read article and answer question

ksssss

Co-optedboardsColesDanielandNaveen.pdf

Home >Business & Finance homework help >read article and answer question

Electronic copy available at: http://ssrn.com/abstract=1699272

Co-opted Boards

Jeffrey L. Colesa, Naveen D. Danielb, Lalitha Naveenc

September 10, 2013

Forthcoming, Review of Financial Studies

Abstract

We argue that not all independent directors are equally effective in monitoring top management. Specifically, directors who are appointed by the CEO are likely to have stronger allegiance to the CEO and will be weaker monitors. To examine this hypothesis, we propose and empirically deploy two new measures of board composition. Co-option is the fraction of the board comprised of directors appointed after the sitting CEO assumed office. Consistent with Co- option serving to measure board capture, as Co-option increases board monitoring intensity decreases: turnover-performance sensitivity diminishes; pay level increases but without a commensurate increase in pay-performance sensitivity; and investment in hard assets increases. Further analysis suggests that even independent directors who are co-opted are less effective monitors. Non-Co-opted Independence––the fraction of the board comprised of independent directors who were already on the board before the CEO assumed office––has more explanatory power for monitoring effectiveness than the traditional measure of board independence. JEL Classifications: G32; G34; K22 Keywords: Corporate governance; Board co-option; CEO entrenchment; Board composition; Board independence ______________________________________________________________ a W. P. Carey School of Business, Arizona State University, Tempe, AZ., 85287, USA, [email protected] bLeBow College of Business, Drexel University, Philadelphia, PA., 19104, USA, [email protected] cFox School of Business, Temple University, Philadelphia, PA., 19122, USA, [email protected] The authors are grateful to an anonymous referee, Renee Adams, Christa Bouwman, Vidhi Chhaochharia, Rachel Diana, Dave Denis, Diane Denis, Ben Hermalin, Yan Li, Antonio Macias, David Maber, John McConnell, Darius Palia, Raghu Rau, David Reeb, Oleg Rytchkov, Partha Sengupta, Mike Weisbach (the editor), Jun Yang, and seminar participants at Case Western Reserve University, Lehigh University, Northeastern University, Purdue University, Rutgers University, the Securities and Exchanges Commission, Villanova University, the 2008 American Financial Association meeting, the 2008 Conference on Corporate Governance and Fraud Prevention at George Mason University, the 2008 Financial Management Association meeting, the 2008 Summer Research Conference at Indian School of Business, the 2010 Weinberg Center for Corporate Governance Conference at the University of Delaware, the 2011 SFS Finance Cavalcade, and the 2011 Finance Down Under Conference at the University of Melbourne for helpful comments.

Electronic copy available at: http://ssrn.com/abstract=1699272

1. Introduction

The board of directors of a corporation is meant to perform the critical functions of

monitoring and advising top management (Mace (1971)). Conventional wisdom holds that

monitoring by the board is more effective when the board consists of majority of independent

directors. The empirical evidence on the connection between board independence and firm

performance, however, is mixed and weak, as is the evidence on the relation between board

independence and other organizational and governance attributes, such as managerial

ownership.1

One potential reason for the paucity of consistent, significant results is that many

directors are co-opted and the board is captured. In practice, CEOs are likely to exert

considerable influence on the selection of all board members, including non-employee directors.

Carl Icahn, activist investor, asserts quite directly (Business Week Online, 11/18/2005) that

“…members of the boards are cronies appointed by the very CEOs they're supposed to be

watching.” Likewise, Finkelstein and Hambrick (1989) allege that CEOs can co-opt the board

by appointing “sympathetic” new directors. Hwang and Kim (2009) suggest that CEOs favor

appointees who share similar views or social ties or because there is some other basis for

alignment with the CEO.

Reflecting similar concerns about board capture, subsequent to the Sarbanes Oxley Act of

2002 (SOX) NYSE and Nasdaq adopted listing requirements that substantially reduced the direct

influence of the CEO in the nominating process. Nonetheless, CEOs are likely to continue to be

able to exert some influence on the board nomination process. At the very least, they approve

the slate of directors, and this slate is almost always voted in by shareholders (Hermalin and

1 See Coles, Daniel, and Naveen (2008), Adams, Hermalin, and Weisbach (2010), and Coles, Lemmon, and Wang (2011), for example.

Weisbach (1998), Cai, Garner, and Walkling (2009)).2

In this paper, we propose and implement two new measures of board composition, which

we term Co-option and Non-Co-opted Independence. Co-option is meant to capture board

capture. Non-Co-opted Independence, on the other hand, is meant to refine the traditional

measure of board independence as a proxy for the monitoring effectiveness of the board.

We define Co-option as the ratio of the number of “co-opted” (or captured) directors,

meaning those appointed after the CEO assumes office, to board size. The idea is that such co-

opted directors, regardless of whether they are classified as independent using traditional

definitions, are more likely to assign their allegiance to the CEO because the CEO was involved

in their initial appointment. Our measure is meant to reflect the additional behavioral latitude

and managerial discretion afforded a CEO when that CEO has significant influence over some

directors on the board. A related interpretation of Co-option is that it captures the disutility to

the board from monitoring the CEO. Along these lines, Hermalin and Weisbach (1998), in their

model of CEO bargaining with the board, specify director utility as a function of, among other

things, a distaste for monitoring (κ in their model), which for a director is reflected in a “… lack

of independence, at least in terms of the way he or she behaves” (p. 101). Co-option can be

thought of as capturing director aversion to monitoring and lack of independence aggregated to

the board level. Intuitively, Co-option reflects what the CEO can get away with.

Co-option ranges from 0 to 1, with higher values indicating greater co-option and board

capture and greater insulation of the CEO from various efficiency pressures. In our sample,

mean Co-option is 0.47, indicating that on average nearly half of the directors on a board joined

the board after the CEO assumed office.

2 Of course, CEO influence on the nomination process is substantially lower in the relatively few instances where directors are put up for election by dissident shareholders in proxy fights.

We predict that a CEO who has co-opted a greater fraction of the board will be less likely

to be fired following poor performance, will receive higher pay, will have lower sensitivity of

pay to performance, and will be able to implement preferred or pet projects even if they are

suboptimal from a shareholder-value perspective. Our findings generally are consistent with

these hypotheses.

First, we find that the sensitivity of forced CEO turnover to firm performance decreases

with co-option. For example, our parameter estimates indicate that CEO-turnover-performance

sensitivity is attenuated by about two-thirds for a one-standard-deviation increase in Co-option.

Second, we find that CEO pay levels increase with board co-option. Of course, higher pay being

associated with higher co-option is not symptomatic of entrenchment if it is compensation for

higher risk borne by the CEO through higher pay-performance sensitivity. Additional evidence,

however, suggests that this is not the case: we find that the sensitivity of CEO pay to firm

performance is generally unrelated to board co-option and even is negatively related to co-option

in some specifications. Finally, we find that investment in tangible assets (the ratio of capital

expenditure to assets) increases with co-option. This is consistent with the idea that CEOs who

have co-opted the board can invest in ways they otherwise would not. For example, in the

absence of effective board monitoring, executives are likely to satisfy their preferences for scale

and span of control, preferences that arise in larger firms for reasons of higher compensation,

control over more resources, and enhanced stature in the industry and community (Jensen

(1986)). Overall, the evidence on turnover, pay, and investment is consistent with the idea that

co-option reduces the monitoring effectiveness of the board.

In all specifications we control for the proportion of independent directors on the board

(Independence), which traditionally has been understood to be a measure of board monitoring.3

We find that Independence has little power to explain CEO turnover-performance-sensitivity,

CEO pay, CEO pay-performance-sensitivity, and investment. If there were a statistical horse

race between Co-option and Independence, Co-option would appear to be more successful.

In light of this result, a natural question is whether independent directors who are co-

opted by the CEO are different in monitoring effectiveness from those who are not co-opted. To

address this question, we calculate the fraction of the board that is comprised of independent

directors appointed after the CEO assumed office (“Co-opted Independence”). Our results using

this measure as an explanatory variable are similar to what we find with Co-option. Specifically,

we find that Co-opted Independence is associated with lower sensitivity of CEO turnover to

performance, higher pay levels, lower sensitivity of pay to performance, and greater investment.

Thus, co-opted independent directors, though independent of the CEO in the conventional and

legal sense, behave as though they are not independent in the function of monitoring

management. This is likely to explain why the literature has not found consistent evidence with

respect to the monitoring effectiveness of independent directors.

To formally test the monitoring effectiveness of independent directors who are not co-

opted, we introduce a second new measure of board composition: Non-Co-opted Independence.

We define this measure as the fraction of the board comprised of independent directors who were

already on the board when the CEO assumed office. In our sample, mean Non-Co-opted

Independence is 0.35, indicating that on average about a third of the board is comprised of

independent directors who are truly independent, having not been co-opted by the CEO. Of

3 See, for example, Weisbach (1988), Byrd and Hickman (1992), Brickley, Coles, and Terry (1994), Dahya, McConnell, and Travlos (2002), Hermalin and Weisbach (2003), Dahya and McConnell (2007), Coles, Daniel, and Naveen (2008), and Dahya, Dimitrov, and McConnell (2008).

course, on most issues the board faces, the majority rules, so there is a significant possibility that

the subset of independent directors who are not co-opted is not influential. Nonetheless,

consistent with our conjecture that independent directors who are not co-opted are the monitors

that matter, we find Non-Co-opted Independence is associated with higher sensitivity of CEO

turnover to performance, lower pay levels, higher sensitivity of pay to performance, and lower

investment.

In sum, not all independent directors are equally effective at monitoring. Those who are

co-opted by the CEO are associated with weaker monitoring, while the independent directors

who join the board before the CEO assumes office, that is, the directors who hired the CEO, are

associated with stronger monitoring.

Our results on board capture are robust to two alternative definitions of Co-option. Our

first alternative proxy, Tenure-Weighted Co-option (TW Co-option), accounts for the possibility

that directors appointed by the CEO become even more co-opted through time and that the

influence of co-opted directors increases with their tenure on the board.4 We define TW Co-

option as the sum of the tenure of co-opted directors divided by the total tenure of all directors,

so an increase likely indicates higher board co-option. Our second alternative proxy is designed

to address the possible concern that co-option increases mechanically with CEO tenure and that

our results on co-option may be capturing the effect of CEO tenure. We estimate Residual Co-

option as the residual from a regression of Co-option on CEO tenure. We similarly estimate

Residual TW Co-option as the residual from a regression of TW Co-option on CEO tenure. By

construction, these residual measures are uncorrelated with CEO tenure. We find qualitatively

similar results using these alternative definitions of co-option.

4 Per Nell Minow, quoted in Hymowitz and Green (2013), “What you want from directors is for them to really push the CEO for answers and, just by human nature, that gets harder the longer they’re on a board.”

Our results also are robust to our best attempts to address endogeneity. All of our base-

case regressions include firm-fixed effects to control for biases introduced by unobserved, firm-

specific, time-invariant, omitted variables that are correlated with co-option. Endogeneity could

still arise, however, either because the omitted variable is not firm-specific or varies through

time, or because reverse causation runs from our firm policy variables, such as pay, to co-option.

We exploit exchange-rule changes enacted in 2002 to address such concerns. Since these rules

were adopted shortly after the passage of Sarbanes-Oxley (SOX), we refer to the post-rules

period as the post-SOX period. Firms that pre-SOX were not compliant with subsequent listing

requirements to have a majority of independent directors on the board chose to appoint new

independent directors (Linck, Netter, and Yang (2009)), thereby causing an exogenous increase

in board co-option for such firms. To isolate the causal impact of co-option, we apply a

modified difference-in-difference approach. We continue to find results on the effects of co-

option that by-and-large are consistent with the evidence described above.

2. Motivation, Related Literature, and Hypotheses Development

2.1. CEO turnover-performance sensitivity

One of the key functions of the board is to evaluate the CEO and to replace him if his

performance is poor (Mace (1971)). While early studies find that the likelihood of CEO turnover

decreases in firm performance, subsequent studies suggest that this relation between turnover

and performance is weaker when the firm’s governance is weaker.5 Along similar lines,

Hermalin and Weisbach (2003) suggest that turnover-performance sensitivity is weaker if the

CEO captures the board. This implies that, for a given level of performance, CEOs of firms with

5 See Coughlan and Schmidt (1985), Warner, Watts, and Wruck (1988), Weisbach (1988), Huson, Parrino, and Starks (2001), Kang and Shivdasani (2005), and Kaplan and Minton (2012).

more co-opted boards should be less likely to be fired. Thus, we expect that:

H1: All else equal, the sensitivity of forced CEO turnover to firm performance decreases

with co-option. 2.2. CEO pay level

A second important function of the board is to set the structure of CEO pay. Many

studies argue that entrenched CEOs and CEOs of firms with weaker monitoring receive higher

pay (Borokhovich, Brunarski, and Parrino (1997) and Core, Holthausen, and Larcker (1999)).

We extend this reasoning to argue that if co-opted boards are more sympathetic to the CEO, then

CEO pay should increase with co-option. This leads to our second hypothesis:

H2: All else equal, CEO pay level increases with co-option. 2.3. CEO pay-performance sensitivity

Pay contingent on performance is a means to align executive incentives with shareholder

interests (e.g., Jensen and Murphy (1990), Bizjak, Brickley, and Coles (1993)). Thus, we also

examine the impact of co-option on CEO pay-performance sensitivity (PPS or “delta”). Hartzell

and Starks (2003) show that the CEO pay-performance sensitivity is higher when institutions

hold more shares and argue that this is consistent with higher institutional holdings being good

for shareholders. Faleye (2007) finds lower PPS for CEOs of firms with staggered boards and

argues that staggered boards are associated with CEO entrenchment. Thus, we expect that, if co-

option results in lower efficiency pressures on the management team, then pay-performance

sensitivity should decrease in co-option.6

6 Empirically, the papers mentioned in this subsection use varying methodologies to capture PPS. For example, Hartzell and Starks (2003) use PPS from new option grants only as the dependent variable. Coles, Lemmon, and

H3: All else equal, CEO pay-performance sensitivity decreases with co-option. 2.4. Investment policy

A long literature addresses managerial incentives to overinvest and to engage in empire

building. Jensen (1986, pg. 323), for example, notes that “growth increases managers' power by

increasing the resources under their control. It is also associated with increases in managers'

compensation, because changes in compensation are positively related to the growth (see Kevin

Murphy (1985)).” Moreover, scale and span of control can enhance the stature of the CEO in the

industry and community. When the CEO has significant influence over some directors on the

board and, accordingly, is permitted additional behavioral latitude and managerial discretion,

such overinvestment is more likely. All else equal, co-option will be associated positively with

investment.

H4: All else equal, firm investment increases with co-option.

3. Data and Summary Statistics

We start with the RiskMetrics database, with coverage of directors of S&P 500, S&P

MidCap, and S&P SmallCap firms over the period 1996-2010. RiskMetrics does not provide a

unique firm-level or director-level identifier over the entire time period. In the Appendix we

describe how we associate unique identifiers with each record on RiskMetrics.7 We obtain

Wang (2011) use the pay performance sensitivity derived from the total portfolio of accumulated stock and option holdings net of dispositions. Falaye (2007) uses Aggarwal and Samwick (1999) type regressions of changes in annual pay on dollar returns and interprets the coefficient on dollar returns as PPS. 7 RiskMetrics provides two different director identifiers, neither of which is fully populated for all directors. Between 23 – 27% of director-years have missing identifiers. We combine both to create a unique identifier for all director-year observations. Importantly, if only one of these identifiers is used, it will result in incorrect estimates of

compensation data from Execucomp, accounting data from Compustat, and stock return data

from CRSP. We exclude firms incorporated outside the U.S. We define below our key

variables.

3.1. CEO forced turnover

The logic underlying our measure of co-option is most applicable for forced turnover.

Unfortunately, it is difficult to classify turnover as forced or voluntary. Very often, even forced

turnovers are reported to the press as voluntary. Nevertheless, we use an approximate

classification scheme, similar to that used in other papers (such as Denis and Denis (1995)) to

separate turnovers into forced or voluntary. We define Forced Turnover as one if the departing

CEO is less than 60 years old, and zero otherwise.

3.2. CEO pay

Our measure of CEO pay is total annual compensation (Execucomp variable TDC1).

This includes the value of annual stock option grants, salary and bonus, value of annual restricted

stock grants, other annual compensation, long-term incentive payouts, and all other

compensation. We discuss in the Appendix how the changes in compensation reporting

following FAS 123R and new SEC disclosure requirements affect the reporting of pay. We

compute an adjusted pay measure (discussed in more detail in the Appendix) that accounts for

these changes in reporting. Our results are similar using this adjusted pay measure.

3.3. CEO pay-performance sensitivity

Pay-performance sensitivity is estimated as the sensitivity of CEO wealth to stock price,

otherwise termed as CEO delta, based on the entire portfolio of stock and options held by the

CEO. Specifically, the semi-elasticity form of delta is the expected $ change in CEO wealth for

board size, independence, co-option etc. Upon request, the authors can provide the unique director identifier that we create as well as the unique firm identifiers (GVKEY and PERMNO) for each record on RiskMetrics.

a 1% change in stock price. We calculate delta using the approach of Core and Guay (2002) but

with adjustments to Execucomp data as specified in the Appendix. Also see Coles, Daniel, and

Naveen (2013) for details on data and on calculation of incentive measures in the presence of

changing financial reporting requirements and formats.

3.4. Investment

Our proxy for investment is capital expenditures scaled by book value of assets.

3.5. Co-option

Our principal measure of co-option is based on the number of directors elected after the

CEO takes office. We refer to such directors as “co-opted” directors.

sizeBoard

directorsoptedCo optionCo

 

This variable ranges from 0 to 1, with higher values indicating greater co-option.8

In some specifications, we use an alternative measure of co-option, Tenure-weighted Co-

option (TW Co-option), which is the sum of the tenure of co-opted directors divided by the total

tenure of all directors. Thus,





  sizeboard

i i

sizeboard

i ii

Tenure

DummyDirectoroptedCoTenure optionCoTW

where Co-opted Director Dummyi equals 1 if the director ‘i’ is a co-opted director, and equals 0

otherwise. Tenurei refers to the tenure of the director ‘i’ on the board. This alternative measure

8 In contemporaneous work independent of ours, Morse, Nanda, and Seru (2011) develop a measure of CEO power based on three elements, one of which is similar to our measure of co-option. They show that more powerful CEOs (CEOs who have titles of Chairman, CEO, and President, CEOs of firms with insider-dominated boards, or CEOs with a greater proportion of directors appointed during their tenure) rig their pay contracts by increasing the weights on the better performing measures.

accounts for the increase of influence of co-opted directors on board decisions through time, as

such directors work alongside the CEO and previously appointed directors. This measure

assumes that the greater the tenure of co-opted directors, the greater their influence on board

decisions. Again, this measure can vary from 0 to 1, with a higher value indicating greater board

capture.

Our third measure of co-option is Residual Co-option, which is defined as the residual

from a regression of Co-option on CEO tenure. Our final measure of co-option is Residual TW

Co-option, which is the residual from a regression of TW Co-option on CEO tenure. These two

measures remove the positive correlation between CEO tenure and co-option.

For each firm-year, RiskMetrics provides the date of the annual meeting and the slate of

directors up for election. The directors on the slate almost always obtain sufficient support to be

elected (Hermalin and Weisbach (1998) and Cai, Garner, and Walking (2009)). The majority of

the sample firms hold their annual meeting during the first 3 – 4 months of the fiscal year. Thus,

because these directors constitute the board for the majority of the fiscal year, we assign directors

on the slate at the annual meeting in a given fiscal year as the directors for that year.

For CEO turnover events, we are careful to identify the board in place before the CEO

was dismissed since this board is the one responsible for replacing the CEO. Thus the CEO

turnover date relative to the meeting date is important for our purpose. Figure 1 illustrates the

timeline. If a CEO turnover occurred after the annual meeting date, then the board that

determined the replacement was the board elected for that year. That is, turnover and co-option

are measured contemporaneously. If a CEO turnover occurred before the annual meeting date,

then the board responsible for replacing the CEO is the one elected in the previous year so we

use lagged measures of co-option in the turnover regression. In non-turnover years, since both

the lagged and contemporaneous boards decide on the CEO’s ‘non-replacement,’ we use the

average of the lagged and contemporaneous values of co-option.

For regressions explaining variation in CEO pay, CEO delta, and investment, we use the

contemporaneous co-option measure, because this is based on the board that is in place for the

majority of the year and also because performance-based pay (which is a significant component

of overall pay) will be decided by the board at the end of the fiscal year.

3.6. Independence

Independence is the ratio of the number of independent directors on the board to total

board size. Independent directors are those who are neither inside nor grey directors (Weisbach

(1988), Byrd and Hickman (1992), Brickley, Coles, and Terry (1994)).

3.7. Summary statistics

Table 1 provides the summary statistics. To minimize the influence of outliers,

throughout the paper we winsorize all variables at the 1st and 99th percentiles.9 The average firm

in the sample is large, with sales of $5.3 billion. This is not surprising given that our sample is

S&P 1500 firms. The average board has about 10 directors. Co-option has a mean value of 0.47,

while mean Independence is 0.69. Thus, on average, although more than two-thirds of the

directors are technically independent, our calculations indicate that nearly half of the board has

been co-opted by the CEO. Average Tenure-Weighted (TW) Co-option is 0.31, implying that

while co-opted directors make up nearly half the board, their influence, after accounting for their

tenure on the board, is a bit lower at 31%. Not surprisingly, Co-option and TW Co-option are

similar, with a correlation of 0.93 (p <0.0001). Co-option and TW Co-option are dissimilar to

board independence (ρ = – 0.07 and ρ = – 0.09 respectively).

9 Our results are similar if we winsorize all variables at the 0.5 and 99.5 percentiles instead. Our results are also similar if we drop the observations in the top and bottom 0.5 percentiles from the analyses.

The unconditional rate of forced turnover is 0.025. For comparison, the equivalent

number is 0.019 in Hazarika et al. (2012) (inferred from their Table 1) and is 0.030 in Mobbs

(2012). On average, CEOs receive $4.9 million in total annual compensation, have a delta of

$789,000, and have about 8 years of tenure. On average, investment is 5.2% of total book assets.

4. Co-option and Monitoring Ineffectiveness: Empirical Results

4.1. Co-option and CEO turnover-performance sensitivity

Our first hypothesis, H1, is that the sensitivity of CEO turnover to performance decreases

with co-option. To test this, we estimate the following logistic regression:

ln[Prob(Forced Turnover)/(1 - Prob(Forced Turnover))] = α0 + α1 Co-option  Performance + α2 Performance + α3 Co-option + α4 Independence + f(Other Controls) + ε1

Our proxy for performance is Prior Abnormal Return. For turnover years, this is

measured as the firm stock return (including dividends) in the year leading up to the actual date

of CEO turnover minus the value-weighted market return over that period. For non-turnover

years, this is measured as the stock return over the previous fiscal year minus the value-weighted

market return over that period. It is well-documented that, in practice, performance is negatively

related to the likelihood of CEO turnover, or that α2 is negative (Weisbach (1988), Warner,

Watts, and Wruck (1988), Parrino (1997), and Kaplan and Minton (2012)). Our hypothesis is

that turnover-performance sensitivity is attenuated by co-option, or that α1 is positive. All

specifications include Independence. Other control variables (Other Controls) include: firm size;

CEO tenure; and governance variables. The governance variables are: CEO ownership; CEO

duality, an indicator variable that equals 1 if the CEO has title of chairman also; outside director

ownership; GIM index, the governance index of Gompers, Ishii, and Metrick (2003); board size;

female director, an indicator variable that equals 1 if the firm has a female director on board; and

(in some models) terms interacting governance variables with prior performance.10 We include

firm- fixed effects to control for any omitted firm-specific and time-invariant variables that are

correlated with co-option. We include year fixed effects to control for variation in common

influences through time. In general, our control variables are based on those in Adams and

Ferreira (2009), Hwang and Kim (2008), Fich and Shivdasani (2007), and Dittmar and Mahrt-

Smith (2007).

Table 2 reports the results. In Models 1 and 2, the key independent variable is the

interaction term of Co-option with Prior Abnormal Return. For each independent variable, we

report the coefficient estimates (Row 1), z-statistics (Row 2), and the marginal effects (Row 3).

We report the marginal effects because there is no ready economic interpretation of the

coefficients in non-linear regressions. The marginal effect is presented in semi-elasticity form.

For continuous variables, the marginal effect represents the percentage change in the probability

of Forced Turnover for a one unit change in the underlying variable, holding all other variables

at their mean values. For indicator variables, we report the percentage change in the probability

of Forced Turnover when the indicator variable moves from zero to one (holding other variables

at their mean values).11

Consistent with our hypothesis, the coefficient on the interaction term of Co-option and

Prior Abnormal Returns (α1) is positive and statistically significant (= 2.021, z-statistic = 3.4),

indicating that an increase in Co-option is associated with a decrease in the sensitivity of CEO

turnover to firm performance. To estimate the effect of the interaction term, we compute the

10 For the four CEO-related variables, the values correspond to the departing CEO in the year of turnover. Also, we do not include CEO age because Forced Turnover is automatically zero when the CEO is over 60. 11 Ai and Norton (2003) note that interpretation of interacted variables in non-linear models is not straight-forward. Stata (Version 11) has since introduced the margins statement which correctly computes the marginal effects in non- linear models with interaction terms. We use this statement to compute all reported marginal effects.

marginal effect of Prior Abnormal Return at two different levels of Co-option: at the mean as

well as mean plus one standard deviation (holding all other variables at their mean values). The

difference indicates how the sensitivity of turnover to performance changes with co-option. As

can be seen in Model 1, the sensitivity of turnover to firm performance decreases by 0.856, from

–1.331 at the mean value of Co-option, to –0.476 when Co-option increases by one standard

deviation from its mean value (for ease of presentation in the table we report only the

difference). In other words, the sensitivity of turnover to performance goes down by almost two-

thirds when Co-option moves by one standard deviation from its mean value. If Co-option

increases even further to the maximum possible value of 1, then the sensitivity of forced turnover

to performance is even smaller (= –0.092; see the last row of Table 2).12 Thus, the results in

Model 1 are consistent with H1. Turnover-performance sensitivity decreases as co-option

increases.

In Model 1, we allow only Co-option to affect the turnover-performance sensitivity (that

is, we include only the interaction term of Co-option with Prior Abnormal Return). In Model 2

we allow all governance-related variables (Independence, CEO ownership, CEO duality, outside

director ownership, GIM index, board size, and female director) to affect the turnover-

performance sensitivity. Two results are worth noting. First, the coefficient on the interaction of

Co-option with Prior Abnormal Return remains significantly positive. Second, Independence

does not appear to have a significant impact on turnover-performance sensitivity. Board co-

option, rather than board independence, has explanatory power for turnover-performance

sensitivity.

12 In the model, α2 represents the effect of Prior Abnormal Return on Forced Turnover when Co-option is zero. When STATA reports the marginal effect of Prior Abnormal Return, however, it reports the total effect of Prior Abnormal Return on Forced Turnover at the mean of all variables.

In Models 3 and 4, we use the same specifications as in Models 1 and 2 respectively, but

include TW Co-option rather than Co-option. The estimated coefficient on the interaction of TW

Co-option with Prior Abnormal Return remains significantly positive in both specifications. In

terms of economic significance, the results in Model 3 indicate that when TW Co-option

increases by one standard deviation from its mean value, the sensitivity of turnover to

performance changes from –1.539 to –0.577 (the table reports the difference = 0.962) . The last

row in the table shows that when TW Co-option increases to 1, the sensitivity of turnover to

performance is altered further to 0.213, which is positive. Results from Model 4 are similar.

A potential issue arises because our two co-option measures (Co-option and TW Co-

option) are positively correlated with CEO tenure. Thus, multicollinearity could be a concern.

To address this concern, we replace Co-option with Residual Co-option, which is the residual

from a regression of Co-option on CEO tenure. Model 5 reports the results. The coefficient on

the interaction of Residual Co-option with Prior Abnormal Returns is significantly positive,

indicating that the effect of Co-option on CEO turnover-performance sensitivity documented in

Model 1 is not due to the correlation between Co-option and CEO tenure. Finally, in Model 6,

we replace Co-option with Residual TW Co-option, which is the residual from a regression of

TW Co-option on CEO tenure. Once again, our results are similar to those in Model 3.13

In terms of the other control variables, our results across the various models show that

CEO duality is significantly negatively related to CEO turnover (as in Goyal and Park (2002)).

In contrast, the other governance variables, in general, are not consistently significant across the

13 We also estimate Models 5 and 6 including interactions of all governance variables with prior abnormal returns (as in Models 2 and 4). When we re-estimate Model 6 in this manner, the results are statistically and economically similar to our main results. When we re-estimate Model 5, the results are economically similar but statistically weaker. The sensitivity of turnover to performance decreases from –1.62 at the maximum value of Residual Co- option to –0.26 at the minimum value. The interaction of Residual Co-option with prior abnormal returns is positive, but is insignificant (p = 0.195).

various specifications.

The number of observations is much smaller in our turnover regressions because the use

of firm-fixed effects means that firms that never had a forced turnover during the sample period

are excluded from the regression. To ensure that our results are not driven by any sample

selection, we estimate the same regression models without firm-fixed effects, but with industry-

fixed effects, and obtain very similar results for all six specifications on a much larger sample.

In all tables that follow, we report t-statistics based on standard errors adjusted for

heteroskedasticity and clustering at the firm level (Petersen (2009)). This option, however, is not

available for the fixed-effects logistic regression models in Table 2. As a robustness check, we

bootstrap the standard errors using 200 replications. We find qualitatively similar results using

the bootstrap.

Overall, the results indicate that, consistent with H1, turnover-performance sensitivity is

attenuated as measures of board co-option increase.

4.2. Co-option and CEO pay level

Our second hypothesis, H2, predicts that CEO pay increases with co-option. To test this,

we estimate regressions of CEO pay on co-option and controls.

CEO Pay = θ0 + θ1 Co-option + θ2 Independence + g(Other Controls) + ε2.

Hypothesis H2 asserts that the coefficient on Co-option (θ1) will be positive. The control

variables, based on prior literature (see Murphy (1999) for a comprehensive review of CEO

compensation), include board independence, firm size, firm performance (both stock and

accounting), CEO tenure, governance variables, and firm and year dummies.14 We do not

14 The results are robust to using industry and year fixed-effects instead.

include CEO turnover years and require that the CEO’s tenure be at least one year. This is

because CEO pay in a turnover year is likely to reflect compensation only for part of the year.

Also, CEOs in their first year may receive higher than average stock compensation (to align their

incentives) and higher bonus (including signing bonuses). We use the logarithm of annual

compensation as the dependent variable because compensation data are skewed.15

Table 3 presents the results. In Model 1, the coefficient on Co-option is significantly

positive, implying that CEO pay increases with co-option.16 The coefficient of 0.223 on Co-

option indicates that moving from zero to full co-option would be associated with an increase in

CEO pay of 22.3%. A less extreme measure of economic significance is the change in pay when

Co-option increases by one standard deviation. In this case, we find that CEO pay increases by

7% relative to the mean pay. This corresponds to about $345,380 annually for the CEO.

In Model 2, we use TW Co-option rather than Co-option. As with Co-option, we find that

the coefficient on TW Co-option is significantly positive. Finally, in Models 3 and 4 we use

Residual Co-option and Residual TW Co-option. The results are similar. The coefficients on

both measures are significantly positive, indicating that co-option is associated with higher pay,

and this effect is not driven by the positive correlation between co-option and tenure.

Board independence has no explanatory power for CEO pay in two of the four

specifications. In the other two models, the coefficient on Independence is positive, which is

inconsistent with greater independence leading to better monitoring of rent extraction.17 For the

other control variables, as expected firm size and performance are strongly positively associated 15 We obtain similar results using unlogged compensation. 16 This result is consistent with Core, Holthausen, and Larcker (1999), who, using a sample of 495 firm-years over 1982-1984, find that CEO total pay is positively related to the proportion of the board composed of new outside (both independent and affiliated) directors. 17 As a robustness check, instead of using contemporaneous values of our co-option measures, we also use the average of the contemporaneous and the lagged values, since the lagged board may also be partly responsible for CEO compensation. Our results are robust to this change.

with pay. Overall, the evidence is consistent with CEO pay increasing in co-option (H2).

4.3. Co-option and CEO pay-performance sensitivity

Pay-performance-sensitivity –– otherwise known as delta –– is seen as aligning the

incentives of managers with the interests of shareholders. Higher delta can mean that managers

will work harder or more effectively because managers share gains and losses. Thus, we now

examine the influence of co-option on CEO delta. The representative specification is

CEO Pay-Performance Sensitivity = γ0 + γ1 Co-option + γ2 Independence + h(Other Controls) + ε3.

Our control variables are based on the prior literature on the determinants of delta (Core and

Guay (1999) and Coles, Daniel, and Naveen (2006)) and the governance variables used in the

preceding regressions.

Table 4 presents the results. As in Table 3, our independent variables are Co-option

(Model 1), TW Co-option (Model 2), Residual Co-option (Model 3), and Residual TW Co-option

(Model 4). In Models 1 and 3, the estimated coefficients on Co-option and Residual Co-option

are negative (consistent with our hypothesis), but insignificant at conventional levels (p = 0.107

and 0.103 respectively). In Models 2 and 4, the estimated coefficient on TW Co-option and

Residual TW Co-option are negative and significant, albeit at the ten percent level. The

coefficient on Co-option in Model 1 indicates that when Co-option increases by one standard

deviation from its mean, pay-performance sensitivity decreases by 12% from its mean value.

When Co-option increases from zero to one, sensitivity of pay to performance decreases by

$296,485, or by 38% from its mean value.

The coefficient on board independence is negative and significant in Models 1 – 4. This

result, which is similar to the result in Coles, Lemmon, and Wang (2011), suggests that board

monitoring and CEO delta may well be substitutes in organization design.

For robustness, as we do with CEO pay, we use the average of the contemporaneous and

the lagged values of the co-option measures instead of the contemporaneous values alone. Also,

we use industry-year fixed effects instead of firm-year fixed effects. The results in both cases

are similar to our base-case result, in that the coefficient on the co-option measures continues to

be negative but insignificant at conventional levels.

In sum, we find weak evidence in support of the hypothesis (H3) that higher co-option is

associated with lower CEO pay-performance sensitivity (PPS). CEO pay and PPS, however,

cannot be viewed as independent of each other. CEOs would demand higher pay if greater risk

is imposed on them in the form of higher PPS. Instead, if anything, co-option is associated with

lower exposure of CEO wealth to risk. Thus our finding that co-option is associated with higher

pay, but similar or even lower PPS, is consistent with co-opted boards adopting more liberal

compensation policies that are favorable to the CEO.

4.4. Co-option and investment

Hypothesis H4 proposes that co-option is positively associated with investment. We

examine this using the following specification:

Investment = μ0 + μ1 Co-option + μ2 Independence + j(Other Controls) + ε4.

The dependent variable is capital expenditure scaled by assets. In addition to board

independence, the other key independent variables are based on Coles, Daniel, and Naveen

(2006) and include vega, delta, cash compensation, CEO tenure, Tobin’s q, firm size, free cash

flow to assets, sales growth, leverage, and stock return.

Table 5 shows the results. In Model 1, we use Co-option as our key variable of interest.

The coefficient on Co-option is positive and statistically significant (= 0.005, p-value = 0.014).

In terms of economic significance, the coefficient indicates that when Co-option increases by one

standard deviation, investment increases by 3% relative to the mean. When Co-option increases

from zero to one, investment increases by 10% relative to its mean value.

In Column 2, as the dependent variable we use TW Co-option instead of Co-option. Once

again, the coefficient on TW Co-option is positive and statistically significant. In the last two

columns we use Residual Co-option and Residual TW Co-option respectively. Our results are

similar in sign and significance.

In all specifications, we find that the fraction of independent directors is negatively

associated with investment. The results on the other control variables are consistent with prior

literature. Consistent with Coles et al. (2006), we find the coefficient on vega is negative

(although not significant at conventional levels), and the coefficient on delta is positive. Higher

Tobin’s q and higher free cash flow are associated with more investment.

As with CEO pay and pay-performance sensitivity, we confirm that the results are

qualitatively similar if we use the average of the contemporaneous and the lagged values of the

co-option measures instead of the contemporaneous values alone. When we use industry-year

fixed effects instead of firm-year fixed effects, however, we find that the coefficient on co-option

is not significant.

Overall, the results in this subsection support the hypothesis that CEOs that have captured

the board to a greater extent are able to invest more than otherwise would have been the case. At

this juncture, based on our results, we are unable to discern whether such investment, which

likely increases firm size and the economic span of control of top management, is necessarily

inconsistent with shareholder interests. On this question, however, in separate, independently

developed work, Pan, Wang, and Weisbach (2013) document that investment increases with the

extent of CEO control of the board, as proxied by a measure of co-option similar to ours. They

also find that the quality of investment (captured by the market reaction to acquisition

announcements) deteriorates over the CEO’s tenure and that this deterioration is related to the

CEO’s control of the board. Thus, the relation we document between CEO investment and Co-

option potentially arises because (Pan et al. (2013, abstract)) “… the CEO overinvests when he

gains more control over his board.”

5. Endogeneity

Endogeneity is an important concern in any study on corporate governance (Coles,

Lemmon, and Wang (2011)). In particular, it is possible that both co-option and pay are high

due to an unobserved (and hence omitted) variable. Since we include firm-fixed effects in all our

specifications, we control for omitted variables that are firm-specific and time-invariant. If the

omitted variable is time-varying or not firm-specific, however, and is correlated with co-option,

this would cause the error term in the outcome equation to be correlated with co-option,

rendering OLS invalid. Another source of endogeneity is that both co-option and our variables

of interests, such as pay, are determined in equilibrium simultaneously. One solution would be a

valid instrument for the endogenous variable (Co-option). It is difficult, however, to find an

instrument that is related to co-option, but is not related to CEO pay or the other outcomes we

examine. As an alternative to firm-fixed-effects specifications, we turn to a natural experiment

to help us address endogeneity concerns.

We exploit the rules enacted in 2002 by the Nasdaq and NYSE, requiring all listed firms

to have a majority of independent directors on their board.18 Since these rules were adopted

shortly after the passage of SOX, we refer to the period following the proposal of the new stock

18 A detailed timeline is available in Chhaochharia and Grinstein (2007).

exchange rules (2002–2010) as the post-SOX period. Pre-SOX non-compliant firms were

required to increase board independence after implementation of the new listing requirements,

and these firms chose to add new independent directors onto the board (Linck, Netter, and Yang

(2009)). This resulted in an exogenous increase in co-option.19

To isolate the causal impact of co-option, we modify somewhat the Bertrand and

Mullainathan (2003) difference-in-difference (DID) methodology. The key difference is that we

allow for the possibility that SOX and associated exchange provisions have a direct effect on

turnover-performance-sensitivity, pay, pay-performance-sensitivity, and investment, as well as

an effect through co-option. This is because other regulations and political pressure arising from

SOX were likely to have affected monitoring through numerous channels.20 For example, under

SOX and the associated exchange provisions: complete independence was mandated for the

compensation, audit, and monitoring committees; a director with financial expertise was required

on the audit committee; in addition to their regular sessions, boards were required to meet

without management; CEO/CFO certification of accounting statements was required; and there

was a general increase in media scrutiny of all firms.

Because of this complication, we modify the typical DID setup to isolate the effect of co-

option (we term this the “clean” effect). The typical DID set up for examining pay, for example,

would be to regress pay on three dummy variables: Post-SOX, Non-Compliant, and the

interaction term Post-SOX × Non-Compliant, where Post-SOX is an indicator variable that

equals 1 if the year is 2002 or later, and equals 0 otherwise, and Non-Compliant is an indicator

variable that equals 1 if the firm was not in compliance in 2001, and equals 0 otherwise. Co-

19 We are particularly grateful to the referee for suggesting this specific line of attack and for shaping some of the other aspects of our approach to ameliorating endogeneity concerns. 20 Indicative evidence on the effects of SOX on pay and turnover is presented in Chhaochharia and Grinstein (2007), Carter, Lynch, and Zechman (2009), and Kaplan and Minton (2012).

option is not included in the above specification, and the focus is on the coefficient on Post-SOX

× Non-Compliant. This coefficient, however, captures both the effect we want to isolate

(through the exogenous shock to co-option) and direct effect (through other channels) of SOX.

To assess the impact of co-option, we estimate the modified regression, which includes Co-

option and the interaction of Co-option with the three dummy variables:

Pay = β0 + β1 Co-option + β2 Post-SOX  Co-option + β3 Non-Compliant  Co-option + β4 Post-SOX  Non-Compliant  Co-option + β5 Post-SOX + β6 Non-Compliant + k(Other Controls) + ε5

The controls in the specification include the independent variables used in the pay regressions in

Table 3 and the individual dummies, as well as the interactions of all the independent variables

with the three key dummy variables: Post-SOX, Non-Compliant, and Post-SOX × Non-

Compliant.

Panel A of Table 6 provides an estimate of the sensitivity of pay to co-option for the four

subsamples of firms: compliant firms in the pre-SOX period, non-compliant firms in the pre-

SOX period, compliant firms in the post-SOX period, and non-compliant firms in the post-SOX

period. The effects are estimated by taking the partial derivative of Pay with respect to Co-

option in the equation above. As can be seen from the table, β1 and β1+β3 represent the

sensitivities for compliant and non-compliant firms respectively in the pre-SOX period. Both

sensitivities include the bias due to endogeneity. The sensitivity of Pay to Co-option for

compliant firms in the post-SOX period is given by β1 + β2 and this includes not only the effect

of bias but, in addition, the direct effects of SOX. The sensitivities for firms in all three groups

are subject to bias due to the standard set of reasons that give rise to the endogeneity problem.

We allow this bias to differ by whether the firm was compliant (superscript C) or not compliant

(superscript NC) pre-SOX, though we do restrict BiasC to be the same both pre- and post-SOX.

The subsample of primary interest is the non-compliant post-SOX group. This group

contains firms facing the exogenous shock to co-option. The sensitivity for this subsample (= β1

+ β2 + β3 + β4) is contaminated by the SOX effects through channels other than co-option, and

thus represents the combined effect of both co-option and SOX on the variable of interest (=

“Clean + SOX”). As can be seen from the table, the typical DID estimate reported in the lower

right cell (β4) does not yield the clean estimate, but rather the negative of BiasNC. The “clean”

estimate, arising from the exogenous increase in co-option, forced on noncompliant firms

through a mandated increase in board independence, is given by β1 + β3 + β4.

Panel B of Table 6 provides the results. For brevity, we present only the clean estimates

for the total impact of co-option on each of the four variables of interest (turnover-performance

sensitivity, pay, pay-performance sensitivity, and investment). For ease of comparison, we

report results from the base-case regressions (Model 1 of Tables 2–5). In terms of the notation in

the regression specifications defined above, we report clean estimates of α1, θ1, γ1, and μ1.

For turnover and pay level, relative to the base case the estimates based on an exogenous

shock to co-option have the same sign and similar (though not quite as strong) statistical

significance. The clean estimate pertaining to the effect of co-option on CEO PPS, like that in

the base case, is statistically insignificant. The clean effect of co-option on investment policy is

still positive, but statistically weaker (t-statistic = 1.0 versus 2.5) than the estimated coefficient

from the base case.

6. Are All Independent Directors Equally Relevant for Board Monitoring?

In this section, we examine whether monitoring effectiveness of directors varies

depending on whether or not they are independent and by whether or not they are co-opted.

6.1. Co-option: independent versus non-independent directors

Our results thus far indicate that board capture is associated with weaker monitoring.

The measures of co-option used above, however, do not differentiate between directors who are

independent versus those that are not. Indeed, the notion that employee and affiliated directors

are co-opted is the basis for using board independence as a measure of monitoring in the first

place (e.g., Weisbach (1988) and Byrd and Hickman (1992)). The question remains as to

whether co-option blunts the monitoring effectiveness of independent directors. If we find that

co-opted directors who are independent are also weak monitors, then it would suggest that the

independence measure traditionally used in the literature does not capture the disposition of the

board to provide effective oversight and monitoring and can be improved. To examine this

question, we further refine Co-option. Co-opted Independence is defined as the proportion of the

board that consists of co-opted directors who are independent, while Co-opted Non-

Independence is defined as the proportion of the board that consists of co-opted directors who

are not independent. These two measures differentiate between directors who are employees or

affiliated versus those who are supposedly independent.

Table 7 documents the results. In Panel A, for ease of comparison, we reproduce our

base-case results on Co-option (Model 1 of Tables 2 – 5). We report the coefficients and

associated t-statistics only for our primary variables. Panels B and C report our results wherein

we replace Co-option by Co-opted Independence and Co-opted Non-Independence respectively.

In Panel B, we find that Co-opted Independence is associated with attenuated turnover-

performance sensitivity, higher pay, lower CEO delta, and higher investment. Per Panel C, Co-

opted Non-Independence is associated with attenuated turnover-performance sensitivity and

higher investment, but has no effect on pay or pay-performance sensitivity. Thus, our overall

results on co-option appear to be driven by independent co-opted directors, rather than by non-

independent co-opted directors. We conclude that, once a director is co-opted, the independence

of the director does not matter from a monitoring perspective. This likely explains why the

literature has found little uniform evidence on the relation between board independence and

various measures of firm performance and structure.

6.2. Independence: co-opted versus non-co-opted directors

Our results to this point suggest that: (i) independent directors typically do not have an

effect on monitoring effectiveness in the presence of co-option (Tables 2 – 5), and (ii) co-opted

directors, even those that are independent, are weak monitors (Table 7). It is likely, therefore,

that only independent directors who are not co-opted by the CEO are effective monitors. To test

this formally, we introduce a second new measure of board composition, Non-Co-opted

Independence. We define this as the proportion of the board that consists of independent

directors who were already on the board when the CEO assumed office.

Panel A of Table 8 depicts the overlap and dissimilarity between our measures of co-

option and independence. As can be seen, the sum of Co-opted Independence and Co-opted

Non-Independence equals Co-option, while the sum of Co-opted Independence and Non-Co-

opted Independence equals Independence.

Figure 2 plots how the various board composition measures described above change over

CEO tenure. As expected, Co-option increases with CEO tenure. This is because in each

director election cycle the CEO has the opportunity to affect the nomination of directors to the

board. Independence, however, remains more or less constant (at around 69% in our sample).

Thus, while on the surface it appears that board independence is high, the composition of the

board as represented by co-option gradually tilts in the CEO’s favor over time. Further, a closer

look at the two components of Independence indicates that as CEO tenure increases, Co-opted

Independence increases while Non-Co-opted Independence decreases. This arises as the CEO

replaces previously appointed independent with new independent directors. This suggests that

the monitoring effectiveness of the board weakens over the CEO’s tenure. We explore this issue

by estimating our base-case regressions with both Co-option and Independence replaced by Non-

Co-opted Independence as the key dependent variable.

Panel B of Table 8 reports the results. Consistent with the idea that independent directors

who are not co-opted are better monitors, we find Non-Co-opted Independence is associated with

higher sensitivity of CEO turnover to performance, lower pay levels, higher sensitivity of pay to

performance, and lower investment.

Overall, these results are consistent with non-co-opted directors being effective monitors.

Moreover, it appears that not all independent directors are the same. Differentiating among

independent directors by whether or not they are co-opted appears to be a more incisive way to

explain monitoring intensity of the board. Independent directors whose selection was influenced

by the CEO appear to be more sympathetic to the CEO. On the other hand, non-co-opted

independent directors appear to be effective monitors. Relative to aggregate board

independence, the representation of non-co-opted independent directors on the board appears to

be a sharper measure of monitoring effectiveness.

7. Alternative Interpretations and Other Robustness Checks

7.1. Is Co-option capturing the effect of CEO tenure?

Figure 2 shows that Co-option increases over the CEO’s tenure. It is likely that CEO

power increases with tenure (e.g., see Weisbach (1988) and Ryan and Wiggins (2004)). Thus,

CEO tenure may be correlated with both power and Co-option and it is possible that our results

on the effect of co-option are due to the positive association between co-option and CEO tenure.

We perform three tests and conclude that CEO tenure is not causing our results.

First, our base case specifications include CEO tenure as an additional control variable in

all our regression specifications. Thus the effect of co-option that we document earlier is after

controlling for CEO tenure.

Second, in the specifications in Tables 2 – 5 (Model 1), instead of Co-option, we use

Residual Co-option, which is the residual from a regression of Co-option on CEO tenure. The

residual now no longer proxies for power arising from tenure but is a proxy for power related to

co-option of the board. We find that all results on Residual Co-option are similar to our results

on Co-option.21

Third, we drop Co-option from all specifications and use only tenure as our measure of

the CEO’s power over the board. If it is true that our results on Co-option somehow obtain only

because of the positive correlation between tenure and Co-option, and do not reflect the true

effect of board capture, then when we drop Co-option from the regressions, our results on tenure

should be similar to what we reported earlier with Co-option. That is, we should find that tenure

decreases CEO turnover-performance sensitivity, increases CEO pay, decreases CEO pay-

performance sensitivity, and increases investment. The results, however, are not supportive of

the idea that co-option is only capturing the effect of CEO tenure on monitoring.22 We find that

CEO tenure has a positive effect on turnover-performance sensitivity, similar to the effect of co-

option. In contrast to the effect of co-option, however, CEO tenure has no effect on pay or

21 The results are similar when we use Residual TW Co-option (the residual from a regression of TW Co-option on CEO tenure) instead. 22 In the interests of conciseness, we do not tabulate the results here or below. All results are available from the authors on request.

investment and a positive association with pay-performance sensitivity.

Overall, the analysis in this section suggests that our results on Co-option are not driven

by the effects of CEO tenure.

7.2. Is Co-option capturing the effect of director inexperience?

Another possibility is that our measures of co-option reflect the inexperience of co-opted

directors, as indicated by the average tenure of directors on the board. Higher co-option could

mean that there are several new directors and these new directors are less experienced (at least on

that board). They may not be sympathetic to the CEO but, perhaps, they are slower to react to

poor performance and more ready to award favorable pay packages and approve investment

projects proposed by the CEO. Average director tenure, thus, would be correlated with co-

option as well as with director inexperience. To isolate the effect of co-option, as with CEO

tenure, we perform three different tests. First, we include average director tenure in our baseline

specifications. The results are similar to our base-case regressions. Second, we estimate the

residual from a regression of Co-option on average director tenure and then include this residual

(termed “Director-Tenure-Adjusted Co-option”) in the regressions instead of Co-option, along

with necessary interactions. We find that our results on the adjusted co-option measure are

similar to our results on Co-option. Finally, we exclude Co-option from our baseline

specifications but include average director inexperience, as well as the necessary interaction

terms. We find that average director tenure has no effect on turnover-performance sensitivity,

pay-performance sensitivity, or investment, and has a negative effect on pay.

Overall, there is no evidence that director inexperience causes a spurious relation

between co-option and monitoring effectiveness.

7.3. Is Co-option capturing the “Involved CEO” effect of Shivdasani/Yermack?

Shivdasani and Yermack (1999) construct an indicator variable, Involved CEO, which

equals 1 if the firm has a nominating committee and the CEO is part of the committee or if the

firm has no nominating committee. In the latter instance, the presumption is that if there is no

nominating committee the CEO will be more centrally involved in director selection. Shivdasani

and Yermack show that when a CEO is involved in director selection as part of the nominating

committee, then he or she tends to appoint more gray outside directors and fewer independent

outside directors. Shivdasani and Yermack do not examine the effect of Involved CEO on CEO

turnover-performance sensitivity, pay, pay-performance sensitivity, or firm investment. It is

possible, however, that our co-option variable does not provide additional explanatory power

relative to Involved CEO because both variables proxy for the same underlying construct, the

extent of the CEO’s influence on the board.

The data do not support this conjecture. First, following the 2002 NYSE and the

NASDAQ listing requirements for completely independent nominating committees, the

Shivdasani-Yermack measure should, by definition, be zero for all firms. Consistent with this,

we find that the mean of Involved CEO drops from about 54% in 1998 (the first year for which

RiskMetrics has data on sub-committees) to 1.9% in 2010, and the correlation between Co-

option and Involved CEO over the full time interval is only 0.11.

To verify that our co-option measure provides additional explanatory power relative to

Involved CEO, we perform three different tests. First, we include Involved CEO as an additional

explanatory variable in all our regressions. We find that the results on Co-option are

qualitatively similar. Second, we estimate the residual from a regression of Co-option on

Involved CEO. The residual, which we term Involved-CEO-Adjusted Co-option, is uncorrelated

with Involved CEO by construction. We include this residual along with the usual interactions in

Model 1 of Tables 2 – 5. We find that the results using Involved-CEO-Adjusted Co-option are

qualitatively similar to our results using Co-option. Third, we drop Co-option from our

regressions but include Involved CEO as well as relevant interactions. We find that Involved

CEO is not associated with turnover-performance sensitivity, pay levels, and pay-performance

sensitivity, but is positively associated with investment. Overall, the evidence does not support

the notion that Involved CEO drives our results. This in no way negates the results found in

Shivdasani and Yermack. Indeed, our findings on co-option are consistent with the premise of

their paper. Our results suggest that rule changes enacted subsequent to the sample period they

analyze reduce the relevance and explanatory power of their measure.

8. Conclusion

The economics of director selection and performance are more complex than standard,

widely used measures of board structure, such as board independence, would suggest. Director

co-option, by way of loyalty and allegiance to the CEO who is associated with appointment of a

director, is likely to affect what that director actually does. Directors who are appointed by a

CEO, regardless of whether they are independent in the legal sense, are more sympathetic to that

CEO (Mace (1971), Lorsch and MacIver (1989), Shivdasani and Yermack (1999)).

We propose and empirically examine two new measures of board composition that reflect

the extent of the CEO’s influence on directors. Co-option is the fraction of the board comprised

of directors appointed after the CEO assumes office. We provide evidence that co-option

reduces the monitoring effectiveness of the board of directors. We find that: (i) the sensitivity of

CEO turnover to performance decreases with co-option; (ii) CEO pay increases with co-option,

but the sensitivity of CEO pay to firm performance is unaffected; and (iii) investment increases

with co-option.

Testing robustness, the results are similar for various alternative measures of co-option.

These results obtain after we control for the traditional measure of board independence (the

fraction of independent directors). Our results are robust to our best attempts to address

endogeneity. We use firm-fixed effects in our main specifications. Alternatively, we use the

changes in NYSE and Nasdaq listing requirements, which led to an exogenous increase in co-

option in firms that were non-compliant as of the announcement. Using this clean source of

identification, we find qualitatively similar results on the effects of co-option on turnover, pay,

and pay-performance sensitivity, although a weaker estimated effect on investment.

Further analysis reveals that independent directors that are co-opted behave as though

they are not independent. Thus, we refine the conventional measure of board independence to be

the proportion of board members who are independent directors but are not co-opted. This

second measure, Non-Co-opted Independence, focuses on independent directors who are most

likely to be truly independent. We find that boards with a higher proportion of non-co-opted

independent directors provide more effective monitoring. Higher Non-Co-opted Independence is

associated with higher sensitivity of CEO turnover to performance, lower pay levels, higher

sensitivity of pay to performance, and lower investment.

In terms of future research, it is possible that there are offsetting economic benefits to co-

option. For example, CEOs with more co-opted boards may be more insulated from the threat of

dismissal and, consequently, may invest more in firm-specific human capital (Burkart, Gromb,

and Panunzi (1997)) or in risky projects with long term payoffs (DeAngelo and Rice (1983) and

Stein (1988)). Proxies for such human capital intensity (HCI) include within-industry

heterogeneity in growth rates, R&D intensity, and firm-specific risk. In regression results not

reported here, we find some preliminary evidence that Co-option is positively associated with

these HCI proxies and that Tobin’s q increases in Co-option for high–HCI firms. We defer

presenting those empirical results in detail until we develop defensible conceptual and

econometric foundations for the empirical analysis.

More broadly, beyond turnover-performance sensitivity, pay level, executive pay-

performance-sensitivity, and investment policy––the issues we examine herein––there is a large

literature on the importance of board monitoring in empirical corporate finance. When

appropriate, researchers can reexamine prior research questions and the corresponding

experiments using new measures, Co-option and Non-Co-opted Independence. For example,

experiments that regress structure on structure would relate our measures of board capture to

takeover bids (hostile or friendly, as bidder or target), takeover defenses (such as poison pills,

antitakeover charter amendments, staggered boards, blank-check preferred stock; and

supermajority voting rules), state takeover law (as determined by domicile), product market

characteristics (including monopoly power and regulatory structure), financial policy (e.g.,

capital structure, cash holdings, and hedging), institutional ownership, investment policy (such as

firm focus and asset intangibility), dividend policy, and other firm characteristics (such as size

and whether it is organized around product markets or functional areas). Performance-on-

structure designs, assuming endogeneity concerns can be overcome, would examine the relation

between Tobin’s q, accounting performance, or stock market performance, for example, and Co-

option or Non-Co-opted Independence. Finally, our co-option measures may well have power to

explain cross-sectional variation in event study returns to announcement of M&A events or

adoption of defensive devices. There appears to be significant remaining empirical opportunity

for researchers interested in board monitoring.

Appendix A.1. Variable definitions

The governance data used in this study are from RiskMetrics (formerly Investor Responsibility Research Center or IRRC) for the period 1996-2010. Compensation data are from Execucomp, accounting data are from Compustat, and stock return data are from CRSP. We also give the mnemonics used by Compustat to define these variables. A.1.1. Co-option-related variables Variable Definition

Co-opted director Director who joined the board after the CEO assumed office

Co-option Number of co-opted directors / Board Size = Co-opted Independence + Co-opted Non-Independence

TW Co-option (Tenure-Weighted Co-option)

Sum of tenure of co-opted directors divided by the sum of tenure of all directors

Residual Co-option Residual from a regression of Co-option on CEO tenure

Residual TW Co-option Residual from a regression of TW Co-option on CEO tenure

Co-opted Independence Number of co-opted independent directors / Board Size

Co-opted Non-Independence Number of co-opted non-independent directors / Board Size

Non-Co-opted Independence Number of independent directors who were already on the board before the CEO assumed office / Board Size

A.1.2. Governance-related variables Variable Definition

Board Size Total number of directors on the board

Independence Number of independent directors / Board size = Co-opted Independence + Non-Co-opted Independence

Outside Director Ownership Cumulative share ownership of outside directors (available from 1998)

GIM Index Governance index as defined in Gompers, Ishii, and Metrick (2003)

Female Director = 1 if board has a female director; = 0 otherwise (available from 1997)

Average Director Tenure Average tenure of all directors on board

CEO Ownership Shares held by the CEO / Number of shares outstanding

CEO Duality = 1 if CEO is also the Chairman; = 0 otherwise

A.1.3. CEO-related variables Variable Definition Forced Turnover = 1 if the departing CEO was younger than 60 years of age; = 0

otherwise

CEO Pay Total annual pay (Execucomp: TDC1)

CEO Pay-Performance Sensitivity (Delta)

Expected dollar change in CEO wealth for a 1% change in stock price (using entire portfolio of stocks and options) computed as in Core and Guay (2002)

CEO Vega Expected dollar change in CEO wealth for a 0.01 change in stock return volatility (using entire portfolio of options) computed as in Guay (1999)

CEO Tenure Tenure as CEO of the firm = fiscal year end date minus date became CEO

CEO Cash Compensation Sum of salary and bonus (Execucomp: TCC)

Involved CEO = 1 if the firm has no nominating committee, or if the firm has a nominating committee and the CEO is a member; = 0 otherwise

A.1.4. Firm-level variables Variable Definition Mnemonic Investment Capital Expenditure / Assets capx / at

Return Stock return over the fiscal year RET variable in Execucomp

ROA Return on Assets = EBITDA/Assets oibdp / at

FCF/Assets Free cash flow / Assets (oancf – dvc – dvp) / at

Leverage Total Debt / Assets (dltt + dlc) / at

Firm Size Log(sales) Log(sale)

Tobin’s q (Assets – Book equity + Market equity) / Assets

(at – ceq + prcc_f×csho) / at

Unsystematic Risk Standard deviation of residuals from Fama-French regression using daily stock returns

R&D/Assets R&D expenditure / Assets xrd / at

Sales Growth Sales / Lagged Sales sale / Lagged(sale)

A.2. RiskMetrics adjustment

Our governance data are drawn from RiskMetrics. One complication with this dataset is that there is no unique firm identifier or unique director identifier across all the years. Although RiskMetrics has CUSIP, TICKER, NAME, and RT_ID as company identifiers, none of these are unique across the entire sample period. Similarly, RiskMetrics has LEGACY_DIRECTOR_ID and DIRECTOR_DETAIL_ID as director identifiers, but these are also not unique or consistent across the entire sample period. WRDS states on its website “Each time IRRC - ISS - RiskMetrics was acquired, new identification methods were introduced…Constructing a complete time series in the directors data is not easy. There is no single variable that is populated for all companies and all years.” We explain below how we resolve the lack of unique firm and director identifiers. In the process, we also associate each firm-year with unique PERMNO and GVKEY, which enables merging RiskMetrics dataset with CRSP and Compustat. In what follows below, variables we create are indicated in italics, while variables drawn from standard databases are in upper case.

The authors are happy to provide the exact GVKEY – PERMNO – YEAR – CUSIP – MEETINGDATE – LEGACY_DIRECTOR_ID – DIRECTOR_DETAIL_ID – Our Director ID combinations on request, where

YEAR = fiscal year of data (from Compustat)

CUSIP = firm CUSIP (from RiskMetrics)

MEETINGDATE = annual meeting date (from RiskMetrics)

LEGACY_ DIRECTOR_ID, DIRECTOR_DETAIL_ID = Director ID variables (from RiskMetrics)

Our Director ID = unique director ID that we create

Researchers can then merge our dataset with the RiskMetrics dataset using CUSIP – MEETINGDATE – LEGACY_DIRECTOR_ID – DIRECTOR DETAIL ID to obtain director data, use GVKEY – YEAR to merge with Compustat, use PERMNO (and MEETINGDATE if required) to merge with CRSP, use GVKEY or PERMNO to uniquely identify firms, and use Our Director ID to uniquely identify directors.

A.2.1. Issues with firm identifier

Starting in 2007, RiskMetrics changed the methodology used to collect data. As a result, data on directors are provided in two datasets in WRDS: a “Directors Legacy” dataset that provides data from 1996-2006 and a “Directors” dataset that provides information from 2007 onwards.

For records from 1996-2006, the CUSIP provided is the 6-digit CUSIP (we refer to this as Cusip6) and for records on or after 2007 the CUSIP provided is the 9-digit CUSIP. We use only the first 8 digits of this (we refer to this as Cusip8) because CRSP provides only 8-digit CUSIP. Additionally, Cusip6 is the header CUSIP, which means that each time the database was updated the most recent CUSIP was applied to the entire history for each company. Since this dataset was not updated after 2007, the Cusip6 was the 6-digit CUSIP for the firm as of 2006 (or, as of the last date the firm was included in the database). Cusip8, on the other hand, is the actual CUSIP for the firm for that year. In both datasets, RiskMetrics also provides the ticker

(TICKER) and company name (NAME) associated with the firm for that year. We use these in addition to CUSIP to identify firms.23

We obtain the PERMNOs from CRSP for each firm using the methodology below. We start with the dataset DSENAMES from CRSP, which contains, for each PERMNO, the full history of changes in NCUSIP (the historic CUSIP), firm ticker (TICKER), and firm name (COMNAM). From this dataset, we create a dataset of unique NCUSIP - PERMNO combinations along with the corresponding date range over which this combination is valid (denoted by us as Start Date and End Date). These dates are obtained using NAMEDT and NAMEENDT variables from CRSP. We do this separately for 8-digit NCUSIPs (NCusip8 – PERMNO – Start Date – End Date) as well as for 6-digit NCUSIPS (NCusip6 – PERMNO – Start Date – End Date). Similarly, we obtain, for each PERMNO, all the tickers associated with it and its relevant date range (TICKER – PERMNO – Start Date – End Date) and all the firm names associated with it and its relevant date range (COMNAM – PERMNO – Start Date – End Date).

We start with 23,242 unique firm-year observations in the RiskMetrics dataset (downloaded as of Nov 2011 – recent downloads will clearly have a different number of observations). Of this, 5,810 observations correspond to the post-2007 period. We match the post-2007 RiskMetrics data with the NCusip8 - PERMNO – Start Date – End Date dataset that we created above using Cusip8 and MEETINGDATE variables of RiskMetrics. We obtain a unique PERMNO for each firm by ensuring that MEETINGDATE is between the Start Date and End Date. We obtain a match for 5366 out of the 5810 observations.

For the remaining 444 (= 5810 - 5366) observations that do not match using Cusip8, we match using Cusip6. Specifically, we match the RiskMetrics data with the NCusip6 - PERMNO – Start Date – End Date dataset. Again we impose the condition that the MEETINGDATE should be between the Start Date and End Date for which the NCusip6 is valid. We find duplicate matches for firms that have multiple traded securities, particularly dual class shares. We identify the correct PERMNO in these cases using data on dual-class shares from Gompers, Ishii, and Metrick, 2010 (available through WRDS) and through web searches. This yields 109 firm-year observations that match on Cusip6.

For the remaining 335 (= 444 - 109) observations, we match using RiskMetrics variable TICKER. We then identify firms where the RiskMetrics TICKER matches with the CRSP TICKER.24 After imposing the condition that MEETINGDATE should be between the Start Date and End Date for which the TICKER is valid, and eliminating duplicates as before, we obtain a match for 319 firm-year observations.

Finally, for the remaining 16 (= 335 - 319) observations, we match the RiskMetrics company name (NAME) with the CRSP name (COMNAM). We do not find any matches after 23 RiskMetrics includes 2 company identifiers: LEGACY_PPS_ID (before 2004) and RT_ID (after 2004). Neither of these uniquely identify firms across the entire time period. Further, they cannot be used to merge with other databases such as CRSP etc. because they are created by IRRC and by WRDS. Therefore we do not use these IDs. 24 RiskMetrics ticker includes a period character in the case of dual class securities. For example, Berkshire Hathaway’s ticker is BRK.B. CRSP does not follow the same nomenclature. We therefore remove the period and any characters after the period (“.B” in the Berkshire example) from the RiskMetrics ticker before matching with CRSP ticker. Also, before matching by name, we remove the following characters – space, period, ampersand, and comma – in both CRSP and RiskMetrics names.

imposing the condition that MEETINGDATE should be between the Start Date and End Date for which COMNAM is valid.

We attempt to populate these 16 observations as follows. If two firm-years have the same Cusip8 around the year for which we have not been able to find a PERMNO, then we assign the PERMNO of the adjoining year for the missing observation. This procedure yields a match for another 10 observations.

For the pre-2007 period, we start with 17,432 firm-year observations on RiskMetrics. We follow a very similar procedure as we did for the post-2007 data, but with one minor difference. Since the Cusip6 is the header CUSIP, we cannot match based on the actual MEETINGDATE as we did with the post-2007 data. We therefore first identify the last year for which the firm was in the RiskMetrics legacy dataset. We label this as LastMeetingDate and we assign this for all years for that firm.

We then merge the RiskMetrics data with our NCusip6 - PERMNO – Start Date – End Date dataset and require the LastMeetingDate to be between the Start Date and End Date. We obtain a match for 17,089 observations. We pick up duplicates (308 firms on RiskMetrics match with more than one PERMNO on CRSP) because of dual-class shares. As before, we check each of these manually to identify the correct PERMNO and eliminate all duplicates.

We next match the 343 (= 17,432 - 17,089) remaining observations using TICKER. We require the MEETINGDATE to be between the Start Date and End Date (unlike Cusip6, which is the header CUSIP on RiskMetrics, TICKER and NAME are historic). After eliminating duplicate matches by manually checking the observations, we identify the correct PERMNOs for 312 firm-years. That leaves 31 observations, which we match using NAME and MEETINGDATE. We find 1 match here. Finally, as before, for the 30 observations for which we could not find a match, we identify the correct PERMNOs for 23 observations by imposing the condition that if two firm-years have the same NCusip6, then they should have the same PERMNO.

Overall, we are able to find a unique PERMNO for 5,804 of the 5,810 firm-year observations post-2007, and for 17,425 of the 17,432 observations pre-2007. We check these 13 observations by hand and find that many of them do not match because of data-entry errors by RiskMetrics, and in some cases, the MEETINGDATE precedes the Start Date by a few days. Once we correct for these, we are able to obtain matching PERMNOs for all 13 of these observations. Thus our procedure finds a matching PERMNO for every observation on RiskMetrics.

We finally match this data with the CRSP–Compustat combined database to obtain GVKEY and YEAR (fiscal year of the data) associated with each firm-year on RiskMetrics using PERMNO and MEETINGDATE by ensuring that MEETINGDATE falls between the fiscal year beginning and ending dates. Either the GVKEY or PERMNO then uniquely identifies the sample firms.

A.2.2. Issues with director identifier

There is no unique director identifier (ID) across the two RiskMetrics directors’ datasets. Two sets of director IDs are maintained. The first is LEGACY_DIRECTOR_ID and the second is DIRECTOR_DETAIL_ID. The latter variable is supplied by Wharton Research Data Services (WRDS) starting from 2004 onwards. All directors in the RiskMetrics database as of 2004 (both

directors who started in 2004 and directors who started prior to 2004 but are still on the board as of 2004) have a valid DIRECTOR_DETAIL_ID from 2004 onwards. WRDS assigned the same ID to directors in earlier years where possible through matching of names, director age, etc. Directors in the database who quit before 2004 do not have a valid DIRECTOR_DETAIL_ID. Thus, prior to 2004, some directors have no DIRECTOR_DETAIL_ID. Over the sample period, 23% of the director-year observations have missing DIRECTOR_DETAIL_ID.

Pre-2004, the unique director ID is LEGACY_DIRECTOR_ID. In contrast to DIRECTOR_DETAIL_ID, this variable is populated only for those directors who were on the database prior to 2004. For all years the directors are on the database, they will continue to have LEGACY_DIRECTOR_ID. Directors who joined on or after 2004 will not have LEGACY_DIRECTOR_ID. Over the sample period, 27% of the director-year observations have missing LEGACY_DIRECTOR_ID. Eventually, as more years of data are added and directors who started serving prior to 2004 are no more on the boards of firms, this data item will not be relevant.

To sum up, for directors who enter the database on or after 2004, only DIRECTOR_DETAIL_ID is available. For directors who exit the database before 2004 only LEGACY_DIRECTOR_ID is available. For directors who joined before 2004 and are still on the board after 2004, both IDs are available. Thus, it is imperative that both IDs are used to create a unique director ID to ensure that all directors are included in the study.

We first confirm that for directors for whom both IDs are available, there is a 1-for-1 match between the two IDs. Second, when both IDs should be available, but only one is given in the data base, we make necessary adjustments. There are times where WRDS should have been able to assign a DIRECTOR_DETAIL_ID to a given director based on the director’s name and age, but they do not do so. For example, director Joe Smith may have served on firm ABC Inc from 2003–2005. Suppose his LEGACY_DIRECTOR_ID is 5555 and his DIRECTOR_DETAIL_ID is 6666. Then WRDS should have assigned this DIRECTOR_DETAIL_ID for Joe Smith for every year from 2003–2005. But we find that this is not always true – sometimes Joe Smith has a missing DIRECTOR_DETAIL_ID in 2004 (say). We know he is the same director because the LEGACY_DIRECTOR_ID is the same. Therefore we assign Joe Smith the same DIRECTOR_DETAIL_ID of 6666 in 2004.

year LEGACY_DIRECTOR_ID DIRECTOR_DETAIL_ID Corrected DIRECTOR_DETAIL_ID

2003 5555 6666 6666 2004 5555 - 6666 2005 5555 6666 6666

The reverse is also true – sometimes the LEGACY_DIRECTOR_ID is not filled in completely even though DIRECTOR_DETAIL_ID is available for all the years. For these director-year observations, we do a similar assignment and correct the LEGACY _DIRECTOR_ID (we call this Corrected LEGACY_DIRECTOR_ID).

We then create our own unique director ID, which we term as Our Director ID, as follows: if Corrected DIRECTOR_DETAIL_ID is available, then we use this as the unique ID, because this is the ID that WRDS will maintain going forward. If Corrected DIRECTOR_DETAIL_ID is not available (which implies the director exited the database prior

to 2004), we define an ID that is Corrected LEGACY_DIRECTOR_ID×100. The multiplication by 100 ensures that the ID we create is unique and will not clash with the WRDS ID. An example of what we do follows below.

Joe Smith, director ABC Inc from 2003–2005

Year Corrected LEGACY_DIRECTOR_ID

Corrected DIRECTOR_DETAIL_ID

Our Director ID

2003 5555 6666 6666 2004 5555 6666 6666 2005 5555 6666 6666 Jane Adams, director ABC Inc from 2004–2006 Year Corrected

LEGACY_DIRECTOR_ID Corrected DIRECTOR_DETAIL_ID

Our Director ID

2004 - 6667 6667 2005 - 6667 6667 2006 - 6667 6667 John Doe, director ABC Inc from 2001–2003 Year Corrected

LEGACY_DIRECTOR_ID Corrected DIRECTOR_DETAIL_ID

Our Director ID

2001 5556 - 555600 2002 5556 - 555600 2003 5556 - 555600

By using both IDs, we ensure that we account for all directors in the database. The final dataset (after cleaning) has 220,963 director-years. If a researcher uses only LEGACY_DIRECTOR_ID, she would have only 195,547 director-year observations. Similarly, if a researcher uses only DIRECTOR_DETAIL_ID, she would have only 170,068 director-year observations. Importantly, if one does not consider both IDs, then board size and co-option and all of the board measures (independence, busyness) will be incorrectly estimated as we would be ignoring a large sample of directors.

A.2.3. Other Issues with DIRECTOR_DETAIL_ID

There are some problems specifically with the DIRECTOR_DETAIL_ID of 35025. It appears to have been wrongly assigned in certain years. We find that this ID appears in 772 observations, of which 719 are in 2004, 43 are in 2005, and the remaining 10 occur in years 2006 – 2010. We examine these 10 observations and find that they all belong to the same director, H. Paulett Eberhart (or slight variations of this name, such as H. Eberhart).

To fix this, we flag observations where the firm (identified using GVKEY), director name (FULLNAME), and the director’s joining year (DIRSINCE) is the same in two consecutive years, but the DIRECTOR_DETAIL_ID is different. This happens because there is wrong assignment of DIRECTOR_DETAIL_ID of 35025 to directors other than H. Paulett Eberhart. For example, in Metlife and General Dynamics, director JOHN M KEANE appears in 2004 with a DIRECTOR_DETAIL_ID of 35025, but in subsequent years with a

DIRECTOR_DETAIL_ID of 35937. For this director, we assign a Corrected DIRECTOR_DETAIL_ID of 35937 in 2004. Where we are not able to correct this mistake, we drop the firm completely from the analysis. Finally, there are data entry errors in RiskMetrics. For example, MARY L GOOD, director of ACXIOM Corp is named as MARY L GOOD PHD in one year and MARY L GOOD in another, but she is given different DIRECTOR_DETAIL_IDs of 112331 and 9612. We correct such data errors to the best of our ability. The manual correction provides the correct DIRECTOR_DETAIL_ID to some of the directors who are mistakenly assigned a DIRECTOR_DETAIL_ID of 35025, and also corrects mistakes where the same director in the same firm is assigned multiple DIRECTOR_DETAIL_IDs.

A.2.4. Issues with MEETINGDATE

RiskMetrics provides two variables to identify the year for which the observation is valid. One is the MEETINGDATE (date of the annual meeting) and the other is YEAR. In theory, YEAR in RiskMetrics is the year of the MEETINGDATE (this is not to be confused with YEAR in Compustat mentioned earlier in the Appendix, which refers to the fiscal year). We find that 482 director-year observations have the YEAR variable not equal to the year indicated by the MEETINGDATE. Of this, 427 observations have the year of MEETINGDATE equal to 1960 whereas the YEAR is given as 2008. For example, for Zimmer Holdings, the YEAR is given as 2008, but the MEETINGDATE is given as 1/1/1960. We correct the year indicated by the MEETINGDATE variable for such observations to be 2008. That is, in this example we replace MEETINGDATE by 1/1/2008. This is important because when we obtain PERMNOs from CRSP, we use the MEETINGDATE variable. The remaining 55 observations (= 482 - 427) have a difference of either one year or three years between the year indicated by MEETINGDATE and YEAR. We check each of these and find that the YEAR variable is correct, but not the MEETINGDATE. We therefore change the year of the MEETINGDATE to be the same as YEAR.

A.3. Discussion of Changes in Reporting of Annual Pay (TDC1)

Execucomp recently changed its compensation data reporting in line with accounting changes imposed by the Financial Accounting Standards Board (FASB) as well as expanded compensation disclosure requirements (for pension, severance, change-in-control payouts, and equity based compensation) imposed by the Securities and Exchanges Commission (SEC). For fiscal years 1992–2005, all companies on Execucomp report using the old reporting format. The data format reporting code is identified using Execucomp variable OLD_DATAFMT_FLAG. For fiscal years 2007 and later, all firms on Execucomp report compensation using the new format. For 2006, 16% of firms report using the old format while the rest report under the new format. We loosely refer to these different reporting formats as the pre-2006 format (OLD_DATAFMT_FLAG = 1) and the post-2006 format (OLD_DATAFMT_FLAG = 0).

The annual compensation (TDC1) pre- and post-2006 is not strictly comparable for at least three reasons. (i) The components and the composition of the components included in TDC1 differ pre- and post-2006. (ii) Pre-2006, equity-based long-term (multi-year) incentive awards were reported based on actual payouts during the fiscal year, while post-2006, they are reported based on target (or expected) levels of payout. (iii) Pre-2006, Execucomp computed the Black-Scholes value of option grants (which is one component of TDC1), whereas post-2006,

Execucomp provides the fair value (which may or may not be the Black-Scholes value) of option grants as reported by firms. Unfortunately, it is impossible to reconcile pre- and post-2006 TDC1 numbers without making some major assumptions. We elaborate below.

Pre-2006, the Execucomp definition of TDC1 was

TDC1 = SALARY + BONUS + LTIP + RSTKGRNT +

OPTION_AWARDS_BLK_VALUE + OTHANN + ALLOTHTOT (A1)

Post-2006, Execucomp definition of TDC1 is

TDC1 = SALARY + BONUS + NONEQ_INCENT + STOCK_AWARDS_FV +

OPTION_AWARDS_FV + OTHCOMP + DEFER_RPT_AS_COMP_TOT (A2)

We provide below a quick recap of the key changes in each of these components from pre- to post-2006. Our understanding is based on talks with Standard & Poor officials (who supply the Execucomp database), on “A User’s Guide to the SEC’s New Rules for Reporting Executive Pay” issued by Moody’s Investors Service (2007), and on explanations provided by Wharton Research Database Services (WRDS).

Execucomp variable Pre-2006 definition Post-2006 definition

SALARY Base annual salary Base annual salary

BONUS Performance-based pay that is:

 Earned during the year  Based on non-formulaic (also

called “discretionary” or “guaranteed”) as well as formulaic plans

 Based on annual (as opposed to multi-year) performance targets

 Cash-based (as opposed to equity-based)

Performance-based pay that is:

 Earned during the year  Based on non-formulaic

plans  Based on annual performance

targets  Cash-based

NONEQ_INCENT Not reported pre-2006 Performance-based pay that is:

 Earned during the year  Based on formulaic plans  Based on annual as well as

multi-year performance targets

 Cash-based

Execucomp variable Pre-2006 definition Post-2006 definition

LTIP Performance-based pay that is:

 Earned during the year  Based on formulaic plans  Cash-based when using

multi-year performance targets or equity-based when using either annual or multi- year performance targets

Not reported post-2006

RSTKGRNT Value of restricted stock granted during the year (estimated by Execucomp assuming Jul 1st as grant date)

Not reported post-2006

STOCK_AWARDS_ FV

Not reported pre-2006 Fair value (estimated by the company as of the grant date) of restricted stock grants and performance-based pay that is yet unearned but will result in stock awards in the future. For performance-based pay, the fair value is, typically, based on target payouts.

OPTION_AWARDS_ BLK_VALUE

Value of options granted during year (estimated by Execucomp using Black-Scholes methodology assuming Jul 1st as grant date)

Not reported post-2006

OPTION_AWARDS_ FV

Not reported pre-2006 Fair value (estimated by the company as of the grant date) of option grants and performance- based pay that is yet unearned but will result in option awards in the future. For performance-based pay, fair value is, typically, based on target payouts.

Execucomp variable Pre-2006 definition Post-2006 definition

OTHANN Execucomp defines this as “perquisites and other personal benefits, above market earnings on restricted stock, options/SARs or deferred compensation paid during the year but deferred by the officer, earnings on long-term incentive plan compensation paid during the year but deferred at the election of the officer, tax reimbursements, dollar value of difference between the price paid by the officer for company stock and the actual market price of the stock under a stock purchase plan that is not generally available to shareholders or employees of the company.”

Not reported post-2006

ALLOTHTOT Execucomp defines this as “severance payments, debt forgiveness, imputed interest, payouts for cancellation of stock options, payment for unused vacation, tax reimbursements, signing bonuses, 401K contributions, and life insurance premiums.”

Not reported post-2006

OTHCOMP Not reported pre-2006 Execucomp defines this as “other compensation received by the executive including perquisites and other personal benefits, termination or change-in-control payments, contributions to defined contribution plans (e.g. 401K plans), life insurance premiums, gross-ups and other tax reimbursements, discounted share purchases etc.”

DEFER_RPT_AS _COMP_TOT

Not reported pre-2006 Execucomp defines this as “deferred compensation earnings as reported in “summary compensation table.”

As seen from the discussion above, the only component of compensation that is directly comparable across the two reporting regimes is SALARY.

A.3.1. Performance-based pay

A big difference pre- and post-2006 is in the reporting of performance-based pay. Below for ease of understanding we present the information in a different manner, where the focus is on how the various components of performance-based pay are reported pre- and post-2006.

Component of Performance-Based Pay (Plan, Performance target, Payout type)

Pre-2006 Post-2006

Non-formulaic, Annual, Cash BONUS BONUS

Formulaic, Annual, Cash BONUS NONEQ_INCENT

Formulaic, Multi-year, Cash LTIP NONEQ_INCENT

Formulaic, Annual/Multi-year, Stock LTIP STOCK_AWARDS_FV

Formulaic, Annual/Multi-year, Option LTIP OPTION_AWARDS_FV

Clearly comparing pre- and post-2006 is difficult. BONUS is comparable across the two- periods only if we assume that all cash-based payments from formulaic annual plans are zero post-2006. Similarly BONUS pre-2006 is comparable to BONUS + NONEQ_INCENT post- 2006 only if we assume that cash-based payments from formulaic multi-year plans are zero in the pre-2006 and/or in the post-2006 period. There is no way to compare performance-based pay that is paid in the form of shares or options because pre-2006 such non-cash performance-based pay was recorded in the year it was actually paid, whereas post-2006 it is recorded at fair value when the performance-based plan is introduced and not when the payouts are earned by the executive. Firms report the threshold, target, and maximum number of units (cash, stock or options) that can be earned by the executive under the plan, as well as a fair value estimated based on target-level grants. The threshold is the number of units that will be received by the executive if the performance crosses a pre-specified threshold. The plan also specifies the maximum awards that can be received by the executive and level of performance necessary to achieve this maximum. The target is the level of awards that the board expects the executive to earn. We read several proxy statements that stated that executives did not actually earn these awards because of the failure of the executives to meet targets.

The data needed to evaluate the value of these awards are not available in Execucomp, and sometimes not available in the proxies either.25 Adding to this lack of data, there are several instances where Execucomp makes mistakes in terms of recording these awards. There appear to

25 For example, IBM indicates in its 2013 proxy statement that the performance criteria for its executives are revenue growth, operating net income, operating EPS, and free cash flow, but does not indicate what levels of these measures have to be achieved for the awards to vest. The firm states “IBM is not disclosing specific targets under the annual and long-term plans because it would signal IBM’s strategic focus areas and impair IBM’s ability to leverage these areas for competitive advantage.”

be at least three types of mistakes: (i) The firm actually reports the dollar value of threshold, target, and maximum award levels but Execucomp wrongly records these as number of threshold, target, and maximum award levels (examples: Briggs & Stratton for 2008 and 2010, Nacco Industries for 2006). (ii) The firm gives performance-based option awards but Execucomp records these as stock grants (examples: SLM Corp in 2008, Epicor in 2008, Mentor Graphics in 2006). (iii) The award is in performance units, where 1 unit is not equivalent to 1 share, but Execucomp records these as number of shares (for example, JM Smucker in 2007– 2010).

To estimate the correct fair value for such awards, we need to know the performance measure being used by the firm and the level of performance needed to earn these awards. Since this information is not available in Execucomp, following the methodology adopted by some firms, we estimate the fair value of performance-based stock awards as the target level of awards multiplied by the closing stock price on the date the plan is put in place. Similarly, for performance-based option awards, we estimate the fair value using the target number of options, the reported exercise price, time-to-maturity, and other variables needed for the Black-Scholes value such as volatility, dividend yield, and risk-free rate. The volatility is the annualized standard deviation of stock returns estimated over the previous 60 months prior to the beginning of the fiscal year period, the dividend yield is the average dividend yield over the current year and the previous two years, and the risk-free rate is the rate of return on a Treasury security with comparable maturity. We then compare our estimates with the fair value estimates provided by the firms. Not surprisingly, we find only a 23% correlation between the two values.

The low correlation is not only because of the data errors mentioned above, but also because of variation in the way firms estimate the value of these awards. (i) While most firms evaluate the fair value at the target payout level, some like AIG (see proxy for 2008) evaluate the fair value at the maximum payout level.26 (ii) Firms may estimate the fair value based on the probability that the executive will achieve the performance level that will trigger the target payout (iii) Some firms assume that the awards will be paid out at the target level three years from the grant date (the typical performance evaluation period) and discount this back at some appropriate discount rate. (iv) Some firms adjust for the dividend paid on the shares as the unearned shares do not qualify for dividends.

A.3.2. Stock grants and option grants

Not only is performance-based pay not comparable, the value of stock and option grants is also not comparable. RSTKGRNT cannot be compared to STOCK_AWARDS_FV and OPTION_AWARDS_BLACK_VALUE cannot be compared to OPTION_AWARDS_FV because of two reasons. (i) RSTKGRNT and OPTION_AWARDS_BLACK VALUE are the value of annual stock and options grants only. STOCKS_AWARDS_FV includes the value of annual stock grants as well as unearned stock (which will be earned only when certain performance conditions are met). Similarly, OPTION_AWARDS_FV includes the value of annual option grants as well as unearned options. This strictly is not a problem because firms

26 The SEC amended its reporting rules in 2009 by requiring that performance-based awards be reported on the basis of the probable outcome of the performance condition, rather than the amount payable for maximum performance.

break out fair values of earned and unearned awards separately.27 (ii) Pre-2006, Execucomp estimated the value of grants whereas post-2006, firms estimate and report the fair value themselves. This creates problems because the methodology used by firms need not be the same across firms and need not be consistent within firms over time. For options, even if all firms used the most-widely used Black-Scholes model, firms could use different assumptions about maturity because some firms could give a maturity haircut to reflect early exercise (e.g., due to executive risk aversion). Moreover, firms could use different assumptions about volatility.

To understand the magnitude of the problem, we first identify the restricted stock grants portion of STOCK_AWARDS_FV. We then value these restricted stock grants as the number of stocks granted times the closing stock price as on grant date. We find a 92% correlation between our estimates and the corresponding fair values reported by firms. Similarly, we first identify the option grants portion of OPTION_AWARDS_FV. We then estimate the Black- Scholes value of these option grants as described earlier. We find a 91% correlation between our estimates and the fair values reported by firms for option grants.

A.3.3. Adjusted TDC1

Finally, we compute an adjusted measure of TDC1. We use the equations (A1) and (A2) given earlier for TDC1 computation. For firm-years using the pre-2006 reporting format, we compute our own estimate of the value of restricted stock grants (in lieu of RSTKGRNT) and the value of option grants (in lieu of OPTION_AWARDS_BLK_VALUE) as discussed in Section A.3.2. The values of other variables needed for TDC1 are taken from Execucomp. For firm- years using the post-2006 reporting format, we estimate the fair value of stock grants and unearned stock awards (in lieu of STOCK_AWARDS_FV) and the fair value of option grants and unearned option awards (in lieu of OPTION_AWARDS_FV) using the method discussed in Section A.3.1 and Section A.3.2. Again, the values of other variables are taken from Execucomp. Overall the correlation between our adjusted TDC1 estimate and the TDC1 reported by Execucomp is over 99% in the pre-2006 period and 44% in post-period.

That TDC1 in the pre- and post-2006 period are so differently defined matters if the annual change in TDC1 is of interest to the researcher. In 2006, with the vast majority of firms (86%) switching to the new reporting format, the change in TDC1 relative to the previous year is misleading. This is true even in 2007, when the remaining 14% of firms reported under the new format. One way around this issue is to compute the adjusted TDC1 as we describe here. But we acknowledge that this measure still has drawbacks, particularly, our inability to estimate the value of performance-based pay using Execucomp data. Another way is to ignore 2006 altogether, as well as 2007 if the firm had reported in the old format in 2006.

27 Execucomp provides details of each type of award (stock grants, option grants, and unearned stock awards, unearned option awards) in the “Plan Based Awards” table along with their fair value (FAIR_VALUE). The sum of this variable across all stock grants and unearned stock awards is indicated as STOCK_AWARDS_FV and the sum across all option grants and unearned option awards is indicated as OPTION_AWARDS_FV in the “Annual Compensation” table.

References

Adams, R., and D. Ferreira. 2009. Women in the Boardroom and their Impact on Governance and Performance. Journal of Financial Economics 94, 291-309.

Adams, R., B. Hermalin, and M. Weisbach. 2010. The Role of Boards of Directors in Corporate Governance: A Conceptual Framework and Survey. Journal of Economic Literature 48: 58- 107.

Aggarwal, R., and A. Samwick. 1999. The Other Side of the Trade-off: the Impact of Risk on Executive Compensation. Journal of Political Economy 107: 65-105.

Ai, C., and E. Norton. 2003. Interaction Terms in Logit and Probit Models. Economics Letters 80: 123-129.

Bertrand, M., and S. Mullainathan. 2003. Enjoying the Quiet Life? Corporate Governance and Managerial Preferences. The Journal of Political Economy 111: 1043-1075.

Bettis, J., J. Bizjak, J. Coles, and S Kalpathy. 2010a. Stock and Option Grants with Performance-Based Vesting Provisions. Review of Financial Studies 23: 3849-3888.

Bettis, J., J. Bizjak, J. Coles, and S Kalpathy. 2010b. Supplement to “Stock and Option Grants with Performance-Based Vesting Provisions”: Preliminary Evidence on the Size, Delta, and Vega of Stock and Option Grants with Performance-Based Vesting Provisions. Online Review of Financial Studies.

Bettis, J., J. Bizjak, J. Coles, and S Kalpathy. 2010c. Supplement to “Stock and Option Grants with Performance-Based Vesting Provisions”: Evidence on Whether Stock and Option Grants with Performance-Based Vesting Provisions Are Associated with Earnings Management, Accounting Restatements, and Shareholder Litigation. Online Review of Financial Studies.

Bettis, J., J. Bizjak, J. Coles, and S Kalpathy. 2013. Performance-Vesting Provisions in Executive Compensation. Working paper, Texas Christian University.

Bizjak, J., J. Brickley, and J. Coles. 1993. Stock-based Incentive Compensation and Investment Behavior. Journal of Accounting and Economics 16: 349-372

Borokhovich, K., K. Brunarski, and R. Parrino. 1997. CEO Contracting and Antitakeover Amendments. Journal of Finance 52: 1495-1517.

Brickley, J., J. Coles, and R. Terry. 1994. Outside Directors and the Adoption of Poison Pills. Journal of Financial Economics 35: 371-390.

Burkart, M., D. Gromb, and F. Panunzi. 1997. Large Shareholders, Monitoring, and the Value of the Firm. Quarterly Journal of Economics 112: 693-728.

Byrd, J., and K. Hickman. 1992. Do Outside Directors Monitor Managers? Evidence from Tender Offer Bids. Journal of Financial Economics 32: 195-222.

Cai, J., J. Garner, and R. Walkling. 2009. Electing Directors. Journal of Finance 64: 2389- 2421.

Carter, M.E., L. Lynch, and S. Zechman. 2009. Changes in Bonus Contracts in the Post- Sarbanes-Oxley Era. Review of Accounting Studies 14: 480-506.

Chhaochharia, V., and Y. Grinstein. 2007. Corporate Governance and Firm Value: The Impact of the 2002 Governance Rules. Journal of Finance 62: 1789-1825.

Coles, J., N. Daniel, and L. Naveen. 2006. Managerial Incentives and Risk-taking. Journal of Financial Economics 79: 431-468.

Coles, J., N. Daniel, and L. Naveen. 2013. Calculation of Compensation Incentives and Firm- related Wealth using Execucomp: Data, Program, and Explanation. Working paper, Temple University.

Coles, J., M. Lemmon, and Y. Wang. 2011. The Joint Determinants of Managerial Ownership, Board Independence, and Firm Performance. Working paper, University of Utah.

Core, J., and W. Guay. 1999. The Use of Equity Grants to Manage Optimal Incentive Levels. Journal of Accounting and Economics 28: 151-184.

Core, J., and W. Guay. 2002. Estimating the Value of Employee Stock Option Portfolios and Their Sensitivities to Price and Volatility. Journal of Accounting Research 40: 613-630.

Core, J., R. Holthausen, and D. Larcker. 1999. Corporate Governance, chief Executive Officer Compensation and Firm Performance. Journal of Financial Economics 51: 371-406.

Core, J., W. Guay, and D. Larcker. 2008. The Power of the Pen and Executive Compensation. Journal of Financial Economics 88: 1-25.

Coughlan, A., and R. Schmidt. 1985. Executive Compensation, Management Turnover, and Firm Performance: An Empirical Investigation. Journal of Accounting and Economics 7: 43- 66.

Dahya, J., O. Dimitrov, and J. McConnell. 2008. Dominant Shareholders, Corporate Boards and Corporate Value: A Cross-country Analysis. Journal of Financial Economics 87: 73-100.

Dahya, J., and J. McConnell. 2007 Board Composition, Corporate Performance, and the Cadbury Committee Recommendation. Journal of Financial and Quantitative Analysis 42: 535-564.

Dahya, J., J. McConnell, and N. Travlos. 2002. The Cadbury Committee, Corporate Performance, and Top Management Turnover. Journal of Finance 57: 461-483.

Deangelo, H., and E. Rice. 1983 Antitakeover Charter Amendments and Stockholder Wealth. Journal of Financial Economics 11: 329-360.

Denis, D., and D. Denis. 1995. Performance Changes Following Top Management Dismissals. Journal of Finance 50: 1029-1057.

Dittmar, A., and J. Mahrt-Smith. 2007. Corporate Governance and the Value of Cash Holdings. Journal of Financial Economics 83: 599-634.

Falaye, O. 2007. Classified Boards, Firm Value, and Managerial Entrenchment. Journal of Financial Economics 83: 501-529.

Fich, E., and A. Shivdasani. 2007. Financial Fraud, Director Reputation, and Shareholder Wealth. Journal of Financial Economics 86: 306-336.

Finkelstein, S., and D. Hambrick. 1989. Chief Executive Compensation: A Study of the Intersection of Markets and Political Processes. Strategic Management Journal 10: 121-134.

Gompers, P., J. Ishii, and A. Metrick. 2003. Corporate Governance and Equity Prices. Quarterly Journal of Economics 118: 107–155.

Gompers, P., J. Ishii, and A. Metrick, 2010. Extreme Governance: An Analysis of Dual-Class Firms in the United States. Review of Financial Studies, 23, 1051-1088.

Goyal, V., and C. Park. 2002. Board Leadership Structure and Chief Executive Turnover. Journal of Corporate Finance 8: 49-66.

Guay, W. 1999 The Sensitivity of CEO Wealth to Equity Risk: An Analysis of the Magnitude and Determinants. Journal of Financial Economics 53: 43-71.

Hartzell, J., and L. Starks. 2003. Institutional Investors and Executive Compensation. Journal of Finance 58: 2351-2374.

Hazarika, S., Karpoff, J., and R. Nahata. 2012. Internal Corporate Governance, CEO Turnover, and Earnings Management. Journal of Financial Economics 104: 44-69.

Hermalin, B., and M. Weisbach. 1998. Endogenously Chosen Boards of Directors and Their Monitoring of the CEO. American Economic Review 88: 96-118.

Hermalin, B., and M. Weisbach. 2003. Boards of Directors as an Endogenously Determined Institution. Federal Reserve Bank of New York Economic Policy Review 9: 1-20.

Huson, M., R. Parrino, and L. Starks. 2001. Internal Monitoring Mechanisms and CEO Turnover: A Long-term Perspective. Journal of Finance 56: 2265-2297.

Hwang, B., and S. Kim. 2009. It Pays to Have Friends. Journal of Financial Economics 93: 138-158.

Hymowtiz, C., and J. Green. 2013. Not Going Anywhere. Business Week, May 27 – June 2, 23- 24.

Jensen, M. 1986. The Agency Costs of Free Cash Flow: Corporate Finance and Takeovers. American Economic Review 76: 323-329.

Jensen, M., and K. Murphy, 1990, Performance Pay and Top-management Incentives. Journal of Political Economy 98: 225-264.

Kang, J. and A. Shivdasani. 1995. Firm Performance, Corporate Governance, and Top Executive Turnover in Japan. Journal of Financial Economics 38: 29-58.

Kaplan, S., and B., Minton. 2012. How has CEO Turnover Changed? International Review of Finance 12: 57-87.

Linck, J., J. Netter, and T. Yang. 2009. Effects and Unintended Consequences of the Sarbanes- Oxley Act on Corporate Boards. Review of Financial Studies 22: 3287-3328.

Lorsch, J., and E. MacIver. 1989. Pawns or Potentates: The Reality of America’s Corporate Boards. Boston: Harvard Business School Press.

Mace, M. 1971. Directors, Myth, and Reality. Boston: Harvard Business School Press. Mobbs, S. 2012. CEOs Under Fire: The Effects of Directors on CEO Compensation and

Turnover. Forthcoming, Journal of Financial and Quantitative Analysis. Morse, A., Nanda, V., and Seru, A., 2011. Are Incentive Contracts Rigged by Powerful CEOs?

Journal of Finance 66, 1779-1821. Murphy, K. 1999. Executive Compensation. In Orley Ashenfelter and David Cards, Eds:

Handbook of Labor Economics, Volume 3, North Holland: New York. Pan, Y., Wang, T., and Weisbach, M., 2013. CEO Investment Cycles. Working Paper,

University of Utah, University of Minnesota, and Ohio State University. Parrino, R. 1997. CEO Turnover and Outside Succession: A Cross-sectional Analysis. Journal

of Financial Economics 46: 165-197. Petersen, M. 2009. Estimating Standard Errors in Finance Panel Data Sets: Comparing

Approaches. Review of Financial Studies 22: 435-480. Ryan, H. E. Jr. and R. A. Wiggins, III. 2004. Who is in Whose Pocket? Director Compensation,

Board Independence, and Barriers to Effective Monitoring. Journal of Financial Economics 73: 497-524.

Shivdasani, A., and D. Yermack. 1999. CEO Involvement in the Selection of New Board Members: An Empirical Analysis. Journal of Finance 54: 1829-1853.

Stein, J. 1988 Takeover Threats and Managerial Myopia. Journal of Political Economy 96: 61- 80.

Warner, J., R. Watts, and K. Wruck. 1988. Stock Prices and Top-management Changes. Journal of Financial Economics 20: 461-92.

Weisbach, M. 1988. Outside Directors and CEO turnover. Journal of Financial Economics 20: 421-460.

Table 1: Summary statistics Co-option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. TW Co-option is the sum of tenure of co-opted directors divided by the sum of tenure of all directors. Residual Co-option and Residual TW Co-option are the residuals from regressions of Co-option on CEO tenure and TW Co-option on CEO tenure. Co-opted Independence is the proportion of the board that consists of co-opted directors who are independent. Co-opted Non-Independence is the proportion of the board that consists of co-opted directors who are not independent. Non-Co-opted Independence is the proportion of the board that consists of independent directors who were already on the board when the CEO assumed office. Independence is the proportion of the board that consists of independent directors. Forced Turnover equals 1 if the CEO departs and is younger than 60 years of age, and equals 0 otherwise. CEO Pay is the total annual pay (Execucomp: TDC1). CEO Delta is the expected dollar change in CEO wealth for a 1% change in stock price, where components of pay- performance sensitivity (delta) arise from CEO holdings of own-firm stock and options, per Core and Guay (2002). Investment is capital expenditure scaled by assets. Reported values are based on data winsorized at the 1st and 99th percentiles.

Obs. Mean Median Std. Dev.

Firm-level variables Sales ($ Million) 18,987 5,341 1,505 11,426 Investment 18,405 0.052 0.038 0.051 Board-related variables Board Size 18,987 9.6 9.0 2.7 Co-option 18,196 0.47 0.44 0.32 TW (Tenure-weighted) Co-option 18,196 0.31 0.17 0.33 Residual Co-option 18,196 0.00 -0.03 0.22 Residual TW Co-option 18,196 0.00 -0.07 0.23 Co-opted Independence 18,196 0.35 0.33 0.25 Co-opted Non-Independence 18,196 0.12 0.09 0.15 Non-Co-opted Independence 18,196 0.34 0.33 0.26 Independence 18,987 0.69 0.71 0.17 CEO-related variables Forced Turnover 18,860 0.025 0.000 0.156 CEO Pay ($ ‘000s) 18,870 4,934 2,853 6,088 CEO Pay-performance Sensitivity (Delta; $ ‘000s) 17,436 789 250 1,802 CEO Tenure (Years) 18,401 8.1 6.0 7.1

Table 2: Effect of co-option on CEO turnover-performance sensitivity The table presents results from logistic regressions of Forced Turnover, which equals 1 if the CEO departs and is younger than 60 years of age, and equals 0 otherwise. Co-option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. TW Co-option is the sum of tenure of co-opted directors divided by the sum of tenure of all directors. For a departing CEO, co-option effectively is the co-option of the board at the time of dismissal. In the year in which turnover occurs, if the turnover occurs after the date of the firm’s annual meeting, then the contemporaneous co-option measure is used; if the turnover occurs before the firm’s annual meeting date, then the lagged co-option measure is used. In non-turnover years, we take the average of contemporaneous and lagged values of co-option measure. Residual Co-option and Residual TW Co- option are the residuals from regressions of Co-option on CEO tenure and TW Co-option on CEO tenure. For turnover years, Prior Abnormal Return is measured as the annual stock returns in the year leading up to the actual date of CEO turnover minus the value-weighted market returns over that period. For non- turnover years, Prior Abnormal Return is measured as the stock returns over the previous fiscal year minus the value-weighted market returns over that period. In the year of turnover, all CEO variables correspond to that of the departing CEO; in non-turnover years, they are measured contemporaneously. All other control variables are defined in the Appendix. For each independent variable, we present the co-efficient estimate (row 1), the z-statistics (in parentheses; row 2), and the marginal effect (in italics; row 3). The marginal effect for a continuous variable is the percentage change in the conditional expected probability of Forced Turnover for a one percent change in the continuous variable, holding all other variables at their means (i.e., it is the elasticity). The marginal effect for an indicator variable is the percentage change in the conditional expected probability of Forced Turnover when the indicator variable goes from 0 to 1, holding all other variables at their means. The marginal effect for the interaction term of Prior Abnormal Return with a continuous variable is the change in the elasticity of Forced Turnover with Prior Abnormal Return when the continuous variable moves by one standard deviation from its mean value, holding all other variables at their means. The marginal effect for the interaction term of Prior Abnormal Return with an indicator variable is the change in the elasticity of Forced Turnover with Prior Abnormal Return when the indicator variable moves from 0 to 1, holding all other variables at their means. The last row of the table gives the elasticity of Forced Turnover with Prior Abnormal Return for a fully co-opted board (i.e., at Co-option = 1 for columns 1 and 2, TW Co-option =1 for columns 3 and 4, Residual Co-option = 0.67 for column 5, and Residual TW Co-option = 0.84 for column 6). ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels.

Dependent variable = Forced Turnover

Co-option measure used:

Coefficient estimate (z-statistic) Co-option Co-option

TW Co-option

Residual Co-option

Residual TW Co-option

Marginal effect (1) (2) (3) (4) (5) (6) Co-option measure × 2.021*** 1.610** 2.304*** 2.557*** 1.558* 2.250*** Prior Abnormal Return (α1) (3.4) (2.5) (4.4) (4.1) (1.8) (3.4) 0.856 0.535 0.961 0.751 0.327 0.407

Prior Abnormal Return (α2) -2.153*** -2.473*** -2.062*** -2.974** -1.163*** -1.337*** (-5.3) (-2.6) (-6.1) (-2.6) (-5.6) (-5.7) -1.331 -1.257 -1.539 -1.550 -1.165 -1.347

Co-option measure 3.817*** 3.804*** 2.475*** 2.517*** 3.832*** 2.539*** (6.9) (6.9) (3.8) (3.8) (6.9) (3.8) 3.797 3.723 2.571 2.628 3.854 2.629

Independencet-1 -0.674 -0.723 -0.555 -0.555 -0.655 -0.527 (-0.8) (-0.9) (-0.6) (-0.6) (-0.8) (-0.6)

-0.650 -0.597 -0.545 -0.462 -0.643 -0.517

Firm Sizet -0.264 -0.235 -0.289 -0.296 -0.248 -0.274 (-0.9) (-0.8) (-0.9) (-1.0) (-0.9) (-0.9)

-0.255 -0.221 -0.284 -0.288 -0.244 -0.269

CEO Tenuret 0.018 0.021 0.106*** 0.106*** 0.132*** 0.174*** (0.7) (0.8) (3.2) (3.2) (6.8) (7.3)

0.018 0.020 0.104 0.103 0.130 0.171

CEO Ownershipt -0.053 -0.090* -0.066 -0.071 -0.042 -0.054 (-1.2) (-1.7) (-1.5) (-1.5) (-1.0) (-1.3)

-0.051 -0.078 -0.064 -0.067 -0.041 -0.053

CEO Dualityt -1.148*** -1.127*** -1.344*** -1.370*** -1.162*** -1.370*** (-5.3) (-5.2) (-5.8) (-5.8) (-5.4) (-5.9)

-1.110 -1.066 -1.321 -1.378 -1.140 -1.350

Outside Director Ownershipt-1 0.015 0.016 0.023 0.023 0.016 0.025 (0.8) (0.8) (1.2) (1.2) (0.9) (1.3)

0.015 0.014 0.022 0.023 0.016 0.024

GIM Indext-1 -0.079 -0.085 -0.093 -0.091 -0.085 -0.099 (-0.7) (-0.8) (-0.8) (-0.8) (-0.8) (-0.9)

-0.076 -0.078 -0.091 -0.083 -0.084 -0.097

Board Sizet-1 -0.009 -0.006 -0.032 -0.021 -0.011 -0.026 (-0.1) (-0.1) (-0.4) (-0.3) (-0.2) (-0.4)

-0.009 0.001 -0.032 -0.012 -0.011 -0.026

Female Directort-1 -0.072 -0.087 0.069 0.057 -0.031 0.101 (-0.3) (-0.3) (0.2) (0.2) (-0.1) (0.3) -0.070 -0.092 0.068 0.038 -0.03 0.099

Independencet-1 × 0.295 0.278 Prior Abnormal Return (0.3) (0.2) 0.108 0.082

CEO Ownershipt-1 × 0.099* 0.013 Prior Abnormal Return (1.8) (0.2) 0.259 -0.003

CEO Dualityt-1 × 0.100 -0.449 Prior Abnormal Return (0.3) (-1.0) 0.059 -0.490

Outside Director Ownershipt-1 × -0.035 -0.023 Prior Abnormal Return (-1.2) (-0.7) -0.198 -0.116

GIM Indext-1 × -0.012 0.039 Prior Abnormal Return (-0.2) (0.4) -0.003 0.142

Board Sizet-1 × 0.045 0.056 Prior Abnormal Return (0.5) (0.5) 0.010 0.192

Female Directort-1 × -0.051 -0.020 Prior Abnormal Return (-0.1) (-0.0) -0.108 -0.104

Fixed Effect Firm, Year

Firm, Year

Observations 2,152 2,152 1,899 1,899 2,152 1,899

Turnover-Performance Sensitivity at Co-option measure = Max Value -0.092 -0.203 0.213 0.415 -0.088 0.465

Table 3: Effect of co-option on CEO pay The table presents regression results where the dependent variable is logarithm of CEO Pay. We drop firm-years that had a turnover and require that the CEO’s tenure be at least one year. This ensures that we do not consider pay for fractional years. CEO Pay is the total annual pay (Execucomp: TDC1). Co-option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. TW Co-option is the sum of tenure of co-opted directors divided by the sum of tenure of all directors. Residual Co-option and Residual TW Co-option are the residuals from regressions of Co-option on CEO tenure and TW Co-option on CEO tenure. All other control variables are defined in Appendix. Intercept is included but not reported. t-statistics given in parentheses are based on standard errors corrected for heteroskedasticity and firm-level clustering. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels. Dependent variable = Logarithm of CEO Pay Co-option measure used:

Coefficient estimate (t-statistic)

Co-option TW Co-option Residual Co-option

Residual TW Co-option

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

Co-option measuret (θ1) 0.223*** 0.133* 0.227*** 0.130*

(3.9) (2.0) (3.9) (1.9)

Independencet 0.141 0.163* 0.141 0.163*

(1.5) (1.7) (1.5) (1.7)

Firm Sizet 0.327*** 0.324*** 0.327*** 0.324***

(10.1) (9.9) (10.1) (9.9)

Stock Returnt 0.103*** 0.104*** 0.103*** 0.104***

(5.9) (5.9) (5.9) (5.9)

ROAt 1.220*** 1.226*** 1.220*** 1.226***

(7.2) (7.2) (7.2) (7.2)

CEO Tenuret -0.008** -0.004 -0.001 -0.000

(-2.4) (-1.3) (-0.4) (-0.1)

CEO Ownershipt -0.012** -0.013*** -0.012** -0.013***

(-2.6) (-2.6) (-2.6) (-2.6)

CEO Dualityt 0.037 0.045* 0.037 0.045*

(1.5) (1.9) (1.5) (1.9)

Outside Director Ownershipt -0.001 -0.001 -0.001 -0.001

(-1.4) (-1.3) (-1.4) (-1.3)

GIM Indext 0.011 0.011 0.011 0.011

(1.0) (1.0) (1.0) (1.0)

Board Sizet -0.005 -0.003 -0.005 -0.003

(-0.7) (-0.4) (-0.7) (-0.4)

Female Directort 0.013 0.014 0.013 0.014 (0.5) (0.5) (0.5) (0.5)

Fixed Effects Firm, Year Firm, Year Firm, Year Firm, Year Observations 12,036 12,036 12,036 12,036

R2 0.167 0.166 0.168 0.166

Table 4: Effect of co-option on CEO pay-performance sensitivity (CEO delta) The table presents regressions of CEO Pay-performance sensitivity (or Delta), defined as the expected dollar change in CEO wealth for a 1% change in stock price, where components of delta arise from current CEO holdings of own- firm stock and options, per Core and Guay (2002). Co-option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. TW Co-option is the sum of tenure of co-opted directors divided by the sum of tenure of all directors. Residual Co-option and Residual TW Co-option are the residuals from regressions of Co-option on CEO tenure and TW Co-option on CEO tenure. Other control variables are defined in Appendix. Intercept is included but not reported. t-statistics given in parentheses are based on standard errors corrected for heteroskedasticity and firm-level clustering. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels.

Dependent variable = CEO Pay-performance sensitivity

Co-option measure used:

Coefficient estimate (t-statistic) Co-option TW Co-option

Residual Co-option

Residual TW Co-option

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

Co-option measuret (γ1) -296.485 -366.150* -299.503 -363.054* (-1.6) (-1.7) (-1.6) (-1.8)

Independencet -558.298** -578.303** -558.695** -578.854** (-2.4) (-2.5) (-2.4) (-2.5)

Firm Sizet 199.686*** 201.071*** 199.643*** 201.065*** (3.1) (3.1) (3.1) (3.1)

Tobin’s qt 378.526*** 378.745*** 378.469*** 378.609*** (7.1) (7.2) (7.1) (7.2)

R&D/Assetst -1,644.191 -1,729.813 -1,640.445 -1,722.186 (-1.4) (-1.5) (-1.4) (-1.5)

Investmentt 1,338.528* 1,355.289* 1,337.425* 1,350.299* (1.7) (1.7) (1.7) (1.7)

Leveraget -200.384 -197.445 -200.717 -197.515 (-1.0) (-1.0) (-1.0) (-1.0)

Log(Unsystematic Riskt) -93.625*** -91.989*** -93.651*** -91.938*** (-2.9) (-2.9) (-2.9) (-2.9)

CEO Tenuret 82.321*** 83.992*** 72.924*** 72.232*** (5.1) (5.2) (6.5) (6.6)

CEO Dualityt 156.116*** 152.332*** 156.272*** 151.993*** (2.7) (2.7) (2.7) (2.7)

Outside Director Ownershipt 0.083 0.062 0.084 0.060 (0.1) (0.1) (0.1) (0.1)

GIM Indext 11.241 9.324 11.222 9.349 (0.3) (0.3) (0.3) (0.3)

Board Sizet 2.524 -1.028 2.514 -1.058 (0.1) (-0.1) (0.1) (-0.1)

Female Directort -11.763 -14.439 -11.837 -14.363 (-0.2) (-0.3) (-0.2) (-0.3) Fixed Effects Firm, Year Firm, Year Firm, Year Firm, Year

Observations 11,539 11,539 11,539 11,539

R2 0.196 0.196 0.196 0.196

Table 5: Effect of co-option on investment The table presents regressions of Investment, defined as capital expenditure scaled by assets. Co-option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. TW Co-option is the sum of tenure of co-opted directors divided by the sum of tenure of all directors. Residual Co-option and Residual TW Co-option are the residuals from regressions of Co-option on CEO tenure and TW Co- option on CEO tenure. Other control variables are defined in Appendix. Intercept is included but not reported. t- statistics given in parentheses are based on standard errors corrected for heteroskedasticity and firm-level clustering. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels. Dependent variable = Investment

Co-option measure used:

Coefficient estimate (t-statistic) Co-option TW Co-option

Residual Co-option

Residual TW Co-option

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

Co-option measuret (μ1) 0.005** 0.006** 0.005** 0.006** (2.5) (2.2) (2.4) (2.0)

Independencet -0.007* -0.006* -0.007* -0.006* (-1.9) (-1.8) (-1.9) (-1.7)

CEO Vegat-1 (×10-6) -2.730 -2.60 -2.730 -0.260 (-1.2) (-1.2) (-1.2) (-1.2)

CEO Deltat-1 (×10-6) 0.830** 0.830** 0.831** 0.825** (2.0) (2.0) (2.0) (2.0)

CEO Tenuret -0.000 -0.000 0.000 0.000 (-1.5) (-1.4) (0.0) (0.2)

CEO Cash Compensationt (×10-6) 0.509** 0.504** 0.509** 0.506** (2.2) (2.2) (2.2) (2.2)

Tobin’s qt 0.005*** 0.005*** 0.005*** 0.005*** (8.6) (8.5) (8.6) (8.5)

Firm Sizet 0.002 0.002 0.002 0.002 (1.6) (1.6) (1.6) (1.6)

FCF/Assetst 0.034*** 0.034*** 0.034*** 0.034*** (5.3) (5.3) (5.3) (5.3)

Sales Growtht 0.001 0.001 0.001 0.001 (0.9) (0.9) (0.9) (0.9)

Leveraget -0.017*** -0.017*** -0.017*** -0.017*** (-3.6) (-3.6) (-3.6) (-3.6)

Stock Returnst -0.010*** -0.010*** -0.010*** -0.010*** (-14.4) (-14.4) (-14.4) (-14.4)

Fixed Effects Firm, Year Firm, Year Firm, Year Firm, Year

Observations 16,041 16,041 16,041 16,041

R2 0.148 0.147 0.147 0.147

Table 6: A natural experiment This table reports the effect of co-option on turnover-performance sensitivity, pay, pay-performance sensitivity, and investment using a natural experiment. As a basis for a natural experiment, we exploit the 2002 NYSE/Nasdaq listing requirement that a majority of the board be comprised on independent directors. Firms reacted by adding independent directors, thus imposing an exogenous shock to co-option on firms that were non-complaint as of 2001. We essentially follow the difference-in-difference methodology, but with one key difference. We allow for the possibility that SOX and associated listing provisions have a direct effect on turnover-performance sensitivity, pay, pay-performance sensitivity, and investment, in order to isolate the “clean” or causal effect of co-option on these four attributes. We illustrate our methodology using pay.

Pay = β0 + β1 Co-option + β2 Post-SOX  Co-option + β3 Non-Compliant  Co-option + β4 Post-SOX  Non-Compliant  Co-option + β5 Post-SOX + β6 Non-Compliant + k(other controls) + error

Post-SOX is an indicator variable that equals 1 if it is the post-SOX period, 2002–2010. Non-Compliant is an indicator variable that equal 1 if the firm was non-compliant as of 2001. The controls in the specification include the usual independent variables used in the pay regressions, year fixed effects, and the interactions of all the independent variables with the three key dummy variables: Post-SOX, Non-Compliant, and Post-SOX × Non- Compliant. Panel A of Table 6 provides an estimate of the sensitivity of pay to co-option for the four subsamples of firms: compliant firms in the pre-SOX period, and non-compliant firms in the pre-SOX period, compliant firms in the post-SOX period, and non-compliant firms in the post-SOX period. The firms in the non-compliant post-SOX group face the exogenous shock to co-option. The sensitivities are estimated by taking the partial derivative of Pay with respect to Co-option. “SOX” represents how changes in the listing requirements affected pay through avenues other than through co-option. “Bias” arises from the standard endogeneity problem. We allow this bias to differ by whether the firm was compliant (superscript C) or not compliant (superscript NC) pre-SOX, though we do restrict BiasC to be the same both pre- and post-SOX. Panel B estimates the “clean” effect. The corresponding base-case estimates from Model 1 from Tables 2–5 are reported for comparison purpose. t-statistics given in parentheses are based on standard errors corrected for heteroskedasticity and firm-level clustering. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels. Panel A: Representative example for regression of CEO pay on co-option Estimated coefficient ( = ∂Pay/∂Co-option): Pre-SOX period

(1996–2001) Post-SOX period

(2002–2010) Difference

Compliant β1 (Clean + BiasC)

β1 + β2 (Clean + BiasC + SOX)

β2 (SOX)

Non-Compliant β1 + β3 (Clean + BiasNC)

β1 + β2 + β3 + β4 (Clean + SOX)

β2 + β4 (SOX – BiasNC)

Difference β3 (BiasNC – BiasC)

β3+ β4

(–BiasC)

β4 (–BiasNC)

Clean = (Clean + SOX) – SOX = β1 + β2 + β3 + β4 – β2 = β1+ β3 + β4

Panel B: Estimates of the clean effects of co-option

Coefficient estimate (z-statistic/t-statistic)

Results from base-case

“Clean” Estimate

Table 2: Model 1, α1 (turnover-performance sensitivity)

2.00***

(3.4) 8.92*

(1.7) Table 3: Model 1, θ1 (pay)

0.23*** (4.0)

0.84*** (3.3)

Table 4: Model 1, γ1 (pay-performance sensitivity)

-284.85 (-1.6)

836.34 (1.2)

Table 5: Model 1, μ1 (investment)

0.006** (2.5)

0.010 (1.0)

Table 7: Are all co-opted directors similar? Co-opted independent versus co-opted non- independent directors The table reports results for experiments in Model 1 of Tables 2 – 5 re-estimated using Co-opted Independence (Panel B) and Non-co-opted Independence (Panel C) rather than Co-option. For ease of comparison, Panel A reproduces the results from base-case results. Co-option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. Co-opted Independence is the proportion of the board that consists of co-opted directors who are independent. Co-opted Non-Independence is the proportion of the board that consists of co-opted directors who are not independent. Column 1 reports results on turnover- performance sensitivity. Column 2 reports results on CEO pay. Column 3 addresses CEO pay-performance sensitivity. Column 4 estimates specifications for investment. t-statistics given in parentheses are based on standard errors corrected for heteroskedasticity and firm-level clustering. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels.

Dependent variable:

Coefficient estimate (z-statistic/t-statistic)

Forced Turnover

(1)

log(CEO Pay)

(2)

CEO Delta

(3) Investment

(4)

Panel A: Base-Case (Co-option)

Co-option 

2.021*** Prior Abnormal Return (3.4)

Co-option 3.817*** 0.223*** -296.485 0.005**

(6.9) (3.9) (-1.6) (2.5)

Panel B: Co-opted Independence

Co-opted Independence 

3.138*** Prior Abnormal Return (3.7)

Co-opted Independence 4.137*** 0.262*** -499.608** 0.005**

(6.1) (3.9) (-2.3) (2.1)

Panel C: Co-opted Non-Independence

Co-opted Non-Independence 

2.306*** Prior Abnormal Return (2.7)

Co-opted Non-Independence 2.245** 0.130 350.636 0.010**

(2.0) (1.1) (1.1) (2.0)

Other controls for all Panels as in: Table 2,

Model 1 Table 3, Model 1

Table 4, Model 1

Table 5, Model 1

Observations 1,899 12,036 11,539 16,041

Table 8: Are non-co-opted independent directors more effective monitors? Panel A depicts the overlap and dissimilarity between co-option and independence. Co-option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. Independence is the proportion of the board that consists of independent directors. Co-opted Independence is the proportion of the board that consists of co-opted directors who are independent. Co-opted Non-Independence is the proportion of the board that consists of co-opted directors who are not independent. The sum of Co-opted Independence and Co- opted Non-Independence equals Co-option. Non-Co-opted Independence is the proportion of the board that consists of independent directors who were already on the board when the CEO assumed office. The sum of Co-opted Independence and Non-Co-opted Independence equals Independence. Non-Co-opted Non-Independence is the proportion of the board that consists of non-independent directors who were already on the board when the CEO assumed office. Panel B reports results for experiments in Model 1 of Tables 2 – 5 re-estimated by replacing Co- option and Independence using Non-Co-opted Independence as the key explanatory variable. Column 1 reports results on turnover-performance sensitivity. Column 2 reports results on CEO pay. Column 3 addresses CEO pay- performance sensitivity. Column 4 estimates specifications for investment. t-statistics given in parentheses are based on standard errors corrected for heteroskedasticity and firm-level clustering. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels. Panel A: Comparison of co-option and independence measures

Co-opted?

Yes No

In de

pe nd

en t?

Yes Co-opted Independence + Non-Co-opted Independence = Independence

No Co-opted Non-Independence

Non-Co-opted Non-Independence

= Co-option

Panel B: Results for Model 1 (Tables 2–5) using Non-Co-opted Independence as primary explanatory variable

Dependent variable:

Coefficient estimate (z-statistic/t-statistic)

Forced Turnover

(1)

log(CEO Pay)

(2)

CEO Delta

(3) Investment

(4)

Non-Co-opted Independence 

-2.899*** Prior Abnormal Return (-3.4)

Non-Co-opted Independence -4.314*** -0.210*** 358.878* -0.006**

(-6.2) (-3.2) (1.9) (-2.4)

Other controls as in: Table 2, Model 1 Table 3, Model 1 Table 4, Model 1 Table 5, Model 1

Observations 1,899 12,036 11,539 16,041

Figure 1. Timeline The example above illustrates how we estimate the relevant board co-option associated with any given outcome (such as turnover-performance sensitivity, pay, pay-performance sensitivity, and investment). Consider a firm that has a December fiscal year end and has its annual meeting in March. Co-option for year 2001 is based on the slate of directors presented at the March 2001 annual meeting, because this is the board in place for the majority of the year. Thus, by definition, co-option is based on the board prevailing at the end of the year rather than at the beginning of the year. Similarly, co-option for the year 2002 is based on the slate of directors presented at the March 2002 meeting. Assume a CEO turnover event occurs after the annual meeting date (say, August 2002). Then the board that determined the CEO turnover is the board in place since March 2002, and hence turnover and co-option are measured contemporaneously. If the turnover occurs before the annual meeting (say, February 2002) then the board that determined the turnover is the board elected as of March 2001, and the lagged co-option measure is used in regressions. For non-turnover years, because both the lagged and contemporaneous boards decide on the CEO’s ‘non-replacement,’ we use the average of the lagged and contemporaneous values of co-option. For regressions of CEO pay, CEO pay-performance sensitivity, and investment, we use the contemporaneous co-option measure because this is based on the board that is in place for the majority of the year and also because performance-based pay will be decided by the board at the end of the fiscal year.

March 2001: Annual Meeting

March 2002: Annual Meeting

Dec 2001 Dec 2002Dec 2000

Aug 2002: CEO Turnover

Feb 2002: CEO Turnover

Figure 2. Board composition and CEO tenure The figure plots various measures of board composition and co-option against CEO tenure. Co- option is the number of directors appointed after the CEO assumed office (“co-opted” directors) divided by the board size. Independence is the proportion of the board that consists of independent directors. Co-opted Independence is the proportion of the board that consists of co- opted directors who are independent. Non-Co-opted Independence is the proportion of the board that consists of independent directors who were already on the board when the CEO assumed office. The sum of Co-opted Independence and Non-Co-opted Independence equals Independence. CEOs are put in 11 groups based on their tenure, where CEO tenure is measured as fiscal year end date minus the date the executive became the CEO. All CEOs with  1 year are put in the 1st group,…, CEOs with tenure greater than 9 years but  10 years are put in the 10th group, and CEOs with greater than 10 years are put in the 11th group.

0.2

0.4

0.6

0.8

0 1 2 3 4 5 6 7 8 9 10 11

B oa

rd C

om po

si ti

CEO Tenure

Co-optionIndependence

Co-opted Independence

Non-Co-opted Independence

+≤