + 0.225*(1.68)AGE + -0.048(-1.61) + 0.033*(1.67) + 0.162(0.63)DIVER + 0.596***(3.36) + �0.115(�0.62) + 2.568*(1.68)EINDEX + 0.05**(2.09 + �0.016(�0.74) � �0.039(�0.18)CV � �0.367*(�1.68) � �0.350**(�2.02) + 1.892**(1.97)RISK � �0.005(�0.50) � �0.034*(�1.75) + 0.0413(0.47)ROAt�1 � �0.008(�0.29) + �0.127(�0.70)CEOT � �0.001(�0.58) + 0.069***(3.10)CEOWN � �0.004***(�2.59) � �0.001(�0.72) + 0.035**(1.75)CEOAGE � 0.037(0.30) + 2.416**(2.23)OUTSIDEOWN + 0.009***(4.53) + �0.009***(�5.09) � �0.0339**(�2.53)NBLOCKOWN + 0.0003(0.14) � �0.016(�1.16)BSt�1 + 0.123***(2.77)INDIRt�1 � 0.167***(3.11) + 0.213(0.48)CAPITAL + 0.013*(1.72) + 0.0009(0.14) � �0.144**(�2.16)MERGERt�1 + 0.095***(2.96) + 0.007(0.37) � �0.487**(�2.25)DSOX +/� �0.06***(�3.62) + 0.009*(1.78) +/� �0.301**(�2.32)Constant 1.164***(2.97) 4.042***(8.2) �14.19***(�2.80)

Panel B: Model fitsAdjusted R2/pseudo R2 0.2755 0.2125 0.1674F-stats. (13|16|17, 1309) 55.57***[0.00] 19.81***[0.00] 29.84***[0.00]No. of pooled obs. 1322 1322 1322

Panel C: Regression diagnosticsAVIF (max.) 1.34 (2.25) 1.41 (2.50) 1.39 (2.28)P1: LM-stats (v2 = 2) 13.33*** [0.00] 89.03*** [0.00] 85.71*** [0.00]P2: F-stats (1, 211) 47.94*** [0.00] 29.75*** [0.00] 74.61*** [0.00]

Panel D: Wald test (F stat) for joint significance ofSCOPE, F (3, 1309|1305) 33.72***[0.00] 3.15** [0.02] 6.90*** [0.00]MONCOST, F (2, 1309|1305) 1.63 [0.20] 2.89* [0.06] 1.59* [0.07]INCENTIVE, F (2, 1309) 17.20*** [0.00] N/A N/ACEOPOWER & SUCCESSION, F (4, 1305) N/A .40 [0.81] 6.86*** [0.00]Constraints on CEO power, F (2, 1305) N/A 13.03*** [0.00] 3.67** [0.03]

This table presents the results of the pooled ordinary least squares (OLS) estimates of Eqs. (1) and (2) and LOGIT estimates of Eq. (3). Standard errors in all estimations areclustered by banks. BS is the number of directors on the board. INDIR is the independent directors as a percentage of board size. DUAL is the dummy variable which equals 1 ifthe CEO also chairs the board. TA is the total assets at fiscal year-end. AGE is the number of years since the BHC was listed in the DATASTREAM database. DIVER is the revenuediversification index calculated following Stiroh and Rumble (2006). EINDEX is the sum of the two variables – staggered board and poison pill, and is an approximation ofBebchuk et al. (2009) entrenchment index. CV is the charter value of the bank calculated (following Keeley (1990)) as the book value of total assets plus market value of equityminus book value of equity, all divided by the book value of total assets. RISK is the standard deviation of the bank’s daily stock returns over a year. ROA is the net income aftertax as a percentage of average total assets. CEOT is the number of years the CEO has held this position. CEOWN is the percentage of shares owned by the CEO. CEOAGE is theage of the CEO in years. OUTSIDEOWN is the percentage of shares owned by the directors and top executives excluding CEOWN. NBLOCKOWN is the percentage of sharesowned by non-affiliated persons/institutions with 5% or more of the bank’s equity. CAPITAL is the bank equity as percentage of total assets. MERGER is the dummy variablewhich equals 1 if the bank has any M&A in the period. DSOX is the dummy variable which equals 1 if the period is either 2003 or 2004. AVIF is the average ‘variance inflationfactor’ shows the degree of collinearity problem among the regressors. P1 is the White (1980) test for heteroskedasticity which provides the Lagrange Multiplier (LM)statistics based on alternative procedure explained in Wooldridge (2006, pp. 282–283). P2 is the test for first-order serial correlation which provides an F-statistic based onWooldridge (2002, pp. 282–283). Finally, the Wald (1943) test is used to assess the joint significance of the respective estimates. Figures in parentheses show the robust t-statistics based on standard errors clustered by both bank and year. Figures in brackets present the p-values of the respective F-statistics from the Wald (1943) tests.

* Statistical significance at 10% levels.** Statistical significance at 5% levels.

*** Statistical significance at 1% levels.


has stronger incentives to monitor. Thus, as anticipated, Hypothe-sis 3 is not supported for banks. That is, bank board independenceneither decreases with CEO power and closeness of CEO retire-ment, nor increases with constraints on such power.

With regard to the hypothesis related to ownership incentives(Hypothesis 4) in BS regression, as anticipated, the coefficient onCEOWN is negative and statistically significant at the 1% level.Likewise, the coefficient on OUTSIDEOWN is positive and statisti-cally significant at the 1% level. The Wald (1943) test for the jointsignificance of the two variables, CEOWN and OUTSIDEOWN (inPanel D of Table 4), shows a statistically significant F-statistic of17.20 (p-value <0.01). Thus, Hypothesis 4 is well-supported anddemonstrates that bank board size is negatively related to insiders’incentives alignment and positively related to outsiders’ incentivesalignment.

The results for DUAL regression, i.e. for Eq. (3), show that thecoefficients on all three measures of scope of bank operations, TA,AGE and DIVER, are positive and statistically significant for TA

and DIVER. The statistically significant F-statistics with Wald(1943) test (in Panel D of Table 4) for the joint significance of TA,AGE and DIVER also confirms Hypothesis H5A that the probabilityof CEO duality increases with bank’s scope of operation. However,the coefficient on EINDEX is negative, but not statistically signifi-cant. The coefficient on CV is positive and statistically significantat the 5% level, while the coefficient on RISK is not statistically sig-nificant. Yet, for non-bank firms, Linck et al. (2008) find no relationbetween CEO duality and directors’ monitoring costs. The Wald(1943) test indicates the joint significance of the CV and RISK, asshown in Panel D of Table 4, with an F-statistic of 1.59 (p-value<0.10). These findings lend partial support to Hypothesis H5B thatCEO duality is positively related to the monitoring costs, but notnegatively related to the monitoring of private benefits. Withregard to Hypothesis H5C, the coefficients on all the four mea-sures of both CEO power and succession planning (lag ROA, CEOT,CEOWN and CEOAGE) are positive and statistically significantat the 5% level or greater for CEOT, CEOWN and CEOAGE. As


anticipated, the coefficients on OUTSIDEOWN and NBLOCKOWN,the proxy for constraints on CEO power, are both negative and sta-tistically significant at 5% for OUTSIDEOWN. The statistically signif-icant Wald (1943) test statistics for the joint significance of all fourvariables related to CEO power and succession planning, and con-straints thereof (in Panel D of Table 4) further supports the results.Taken together, Hypothesis H5B is well-supported. That is, banksare more likely to combine both CEO and board chair positions,when the CEO has greater power and closeness to retirement, andconstraints on such power are restricted.

The results for control variables yield further insights into theforces shaping bank boards. The statistically significant positivecoefficient on lag BS in INDIR regression indicates that the percent-age of independent directors increases with the prior period direc-tor numbers (BS). Similarly, the positive and statistically significantcoefficients on lag INDIR in BS regression show that bank boardsize is positively related to the prior period percentage of indepen-dent directors. The statistically significant coefficients on MERGERin BS and DUAL regressions illustrate that any recent M&A activity(MERGER) is associated with larger boards and the lower probabil-ity of combining both CEO and board chair roles. This occurs in caseacquisitions representatives from the target bank’s board are likelyto be added to the acquirer bank’s board, in order to attract the tar-get to accept the bid offer and also to share the board leadership

Table 53SLS regression results for the determinants of the bank board structure.

Variables Pred. sign BS (Eq. (1))

Panel A: Explanatory variablesTA + 0.084***(11.13)AGE + �0.052(�1.11)DIVER + 0.609***(6.03)EINDEX + 0.052***(4.42)CV � �0.319***(2.64)RISK � 0.008(1.13)ROAt�1

CEOTCEOWN � �0.004***(�4.43)CEOAGEOUTSIDEOWN + 0.01***(10.13)NBLOCKOWNBSt�1

INDIRt�1 � 0.26***(8.99)CAPITAL + 0.012***(2.85)MERGERt�1 + 0.096***(3.62)DSOX +/� �0.06***(�3.22)Constant 0.741*(3.88)

Panel B: Model fitsAdjusted R2/pseudo R2 0.2598v2-stats. (13|16|16, 1534) 525.55***[0.00]No. of banks 212No. of pooled obs. 1532

Panel C: Wald test (F-stat.) for joint significance ofSCOPE, F (4|3|3, 211) 90.12***[0.00]MONCOST, F (2, 211) 4.25***[0.01]INCENTIVE, F (2, 211) 63.78***[0.00]CEOPOWER & SUCCESSION, F (4, 211) N/AConstraints on CEO power, F (2, 211) N/A

This table presents the results of the three-stage least squares (3SLS) estimation of the syis the independent directors as a percentage of board size. DUAL is the dummy variable wAGE is the number of years since the BHC was listed in the DATASTREAM database. DIVEREINDEX is the sum of the two variables – staggered board and poison pill, and is an approbank calculated (following Keeley (1990)) as the book value of total assets plus market vaRISK is the standard deviation of the bank’s daily stock returns over a year. ROA is the netthe CEO has held this position. CEOWN is the percentage of shares owned by the CEO. CEOby the directors and top executives excluding CEOWN. NBLOCKOWN is the percentage oequity. CAPITAL is the bank equity as percentage of total assets. MERGER is the dummyvariable which equals 1 if the period is either 2003 or 2004. Finally, the Wald (1943)parentheses show the t-statistics while those in brackets presents the p-values of the re



structure. Finally, the statistically significant coefficients on DSOXin all three regressions suggest that in the post-SOX period, bankboard size decreased, the percentage of independent directors in-creased, and the likelihood of banks combining both CEO and chairtitles decreased.

6. Robustness tests

As mentioned earlier in Section 4.3.2, this paper uses severaladditional tests to check the robustness of the results with respectto potential endogeneity problems in some explanatory variables,cross-sectional dependence in residuals, non-normality in residu-als from outliers (if any), and reducing attenuation bias from multi-ple proxies.

6.1. Results for three-stage least-squares (3SLS)

Following Agrawal and Knoeber (1996), Eqs. (1)–(3) for BS, IN-DIR, and DUAL, respectively, are all estimated in a simultaneoussystem using 3SLS technique and the results are reported in Table 5below.

Column 3 of Table 5 reports the determinants for bank boardsize (BS), as specified by Eq. (1). The findings remain the same as

Pred. sign INDIR (Eq. (2)) Pred. sign DUAL (Eq. (3))

+ 0.014**(1.99) + 0.0430***(3.7)+ 0.035***(2.70) + 0.033(1.47)+ �0.161(1.56) + 0.494***(3.17)+ �0.0187(�.76) � 0.0006(0.03)� �0.303**(�2.46) + 0.365*(1.71)� �0.0329***(�5.07) + 0.007(0.63)� �0.009(�0.51) + �0.005(�0.15)� �0.001(�0.95) + 0.0128***(7.14)� �0.001(-1.24) + 0.006***(4.32)� 0.035(0.57) + 0.471***(4.41)+ �0.009***(�11.08) � �0.007***(�4.63)+ �0.0003(�0.33) � �0.003**(�2.29)+ 0.202***(8.57)

+ 0.047(1.06)+ 0.0002(0.05) � �0.029***(�4.04)+ �0.008(�0.33) � �0.10**(�2.51)+ 0.0112(0.68) +/� �0.06**(�2.14)

3.883***(13.63) �2.295***(�4.42)

0.2059 0.2037403.48***[0.00] 337.35***[0.00]212 2121532 1532

6.86***[0.00] 22.37***[0.00]16.24***[0.00] 2.74*[0.07]N/A N/A3.04 [0.17] 49.72***[0.00]62.36***[0.00] 12.45***[0.00]

stem of simultaneous Eqs. (1)–(3). BS is the number of directors on the board. INDIRhich equals 1 if the CEO also chairs the board. TA is the total assets at fiscal year-end.

is the revenue diversification index calculated following Stiroh and Rumble (2006).ximation of Bebchuk et al. (2009) entrenchment index. CV is the charter value of thelue of equity minus book value of equity, all divided by the book value of total assets.income after tax as a percentage of average total assets. CEOT is the number of yearsAGE is the age of the CEO in years. OUTSIDEOWN is the percentage of shares owned

f shares owned by non-affiliated persons/institutions with 5% or more of the bank’svariable which equals 1 if the bank has any M&A in the period. DSOX is the dummytest is used to assess the joint significance of the respective estimates. Figures inspective F-statistics from the Wald (1943) tests.


with those reported in Table 4 with improve statistical signifi-cance. Column 5 of Table 5 presents the findings for the percentageof independent directors (INDIR). The interpretation of the resultsalso remains the same as those in Table 4, except that the coeffi-cient on DSOX is no longer statistically significant. The findingsfor CEO duality (specified by Eq. (3)) in Column 7 of Table 5 alsodo not show any significant deviations from the results reportedin Table 4. However, the negative coefficient on NBLOCKOWN isnow statistically significant.

Thus, even with direct control for endogeneity using 3SLS, thisstudy finds evidence that the costs and benefits of the board’smonitoring and advising roles could explain differences in bankboard structure (board size, board independence, and CEOduality).

6.2. Results for two-step system generalized method of moments(GMM)

Table 6 below presents Arellano and Bover (1995) and Blundelland Bond (1998) two-step ‘system GMM’ estimates of Eqs ()()()(1)–(3). In the system GMM, first-differenced variables are used asinstruments for the equations in levels and the estimates arerobust to unobserved heterogeneity, simultaneity and dynamicendogeneity (if any). The ‘system GMM’ estimates are obtainedusing the Roodman ‘xtabond2’ module in Stata and Roodman

Table 6Two-step system GMM regression results for the determinants of the bank board structur

Variables Pred. sign BS (1)

Panel A: Explanatory variablesTA + 0.084***(4.39)AGE + �0.052(�1.24)DIVER + 0.48***(2.67)EINDEX + 0.025(0.75)CV � �0.84**(�2.01)RISK � 0.006(0.52)ROAt�1

CEOTCEOWN � �0.007*(�1.89)CEOAGEOUTSIDEOWN + 0.003*(1.72)NBLOCKOWNBSt�1

INDIRt�1 � 0.117*(1.69)CAPITAL + �0.009(�0.67)MERGERt�1 + 0.068***(4.41)DSOX +/� �0.07**(�1.98)Constant 2.28***(3.23)Year dummies Included

Panel B: Model fitsF-statistics (16|20|20, 211) 8.72 [0.00]P1 �3.40***[0.00]P2 �0.22 [0.829]Hansen J-statistics (v2 = 156|60|60) 168.20 [0.239]No. of instruments 174No. of banks 212No. of pooled observations 1322

This table presents the results of the two-step system GMM (OLS) estimates of Eqs. (1)–(3percentage of board size. DUAL is the dummy variable which equals 1 if the CEO also chairthe BHC was listed in the DATASTREAM database. DIVER is the revenue diversification ivariables – staggered board and poison pill, and is an approximation of Bebchuk et al. (2Keeley (1990)) as the book value of total assets plus market value of equity minus bookdeviation of the bank’s daily stock returns over a year. ROA is the net income after tax asthis position. CEOWN is the percentage of shares owned by the CEO. CEOAGE is the age oand top executives excluding CEOWN. NBLOCKOWN is the percentage of shares owned bythe bank equity as percentage of total assets. MERGER is the dummy variable which equal1 if the period is either 2003 or 2004. Finally, P1 and P2 are the test statistics for first-orover-identifying restrictions. Figures in parentheses are t-statistics while p-values are in



(2006) illustrates detail estimation procedure of dynamic paneldata using ‘xtabond2’. The diagnostics tests in Panel B of Table 6show that the model is well fitted with statistically insignificanttest statistics for both second-order autocorrelation in second dif-ferences (P2) and Hansen J-statistics of over-identifying restric-tions. The residuals in the first difference should be seriallycorrelated (P1) by way of construction but the residuals in the sec-ond difference should not be serially correlated (P2). Accordingly,in Panel B of Table 6, we could see statistically significant P1 andstatistically insignificant P2 for all equations. Likewise, the HansenJ-statistics of over-identifying restrictions tests the null of instru-ment validity and the statistically insignificant Hansen J-statisticsfor all equations of board structure determinants indicate thatthe instruments are valid in the respective estimation. Finally,the number of instruments (i.e. 174 for Eq. (1) and 82 for bothEqs. (2) and (3)) used in the model is less than the panel (i.e.212) which makes the Hansen J-statistics more reliable.

The interpretation of the GMM estimates in Panel A of Table 6remains the same as in Table 4 with only few notable differences.In Eq. (1) for BS, the coefficient on EINDEX is no longer statisticallysignificant. The coefficients on both RISK and OUTSIDEOWN are notalso statistically significant in Eq. (2) for INDIR. Likewise, in the Eq.(3) for DUAL, the coefficient on CEOAGE is not statistically signifi-cant while the positive coefficient on CV is now statistically signif-icant. However, these differences do not discredit the overall

e.

Pred. sign INDIR (2) Pred. sign DUAL (3)

+ 0.024*(1.84) 0.069**(2.41)+ 0.053*(1.86) �0.001(�0.04)+ �0.084(�0.43) 0.53*(1.72)+ �0.011(�0.42) � �0.006(�0.10)� �0.553*(�1.70) + 0.191*(1.68)� �0.02(�0.80) + 0.013(0.54)� �0.001(�0.03) + 0.049(1.31)� 0.001(0.04) + 0.0185***(2.61)� �0.0003(�0.07) + 0.001*(�1.67)+ �0.168(�0.60) � 0.486(1.09)+ �0.006(�1.41) � �0.009(�1.26)+ �0.003(�0.81) �0.007(�0.75)+ 0.033*(1.86) �� + �0.059(�0.37)+ �0.008(�0.77) � �0.025*(�1.79)+ �0.002(�0.12) � �0.0457*(�1.69)

0.007*(1.72) �0.153***(�3.47)5.27***(3.90) �1.8984(�0.93)Included Included

2.61 [0.00] 3.99 [0.00]�2.37**[0.018] �3.15***[0.00]�0.54 [0.586] �1.03 [0.304]43.89 [0.941] 58.57 [0.528]82 82212 2121322 1322

). BS is the number of directors on the board. INDIR is the independent directors as as the board. TA is the total assets at fiscal year-end. AGE is the number of years since

ndex calculated following Stiroh and Rumble (2006). EINDEX is the sum of the two009) entrenchment index. CV is the charter value of the bank calculated (followingvalue of equity, all divided by the book value of total assets. RISK is the standarda percentage of average total assets. CEOT is the number of years the CEO has held

f the CEO in years. OUTSIDEOWN is the percentage of shares owned by the directorsnon-affiliated persons/institutions with 5% or more of the bank’s equity. CAPITAL is

s 1 if the bank has any M&A in the period. DSOX is the dummy variable which equalsder and second-order serial correlation respectively. Hansen J-statistics is the test of

brackets.

Table 7Determinants of the board structure for small, medium and large banks.

Variables Pred BS (Eq. (1)) Pred INDIR (Eq. (2))

sign Small Medium Large sign Small Medium Large

SCOPE + .054* .088*** .07* + �0.086* .058** .037*

(1.69) (2.70) �1.74 (�1.87) (2.03) (1.87)EINDEX + �.013 .049* .06* + 0.057 �.026 �.024

(�1.76) �1.68 �1.74 (1.16) (�0.85) (�0.93)MONCOSTS � �.003 .017 .049 � �0.04** �.013* .014*

(�0.11) (0.62) (0.98) (�1.96) (�1.75) (1.67)CEOPOWER � �0.09 �.004 �.023

(�1.58) (�0.15) (�0.98)CEOAGE � 1.02 .13 .12

(0.82) (0.56) (0.55)CEOWN � �.004* �.005* �.011*

(�1.77) (�1.84) (�1.74)OUTSIDEOWN + .009*** .007* .011*** + �0.014*** �.007** �.007***

(2.84) (1.89) (4.07) (�3.53) (�2.32) (�5.80)NBLOCKOWN + 0.013*** .003 �.004***

(4.28) (1.47) (�3.27)BSt�1 + 0.51*** .049 .091

(3.46) (0.87) (1.16)INDIRt�1 +/� .28*** .073 .123

(3.91) (0.81) (0.63)CAPITAL + �.006 .015 .011 + �0.002 .005 �.003

(�0.39) (0.94) (0.48) (�0.12) (0.49) (�0.54)MERGERt�1 + .084 .13*** .15*** + 0.021 .034 .01

(1.57) (4.00) (3.02) (0.38) (1.08) (0.27)DSOX +/� �.07 �.07 �.05 + �0.0003 �.0189 .017

(�0.98) (�1.37) (�1.30) (�0.00) (�0.60) (0.51)Constant 1.32*** 1.96*** 1.82** �1.229 3.56*** 3.577***

(4.34) (4.51) (2.13) (�0.74) (3.92) (3.95)Year dummies Included Included Included Included Included IncludedAdj. R2/pseudo R2 0.1912 0.1023 0.2408 0.3362 0.1249 0.3077F-statistics 4.15*** 3.47*** 8.45*** 5.93*** 1.76** 7.78***

Number of banks 74 130 59 74 130 59No. of pooled obs. 305 672 345 305 672 345

Variables Pred DUAL (Eq. (3))

sign Small Medium Large

SCOPE + .716* .578** �.011*

(1.80) (2.45) (�1.73)EINDEX � �.118 �.163 �.372

(�0.23) (�0.52) (�0.88)MONCOSTS + .469 �.034 �.946***

(1.52) (�0.16) (�3.48)CEOPOWER + 1.23*** .71** 1.061**

(3.00) (2.42) (2.26)CEOAGE + �3.01 �2.07 �.53

(�0.81) (�0.92) (�0.12)CEOWN

OUTSIDEOWN � �.044* �.016 �.076***

(�1.67) (�0.82) (�2.84)NBLOCKOWN � �.03 .016 �.033*

(�1.01) (0.51) (�1.88)BSt�1

INDIRt�1 � 1.23* �.5 1.23(1.74) (�0.78) (0.78)

CAPITAL � .383** �.3*** �.37***

(2.54) (�2.80) (�3.47)MERGERt�1 � �1.14** �.007* �1.2**

(�2.42) (�1.82) (�2.35)DSOX � .234 �.57 �.379

(0.30) (�1.63) (�0.75)Constant 4.88 13.69 2.45

(0.33) (1.45) (0.12)Year dummies Included Included IncludedAdj. R2/pseudo R2 0.3128 0.1214 0.2701F-statistics 3.50*** 1.55* 3.84***

Number of banks 74 130 59No. of pooled obs. 305 672 345

This table presents the results of the pooled ordinary least squares (OLS) estimates of Eqs. (1) and (2) and LOGIT estimates of Eq. (3). Standard errors in all estimations areclustered by banks. BS is the number of directors on the board. INDIR is the independent directors as a percentage of board size. DUAL is the dummy variable which equals 1 ifthe CEO also chairs the board. SCOPE is the PCA factor covering TA, AGE and DIVER. MONCOST is the PCA factor for CV and RISK. CEOPOWER is the PCA factor for lag of ROA,CEOT, and CEOWN. CEOAGE is the age of the CEO in years. OUTSIDEOWN is the percentage of shares owned by the directors and top executives excluding CEOWN.NBLOCKOWN is the percentage of shares owned by non-affiliated persons/institutions with 5% or more of the bank’s equity. CAPITAL is the bank equity as percentage of totalassets. MERGER is the dummy variable which equals 1 if the bank has any M&A in the period. DSOX is the dummy variable which equals 1 if the period is either 2003 or 2004.Figures in parentheses show the robust t-statistics while those in brackets presents the p-values of the respective F-statistics from the Wald (1943) tests.





findings regarding the determinants of board structure in banksgiven bank institutional arrangements as already reported inSection 5.

6.3. Results for other analysis

Other robustness tests include Fama and MacBeth (1973) forcross-sectional dependence, iteratively re-weighted least squares(IRLS) for outliers, and Prais-Winsten (1954) for heteroskedasticity,cross-sectional dependence and first-order serial dependence.

The Fama and MacBeth (1973) estimates of regression Eqs. (1)–(3), are robust to the contemporaneous cross-sectional depen-dence. In the first step, for each year, a cross-sectional regressionmodel is fitted. In the second step, the final coefficients are esti-mated as the average of the first step coefficient estimates. TheFama and MacBeth (1973) estimates are comparable to thepooled-OLS estimates in Table 4, and hence not tabulated. The sta-tistical significances of some estimates have improved. For exam-ple, the positive coefficients on TA and AGE in INDIR regressionEq. (2) are now statistically significant at 1% or better.

Following Coles et al. (2008) and Linck et al. (2008), to resolvedifficulties with non-normality in residuals from outliers, regres-sion Eqs. (1)–(3) are re-estimated using IRLS procedure. IRLS is de-rived from Huber’s (1964) M-estimators (for maximum-likelihood)family and is resistant to any outliers or influential observations.Following Kmenta (1986, pp. 318–320), regression Eqs. (1)–(3)are also estimated with the non-conventional Prais-Winsten(1954) procedure, in which standard errors and the variance-covariance matrix are robust to heteroskedasticity, cross-sectionaldependence and first-order serial dependence. Generally, the re-sults of these two procedures confirm the findings displayed in Ta-bles 4 and 5 with improved statistical significance levels and forthis reason are not in a table.15 The explanatory power of the mod-els now increases substantially, ranging from 22% to 85%.

Finally, as mentioned earlier in Section 5, the use of multipleproxies to measure one variable (such as TA, AGE and DIVER forthe bank’s scope of operations) could bias the coefficient estimatestoward zero (Lubotsky and Wittenberg 2006). This is because thesedifferent proxies could be highly correlated. Therefore, the proxiesfor one variable are combined to form a single factor using princi-pal component analysis (PCA). As such, three new proxy variables,SCOPE, MONCOSTS and CEOPOWER, are created. SCOPE is the prin-cipal factor covering TA, AGE and DIVER, while MONCOST ad-dresses the information contained in CV and RISK andCEOPOWER for lag of ROA, CEOT, and CEOWN. The findings donot deviate with the pooled-OLS estimation of regression Eqs. (1)and (2) and the LOGIT estimation of Eq. (3), with SCOPE, MON-COSTS and CEOPOWER & SUCCESSION in place of multiple proxiesfor the banks’ scope of operations, directors’ monitoring costs, andCEO power (respectively), and hence unreported.

16

6.4. Results for small, medium and large banks

Although TA is incorporated to control for bank size, it could beworth investigating whether the results remain the same fordifferent size groups. Hence, regression Eqs. (1)–(3) are re-esti-mated for three different size groups: small, medium and largebanks. The size groups are formed by ranking the banks into quar-tiles based on their total assets per year. The first quartile banks arelabeled as small, the second and third quartiles as medium, and thefourth quartile as large.

Table 7 reports the results of Eqs. (1)–(3) for these three distinctsize groups. The findings appear to remain the same for SCOPE,

15 The results are available from the author upon request.

MONCOSTS and CEOPOWER across these groups, in relation tobank board structure. However, there are a few significant differ-ences. For example, the positive coefficient on lag INDIR in BSdeterminants (i.e. Eq. (1)) is statistically significant for small banks.Similarly, the negative coefficient on NBLOCKOWN in INDIR deter-minants (Eq. (2)) is statistically significant for small and largebanks, but not the same for medium banks. Non-executive direc-tors’ shareholding appears to lower the probability of CEO dualityfor small and large banks, but not for medium banks, as indicatedby the negative coefficient on OUTSIDEOWN in DUAL regressionEq. (3). In addition, prior merger is associated with larger boardsize for medium and large banks. Taken together, the explanatorypower of the models are comparable for both small and largebanks, rather than for medium banks. In conclusion, the resultsfor this analysis should be cautiously interpreted due to low statis-tical power from the small sample size for each group, compared tonon-bank findings.

7. Impact of reforms with SOX

Several studies have examined the impact of ‘reforms with SOX’on non-financial board structure and firm value (e.g, Linck et al.,2008). Particularly, Linck et al. (2008) show that both board sizeand independence increased during the post-SOX period for non-financial firms. They also demonstrate that the results for boardstructure determinants do not alter notably in the post-SOX. Sim-ilarly, for financial service firms (banks, thrifts institutions, insur-ance and securities), Akhigbe and Martin (2006) find a favorable‘valuation effect’ of SOX for those with more independent boardsand audit committees, financial experts on the audit committees,increased insiders’ incentives and institutional shareholdings.Therefore, though this is not the main focus of the paper, this sec-tion discusses the results for some univariate tests provided toshow whether there are any significant differences in bank boardstructure between the pre- and post-SOX periods. In this regard,both parametric-(paired t-test) and nonparametric tests (Wilcoxonsigned rank sum test) are used. In addition, the researcher hastested in a multivariate framework whether the determinants ofbank board structure are significantly different in the post-SOX.

Table 8 reports on the changes between mean difference tests(paired t-tests) and median difference tests (Wilcoxon signed ranksum test) of the board structure variables between the pre- and thepost-SOX periods. As board variables seem to be less time-varying,banks in 2001 (as pre-SOX period) and 2004 (as post-SOX period)are used to test for these changes.16 The rationale is that somebanks could have adopted SOX provisions prior to 2002, while otherscould have been slow to implement them. Thus, by using only 2001and 2004 as the pre- and post-SOX periods, respectively, the change,if any, should be most notable. The bank board size (BS) decreased by4% (12.93 versus 12.37), with the statistically significant paired t-sta-tistic of �3.326 (p-value <0.01) and Wilcoxon statistic of �2.990 (p-value <0.01). For banks, this decrease in board size could reflect anincrease public awareness and investors’ preference for small boardsbecause the bank size increase was statistically significantly over thesample period. Due to less coordination and communication costs,smaller boards are more efficient than larger boards (Yermack,1996).

It is not surprising to see that the percentage of executive direc-tors on the bank board decreased by 9%, which is statistically sig-nificant at the 1% level (paired t-statistic = �3.094 and Wilcoxonstatistic = �3.436). Similarly, the percentage of independent bank

Similarly, Linck et al. (2008) also considered 2001 and 2004 as pre- and post-SOXperiods, respectively, in their study of the impact of SOX on the non-bank firms boardstructure.

Table 8Univariate results of the changes in bank board structure between pre- and post-SOX periods.

Pre-SOX (2001) Post-SOX (2004) Difference (Post-SOX less Pre-SOX) Paired t-statistics Wilcoxon signed rank sum test -statistics

N 207 207Board size (number) 12.93 12.37 �0.56 �3.326*** �2.990***

Executive directors (%) 16.01 14.56 �1.45 �3.094*** �3.436***

Independent directors (%) 63.40 66.59 3.19 4.665*** 4.263***

CEO duality (dummy) 0.57 0.55 �0.02 �0.666 �0.581Total assets ($ in mil.) 22,683.17 33,495.07 10,811.90 2.794*** 12.121***

This table presents changes in bank board structure variables between pre- and post-SOX periods. The pre- and post-SOX periods are represented by 2001 and 2004,respectively. Paired t-statistics show the significance of the difference in means between two matched pair samples whereas the Wilcoxon signed rank sum test-statisticsshow the significance of the difference in medians between two matched pair samples. The size groups are formed by ranking the banks into quartiles based on their totalassets each year. We label the first quartile banks ‘‘small”, quartiles second and third ‘‘medium” and the fourth quartile ‘‘large”.*** Statistical significance at 1% level.

Table 9Determinants of bank board structure for pre- and post-SOX.

Variables Pred. sign BS (1) Pred. sign INDIR (2) Pred. sign DUAL (3)

Panel A: Explanatory variablesSCOPE + 0.10***(7.12) + 0.027**(2.08) + 0.403***(3.62)EINDEX + 0.03*(1.25) + �0.0167(�0.72) � �0.238(�1.1)MONCOSTS � 0.014(0.94) � �0.0109(�0.57) + �0.122(�0.75)CEOPOWER � �0.0118(�0.67) + 0.513***(2.81)CEOAGE � 0.1099(0.9) + 2.699**(2.17)CEOWN � �0.0047**(�2.12) �OUTSIDEOWN + 0.0087***(3.97) + �0.009***(�4.27) � �0.04***(�2.91)NBLOCKOWN + 0.00048(0.2) � �0.019(�1.21)BSt�1 + 0.138***(2.67)INDIRt�1 + 0.175***(2.93) � 0.1266(0.28)CAPITAL + 0.0067(0.61) + �0.0017(�0.22) � �0.156**(�2.14)MERGERt�1 + 0.1185***(4.91) + 0.0178(0.8) � �0.452*(�1.97)DSOX +/� �.194***(�3.47) + 0.633(1.59) +/� �0.457**(�2.12)SCOPE*DSOX 0.000117(0.01) �0.0004(�0.05) 0.125(1.02)EINDEX*DSOX 0.027*(1.7) 0.0212(1.32) 0.559***(3.06)MONCOST*DSOX 0.0234(1.28) 0.0039(0.2) 0.2105(1.09)CEOPOWER*DSOX �0.0049(�0.55) 0.249*(1.69)CEOAGE*DSOX �0.1624*(�1.67) 0.1853(0.17)CEOWN*DSOX 0.00139(0.73)OUTSIDEOWN*DSOX 0.0028(1.62) 0.00058(0.39) 0.0143(0.99)NBLOCKOWN*DSOX 0.000328(0.28) 0.0027(0.28)CONSTANT 1.489***(5.45) 3.48***(6.6) �10.61*(�1.68)Year dummies Included Included Included

Panel B: Model fitsAdjusted R2/pseudo R2 0.2298 0.1905 0.1699F-statistics (19|23|23, 211) 10.51***[0.00] 4.15***[0.00] 3.29***[0.00]Number of banks 212 212 212No. of pooled obs. 1322 1322 1322

This table presents the results of the pooled ordinary least squares (OLS) estimates of Eqs. (1) and (2) and LOGIT estimates of Eq. (3). Interactions with DSOX terms areincluded in the respective determinants equations. Standard errors in all estimations are clustered by banks. BS is the number of directors on the board. INDIR is theindependent directors as a percentage of board size. DUAL is the dummy variable which equals 1 if the CEO also chairs the board. SCOPE is the PCA factor covering TA, AGEand DIVER. MONCOST is the PCA factor for CV and RISK. CEOPOWER is the PCA factor for lag of ROA, CEOT, and CEOWN. CEOAGE is the age of the CEO in years. OUTSIDEOWNis the percentage of shares owned by the directors and top executives excluding CEOWN. NBLOCKOWN is the percentage of shares owned by non-affiliated persons/institutions with 5% or more of the bank’s equity. CAPITAL is the bank equity as percentage of total assets. MERGER is the dummy variable which equals 1 if the bank has anyM&A in the period. DSOX is the dummy variable which equals 1 if the period is either 2003 or 2004. Figures in parentheses show the robust t-statistics based on standarderrors clustered by both bank and year. Figures in brackets present the p-values of the respective F-statistics from the Wald (1943) tests.




directors (INDIR) increased by 5%, from 63.40% in 2001 to 66.59% in2004 and is statistically significant with the paired t-statistic of4.665 (p-value <.01) and Wilcoxon statistic of 4.263 (p-value<0.01). However, the percentage of banks that combined bothCEO and board chairman titles (DUAL) decreased by 4%, from57% in 2001 to 55% in 2004, but this change is not statistically sig-nificant. In contrast, Linck et al. (2008) reported a statistically sig-nificant decrease in the percentage of non-bank firms with acombined CEO and board chair titles, from 62% in 2001 to 56% in2004.

Table 9 presents changes in the results for the bank board struc-ture determinants between the post-SOX and pre-SOX periods. In

regression Eqs. (1)–(3), the post-SOX dummy, DSOX, interacts witheach of the hypothesized determinants. The results show onlythree notable differences. The coefficients on both EINDEX andits interaction term with DSOX are positive and statisticallysignificant in the BS regression Eq. (1). The Wald (1943) test forthe joint significance of EINDEX and its interaction term with theDSOX also provide statistically significant F-statistics. This indi-cates that EINDEX is associated with larger boards in both periods,but the relation is stronger post-SOX, compared to pre-SOX. Final-ly, the coefficient on CEOAGE is insignificant, but its interactionterms with DSOX in the INDIR regression Eq. (2) is negativeand statistically significant. The Wald (1943) test for the joint


significance of CEOAGE and CEOAGE*DSOX provides statisticallyinsignificant F-statistics. This suggests that CEOAGE do not affectboard independence in both periods, but appear to have some neg-ative impact on INDIR post-SOX. Finally, the results for CEOPOWERand its interaction term in DUAL regression Eq. (3) indicate thatCEOPOWER positively affect bank CEO duality, and this relationis even stronger in the post-SOX. Taken together, there is some evi-dence that bank board structure changed in the post-SOX com-pared to the pre-SOX.

8. Conclusion

This paper investigates the trends and the endogenously chosenbank boards of directors, with specific reference to the costs andbenefits associated with boards’ monitoring and advising func-tions. In particular, the study endeavors to assess whether bankboard structure (board size, composition and CEO duality) is asso-ciated with the bank scope of operation, trade-off between costsand benefits of monitoring, negotiations with the bank CEO andincentives alignments, after controlling for other bank specificcharacteristics.

Using a sample of 212 BHCs over the 1997–2004 period, or 1534bank year observations, this study finds evidence that board struc-ture, even in a regulated industry like banking can be explained byits monitoring and advising roles. Consistent with expectations andexisting non-bank findings, the results demonstrate that larger andmore diversified banks have larger boards, more independentdirectors and combined leadership structure. In the presence ofmanagers’ opportunities to consume private benefits, the studyindicates that banks benefit from larger boards, while banks bene-fit from more independent directors when the costs of monitoringmanagers are low. In contrast to non-bank evidence, it is arguedthat bank managers – including CEOs – do not influence boardselection processes due to constant monitoring by bank regulatorsand fear of severe disciplinary action. Consistent with this view,board independence is found not to be the outcome of negotiationswith bank CEOs. This study also confirms that banks have smallerboards when insiders’ shareholdings are high and the outsiders’shareholdings are low.

The trend analysis shows that bank boards become smaller overthe sample period, specifically for large and medium banks in thepost-SOX period. Small bank board size remains stable over thissample period. The percentage of independent directors increasesremarkably for small banks during the early 2000s. Moreover,there is some evidence that bank board structure changed signifi-cantly in the post-SOX period. For example, the mean percentage ofindependent directors increased with statistical significance in thepost-SOX period, rather than in the pre-SOX period. The averagepost-SOX bank board size is smaller than that of the pre-SOX bankboard.

Taken as a whole, this study provides some evidence that evenin a regulatory industry like banking, the structure of boards isconsistent with the efficiency argument, i.e., banks structure theirboards in a way to maximize shareholders’ wealth. Yet the resultsfor control variables that board size and independence do not re-late to past performance could support the inefficiency argumentput forth by Boone et al. (2007). Therefore, further study on boardstructure determinants is warranted to enhance academic under-standing of this subject. For bank regulators, the findings haveimportant policy implications. For instance, to the extent that bankboard structure adapts to its unique competitive environment, itsuggests that uniform rules or guidelines to reform board gover-nance could prove counter-productive. Thus, bank regulatorsshould cautiously evaluate the effectiveness of SOX and otherrequirements for banks.

Acknowledgements

This paper is prepared from the first empirical chapter of thefirst author’s doctoral thesis at Monash University. The authorswish to thank Barry Williams, Richard Heaney, Renée Adams, ananonymous referee, Ike Mathur (the Editor), Mark Flannery, Mau-reen O’Hara, Benjamin Hermalin, Robert Faff, John Kose, Mark Har-ris, Mohamad Ariff, J. Wickramanayake, Mamiza Haq, PeterVerhoeven, Janice How, Robert Bianchi, Michele Meoli, John Now-land, John Chen, Mohammad Hayat, Colleen Puttee, John Zhang,Alexander Akimov, George Tanewski, Petko Kalev, Noel Gaston,Gulesekaran Rajaguru, Hardjo Koerniadi, Mohamad Belkhir, andseminar/conference participants at Bond University, Deakin Uni-versity, Queensland University of Technology, University of Wes-tern Sydney, Griffith University, 21st Australian Finance andBanking Conference, 2009 Asian Finance Association Conference,and 2009 AFAANZ Conference for their helpful comments. The firstauthor acknowledges the AFAANZ 2009 travel grant to support thisresearch. The usual caveats apply.

References

Adams, R.B., Ferreira, D., 2007. A theory of friendly board. The Journal of Finance 62,217–250.

Adams, R.B., Mehran, H., 2009. Corporate performance, board structure and theirdeterminants in the banking industry. Journal of Financial Intermediation,forthcoming.

Agrawal, A., Knoeber, C.R., 1996. Firm performance and mechanisms to controlagency problems between managers and shareholders. Journal of Financial andQuantitative Analysis 31, 377–397.

Akhigbe, A., Martin, A.D., 2006. Valuation impact of Sarbanes–Oxley: evidence fromdisclosure and governance within the financial services industry. Journal ofBanking and Finance 30, 989–1006.

Alchian, A., 1950. Uncertainty, evolution and economic theory. Journal of PoliticalEconomy 58, 211–221.

Anderson, R.C., Fraser, D.R., 2000. Corporate control, bank risk taking, and the healthof the banking industry. Journal of Banking and Finance 24, 1383–1398.

Arellano, M., Bover, O., 1995. Another look at the instrumental variables estimationof error-components models. Journal of Econometrics 68, 29–51.

Baker, M., Gompers, P.A., 2003. The determinants of board structure at the initialpublic offering. Journal of Law and Economics 46, 569–598.

Bebchuk, L.A., Cohen, A., Ferrell, A., 2009. What matters in corporate governance?Review of Financial Studies 22, 783–827.

Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamicpanel data models. Journal of Econometrics 87, 115–143.

Boone, A.L., Field, C.L., Karpoff, J.M., Raheja, C.G., 2007. The determinants ofcorporate board size and composition: an empirical analysis. Journal ofFinancial Economics 85, 66–101.

Brickley, J.A., Coles, J.L., Jarrell, G., 1997. Leadership structure: separating the CEOand chairman of the board. Journal of Corporate Finance 3, 189–220.

Chatterjee, S., Hadi, A.S., Price, B., 2000. Regression Analysis by Example. John Wileyand Sons, New York.

Chi, D.J., Lee, D.S., 2010. The conditional nature of the value of corporategovernance. Journal of Banking and Finance 34, 350–361.

Coles, J.L., Daniel, N.D., Naveen, L., 2008. Boards: does one size fit all? Journal ofFinancial Economics 87, 329–356.

Cornett, M.M., McNutt, J., Tehranian, H., 2009. Corporate governance and earningsmanagement at large US bank holding companies. Journal of Corporate Finance15, 412–430.

Demirgüç-Kunt, A., Detragiache, E., 2002. Does deposit insurance increase bankingsystem stability? An empirical investigation. Journal of Monetary Economics 49,1373–1406.

Fama, E.F., Jensen, M.C., 1983. Agency problems and residual claims. Journal of Lawand Economics 26, 327–342.

Fama, E.F., MacBeth, J., 1973. Risk, return and equilibrium: empirical tests. Journalof Political Economy 71, 607–636.

Fanto, J.A., 2006. Paternalistic Regulation of Public Company Management: Lessonsfrom Bank Regulation. Brooklyn Law School Legal Studies Paper No. 49 (Viewedon August 29, 2009). <http://ssrn.com/abstract=873667>.

Flannery, M.J., 1998. Using market information in prudential bank supervision: areview of the US empirical evidence. Journal of Money, Credit and Banking 30,273–305.

Galai, D., Masulis, R.W., 1976. The option pricing model and the risk factor of stock.Journal of Financial Economics 3, 53–81.

Harris, M., Raviv, A., 2008. A theory of board control and size. Review of FinancialStudies 21, 1797–1832.

Hermalin, B.E., Weisbach, M.S., 1998. Endogenously chosen boards of directors andtheir monitoring of the CEO. The American Economic Review 88, 96–118.

http://ssrn.com/abstract=873667


Hermalin, B.E., Weisbach, M.S., 2003. Boards of directors as an endogenouslydetermined institution: a survey of the economic literature. FRBNY EconomicPolicy Review 9, 7–26.

Huber, P.J., 1964. Robust estimation of a location parameter. The Annals ofMathematical Statistics 35, 73–101.

Jonghe, O.D., Vennet, R.V., 2008. Competition versus efficiency: what drivesfranchise values in European banking? Journal of Banking and Finance 32,1820–1835.

Keeley, M.C., 1990. Deposit insurance, risk and market power in banking. TheAmerican Economic Review 80, 1183–1200.

Kmenta, J., 1986. Elements of Econometrics, 2nd ed. Macmillan, New York.Lehn, K., Patro, S., Zhao, M., 2009. Determinants of the size and structure of

corporate boards: 1935–2000. Financial Management 38, 747–780.Linck, J.S., Netter, J.M., Yang, T., 2008. A large sample study on board changes and

determinants of board structure. Journal of Financial Economics 87, 308–328.Lubotsky, D., Wittenberg, M., 2006. Interpretation of regressions with multiple

proxies. Review of Economics and Statistics 88, 549–562.Macey, J.R., Miller, J.P., 1993. Corporate stakeholders: a contractual perspective. The

University of Toronto Law Journal 43, 401–424.Macey, J.R., O’Hara, M., 2003. The corporate governance of banks. FRBNY Economic

Policy Review 9, 91–107.Marcus, A.J., 1984. Deregulation and bank financial policy. Journal of Banking and

Finance 8, 557–565.McConnell, J.J., Servaes, H., 1990. Additional evidence on equity ownership and

corporate value. Journal of Financial Economics 27, 595–612.Pathan, S., 2009. Strong boards, CEO power and bank risk-taking. Journal of Banking

and Finance 33, 1340–1350.Petersen, M.A., 2009. Estimating standard errors in finance panel data sets:

comparing approaches. Review of Financial Studies 22, 435–480.

Raheja, C.G., 2005. Determinants of board size and composition: a theory ofcorporate boards. Journal of Financial and Quantitative Analysis 40, 283–306.

Rediker, K.J., Seth, A., 1995. Boards of directors and substitution effects ofalternative governance mechanisms. Strategic Management Journal 16, 85–99.

Roodman, D., 2006. How to do xtabond2: An Introduction to ‘‘Difference” and‘‘System” GMM in Stata. Center for Global Development Working Paper Series.

Shepherd, J., 1992. The liability of officers and directors under the FinancialInstitutions Reform, Recovery and Enforcement Act of 1989. Michigan LawReview 90, 1119–1144.

Stiroh, K.J., Rumble, A., 2006. The dark side of diversification: the case of US financialholding companies. Journal of Banking and Finance 30, 2131–2161.

Tong, Z., 2008. Deviations from optimal CEO ownership and firm value. Journal ofBanking and Finance 32, 2462–2470.

Vancil, R.F., 1987. Passing the Baton: Managing the Process of CEO Succession.Harvard Business School Press, Boston.

Wald, A., 1943. Tests of statistical hypotheses concerning several parameters whenthe number of observations is large. Transactions of the American MathematicalSociety 54, 426–482.

Whidbee, D.A., 1997. Board composition and control of shareholder voting rights inthe banking industry. Financial Management 26, 27–41.

White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and adirect test for heteroskedasticity. Econometrica 48, 817–838.

Wooldridge, J.M., 2002. Econometric Analysis of Cross-Section and Panel Data. MITPress, Boston.

Wooldridge, J.M., 2006. Introductory Econometrics: A Modern Approach, third ed.Thomson/South Western, Mason.

Yermack, D., 1996. Higher market valuation of companies with a small board ofdirectors. Journal of Financial Economics 40, 185–212.

artikel smk (bank)

Documents