the role of boards of directors in corporate governance: a conceptual framework and survey

8/10/2019 The Role of Boards of Directors in Corporate Governance: A Conceptual Framework and Survey

1/53

Journal of Economic Literature 2010, 48:1, 58107http:www.aeaweb.org/articles.php?doi=10.1257/jel.48.1.58

58

1. Introduction

People often question whether corporateboards matter because their day-to-dayimpact is difficult to observe. But when thingsgo wrong, they can become the center of atten-tion. Certainly this was true of the Enron,

Worldcom, and Parmalat scandals. The direc-tors of Enron and Worldcom, in particular,

were held liable for the fraud that occurred:Enron directors had to pay $168 million to

investor plaintiffs, of which $13 million wasout of pocket (not covered by insurance); and

Worldcom directors had to pay $36 million,of which $18 million was out of pocket.1Asa consequence of these scandals and ongoingconcerns about corporate governance, boardshave been at the center of the policy debateconcerning governance reform and the focusof considerable academic research. Because

of this renewed interest in boards, a reviewof what we have and have not learned fromresearch on corporate boards is timely.

Much of the research on boards ulti-mately touches on the question what is therole of the board? Possible answers rangefrom boards being simply legal necessities,

1 Michael Klausner, Bernard S. Black, and Brian R.Cheffins (2005).

The Role of Boards of Directors inCorporate Governance: A ConceptualFramework and Survey

R B. A, B E. H, M S. W*

This paper is a survey of the literature on boards of directors, with an emphasis onresearch done subsequent to the Benjamin E. Hermalin and Michael S. Weisbach(2003) survey. The two questions most asked about boards are what determines

their makeup and what determines their actions? These questions are fundamentallyintertwined, which complicates the study of boards because makeup and actions arejointly endogenous. A focus of this survey is how the literature, theoretical as well asempirical, dealsor on occasions fails to dealwith this complication. We suggest

that many studies of boards can best be interpreted as joint statements about both thedirector-selection process and the effect of board composition on board actions and

firm performance. (JEL G34, L25)

*Adams: University of Queensland and ECGI.Hermalin: University of California, Berkeley. Weisbach:Ohio State University. The authors wish to thank Ji-WoongChung, Rdiger Fahlenbach, Eliezer Fich, John McCon-nell, La Stern, Ren Stulz, and Shan Zhao for helpfulcomments on earlier drafts. The authors are especiallyappreciative of the comments received from three anony-mous referees and the editor, Roger Gordon.


2/53

59Adams, Hermalin, and Weisbach: The Role of Boards of Directors

something akin to the wearing of wigs inEnglish courts, to their playing an activepart in the overall management and controlof the corporation. No doubt the truth liessomewhere between these extremes; indeed,there are probably multiple truths when thisquestion is asked of different firms, in differ-ent countries, or in different periods.

Given that all corporations have boards,the question of whether boards play a rolecannot be answered econometrically as thereis no variation in the explanatory variable.Instead, studies look at differences acrossboards and ask whether these differencesexplain differences in the way firms func-

tion and how they perform. The board dif-ferences that one would most like to captureare differences in behavior. Unfortunately,outside of detailed fieldwork, it is difficult toobserve differences in behavior and harderstill to quantify them in a way useful for sta-tistical study. Consequently, empirical workin this area has focused on structural dif-ferences across boards that are presumed tocorrelate with differences in behavior. Forinstance, a common presumption is that out-side (nonmanagement) directors will behavedifferently than inside (management) direc-tors. One can then look at the conduct ofboards (e.g., decision to dismiss the CEO

when financial performance is poor) withdifferent ratios of outside to inside direc-tors to see whether conduct varies in a sta-tistically significant manner across differentratios. When conduct is not directly observ-able (e.g., advice to the CEO about strategy),

one can look at a firms financial performanceto see whether board structure matters (e.g.,the way accounting profits vary with the ratioof outside to inside directors).

One problem confronting such an empiri-cal approach is that there is no reason tosuppose board structure is exogenous;indeed, there are both theoretical argumentsand empirical evidence to suggest boardstructure is endogenous(see, e.g., Hermalin

and Weisbach 1988, 1998, and 2003). Thisendogeneity creates estimation problems ifgovernance choices are made on the basis ofunobservables correlated with the error termin the regression equations being estimated.In fact, one of our main points in this survey isthe importance of endogeneity. Governancestructures arise endogenously because eco-nomic actors choose them in response to thegovernance issues they face.2

Beyond the implications endogeneity holdsfor econometric analysis, it also has implica-tions for how to view actual governance prac-tice. In particular, when we observe whatappears to be a poor governance structure,

we need to askwhythat structure was chosen.Although it is possible that the governancestructure was chosen by mistake, one needsto give at least some weight to the possibilitythat it represents the right, albeit poor, solu-tion to the constrained optimization problemthe organization faces. After all, competitionin factor, capital, and product markets shouldlead, in Darwinian fashion, to the survivalof the fittest. While admittedly fittest doesnot mean optimal, anything that was sub-optimal for known reasons would be unfitinsofar as there would be pressure to addressthese reasons for suboptimality. In other

words, existing suboptimality is unlikely tolend itself to quick or obvious fixes.

This insight about endogeneity is, however,easy to forget in the face of data. Figure 1shows a plot of two data points.3 On the

2 Harold Demsetz and Kenneth Lehn (1985) were

among the first to make the general point that governancestructures are endogenous. Others who have raised itinclude Charles P. Himmelberg, R. Glenn Hubbard, andDarius Palia (1999), Palia (2001), and Jeffrey L. Coles,Michael L. Lemmon, and J. Felix Meschke (2007). Thepoint has also been discussed in various surveys of theliterature; consider, e.g., Sanjai Bhagat and Richard H.Jefferis (2002) and Marco Becht, Patrick Bolton, and AilsaRell (2003), among others.

3 Figure 1 is presented for illustrative purposes andshould not be read as a critique of any existing research.In particular, none of the studies discussed below are asnaive as figure 1.


3/53

Journal of Economic Literature, Vol. XLVIII (March 2010)60

horizontal axis is an attribute of governance(e.g., board size). On the vertical axis is ameasure of financial performance. One firmhas more of the attribute but weaker perfor-mance, while the other firm has less of theattribute but better performance. A regres-sion line through the points underscores theapparent negative relation between attributeand performance. Without further analysis,one might be tempted to conclude that a firm

would do better if it shrank the size of itsboard. The problem with such a conclusion

is that it fails to considerwhya large boardmight have been chosen in the first place.

Figure 2 replicates figure 1, but it alsoshows the optimization problems faced bythe two firms in question. Observe that,for a

given firm, there is a nonmonotonic relationbetween the attribute and financial perfor-mance. In particular, the relation is concaveand admits an interior maximum. Moreover,each of the two firms is at its maximum.

Consequently, whereas Firm 2 would preferceteris paribus to be on Firm 1s curve, it isntand, thus, would do worse than it is doing ifit were to shrink its board in line with thenaive conclusion drawn from the regressionin figure 1.

Figures 1 and 2 illustrate another issueconfronting the study of governance, namelyheterogeneity in the solutions firms choose fortheir governance problems.4 As illustrated,

4 To be sure, a real empirical study would attempt,

in part, to control for such heterogeneity by putting inother controls, including, if the data permitted, firm fixedeffects. It should be noted, however, that (i) there can stillbe a problem with the specification if the attribute entersinto the specification only linearly (as opposed to nonlin-early as suggested by the parabolas in figure 2); and (ii)if different firms face differently shaped trade-offs (e.g.,if the parabolas arent the same shape for all firms), thenthe coefficients on the attribute, its square, etc., will varyacross firms, suggesting a random-coefficients approach iswarranted. See Hermalin and Nancy E. Wallace (2001)and Bhagat and Jefferis (2002) for a discussion of some ofthese methodological issues.

Figure 1. Relation between a Specific Firm Attribute and Firm Financial Performance

Financial

performance

Governance

attribute

Firm 1

Firm 2


4/53


Firms 1 and 2 face different governanceproblems and, not surprisingly, are drivento different solutions. Almost every model ofgovernance shows that the equilibrium out-come is sensitive to its exogenous parameters;consequently, heterogeneity in those param-eters will lead to heterogeneity in solutions.Moreover, once one takes into account vari-ous sources of nonconvexity, such as thosearising in optimal incentive schemes, onemay find that strategic considerations lead

otherwise identical firms to adopt differentgovernance solutions (see, e.g., Hermalin1994).

Some help with the heterogeneity issuecould be forthcoming from more theoreti-cal analyses. Although a commonand notnecessarily inaccurateperception of theliterature on corporate governance, particu-larly related to boards of directors, is that itis largely empirical, such a view overlooks a

large body of general theory that is readilyapplied to the specific topic of boards. Forinstance, monitoring by the board wouldseem to fit into the general literature onhierarchies and supervision (e.g., Oliver E.

Williamson 1975; Guillermo A. Calvo andStanislaw Wellisz 1979; Fred Kofman andJacques Lawarre 1993, Jean Tirole 1986;Tirole 1992). As a second example, issues ofboard collaboration would seem to fit intothe general literature on free-riding and the

teams problem (see, e.g., Bengt Holmstrom1982).

The teams-problem example serves toillustrate a problem that can arise in apply-ing off-the-shelf theory to boards. It is wellknown that, as a members share of a teamsoutput falls, he or she supplies less effort. Forboards, however, the question is not asingledirectors effort, but what happens to totaleffort (e.g., are larger boards less capable

Figure 2. The Real Decisions Faced by the Firms

Financial

performance

Attribute

Firm 1s optimization problem

Firm 2s optimization problem


5/53


monitors because of the teams problem)?Yet, here, theory cannot provide a definitiveanswerwhether total equilibrium effortincreases or not with board size dependscritically on assumptions about functionalforms.5 While anything goes conclusionscan be acceptable in an abstract theoreticalmodel, they are often less than satisfactory inapplied modeling. The lack of clear definitivepredictions in much of the related generaltheory is, therefore, a hindrance to model-ing governance issues. Conversely, if a spe-cific model makes a definitive prediction,then one can often be left wondering if it isan artifact of particular assumptions rather

than a reflection of a robust economic truth.A second, related point is that, in a sim-

ple, and thus tractable, model, theory can betoo strong; that is, by application of sophis-ticated contracts or mechanisms, the par-ties (e.g., directors and CEO) can achieve amore optimal outcome than reality indicatesis possible. To an extent, that problem can befinessed; for instance, if one restricts atten-tion to incomplete contracts. But as othershave noted, the assumption of incompletecontracts can fail to be robust to minorperturbations of the information structure(Hermalin and Michael L. Katz 1991) or theintroduction of a broader class of mecha-nisms (Eric Maskin and Tirole 1999).

A further issue is that corporations arecomplex, yet, to have any traction, a modelmust abstract away from many features ofreal-life corporations. This makes it difficultto understand the complex and multifaceted

solutions firms use to solve their governanceproblems. For instance, the optimal gov-ernance structure might involve a certain

5 For instance, if a teams total benefit is n=1N

en,where en is the effort of agent n, each agent gets 1/Nof the benefit, and each agentns utility is (m=1

N em)/N

(en+1)/( + 1), then total equilibrium effort is N(1/N)1/,

which is increasing in N if > 1, decreasing in N if (0, 1), and constant if = 1.

type of board, operating in a certain fashion,having implemented a particular incentivepackage, and responding in certain ways tofeedback from the relevant product and capi-tal markets. To include all those features ina model is infeasible, but can we expect theassumption of ceteris paribus with respect tothe nonmodeled aspects of the situation to bereasonable? The constrained answer arrivedat by holding all else constant need not rep-resent the unconstrained answer accurately.

Yet another point, related both to theprevious point and to our emphasis onissues of endogeneity, is that, motivated byboth a desire to simplify and to conform to

institutional details, the modeler is oftentempted to take certain aspects of the gov-ernance structure as given. The problem

with this is that the governance structureis largely endogenous; it is, in its entirety,the solution reached by economic actors totheir governance problems. Of course, cer-tain features, such as the necessity of havinga board of directors, can largely be seen asexogenous (although it should be remem-bered that the decision to make a companya corporation rather than, say, a partnershipis itself endogenous). Furthermore, the tim-ing of events, particularly in the short run,can make it reasonable to treat some aspectsof the governance structure as exogenous forthe purposes of investigating certain ques-tions theoretically.

In this survey, we focus primarily on workthat illustrates the sorts of challenges dis-cussed above, papers that help clarify the

nature of board behavior, or that use novelapproaches. We also attempt to put the workunder the same conceptual microscope,namely how should the results be interpretedin light of governance structures being thesecond-best solution to the governance prob-lems faced by the firm. Our focus is also onmore recent papers, even if they are not yetpublished, because prior surveys by KoseJohn and Lemma W. Senbet (1998) and


6/53


Hermalin and Weisbach (2003) cover manyestablished papers in this field. Although weaim to be comprehensive, it would be impos-sible to discuss every paper in light of therecent explosion in the literature on boards.6Of necessity, we omit many interestingpapers in this area and we apologize to theirauthors in advance. For a more detaileddiscussion of the event-study evidence sur-rounding board appointments, we refer thereader to David Yermack (2006). M. AndrewFields and Phyllis Y. Keys (2003) review themonitoring role of the board, as well as theemerging literature on board diversity (seealso David A. Carter, Betty J. Simkins, and

W. Gary Simpson 2003; Kathleen A. Farrelland Philip L. Hersch 2005; and Rene B.Adams and Daniel Ferreira 2009 on boarddiversity). We do not directly discuss direc-tor turnover; Hermalin and Weisbach (2003)review some of the relevant literature onthis topic.7For the sake of brevity, we do notdiscuss the literature on boards of financialinstitutions. Because this is a survey of cor-

porate boards, we also do not discuss theliterature on boards of organizations such asnonprofits and central banks. Partly becauseof the difficulty in obtaining data, this lit-erature is less developed than the literatureon corporate boards (William G. Bowen1994 discusses some of the similarities anddifferences between corporate and non-corporate boards).8 Similar data limitationsrestrict us to a discussion of boards of pub-licly traded corporations. Finally, we do not

6After searching the literature, we estimate that morethan 200 working papers on boards were written in thefirst five years since Hermalin and Weisbach (2003) pub-lished their board survey (no causal link is implied).

7 See also Eliezer M. Fich and Anil Shivdasani (2007),Tod Perry and Shivdasani (2005), and Yermack (2004), forsome recent work in this area.

8Also see Hermalin (2004) for a discussion of howresearch on corporate boards may inform the study ofuniversity and college boards. James O. Freedman (2004)discusses the relation between universities and collegesboards and their presidents.

consider studies that compare governanceinternationally.

Although this survey primarily consid-ers the economics and finance literatures,boards are a subject of interest in manyother disciplines, including accounting, law,management, psychology, and sociology.9

While there is an overlap in these literatures,there are also differences. For instance, theeconomics and finance literatures focus hastraditionally been on the agency problemsboards solve or, in some instances, create. Incontrast, the sociological and managementliteratures also emphasize that boards can(i) play a role in strategy setting and (ii) pro-

vide critical resources to the firm, such asbuilding networks and connections (see, e.g.,Sydney Finkelstein, Hambrick, and AlbertA. Cannella 2009). Some of the topics previ-ously in the domain of other disciplines arebeginning to be of interest in the econom-ics and finance literature. For example, thisliterature has begun to incorporate issues ofexpertise, trust, diversity, power, and net-

works into their analyses.10

The next section considers the questionof what directors do. The section following,section 3, considers issues related to boardstructure. Section 4 discusses how boardsfulfill their roles. Section 5 examines the

9 Some examples of this broader literature includeLucian A. Bebchuk and Jesse M. Fried (2004), Ada Demband F. Friedrich Neubauer (1992), Anna Grandori (2004),Donald C. Hambrick, Theresa Seung Cho, and Ming-Jer Chen (1996), Jay W. Lorsch (1989), Myles L. Mace

(1971), Jeffrey Pfeffer (1972), Mark J. Roe (1994), JamesD. Westphal and Edward J. Zajac (1995), Westphal (1999),and Zajac and Westphal (1996).

10 For research from an economic perspective on direc-tor diversity, see Carter, Simkins, and Simpson (2003),Farrell and Hersch (2005), and Adams and Ferreira(2009). Director expertise is discussedinfrain section 2.3.Some aspects of power related to boards are captured inHermalin and Weisbach (1998); see Raghuram G. Rajanand Luigi Zingales (1998) for a more general economicanalysis of power in organizations. See Asim Ijaz Khwaja,Atif Mian and Abid Qamar (2008) for work on the value toa firm created by its directors social networks.


7/53


literature on what motivates directors. Weend with some concluding remarks.

2. What Do Directors Do?

To understand corporate boards, oneshould begin with the question of what dodirectors do?11

2.1 Descriptive Studies

One way to determine what directors dois to observe directors; that is, do field work.There is a large descriptive literature onboards (e.g., Mace 1971; Thomas L. Whisler

1984; Lorsch 1989; Demb and Neubauer1992, and Bowen 1994).

The principal conclusions of Mace werethat directors serve as a source of adviceand counsel, serve as some sort of discipline,and act in crisis situations if a change inCEO becomes necessary (p. 178). The natureof their advice and counsel is unclear.Mace suggests that a board serves largelyas a sounding board for the CEO and topmanagement, occasionally providing exper-tise when a firm faces an issue about whichone or more board members are expert. YetDemb and Neubauers survey results findthat approximately two-thirds of directorsagreed that settingthe strategic direction ofthe company was one of the jobs they did(p. 43, emphasis added).12 Eighty percentof the directors also agreed that they were

11 This question is distinct from the question of whatshould directors do? This second question is answered,in part, by the legal obligations imposed by corporate law(both statute and precedent), having to do with fiduciaryobligations (see, e.g., Robert C. Clark 1986, especiallychapters 3 and 4).

12 It is important to note that the Demb and Neubauersurveys and questionnaires sample very few Americandirectors (4.2 percent). The top four nationalities surveyedby them are British (29.6 percent), German (11.3 percent),French (11.3 percent), and Canadian (9.9 percent). Overall43.7 percent of their respondents come from common-lawcountries.

involved in setting strategy for the com-pany (p. 43). Seventy-five percent of respon-dents to another of Demb and Neubauersquestionnaires report that they set strategy,corporate policies, overall direction, mission,

vision (p. 44). Indeed far more respondentsagreed with that description of their job thanagreed with the statements that their jobentailed oversee[ing], monitor[ing] top man-agement, CEO (45 percent); succession,hiring/firing CEO and top management(26 percent); or serving as a watchdog forshareholders, dividends (23 percent).

The disciplinary role of boards is alsounclear from descriptive studies. Perhaps

reflecting the period he studied, Mace sug-gests that discipline stems largely from theCEO and other top management knowingthat periodically they must appear before aboard made up largely of their peers (p. 180).Lorsch takes an even dimmer view, suggest-ing that boards are so passive that they offerlittle by way of discipline (see, especially, p.96). Demb and Neubauers statistics seembroadly consistent with this view, as less thanhalf of their respondents agree that their

job is to oversee, monitor top management,CEO and less than a quarter agree thattheir job is to serve as a watchdog for share-holders, dividends (p. 44).

On the other hand, it has been suggestedthat the board passivity described by Maceand Lorsch is a phenomenon of the past.For instance, Paul W. MacAvoy and Ira M.Millstein (1999) suggest that boards haverecently become less passive; that is, they

have evolved from being managerial rubber-stamps to active and independent monitors.MacAvoy and Millstein provide statisticalevidence in support of that conclusion, find-ing that CalPERS grading of a firms boardprocedures is positively correlated withaccounting-based measures of performance.Another piece of evidence consistent withthe view that boards have become tougheris that CEO dismissal probabilities have


8/53


been trending upward (see Mark R. Huson,Robert Parrino, and Laura T. Starks 2001 forevidence over the period 1971 to 1994 andsee Steven N. Kaplan and Bernadette A.Minton 2006 for more recent evidence).

2.2 The Hiring, Firing, and Assessment ofManagement

One role that is typically ascribed todirectors is control of the process by

which top executives are hired, promoted,assessed, and, if necessary, dismissed (see,e.g., Richard F. Vancil 1987 for a descriptiveanalysis and Lalitha Naveen 2006 for statis-tical evidence).

Assessment can be seen as having twocomponents, one is monitoring of what topmanagement doesand the other is determin-ing the intrinsic abilityof top management.The monitoring of managerial actions can,in part, be seen as part of a boards obliga-tion to be vigilant against managerial mal-feasance. Yet, being realistic, it is difficultto see a board actually being in a positionto detect managerial malfeasance directly;at best, a board would seem dependent onthe actions of outside auditors, regulators,and, in some instances, the news media.Indirectly, a board might guard againstmanagerial malfeasance through its choiceof auditor, its oversight over reportingrequirements, and its control over account-ing practices.

The principal focus of the literature onassessment, at least at a theoretical level, hasbeen on the question of how the board deter-

mines managerial ability and what it does withthat information.13One strategy for studyingthe question of ability assessment has beenthe adaptation of Holmstroms (1999) model,

which analyzes agency and monitoring when

13 Typically, the CEO is a member of the board. Instating the CEO is at odds with the board, we are, likethe literature, using the board as shorthand for the boardminus the CEO.

agents have career concerns, to boards. Withinthat approach, authors have focused on howthe assessment of ability relates to the powerof the CEO (e.g., Hermalin and Weisbach1998); to the selection of projects and strategy(e.g., Silvia Dominguez-Martinez, Otto H.Swank, and Bauke Visser 2008); to the processof selecting the CEO (e.g., Hermalin 2005);among other issues.

2.2.1 Assessment, Bargaining Power, andCEO Control

The first article to apply Holmstromsframework to boards was Hermalin and

Weisbach (1998). In their model, there is

an initial period of firm performance underan incumbent CEO. Based on this perfor-mance, the board updates its beliefs aboutthe CEOs ability. In light of these updatedbeliefs, the board may choose to dismiss theCEO and hire a replacement from the poolof replacement CEOs or it may bargain withthe incumbent CEO with regard to changesin board composition and his future salary.The board, then, chooses whether to obtainan additional, costly signal about CEO abil-ity (either that of the original incumbent ifretained or the replacement if hired).14Basedon this signal, if obtained, the board againmakes a decision about keeping or replacingthe CEO. If replaced, a (another) CEO isdrawn from the pool of replacement CEOs.Finally, second- (and final-) period profits arerealized, with the expected value of the prof-its being a positive function of the then-in-charge CEOs ability.

The boards inclination to obtain an addi-tional signal is a function of its independence

14 An alternative, but essentially equivalent, model-ing strategy for this stage would be to assume the boardalways receives the additional signal, but the board hasdiscretion over the informativeness of the signal, withmore informative signals being costlier to the board thanless informative signals. See the discussion in Hermalin(2005) on this matter.


9/53


from the CEO.15The boards independence atthat stage will depend on the outcome of thebargaining game between the board and theincumbent CEO if he is retained.16Becausethe acquisition of the additional signal canonly increase the risk of being dismissed andthe CEO enjoys a noncontractible controlbenefit, the CEO prefers a less independentboard; that is, a board less likely to acquirethis additional signal. The board, however,prefers to maintain its independence. Whenthe CEO has bargaining powerspecifi-cally when he has demonstrated that hes arare commodity by performing welltheboards independence declines. Intuitively,

a CEO who has shown himself to be aboveaverage bargains on two dimensions: hecan bargain for more compensation and,because he prefers to remain CEO ratherthan be fired, the degree of the boards inde-pendence. At any moment in time, given itsmarginal rate of substitution between firmperformance and disutility of monitoring, aboard views itself as optimally independent(i.e., the directors view any change in theircomposition that may lead to more or less dil-igence in monitoring as moving it away fromthe incumbent boards optimum).17 Hence,a local change in independence representsa second-order loss for the board (the topof the hill is essentially flat), whereas as anincrease in the CEOs salary is a first-orderloss (the marginal cost of a dollar is alwaysa dollar). The board, therefore, is more will-

15 Independence is a complex concept. With respectto monitoring the CEO, one imagines that directors whohave close ties to the CEO (e.g., professionally, socially,or because the CEO has power over them) would findmonitoring him more costly than directors with fewer ties(although see Westphal 1999 for an opposing view). Wediscuss independence at lengthinfra.

16 Hermalin and Weisbach assume there is sufficientcompetition among potential replacement CEOs forthe position that a replacement CEO has no bargainingpower. Their model would be robust to giving a replace-ment CEO some bargaining power as long as it was lessthan that enjoyed by an incumbent CEO who is retained.

ing to budge on the issue of independence(willingness to monitor) than salary, at leastinitially; hence, there is movement on inde-pendence. So a CEO who performs well endsup facing a less independent board. The flipside is that a CEO who performs poorly is

vulnerable to replacement.Malcolm Baker and Paul A. Gompers

(2003), Audra L. Boone et al. (2007), andHarley E. Ryan and Roy A. Wiggins (2004)each find evidence consistent with the ideathat successful CEOs are able to bargain forless independent boards. Boone et al. findthat variables that are reasonably associated

with bargaining power either for the board

or the CEO are significant and have theright sign. In particular, measures of CEObargaining power, tenure, and the CEOsshareholdings, are negatively correlated withboard independence. The tenure findings, inparticular, are precisely what the Hermalinand Weisbach model predicts. Measuresthat indicate that the CEO has relatively lessbargaining power, including outside directorownership and the reputation of the firmsinvestment banker at the time of its IPO, areall positively correlated with board indepen-dence. Similarly, Baker and Gompers findthat measures that reflect the CEOs bar-gaining power, including an estimate of theCEOs Shapley value and the reputation of thefirms venture capitalists, have the predictedsigns (negative for the former and positive forthe latter) with respect to the percentage of

17 For instance, if the boards actions are deter-mined by a median-voter model, then the incum-bent median director (voter) knows that moni-toring will be optimal from her perspective ifthere is no change in board composition. If, how-ever, composition changes so that she is no longer themedian director, then the level of monitoring will nolonger be optimal from her perspective. Provided,though, that the tastes of the new median directorrange on a continuum from the incumbent mediandirectors, then having a new median director withonly slightly different tastes than the incumbentrepresents a second-order loss for the incumbent.


10/53


non-inside directors on the board. At oddswith the Hermalin and Weisbach model andunlike Boone et al., Baker and Gompers findapositivealbeit statistically insignificantrelationship between CEO tenure and per-centage of non-inside directors. Finally, Ryanand Wiggins find that a CEOs pay becomesless linked to equity performance as his con-trol over the board increases (proxied by histenure and the proportion of insiders). Theseauthors interpret these findings as consistent

with the Hermalin and Weisbach bargainingframework, because it suggests that as CEOsbecome more powerful, they use this powerto improve their well-being (e.g., as here,

where this power allows them to reduce thevolatility of their compensation).

Baker and Gompers, Boone et al., andRyan and Wiggins are all sensitive to the issuethat governance structures are endogenous.Baker and Gompers, in particular, provide aconvincing solution to the problem by iden-tifying plausible instruments for the endoge-neous variable in their specification, venturecapital financing, these instruments beingthe state of operation and a time dummy thatcaptures exogenous capital inflows to ven-ture capital funds. Yet none of these paperssheds light on whether successful CEOs areable to bargain for a less independent board

within the same firm, because they all relyon analyses of repeated cross-sections ofdata rather than panel data with firm fixed-effects. Shedding more direct empirical lighton the dynamic nature of the CEO-boardrelationship within firms remains an inter-

esting topic for future research.2.2.2 Assessment and Project Selection

Dominguez-Martinez, Swank, and Visser(2008) is a model similar to Hermalin and

Weisbach (1998). A key difference betweenthe two is that, in Dominguez-Martinez,Swank, and Visser, it is the CEO who deter-mines what information the board learns.An interpretation of Dominguez-Martinez,

Swank, and Vissers model is that there aretwo possible types of CEO, good and bad.In each of two productive periods, a CEOdraws a project at random from a distribu-tion of different projects (conditional onCEO ability, each periods draw is an inde-pendent event). Think of each project beingsummarized by its net present value (NPV).The difference between the two types ofCEOs is that the distribution of projects(distribution of NPVs to be precise) is betterfor the good type than the bad type (e.g.,the good types distribution dominates thebad types in the sense of first-order stochas-tic dominance).

The CEO sees the stamped NPV on theproject he draws, whereas the board doesnot. In the second (final) period, the CEOsincentives are such that he implements theproject he draws if and only if it has a positiveNPV. In the first period, however, the CEOsincentives are possibly misaligned with thatof the shareholders: the CEO values keepinghis job. If his first-period actions or perfor-mance lead the directors to infer he is thebad type and the board is not committed toretain him, then he will be dismissed as it isbetter to draw again from the pool of CEOsthan to continue to the second period witha CEO who is known to be bad. The CEOsconcern about retaining his job makes ittempting, therefore, for him to avoid risk tohis reputation by not pursuing even positiveNPV projects in the first period.

One potential solution would be for theboard to commit to retain the first-period

CEO for the second period. With that com-mitment, CEOs would choose only positiveNPV projects in the first period. This, how-ever, is not necessarily optimal because thedirectors are throwing away the option toreplace the CEO if they infer he is likely tobe bad. That is, as is also noted in Hermalinand Weisbach, the ability to replace a CEOa board infers is probably bad creates a valu-able real option for the firm.


11/53


Given that good-type CEOs are morelikely to have positive NPV projects than badtypes, an alternative strategy for the board

would be to commit to dismiss the CEOonly if he doesnt undertake a project. This,however, is not without cost because now aCEO could be willing to undertake a nega-tive NPV project if it is not so bad that thedisutility resulting from pursuing the projectoutweighs his utility from retaining his job.18

Under this governance rule, some number ofnegative NPV projects will be pursued.

A third strategy might be for the boardto commit to keep the CEO only if he

18 Dominguez-Martinez, Swank, and Visser assume aCEOs first-period utility function is + , where isthe returns from the first-period project, > 0 is his ben-efit of keeping his job, and {0, 1} indicates whether heloses or keeps his job, respect ively.

undertakes a positive NPV project. Thismight seem optimal, insofar as it avoids neg-ative NPV projects and allows some learn-ing, but could nevertheless be suboptimal:how much is learned about the CEOs abil-ity depends on the relative likelihood of thetwo types having projects with a particularNPV. It is possible, therefore, that if a givenNPV is more likely from a good type thana bad type, then it could be worth having

that project undertaken even if the NPV isnegative because seeing the project provides

valuable information about the CEOs abil-ity. Conversely, if a given NPV is more likelyfrom a bad type than a good type, then itcould be worthwhile dismissing the CEOfollowing the realization of the project evenif its NPV is positive. Figure 3 illustrates.Purely from the perspective of optimal infer-ence, the board should retain a CEO if he

Figure 3. Illustration of the Dominguez-Martinez, Swank, and Visser (2008) Model

Notes: The probability density functions over NPV are shown for the two types. From an informational per-spective, the CEO should be retained if and only if the realized value of a project is abovevi. If, however,v0denotes the project with an NPV =0, then the board, to limit first-period losses, may wish to commit to retainthe CEO if and only if the realized value is above some cutoff strictly betweenviandv0.

vi

v0

NPV

Bad type density Good type density


12/53


has a project with an NPV aboveviand dis-miss him otherwise. If, howevervi< 0 =v0,then this cutoff implies first-period costs.Trading off these first-period costs againstthe value of information, the board may

wish to set a cutoff, vc, between vi and v0;that is, a CEO keeps his job if and only if heundertakes a project and that project paysoff at leastvc.

Dominguez-Martinez, Swank, and Visserobserve that their model offers a possibleexplanation for why evidence of poor deci-sion making does not always lead to CEOdismissal. Sometimes it is optimal to let aCEO pursue a bad strategy rather than

stick to the status quo (i.e., better to pur-sue a negative NPV project rather than donothing) because the information revealedfrom that course of action allows the boardto update positively about the CEOs ability.Admittedly, as formulated here, the samemodel would also explain the dismissal ofa CEO after moderate success if moderatesuccess is more associated with low abilitythan high ability.19 Dominguez-Martinez,Swank, and Vissers model also suggests anexplanation for why new CEOs rarely seemto be riding with training wheels whenit comes to managing their companies.Limiting a CEOs range of action, while per-haps a way to avoid risky mistakes, also lim-its how much the board can learn about hisability. Especially early in his career, whenrelatively little is known, the expected valueof information can outweigh the expectedcost of mistakes.

2.2.3 Assessment and CEO Selection

Hermalin (2005) is concerned with thefact that information is more valuable whena board is seeking to infer the ability of a

19 Dominguez-Martinez, Swank, and Visser do notmake this point. This is one of the ways our interpretationof their model could be said to differ from their actualmodel.

relatively unknown CEO than that of amore established veteran. The reason is thatthe option to dismiss a poorly performingCEO and hire a new one is like an exchangeoption. Consequently, its value is greater,ceteris paribus, the greater is the amountof uncertainty. Hermalin builds on thisinsight to examine the relationship betweena boards structure and its propensity to hirea new CEO from the outside (an externalhire) versus from the inside (an internalhire). Presumably an internal hire is a bet-ter-known commodity than an external hire,meaning that an external hire offers greateruncertainty and, thus, a greater option

value. An external hire is, therefore, morevaluable ceteris paribus. How much morevaluable, however, depends on the degree towhich the board will monitor the CEO (itsdegree of diligence). Like the Hermalin and

Weisbach (1998) model, the board makes adecision as to how intensively it will monitorthe CEO, which is reflected in the probabil-ity it will get an additional signal correlatedwith his ability.20Without the signal, there isno option value. Consequently, the value ofuncertainty about a new CEO is greater themore diligent the board (i.e., the more likelyit is to acquire the signal) and, therefore, amore diligent board is more willing to tradeoff other attributes for greater uncertaintythan is a less diligent board. Hermalinargues that this insight offers an explana-tion for why there has been a growing trendtoward both more external hires and shorterCEO tenures: Due to increased pressure

from institutional shareholders, more gov-ernment regulations, greater threats oflitigation, and new exchange requirements,boards have become more independent and

20 Alternatively, and essentially equivalently, the signalis always observed, but its precision is an increasing func-tion of the boards efforts at monitoring. See section 6 ofHermalin.


13/53


diligent.21 Hence, boards are more willingto monitor, which raises the likelihood theyhire externally for the CEO position.22Moremonitoring directly raises the likelihood ofCEO dismissal and indirectly raises it if itleads firms to hire CEOs about whom less isknown.

One response of CEOs to this greater mon-itoring pressure is for them to work harder(which could be interpreted as taking lessperquisites). Both because they are led to

work harder and their jobs are less secure,

21 See Huson, Parrino and Starks (2001) and Stuart L.Gillan and Starks (2000) for evidence on trends towardgreater board independence (technically, boards with agreater proportion of outside directors) and the rise ofinstitutional investors.

22 See Kenneth A. Borokhovich, Parrino, and TeresaTrapani (1996), Huson, Parrino, and Starks (2001), andJay Dahya, John J. McConnell, and Nickolaos G. Travlos(2002) for evidence that the proportion of new CEO hiresthat are external has been increasing; the last providesevidence for this trend outside the United States.

CEOs will demand greater pay in compen-sation. Hence, a consequence of more inde-pendent boards over time could be upwardpressure on CEO compensation.23Figure 4summarizes Hermalins model.24

23As Hermalin notes, the positive correlation betweenboard independence and CEO pay in time series neednot imply a positive correlation in the cross section at anypoint in time. Hermalin sketches an extension of his modelthat would predict anegativecorrelation in cross section,despite a positive correlation over time. See his section 5.

24 It is worth noting that Hermalin is not the only theo-

retical explanation for the trend toward more externalhires and greater CEO compensation. Kevin J. Murphyand Jn Zbojnk (2006); Murphy and Zbojnk (2004)offer a non-boards-based model that takes as its mainpremise that there has been a decline in the value ofmanagers firm-specific knowledge relative to the value oftheir general knowledge. As they show, this will increasethe willingness of firms to hire CEOs externally. GivenMurphy and Zbojnks modeling of the CEO labor mar-ket, this greater willingness to go outside translates into arise in CEO compensation. Hermalin discusses how hismodel can be extended to incorporate the Murphy andZbojnk model, see his section 6.

Greaterpropensityto monitor

Greatervalue ofoption More

externalhires

Moreeffort

Greatercompensation

Shorteraveragetenures

More risk more

incentive

More uncertainty morelikely firedmore

likely to bediscovered lowability

Compensationrequired for

greater effort

Compensationrequired forless job

security

Moreindependentboards

More risk more

incentive

Figure 4. A Graphical Summary of the Hermalin (2005) Model


14/53


2.2.4 Other Assessment Models

A number of other papers examine themechanisms associated with the boardsassessment of the CEO. Clara Grazianoand Annalisa Luporini (2005) also has aboard that seeks to determine CEO ability.Critical to their analysis is the presence of alarge shareholder on the board, one who is

willing to bear the cost of monitoring, butwho also gains private benefits if the com-pany pursues certain strategies (projects).Because only the large shareholder willmonitor, they find there can be advantagesto a dual-board system (e.g., as in much of

continental Europe) because it may beadvantageous to divorce the monitoring rolefrom the power to have a say over the com-panys strategy. David Hirshleifer and Anjan

V. Thakor (1994) assume that boards alwaysreceive signals useful to assessing the CEOsability, but boards differ insofar as some arelax and some are vigilant. Vigilant boardsmay choose to fire the CEO on the basis ofa bad signal. The situation in Hirshleifer andThakor is complicated by the possibility of atakeover bid by an outside party with inde-pendent information about the firm; con-sequently, it may behoove a vigilant boardnot to act on its own information, but waitto see what information can be learned bythe presence (or not) of a takeover bid andthe price bid. This article also exemplifies thefact that board governance is only one sourceof managerial discipline and, more specifi-cally, it captures the notion that internal and

external monitoring can serve as substitutesor complements. Vincent A. Warther (1998)presents another model in which the boardacquires information about managerial abil-ity. Here, unlike the other models weve dis-cussed, each director gets a private signaland aggregation of information is costly inso-far as a director who indicates he received anegative signal is at risk of losing his boardseat if he proves to be in the minority.

A recent strand of the literature hasrecognized that the boards monitoring ofthe CEO can create, in effect, a dangerof opportunism or holdup by the board.25The ability to dismiss the CEO after he hasmade firm-specific investments means theboard can appropriate some of the CEOsreturns, thereby diminishing his originalinvestment incentives. Two papers in thisstrand are Andres Almazan and JavierSuarez (2003) and Volker Laux (2008). Inboth, two critical assumptions are (i) initialcontracts between board and CEO can berenegotiated and (ii) at least some kindsof boards (strong in Almazan and Suarez,

independent in Laux) cannot commit to notbehaving opportunistically or aggressivelyin renegotiation.

In Almazan and Suarez, after being hired,a CEO can, at personal cost, take a discreteaction that raises, by a discrete amount, theprobability that a given strategy or project

will succeed. This action is observable by theboard, but not verifiable, which creates anopportunity for later holdup. After the CEOtakes (sinks) his action, a profitable oppor-tunity for the firm may arise that requires anew CEO to exploit. If the board is strongenough to fire the incumbent CEO in favorof a new CEO, then the board can use thatpossibility to obtain salary concessions fromthe incumbent because losing his job meanshe loses a private benefit. The threat of beingforced to make such concessions can under-mine the CEOs initial incentive to take thecostly action.

To be more concrete, consider a variationon Almazan and Suarezs idea:26 Supposethat the new opportunity has the same

25 Opportunism and holdup problems have beenstudied in a large number of areas of economics sinceWilliamson (1975, 1976).

26 The actual Almazan and Suarez (2003) model ismore complex than what we present here. While thosecomplications lead to a richer and more nuanced analysis,they are not necessary to get the basic idea across.


15/53


expected payoff as keeping the incumbentCEO if he took the action and, thus, ahigher expected payoff than keeping him ifhe didnt take the action. Suppose a weakboard will never fire the CEO when theexpected value of keeping him equals thatof the new opportunity, but can fire him

when the latter is greater. A strong board isalways capable of firing the CEO. Assumeit is possible, when the threat to dismiss theCEO is credible, for the board to capture,in renegotiation, the CEOs private benefitof control and push the CEO to some reser-

vation utility (call it 0). Hence, a CEO witha strong board has no incentive to take the

action: If the new opportunity doesnt arise,he retains his job no matter what he did,there is no renegotiation of his compensa-tion, and he enjoys the control benefit. Butif the new opportunity does arise he gets 0regardless of his action; either he is fired,thus denied both pay and private benefit,or through renegotiation is forced down toa 0 reservation utility (payoff). Because hisultimate payoff is independent of his action,he has no incentive to incur the cost of tak-ing it. The story is, however, different for aCEO who faces a weak board. Now, he isstrictly better off if he has taken the actionand the new opportunity arises: the boardcannot threaten to fire him, so he contin-ues to capture rents (wage plus private ben-efit). If he didnt take the action and thenew opportunity arose, then he would loseboth wage and private benefit. If the newopportunity arises with low frequency, so it

is efficient for the incumbent CEO to takethe action, then having a weak board will bebetter than having a strong board.

In Almazan and Suarez, the distinctionbetween strong and weak boards is a dis-tinction about their bargaining power. InLaux (2008), the board always has all thebargaining power at the renegotiation stage(can make a take-it-or-leave-it offer to theCEO), but boards differ in their degree of

independence. This variation in degree ofindependence acts, however, like a shiftin bargaining power. Consequently, forreasons similar to those in Almazan andSuarez, a firm can be better off with a lessindependent board than a more indepen-dent board.

2.2.5 Additional Empirical Analyses ofAssessment

There is both anecdotal and statisticalevidence that boards dismiss poorly per-forming CEOs. Based on interviews, Mace(1971) and Vancil (1987) conclude that

boards fire, albeit often reluctantly, poorlyperforming CEOs. There are numerousstatistical analyses that show poor perfor-mance, measured either as stock returnsor accounting profits, positively predicts achange in the CEO.27Simply documenting arelationship between poor performance andan increased probability of a CEO turnover,although suggestive of board monitoring, isnonetheless far from conclusive. After all, asense of failure or pressure from sharehold-ers could explain this relationship. To bet-ter identify the role played by the board,

Weisbach (1988) interacts board composi-tion and firm performance in a CEO turn-over equation. His results indicate thatwhen boards are dominated by outsidedirectors, CEO turnover is more sensitiveto firm performance than it is in firms with

27A problem facing empirical work is that firms oftenoffer a face-saving rationale for a change in CEO (e.g., hewishes to spend more time with his family) rather thanadmit the CEO was forced out for doing a bad job. SeeJerold B. Warner, Ross L. Watts, and Karen H. Wruck(1988), Weisbach (1988), Parrino (1997), and Dirk Jenterand Fadi Kanaan (2008) for further discussions of thisissue and strategies for dealing with it. To the extentnon-performance-based CEO turnover is random, itsimply adds noise to turnover regressions, thus reducingthe power of such tests, but leaves them unbiased andconsistent.


16/53


insider-dominated boards.28 This result isconsistent with the predictions of Hermalinand Weisbach (1998) and Laux (2008) underthe presumption that outsider domination isa good proxy for board independence.

Yermack (1996) also seeks to relate boardstructure to CEO turnover. Instead of aninteraction between board composition andperformance, Yermack interacts the log ofboard size with financial performance andfinds a positive and significant coefficient onthis interaction term.29 That the coefficientis positive indicates that firms with smallerboards have a stronger relationship betweenpoor performance and CEO turnover than do

firms with larger boards. This finding is con-sistent with the often-heard view that smallerboards are more vigilant overseers of the CEOthan larger boards. In particular, in responseto poor performance, they may not be para-lyzed by free-riding or otherwise plagued

with inertia in the way that larger boards are.Another of Yermacks findings (supported

by later work by Theodore Eisenberg, StefanSundgren, and Martin T. Wells 1998) isthat board size and firm performance, thelatter measured by average Tobins Q, are

negatively correlated.30 It is not obvious

28 Dahya, McConnell, and Travlos (2002); Dahya andMcConnell (2007) find a similar result in the UnitedKingdom: firms that adopted the recommendations ofthe Cadbury Commission show a greater sensitivity ofCEO turnover to performance than nonadopting firms.Related, Vidhan K. Goyal and Chul W. Park (2002) findthat the sensitivity of CEO turnover to performance isless when the CEO also serves as board chair. Adams

and Ferreira (2009) find that the proportion of women onboards increases the CEO performance-turnover sensi-tivity even after controlling for the proportion of outsidedirectors, which suggests that the proportion of femaleoutside directorsdirectors outside of the old-boy net-workis proxying for board independence.

29 See Olubunmi Faleye (2003) for a similar study.30 Average Tobins Qis the ratio of the market value of

assets to their book value. A presumption in the literatureis that Q > 1 is partially a reflection of the good job man-agement is doing. As long as one controls for book value ofassets, Tobins Q regressions are similar to market-valueregressions.

how to reconcile Yermacks results with therenegotiation-based models discussed previ-ously: These models suggest that too vigilant(here, small) a board is detrimental to a firminsofar as it discourages the CEO from tak-ing valuable actions or it means such actionscan be implemented only at greater cost.Yermacks findings could also be at odds withHermalin and Weisbachs (1998) bargaining-based model: If larger boards are less vigi-lanteffectively less independentthen thelogic of the Hermalin and Weisbach modelsuggests a successful CEO will bargain toincrease the size of his board. This would

yield a prediction consistent with Yermacks

interaction effect: larger boards will be lessresponsive to a signal of poor performancethan smaller boards. However, because itis the more successful CEOs who have thelarger boards, the Hermalin and Weisbachmodel would seem to predict that firms

with larger boards would outperform thosewith smaller boards, which is contrary toYermacks findings.

It may be possible to reconcile Yermacksfinding with the Hermalin and Weisbachmodel if (i) a successful CEO is a CEO thattook successful advantage of valuable growthopportunities his firm had; and (ii) the timeit takes to recognize the CEO was success-ful is sufficiently long that his firm would bemature at the time it is recognized, leadingto a lower Q.31

Such issues led Coles, Naveen D. Daniel,and Naveen (2008) to reestimate Yermack,but with greater attention to heterogeneity

issues. Consistent with the spirit of figure 2and the conceptual framework set forththere, Coles, Daniel, and Naveen seek tocontrol for the possibility that boards havedifferent sizes because firms face differentproblems. In contrast to Yermacks findings,Coles, Daniel, and Naveen find that firm

31 The authors thank Ren Stulz for this insight.


17/53


performance (average Tobins Q) is increas-ing in board size for certain types of firms,namely those that are highly diversified orthat are high-debt firms.

Perry (1999) breaks down the cross-sec-tional relationship between CEO turnoverand firm performance by whether the out-side directors are paid using incentives. Hefinds that the relationship between CEOturnover and firm performance is stronger

when boards have incentives. This findingsuggests that providing explicit incentivesto directors leads them to be more vigilant(act more independently). Beyond incentivereasons, another potential explanation is the

following: in firms that make use of incen-tive pay for directors, the directors have aprofessional rather than a personal relation-ship with the CEO and, thus, are relativelyindependent of him.

To conclude this section, it is worth not-ing that few analyses of CEO turnover con-trol for firm-specific heterogeneity usingfirm effects. As increasingly long panel-datasets become available, future research willbe able to shed more light on within-firmchanges in CEO turnover.

2.3 Setting of Strategy

In addition to making decisions concern-ing the hiring and firing of CEOs, boardsmay also be involved in the setting of strat-egy or, somewhat equivalently, the selectionof projects. Certainly surveys of directorssee the discussion of Demb and Neubauer(1992) aboveindicate that directors believe

themselves to be involved in setting strategy.2.3.1 Theory

To an extent, many of the models discussedabove could be modified to make themabout boards oversight of strategy. Insteadof replacing the CEO, the board compelshim to change strategy. In an adaptation ofAlmazan and Suarez (2003) or Hermalinand Weisbach (1998), the CEO could be

assumed to have an intrinsic preference forthe incumbent strategy versus a replacement(the incumbent strategy provides, e.g., moreopportunity to consume perquisites). In anadaptation of Laux (2008), similar results

would follow if one assumed the financialreturns to the replacement strategy are inde-pendent of the CEOs initial actions.

An alternative modeling approach is toinvestigate the choice of strategy as a gameof information transmission: the CEO (ormanagement more generally) has differentpreferences than the board concerning proj-ects (strategies). A number of observers arecoming to the view that information trans-

mission between the board and the CEOis important for good governance (see, e.g.,Holmstrom 2005). This is particularly true

when the CEO has payoff-relevant privateinformation, insofar as an agency problemarises because the CEO can influence theboards decision through the strategic releaseof information.

Adams and Ferreira (2007) build amodel based on four broad assumptions:(i) the CEO dislikes limits on his actions(loss of control); (ii) advice from the boardraises firm value without limiting a CEOsactions; (iii) the effectiveness of the boardscontrol and the value of its advice are bet-ter the more informed the board is; and (iv)the board depends crucially on the CEOfor firm-specific information. In the Adamsand Ferreira model, the board can learn theamount, a [0, 1], by which a project shouldbe optimally adjusted (e.g., what the appro-

priate level of investment in it should be).The board can do this, however, only if theCEO has informed them about the project. Itis assumed the CEO can withhold that infor-mation, but if he chooses to share it, thenhe must do so honestly (i.e., using the stan-dard terminology of the contracts literature,the information is hard). The CEO has abias, b > 0, such that he likes to increasethe size of projects (e.g., invest more than is


18/53


19/53


loss from the size of the project sometimesbeing distorted (i.e., in those states whenthe CEO retains control). The Adams andFerreira model also implies that it may beoptimal to separate the advisory and moni-toring roles of the board; that is, to have adual board system as in many countries inEurope.

Milton Harris and Artur Raviv (2008)is similar in spirit to Adams and Ferreira.Harris and Raviv assume that the CEO andthe insider directors, like the outside direc-tors in Adams and Ferreira, have informationrelevant to the quadratic loss. The payoffs,net of fixed terms and additively separable

aspects of their respective utilities, are

UO = (saOaI)2 and

UI = (saOaIb)2,

where the subscripts Oand Idenote outsid-ers and insiders, respectively, and at is theinformation that the t group of directorshave about the optimal size of the project.Observe, now, that the optimal size from theshareholders perspective iss = aO+ aI. The

value of at is the private information of thet group of directors. Unlike in Adams andFerreira, now it could be suboptimal, fromthe shareholders perspective, to give controloversto the outsiders: although the insiders

will almost surely not choose the optimal sgiven control, they might get closer if theirinformation is particularly valuable (i.e., the

variance of aI is relatively big). Harris and

Raviv consider two board structures: outsidercontrol and insider control. When group thas control, it has the choice of choosing sor delegating the choice to the other group.

When group tmakes the choice it receivesa message from the other group about thatother groups information. As in Adams andFerreira, the equilibria of these cheap-talkgames do not permit full information rev-elation. When the insiders information is

sufficiently valuable relative to the outsid-ers (i.e., Var(aI)/Var(aO) > 1, a con-stant that depends on parameter values) andinformation is valuable relative to the agencyproblem (specifically, Var(a

I

)/b2 > 1, a constant that depends on parameter val-ues), then insider control is superior to out-sider control. If those conditions arent met,then outsider control is superior.

Like Adams and Ferreira and Harris andRaviv, Charu G. Raheja (2005) wishes tounderstand board structure in the light of theboards need to obtain information about thefirms projects or strategies. Unlike Adamsand Ferreira, where all board members are

equally ignorant, or Harris and Raviv, whereboth inside and outside directors respectivelyhave private information, Raheja assumesthat only the inside directors possess privateinformation. In contrast to most of the litera-ture, Raheja departs from the idea that thenon-CEO inside directors and the CEO havecoincident incentives. Insiders control theCEO through the threat of ratting him outto the outsiders, who will then join with theinsiders in firing the CEO, should the CEOmisbehave.

Although a clever model, it is difficultto reconcile Rahejas model with the evi-dence in Mace (1971) or Vancil (1987).Insubordination by a CEOs managementteam seems exceedingly rare. Moreover,

what evidence there is about whistle-blow-ers (rats) is hardly encouraging for Rahejasmodel. Anecdotal evidence, at least, suggeststhat whistle-blowers tend to suffer, more

than be rewarded, for their actions (see, e.g.,Joann S. Lublin, 2002). Evidence of whistle-blowers going to outside directors is rarethe most prominent recent whistle-blower,Enrons Sherron Watkins, for instance wentto the CEO (Ken Lay) with her concerns.

Fenghua Song and Thakor (2006) alsoconsider information transmission relevantto project selection. Like some other work inthis area, they build on the career-concerns


20/53


notions of Holmstrom (1999). Unlike previ-ous work, they assume that both the boardand the CEO have career concerns. UnlikeHolmstrom, who assumes all actors areequally ignorant about theirs and othersabilities, Song and Thakor assume that boththe CEO and board each know their ownabilities. In the Song and Thakor model,CEO ability means how likely the CEO isto identify a project to undertake; whereasboard ability means how accurate the boardis at assessing the value of any project putforth by the CEO. Independent of his ability,the CEO also obtains a signal of a projectsquality, which he can pass along truthfully

or not to the board. Song and Thakor showthat when the probability of good projectsis low, then the board will be biased towardunderinvestment. If the probability of goodprojects is high, however, then the board willbe biased toward overinvesting. Song andThakor suggest that the probability of goodprojects will be low during economic down-turns and high during economic booms,

which means their model offers an explana-tion of changes in governance over the busi-ness cycle: during downturns, the board willbe tougher and, during upturns, the board

will be more lenient.The Song and Thakor model is rather com-

plex, with many moving parts. To providesome intuition for its results, consider anadaptation of Hermalin and Weisbach (2009)motivated by Song and Thakor. Assumea risk-averse CEO with career concerns la Holmstrom (1999). Assume his abil-

ity, unknown ex ante to all, is N(0, 1/),where N(, 2) denotes a normal distribu-tion with mean and variance 2.33A proj-ect arises that will payoff r + + , whereris a known constant reflecting the currenteconomic environment and N(0, 2). Apublic signal, s, about the CEOs ability is

33While the realization of is unknown by anyone, alldistributions are common knowledge.

realized after the project arises, but beforethe board must commit to the project.AssumesN(, 1/q), where qis a measureof the boards quality. Note the uncondi-

tional distribution of s is N(0, 1/H), where1/H = 1/ + 1/q. Normalize the firms rev-enues if the board decidesnotto pursue theproject to be 0. Using the standard formulafor forming posteriors from normal distri-butions (see, e.g., Morris H. DeGroot 1970,p. 167), the expected value of the projectconditional on the signal is

r +qs

_____

q + .

The board proceeds with the project if that ispositive; that is, if

s (q + )r

_______

q S.

Given the option of blocking a negative NPVproject, the firms expected valuepriorto thereception of the signal is

(1) V =

max

e0, r +

qs

____

q+

f

___

H

___

2expa H__

2s2bds w

=__

H____

(S__

H)

+ (1 (S__

H))rw,

where w is the CEOs compensation, () isthe density function of a standard normalrandom variable (i.e., with mean zero and

variance one), and () is the correspondingdistribution function.

Differentiating V with respect to q, it isreadily shown that the firms expected value,V, is increasing in the quality of the board,q, all else held equal. Intuitively, the abilityto block a bad project creates an option. An


21/53


option that is never exercised is worthless;hence, if the signal were complete noise, as

would be the case if board had zero qual-ity (recall the signals variance is 1/q), there

would be effectively no option. As the qualityof the board and, thus, information improves,the more valuable this option becomesand, therefore, the more valuable the firmbecomes.

It is not, however, costless to increaseboard quality without bound. First, it seemsreasonable that higher quality directors com-mand a premium or that providing a board

with sufficient incentives to do a high-qualityjob is expensive. So the cost of board qual-

ity is increasing in quality. Under suitableassumptions about this cost function (e.g.,that marginal cost be rising in q), there willbe an optimal finite value for q. In addition,if the CEO labor market reacts to the signalso that the CEOs future salary is an increas-ing function of the signal, then the CEO isexposed to more future salary risk the moreinformative the signal is (i.e., the greateris q). Intuitively, the posterior estimate of theCEOs ability is a weighted average of theprior, which is fixed, and the signal, whichis noisy. The more informative the signal isknown to be, the more weight is assigned thesignal. This increases the CEOs risk morethan the lower variance of the signal itselfreduces it (see Hermalin and Weisbach 2009for details). A CEO will require compensa-tion for this greater risk, so his initial salary(win expression (1)) will have to be greater.In light of this cost, under suitable condi-

tions, it will again be the case that a finite qis optimal.

From expression (1), the marginal netreturn to qis

1

____

2q2ar____

__

HbH3/2w___

q

(note S__

H = r/__

H). The change in themarginal net return to q with respect to r,

the measure of the current economic envi-ronment, has the same sign as

d

___

dr

ar

____

__

H b< 0 ,

where the inequality follows because anincrease in r is a move further into the lefttail of the density. Therefore, the marginalnet return to q is falling in r, which meansthat the optimal quality of the board is lower

when economic conditions are good (i.e., rishigh) than when they are bad (i.e., r is low).Intuitively, when times are good, the board

will wish to let mediocre CEOs go aheadwith projects, but they wont when times arebad. Consequently, the value of improvingthe monitoring of projects is greater whentimes are bad than when they are good.

Nina Baranchuk and Philip H. Dybvig(2009) is an interesting article in this areabecause it is not worried about informationtransmission between CEO and board, butamong the various board members them-selves (which, in practice, include the CEO).Each director i has a belief, ai n, as to

what the firm should do. Similar to Adamsand Ferreira (2007) and Harris and Raviv(2008), a director expects to suffer a qua-dratic loss in the distance between his beliefsas to what the firm should do and what thefirms actual course of action, a, is; that is, adirectors utility is

ai a.

The directors arrive at aaccording to a solu-tion concept that the authors call consensus.This solution concept has many desirableproperties, including existence for all suchgames. A weakness of the concept, however, isthat there is no explicit extensive-form gameto which it is a solution (consensus is a coop-erative game-theoretic concept). Anotherissue is there is no scope for directors to


22/53


23/53


and insiders and find that both types ofdirectors earn abnormal profits, but thatinsiders earn better returns than do outsid-ers. These results suggest that both types ofdirectors have access to inside informationbut that outsiders information is strictlyworse than insiders. Thus the finding sup-ports the underlying assumption of theinformation-based models of boards.

Breno Schmidt (2008) considers a situ-ation in which advice could be particu-larly valuable, namely during mergers andacquisitions. On the basis of the social tiesbetween the CEO and other directors, heclassifies boards as friendly (ties exist) and

unfriendly (a continuous measure is alsoemployed). When it is likely that directorspossess valuable information about an acqui-sition (an index measure), the returns of theacquirer are higher on announcement of theacquisition for bidders with more friendlyboards. Conversely, when the need to disci-pline the manager is a greater concern, socialties prove to be a negative.

Although there is a growing empirical lit-erature seeking to estimate the role of direc-tors in strategy setting, it is safe to say thatthis is an area in which much work remainsto be done.

3. How are Boards of DirectorsStructured?

We have discussed some explanations forwhy there are boards, and why one mightexpect endogenously-chosen boards to pro-

vide monitoring of management, despite thefact that management typically has somesay over the boards composition. But thetheories simply provide a stylized descrip-tion of the underlying tensions in the role ofthe board in corporate governance. Actualgovernance is much richer than these bare-bones characterizations.

There are a number of questions thatcan only be answered by looking at data

on real-world boards of directors. How areboards structured in practice? Does thisstructure coincide with the earlier-discussedtheories? How has it changed over time, bothin response to changes in the economy andregulatory environments?

3.1 Some Facts

Observers typically divide directors intotwo groups: inside directors and outsidedirectors. Generally, a director who is afull-time employee of the firm in questionis deemed to be an inside director, while adirector whose primary employment is not

with the firm is deemed to be an outside

director. Outside directors are often taken tobe independent directors, yet the indepen-dence of some directors who meet the defini-tion of an outsider is questionable. Examplesof such directors are lawyers or bankers whodo business with the company. Outsiders ofdubious independence are sometimes put ina third category in empirical work (see, e.g.,Hermalin and Weisbach 1988): affiliated orgray directors. In recent years, public pres-sure and regulatory requirements have ledfirms to have majority-outsider boards.

The characteristics of boards of large U.S.corporations have been described in a numberof studies. For example, Fich and Shivdasani(2006) consider a sample of 508 of the largestU.S. corporations between 1989 and 1995.They find that, on average, outsiders make up55 percent of directors, insiders 30 percent,and affiliated directors the remaining 15percent. The average board contains twelve

directors, each receiving approximately$36,000 in fees (plus stock options), and has7.5 meetings a year. A number of the directorsserved on multiple boards; the outside direc-tors in these firms averaged over three direc-torships. While these data are for large publicfirms, James S. Linck, Jeffry M. Netter, andTina Yang (2008) consider a larger sample of8,000 (necessarily) smaller firms, with similarpatterns in the data.


24/53


While the existence and basic structureof boards have remained relatively constantover time, the way in which they are com-posed has changed. Lehn, Sukesh Patro, andMengxin Zhao (2009) consider a sample of 81firms that have survived as public companiesfrom 1935 until 2000. Survivorship bias com-plicates the interpretation of their findings,nevertheless they reflect some basic trendsthat have affected boards. First, board sizeappears to have a hump pattern over time; itaverages 11 in 1935, peaks at 15 in 1960, anddeclines to 11 in 2000. However, board sizehas become more uniform over time as thestandard deviation of board size drops from

5.5 in 1935 to 2.7 in 2000. These companiesboards have become more outsider-domi-nated as well; insider representation dropsfrom 43 percent in 1935 to just 13 percentin 2000. Part of this drop can be explainedby the typical life cycle of firms. As found-ing families exit and firms become moreprofessionally managed, agency problemscan become worse as those in control are nolonger significant owners. In response, firms

will wish to add outside directors to counter-act the increased agency problems.

Since 2000, there have been significantchanges. SarbanesOxley contained a num-ber of requirements that increased the work-load of and the demand for outside directors(see Linck, Netter, and Yang 2009 for adescription of these requirements). In addi-tion, the scandals at Enron and Worldcomhave led to substantially increased public scru-tiny of corporate governance. Consequently,

boards have become larger, more indepen-dent, have more committees, meet moreoften, and generally have more responsibil-ity and risk (again see Linck, Netter, andYang 2009). These changes both increasedthe demand for directors and decreased the

willingness of directors to serve for a givenprice. It is not surprising, therefore, thatdirector pay and liability insurance premi-ums have increased substantially. From the

shareholders perspective, the net effect ofthis regulation is not clear; future research

will need to address the extent to which theadditional monitoring offsets the incremen-tal costs imposed by SarbanesOxley.

3.2 Factors in Board Composition thatPotentially Affect a Boards Actions

We have already discussed much of theliterature relating board composition (interms of the insider-to-outsider ratio) andboard size to board actions regarding over-sight of the CEO, as well as to overall firmperformance (see section 2.2). Yet beyondthe insider-to-outsider ratio and board size,

other board attributes no doubt play a role.Each board of directors is likely to have itsown dynamics, a function of many factorsincluding the personalities and relationshipsamong the directors, their backgrounds andskills, and their incentives and connections.Some of these factors are readily measured

while others are not. There has been con-siderable research that seeks to estimate theimpact of various board characteristics onboard conduct and firm performance.

3.2.1 CEOChairman Duality

Many CEOs also hold the title of Chairmanof the Board; this duality holds in almost80 percent of large U.S. firms (see PaulaL. Rechner and Dan R. Dalton 1991). Thisstructure is viewed by many as giving CEOsgreater control at the expense of other par-ties, including outside directors. To mitigatethe consequent problems, many observers of

corporate governance have called for a prohi-bition on the CEO serving as chairman (see,e.g., Michael C. Jensen 1993).

A number of recent papers have examinedthe use of dual titles in corporate governanceempirically. James A. Brickley, Coles,and Gregg A. Jarrell (1997) estimates theperformance effects of combined titles. Theseauthors find little evidence that combiningor separating titles affects corporate


25/53


performance. They conclude that the sepa-ration and combination of titles is part ofthe natural succession process described by

Vancil (1987). In contrast, Goyal and Park(2002) find that the sensitivity of CEO turn-over to performance is lower when titles arecombined, consistent with the notion thatthe combination of titles is associated withincreased power over the board. Similarly,Adams, Heitor Almeida, and Ferreira (2005)find evidence consistent with the view thatCEOs also holding the chairman title appearto hold greater influence over corporate deci-sion making.

Overall, these studies are consistent with

the view that combined titles are associatedwith CEOs having more influence in thefirm. However, this relation is not neces-sarily causal. Influence inside an organiza-tion arises endogenously, and with influencegenerally come fancier titles. The Goyal andPark and Adams, Almeida, and Ferreirafindings potentially reflect CEO power thatcame about endogenously through a mannersimilar to that described in the Hermalin and

Weisbach (1998) model. In other words, aCEO who performs well would be rewardedby his being given the chairman title as well.Such a process, especially if the increase inpower arises because of a demonstrated highability, would not necessarily imply perfor-mance changes following shifts in titles, con-sistent with the Brickley, Coles, and Jarrellfindings.

Even if it is true that combining the titles ofCEO and chairman means that an individual

has, on average more influence over his firm,it does not follow that mandating separatetitles would improve corporate performance.In fact, Adams, Almeida, and Ferreirasimilar to Brickley, Coles, and Jarrellfindthat measures of CEO power arenotsystem-atically related to firm performance. This isconsistent with our overarching argumentthat actual corporate-governance practiceneeds to be seen as part of the solution to

the constrained optimization program that iscorporate-governance design. Hence,impos-

ing separate titles would either yield a lessoptimal solution or lead to a, possibly ineffi-cient, work-around that maintained the opti-mal amount of CEO power.36Moreover, asnoted earlier, making the CEOs job worselikely means an offsetting increase in pay ascompensation. Consequently, as with mostpolicy prescriptions in the area of gover-nance, policy makers should be wary of callsfor prohibiting the CEO serving as chairman.

3.2.2 Staggered Boards

A common, yet controversial, governance

arrangement is known as staggered boards.When a firm has a staggered board, insteadof holding annual elections for each direc-tor, directors are elected for multiple yearsat a time (usually three), and only a fraction(usually a third) of the directors are electedin a given year. This practice is typicallyadopted as a way of shielding a firm fromtakeover because a potential acquirer can-not quickly take control of the firms boardeven it controls 100 percent of the votes.This arrangement is more common than onemight imaginein the Faleye (2007) sam-ple, roughly half of the firms have classified(staggered) boards.

While the consequence of the separa-tion of the CEO and chairman positionson firm performance is ambiguous, lessambiguity exists with respect to staggeredboards; the empirical evidence indicatesthis arrangement is not in the sharehold-

ers interests (although, as with much of theempirical work, caution is warranted dueto joint-endogeneity issues). Both Jarrelland Annette B. Poulsen (1987) and JamesM. Mahoney and Joseph T. Mahoney (1993)

36 Recall that, in a number of models of boards, cedingsome control to management is optimal (see e.g., Almazanand Suarez 2003; Laux 2008; Adams and Ferreira 2007;and Harris and Raviv 2008).


26/53


find negative returns when firms announcethey are classifying their boards (althoughJarrell and Poulsens finding is not statisti-cally significant). Bebchuk, John C. Coates,and Guhan Subramanian (2002) find thata classified board almost doubles the oddsthat a firm remains independent when faced

with a hostile takeover. Because some would-be acquirers are no doubt scared off by thestaggered board, the Bebchuk, Coates, andSubramanian findings likely underestimatethe ability of a classified board to resist take-overs. Bebchuk and Alma Cohen (2005) findthat firms with staggered boards have lower

value than other firms, using Tobins Qas a

measure of value. Finally, Faleye (2007) findsthat a staggered board lowers the sensitivityof CEO turnover to firm performance.

An implication of the view that staggeredboards entrench managers and decrease

value is that when firms destagger, returnto annual elections for all directors, valueshould increase. Re-Jin Guo, Timothy A.Kruse, and Tom Nohel (2008) consider asample of firms that destagger and find thatthe value of these firms does, in fact, increase.They also find that destaggering is not typi-cally initiated by managers, but by activistshareholders. Subsequent to the destagger-ing, investor reaction indicates that thesefirms are more likely to be takeover targets.All of these findings reinforce the view thatstaggering boards is a mechanism that servesto protect management by making takeoversdifficult.

All in all, it appears that firms with stag-

gered boards do worse than firms withannual board elections. Of course, someof this effect could be due to endogeneity;firms with already entrenched managers aremore likely to be able to convince sharehold-ers to adopt staggered boards. Or, to take aless sinister view, those managers who provethemselves are in a position to bargain forgreater job security as part of an optimal(second-best) bargain for their continued

service (and those who fail to prove them-selves become vulnerable to destaggeringand takeover). In this light, stock-marketreaction to announcements about whetherthe board will be staggered or not could bedue to the news such announcements convey

vis--vis the bargaining toughness and inde-pendence of the board rather than to simply

whether the board is or isnt staggered.

3.3 The Role of Particular Types of OutsideDirectors

To be considered an outsider, a directorsprimary employment must be with a dif-ferent organization than the firm on whose

board he serves. Outside directors typicallyhave backgrounds that will enable them tobe valuable to a board, or to represent animportant constituency. A small literatureconsiders particular types of directors andtheir specific roles in corporate governance.

3.3.1 Bankers

Many firms have bankers on their boards.Bankers may be added to boards bothbecause they can monitor the firm for thelender for whom they work and becausethey can provide financial expertise. BothJames R. Booth and Daniel N. Deli (1996)and Daniel T. Byrd and Mark S. Mizruchi(2005) consider the extent

the role of boards of directors in corporate governance: a conceptual framework and survey

Documents