Kaplan and Minton_How Has CEO Turnover Changed

7/29/2019 Kaplan and Minton_How Has CEO Turnover Changed

1/31

How Has CEO Turnover Changed?*

STEVEN N. KAPLAN AND BERNADETTE A. MINTONUniversity of Chicago Booth School of Business and NBER and

Ohio State University

ABSTRACT

We study CEO turnover both internal (board driven) and external (throughtakeover and bankruptcy) from 1992 to 2007 for a sample of large UScompanies. Annual CEO turnover is higher than that estimated in previous

studies over earlier periods. Turnover is 15.8% from 1992 to 2007, implyingan average tenure as CEO of less than 7 years. In the more recent period since2000, total CEO turnover increases to 16.8%, implying an average tenure ofless than 6 years. Internal turnover is significantly related to three compo-nents of firm stock performance performance relative to industry, industryperformance relative to the overall market, and the performance of theoverall stock market. The relations are stronger in the more recent periodsince 2000. We find similar patterns for both forced and unforced turnover,suggesting that some, if not most, turnover labeled as unforced is actually notvoluntary. The turnover-performance sensitivity is modestly related to blockshareholder ownership and board independence.

I. INTRODUCTION

In the last decade, corporate governance in the United States has come undergreat scrutiny, if not attack. The corporate governance scandals at the turn ofthe century led to the Sarbanes Oxley (SOX) legislation. Since the legislation,the criticism of corporate governance has continued. CEOs are routinely criti-cized for being overpaid. Boards of directors are commonly criticized as croniesof those overpaid CEOs. Jensen et al. (2004) document the increase in CEO paysince the 1970s. Bebchuk and Fried (2002) and Bebchuk and Grinstein (2005)document a substantial increase in CEO pay accelerated after 1995. All the threepapers criticize boards of directors for the increases in CEO pay and for notdoing a good job monitoring pay practices and CEOs.

While a great deal of work has focused on changes in CEO pay, recentchanges in CEO turnover and board behavior have received little attention.Earlier work and casual empiricism suggest that the CEOs job has becomeriskier over time. Khurana (2002) reports that CEO turnover increased in the

* This research has been supported by the Center for Research in Security Prices, by the Lynde and

Harry Bradley Foundation and the Olin Foundation through grants to the Center for the Study of the

Economy and the State, and the Dice Center for Research in Financial Economics. We thank Stuart

Gillan, Chester Spatt, and seminar participants at the 2007 AFA Meetings, Berkeley, NBER Corporate

Governance Summer Institute, and the University of Chicago for helpful comments.

International Review of Finance, 2011DOI: 10.1111/j.1468-2443.2011.01135.x

2011 The Authors. International Review of Finance International Review of Finance Ltd. 2011. Publishedby Blackwell Publishing Ltd., 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA02148, USA.


2/31

1990s relative to the 1970s and 1980s. Murphy and Zabojnik (2008) and Jensenet al. (2004) also report that turnover has increased in the 1990s, although themagnitude they report is quite small from 10% per year in the 1970s and

1980s to 11% in the 1990s. The samples in these papers do not go beyond theyear 2000 so they are unable to consider the period in which corporate govern-ance and CEO performance and pay have been subject to intense scrutiny.

In this paper, we study CEO turnover from 1992 to 2007 for a sample of largeUS companies. We consider turnover that occurs through takeover and bank-ruptcy as well as turnover in ongoing companies. When takeovers and bank-ruptcies are taken into consideration, the job of CEO in large US companiesappears more precarious than before, particularly since 2000. Annual CEOturnover is 15.8% from 1992 to 2007, implying an average tenure as CEO of lessthan 7 years. From 1992 to 1999, total CEO turnover averages about 12.6%,

implying an average CEO tenure of just less than 8 years. Since 2000, total CEOturnover increases to about 16.8%, implying an average CEO tenure of about6 years. Internal or board-driven turnover also rises substantially, increasingfrom 10.9% in the first part of the sample to 12.4% in the latter part of thesample. Looked at another way, only 21.30% of CEOs in place in 1992 remainedCEO in 1999, while only 16.35% of CEOs in place in 2000 remained CEO in2007.

We then examine how turnover varies with firm performance. Previous worksuggests a modest relation between internal (board-initiated) turnover and firmstock performance (see Murphy 1999 and Jensen et al. 2004). We find stronger

(and more significant) relations. Internal turnover is related to three differentcomponents of total firm stock performance. Turnover is sensitive to the stockperformance of the firm relative to the industry, the stock performance of theindustry relative to the stock market (under certain specifications), and theperformance of the overall stock market. (Jenter and Kanaan forthcomingobtain similar results for forced turnover.) The sensitivity to one-standard devia-tion differences in each measure is economically meaningful. We find similarresults for both forced and unforced turnover.

We also find that internal turnover since 2000 is more strongly related to allthree measures of stock performance. In fact, the sensitivity to stock perform-ance appears to be greater than that in any of the periods between 1970 and1995 studied in Murphy (1999). Ironically, it appears that during the period inwhich boards have been criticized, boards have become increasingly sensitive tofirm stock performance.

We consider four possible explanations for factors that drive the changes inturnover and turnover-performance sensitivity. There is some evidence that theturnover-performance sensitivity is related to increases in block holdings and todirector independence but not to the Gompers et al. (2003) governance index orSOX legislation.

External turnover turnover primarily related to mergers and acquisitions in most regression specifications is unrelated to stock performance, suggestingthat, on average, the takeovers are not disciplinary in nature.

International Review of Finance

2011 The Authors2 International Review of Finance International Review of Finance Ltd. 2011


3/31

As we discuss in more detail in the conclusion, our results have a numberof implications. First, CEO tenures have declined, suggesting the CEO job ismore precarious than in the past. When external takeovers are included, the

average tenure of a CEO has declined to less than 6 years since 2000. Therecent tenures are substantially shorter than those reported in previous workfor the 1970s, 1980s, and 1990s. For individual CEOs, the shorter expectedtenure likely offsets some of the benefits of the increase in CEO pay over thisperiod. This result is consistent with Hermalin (2005) who presents a modelthat predicts increased board vigilance will be associated with shorter CEOtenures (and higher CEO pay). The aforementioned relations with more inde-pendent boards and blockholders also are consistent with the increased vigi-lance in Hermalins model.

Second, our similar results for the turnover-performance sensitivities of

forced and unforced turnover suggest that a number of turnovers labeled asunforced are, in fact, not voluntary. In subsequent work, partially motivated byour results, Jenter and Lewellen (2010) find additional evidence consistent withthis conclusion.

Third, the results suggest an evolving role for boards. In a sample from the1980s, Morck et al. (1989) find that internal turnover is related to industry-adjusted performance while external turnover from hostile takeovers is relatedto industry performance. They interpret this as indicating boards respond wellto poor performance relative to the industry, but do not respond well to poorindustry performance. The external takeover market becomes active in reaction

to poor industry performance and a need for restructuring.Our results suggest that boards respond not only to poor performance relativeto the industry but also to poor industry performance and to poor marketperformance. One interpretation of these results is that boards (perhaps inconcert with shareholders) perform both the role they performed in the 1980sand some of the role that hostile takeovers played then.

Fourth, the shorter expected CEO tenures and sensitivity of those tenures tostock performance have implications for the measurement of CEO pay. Theshorter expected tenures suggest that the estimates of CEO pay used in mostcompensation studies may be overstated.

Finally, shorter CEO tenures, the greater sensitivity to stock performance, aswell as higher CEO pay may have created a greater incentive for CEOs to engagein earnings management or manipulation.

This paper was written contemporaneously with Jenter and Kanaan (forth-coming) who study related issues in a sample of CEO turnovers from 1993 to2001. They focus on forced CEO turnover, rather than all CEO turnovers. Forcedturnovers represent somewhere between 15% and 25% of total internal turno-vers. As we do for forced and unforced turnover, they find that forced CEOturnover is significantly related to industry-adjusted, industry, and marketreturns. They focus most of their paper on verifying this effect for forcedturnover and explaining why boards might behave this way. They also study alarger sample of firms but over a shorter time period. Unlike us, they do not

How Has CEO Turnover Changed?

2011 The Authors3International Review of Finance International Review of Finance Ltd. 2011


4/31

focus on the level of total turnover, the annual variation in that turnover, anddo not consider external turnovers. Given their shorter sample, they also do notconsider how turnover behavior changes over time.

Our paper also is related to that of Mikkelson and Partch (1997) whocompare complete management turnover in US companies in two 5-yearperiods the active takeover market of 19841988 and the less active marketof 19891993. In the active takeover period, they find that 39% of firmsexperience CEO turnover, and 23% of firms experience complete managementturnover; in the less active period, 34% of firms experience CEO turnover,while 16% of firms experience complete management turnover. They findthat the decline in turnover frequency is more pronounced among poorlyperforming firms. They argue that the activity of the external takeover marketaffects the intensity of management discipline. Our results suggest that the

intensity of management discipline has increased since the end of theirsample period and likely exceeds the intensity of the active takeover period.Huson et al. (2001) also examine CEO turnover across subperiods to see if therelation between performance and turnover has changed over time. Usingfour 6-year subperiods from 1971 to 1994, they document that while CEOturnover is negatively related to accounting performance and industry-adjusted stock returns, the relations did not change significantly over time.Our analysis begins at the end of their sample period and shows that, at least,during the 1992 to 2007 period, the relation between turnover and perform-ance has changed.

The paper proceeds as follows. Section II describes our sample. Section IIIpresents the results for turnover levels. Section IV presents the turnover-performance regressions. Section V summarizes the results and discusses theirimplications in more detail.

II. SAMPLE AND DATA

The sample of firms includes all Fortune 500 firms with data on both the Centerfor Research in Security Prices tapes and Compustat files (research and currentfiles). The sample runs from fiscal year-end 1991 to fiscal year-end 2007. Weconstruct the sample using the annual Fortune 500 lists from 1992 to 2008. Eachyear, Fortune ranks firms based on sales at fiscal year-end and publishes the listin an April or May issue of the following year. For example, fiscal year-end 1991rankings are published in an April or May 1992 issue.

We follow the sample firms from the first year they appear on a Fortune listuntil the end of the sample period or until the firm exits the sample becauseof a merger, acquisition, or delisting from a major stock exchange. Thus, wecontinue to follow a firm even if it falls out of the Fortune 500. We identify CEOturnovers using the Fortune 500 and Fortune 1000 lists, 10-K filings, proxystatements, Dun and Bradstreets Million Dollar Directory, the Wall Street

Journal, and Lexis/Nexis business news searches.




5/31

III. TURNOVER LEVELS

Turnover in a given fiscal year, T, means that the CEO in the spring of year Tis

no longer the CEO by the following spring of year T+ 1. We measure turnover,therefore in the years 1992 to 2007.

We consider two types of turnover. Standard or internal turnover is turnoverthat is associated with a companys board of directors. For standard turnover, acompany remains publicly listed over the course of the year, but the CEO in thespring is no longer the CEO the following spring. This is the turnover that isgenerally measured in studies of turnover. For example, see Huson et al. (2001).Nonstandard or external turnover is turnover due to a merger or bankruptcy/delisting. We also consider the CEO to have been turned over if his or hercompany is taken over by another company, and he or she is not CEO of the

combined company. We view this as an instance of turnover because in manymergers, the target CEO leaves the combined company. In those instances inwhich the CEO target remains, the target CEO generally experiences a reduc-tion in pay and power. Total turnover is the sum of internal and externalturnover.

Table 1 presents the level of CEO turnover by year. For total turnover andstandard turnover, we use two definitions of turnover. Definition 1 defines aturnover occurrence if a new CEO is selected. Definition 2 defines a turnoveroccurrence in which the CEO dies as a non-turnover event. Figure 1 presents thetotal and standard turnover for all firms according to definition 1 graphically.

There are three noteworthy patterns across the panels. First, total turnoverlevels are substantially higher than those typically reported. Overall, totalturnover under definition 1 is 15.8% over the entire sample period implying anaverage CEO tenure of 6.3 years. This is substantially higher than that reportedin Jensen et al. (2004) and Murphy and Zabojnik (2008) who study a differentsample of large firms (from the Forbes lists) over three decades from 1970 to2000. They report turnover of 10.2% in the 1970s, 10.0% in the 1980s, and11.3% in the 1990s. All of these measures, however, are for standard or internalturnover. For our sample period of 1992 to 2007, we obtain a standard turnoverof 11.7%, similar to their results for the 1990s. At 11.7%, the estimated averageCEO tenure is 8.5 years, roughly 2 years greater than the actual average tenure(that includes external takeover).

The second noteworthy pattern in Table 1 is the time series variation in thelevels of both total and internal turnover. For example, using definition 1 totalturnover is as low as 6.94% in 2003 (and only 9.35% in 1994), and as high as26.32% in 2000 (and 21.37% in 1999).

Third, turnover increased significantly in the latter part of the sample. In theearlier period from 1992 to 1996, total CEO turnover using definition 1 is12.62% per year implying an average tenure of 7.9 years. In the period from1997 to 2002, total turnover increases to 19.15% per year, implying an averagetenure of just 5.2 years. This period roughly coincides with the large increase inCEO pay described in Bebchuk and Grinstein (2005). The period in which CEO




6/31

pay increased substantially coincides with a period in which CEO tenuredecreases substantially. It is worth adding that the increased level of turnoverbegan well before 2002 (when the Enron and Worldcom scandals becameapparent and Sarbanes-Oxley was passed). In the more recent period from 2003to 2007, total CEO turnover declines from the middle period to 14.70%, imply-ing an average CEO turnover of 6.8 years. Internal or board- driven turnoverfollows a similar trend.

As stated previously, we continue to follow a firm once it is included in theFortune 500, even if it drops from the Fortune 500. This may make our resultsharder to compare with other studies that restrict themselves to firms in the

Fortune 500, S&P 500 or Forbes lists. Accordingly Appendix Table A1 examines

Table 1 CEO turnover

Year Numberof firms

Total turnover Standard (internal) turnover

(1) (2) (1) (2)

n % n % n % n %

1992 460 57 12.39 57 12.39 54 11.74 54 11.741993 486 57 11.73 57 11.75 49 10.10 49 10.101994 727 68 9.35 66 9.08 56 7.70 54 7.431995 741 113 15.25 113 15.25 92 12.42 92 12.421996 739 103 13.94 100 13.53 75 10.15 72 9.741997 736 113 15.35 111 15.08 68 9.24 66 8.971998 722 130 18.01 130 18.01 87 12.05 87 12.051999 716 153 21.37 153 21.37 101 14.11 101 14.112000 703 185 26.32 183 26.03 134 19.06 132 18.78

2001 683 97 14.20 97 14.20 63 9.22 63 9.222002 676 133 19.67 131 19.38 102 15.09 100 14.792003 663 46 6.94 46 6.94 34 5.13 34 5.132004 669 94 14.05 92 13.75 78 11.66 76 11.362005 679 127 18.70 126 18.56 94 13.84 93 13.702006 663 112 16.89 111 16.74 83 12.52 82 12.372007 652 110 16.87 107 16.41 84 12.88 81 12.42Total 10,715 1698 15.84 1680 15.67 1254 11.70 1236 11.5419921996 3153 398 12.62 393 12.46 326 10.34 321 10.1819972002 4236 811 19.15 805 19.00 555 13.10 549 12.9620032007 3331 489 14.70 482 14.49 373 11.21 366 11.00

19921999 5327 794 14.91 787 14.77 582 10.93 575 10.7920002007 5393 904 16.78 893 16.57 672 12.47 661 12.27

CEO turnovers in publicly traded Fortune 500 companies between 1992 and year-end 2007. Totalturnover is all CEO turnovers including turnover due to mergers and acquisitions and delistingsfrom a major stock exchange. Standard (internal) turnover excludes turnover due to mergers andacquisitions and delistings from a major stock exchange. For total and standard turnover,turnover is measured in two ways: (1) and (2). (1) Defines a turnover occurrence if a new CEO isselected. (2) Defines occurrences where the CEO dies as a non-turnover event. Data are fromannual Fortune 500 lists, 10-K filings, proxy statements, and the Wall Street Journal. Year denotesthe fiscal year-end for the sales data on which Fortune ranks firms. (i.e., 1992 corresponds to the1993 April/May Fortune list.)




7/31

turnover separately for firms in the Fortune 500 and those not in the Fortune 500in the particular year. The increase in turnover from the earlier part of thesample to the latter part of the sample is consistent across both Fortune 500 andnon-Fortune 500 companies. While it appears that annual turnover is larger fornon-Fortune 500 firms, annual tests of the differences in average turnover showsthat the turnover across groups is not different in all but 3 years.

Table 2 presents the results of probit regressions estimating the probability ofCEO turnover. The regression includes only an indicator variable equal to oneif the year is 2000 or later. This tests whether turnover in the later period (2000to 2007) is statistically different from that in the earlier period. We report theresults for all firms for total turnover and standard turnover using both defini-tions of turnover. The coefficient estimates on the indicator variable are positiveand statistically significant in all four regressions, suggesting that turnover issignificantly higher in the later period.

Table 3 presents the turnover data in a different way. We compare the frac-tion of the CEOs in 1992 who are no longer CEOs in 1999 to the fraction ofCEOs who are CEOs in 2000 and no longer CEOs in 2007. The Table shows that78.70% of CEOs in 1992 were no longer CEOs in 1999, while over 83% of CEOsin 2000 were no longer CEOs by 2007. Again, this result suggests that the job ofCEO has become increasingly precarious over the sample period.

0.00%

5.00%

10.00%

15.00%

20.00%

25.00%

30.00%

1992 1994 1996 1998 2000 2002 2004 2006

Total and standard CEO turnover

1992-2007

Total turnover Standard turnover

Figure 1 CEO Turnovers in Publicly Traded Fortune 500 Companies between 1992and Year-End 2007.

Total turnover is all CEO turnovers including turnover due to mergers and acquisitionsand delistings from a major stock exchange. Standard (internal) turnover excludesturnover due to mergers and acquisitions and delistings from a major stock exchange.For total and standard turnover, turnover is measured using definition 1, which definesturnover if a new CEO is selected.




8/31

Ta

ble2

Turnoveracrossperiods

Va

riable

Totalturnover

Standard(internal)turnove

r

(1)

(2)

(1)

(2)

DProb

(p-value)

DProb

(p-value)

DProb

(p-value)

DProb(p-value)

20

00orlater

indicator

variable

0.01873***

(0.0071)

0.01800

***

(0.0070)

0.0165

**

(0.0065)

0.0158*

*

(0.0064)

n

10,715

10,715

10,272

10,272

PseudoR2

0.0008

0.0007

0.0023

0.0008

ProbitregressionestimatesofthelikelihoodofCEOturnoverdurin

gtheperiodfrom

1992to2007

totestwhethertheprobabilityo

fturnover

ishigherinthe2000to2007period.TotalturnoverisallCEOturno

versincludingturnoverduetom

ergersandacquisitionsanddelistingsfrom

amajorstockexchange.Standard

(internal)turnoverexcludestur

noverduetomergersandacqu

isitionsanddelistingsfrom

am

ajorstock

exchange.Fortotalandstandardturnover,turnoverismeasuredintwoways:(1)and(2).(1)Definesaturnoveroccurrenceifan

ewCEO

is

sel

ected.(2)DefinesoccurrenceswheretheCEOdiesinofficeasan

on-turnoverevent.ThedependentvariableequalsoneiftheCEO

turnovers

an

dzerootherwise.

DProbrepresen

tsthechangeintheprobability

associatedwithmovingtheindicatorfrom

0to1.Robuststandarderrorsto

controlforheteroskedasticityarereportedinparentheses.

**

and***indicatesignificanceatthe5%

and1%

levels,respective

ly.




9/31

The turnover also is substantially greater than that measured by Mikkelsonand Partch (1997) over two earlier 5-year periods. In the active takeover periodfrom 1983 to 1988, they find that 39% of firms experience CEO turnover (and23% of firms experience complete management turnover); in the less activeperiod from 1989 to 1993, 34% of firms experience CEO turnover (while 16% offirms experience complete management turnover). Unfortunately, these resultsare not directly comparable because the sample in Mikkelsen and Partchconsists of smaller firms.

IV. THE RELATION BETWEEN TURNOVER AND PERFORMANCE

A. Internal turnover

We estimate pooled annual probit regressions to examine the likelihood ofinternal CEO turnover. In all of the probit regressions, the dependent variableis equal to one if a CEO turns over and zero otherwise. Turnover is measuredusing definition 2 (i.e., deaths are coded as non-turnover events). The Tablesreport the marginal changes in the probability of internal CEO turnover impliedby the probit coefficient estimates that result from a unit change in the explana-tory variables. For indicator variables, the coefficient represents the change inthe probability associated with moving the indicator from 0 to 1. These mar-ginal sensitivities, labeled DProb, are economically equivalent to coefficientestimates from ordinary least squares estimation. In the discussion below, wefocus on the significance of these marginal effects. Robust standard errors tocontrol for heteroskedasticity are reported in parentheses.

In all of the probit regressions, three proxies are used to measure stockmarket performance. First, we measure market performance using the annualreturn on the S&P 500 index. Second, relative industry performance is meas-ured at the two-digit SIC code level and equals the difference between thereturn on the median firm in the industry and the return on the S&P 500index. Third, relative firm performance is measured as the industry-adjusted

Table 3 CEO turnover

Year 1992Number of firms in the sample in 1992 460

Number of firms experiencing no turnover between 1992 and 1999 98Percent of firms experiencing no turnover between 1992 and 1999 21.30%

Year 2000Number of firms in the sample in 2000 703

Number of firms experiencing no turnover between 2000 and 2007 115Percent of firms experiencing no turnover between 2000 and 2007 16.35%

Number and percent of firms experiencing no turnover over a 5-year period. Turnover is meas-ured using total turnover. Total turnover is all CEO turnovers including turnover due to mergersand acquisitions and delistings from a major stock exchange. Occurrences where the CEO dies inoffice is defined as a non-turnover event.




10/31

firm stock return, which is equal to the firm stock return minus the return forthe median firm in the same two-digit SIC code. Returns are measured overthe calendar year period. Lagged returns are measured over the previous cal-

endar year. For example, for companies in Fortunes April 2006 issue, wemeasure stock returns for calendar year 2005 and lagged stock returns forcalendar year 2004.

Table 4 reports the results for two sets of probit regressions. In panel A, weinclude measures of stock market performance and an indicator variable equalto one if lagged CEO age is greater than or equal to 60 (CEO age dummy). In theregressions in panel B, we add the change in ROA as a measure of operatingperformance where ROA equals the sum of income before extraordinary incomeplus interest and related expenses divided by total assets. We use change in ROAbecause this measures the change in operating performance. For the two sets of

regressions, we report the results for the full sample period and five subperiods:three 5-year periods, 1992 to 1996, 1997 to 2002, 2003 to 2007; and two longerperiods, 1992 to 1999 and 2000 to 2007.

Table 4 shows that in the full sample regression, all three components ofstock performance are significantly related to internal CEO turnover. Addition-ally, the negative associations between stock performance and the likelihood ofturnover are driven by the later subperiods.

Table 4 documents turnover increases with poor industry-adjusted stock per-formance over the entire sample period, and particularly, in the latter subperi-ods. For example, for the 2000 to 2007 subperiod, a one-standard deviation

(35.4%) decline in a firms industry adjusted stock return is associated with anincrease of 3.2% in the mean likelihood of CEO turnover. From a base turnoverlevel of 12.55%, this change implies a likelihood of 9.35% for a CEO whosefirm performs one-standard deviation above the industry versus 15.75% for aCEO whose firm performs one-standard deviation below the industry. Theseare economically meaningful differences with 9.35% implying a tenure of10.7 years and 15.75% implying a tenure of 6.3 years.

The reported marginal probabilities appear to be greater than the sensitivitiesreported in Murphy (1999) for various subperiods between 1970 and 1995. Yet,similar to Murphy (1999) who finds that turnover is not related to industry-adjusted performance between 1990 and 1995, panel A of Table 4 shows thatturnover is not statistically related to industry-adjusted performance during the1992 to 1996 period.

CEO turnover also is related to poor industry performance. Again, the nega-tive relation for the full sample period appears to be driven by the subperiod1997 to 2002 if we split the time period in three or by the 2000 to 2007subperiod if we divide the period in half. For the 2000 to 2007 subperiod, aone-standard deviation (18.7%) decline in industry performance is associatedwith a 2.7% increase in the mean likelihood of CEO turnover.

Lower overall market performance, as measured by the return on the S&P 500index, is also associated with a higher likelihood of internal CEO turnover forthe full sample and the later subperiods. For the 2000 to 2007 subperiod a




11/31

Ta

ble4

Probitregressionsoftheprobabilityofinternal

CEOturnoveronperformance

Variable

19922007DP

rob

(s.e)

19921996DProb

(s.e)

1997

2002DProb

(s.e)

20032007DProb

(s.e.)

19921999DProb

(s.e.)

20002007DProb

(s.e.)

PanelA

ReturnonS&P

500

-0.1188***

(0.0

203)

0.1121***

(0.0406)

-0.1477***

(0.0311)

-0.3533***

(0.0722)

0.0891**

(0.0366)

-0.1823***

(0.0286)

Industryreturn

returnonS&P

500

-0.0772***

(0.0

163)

0.0700*

(0.0396)

-0.0963***

(0.0252)

-0.0463(0.0385)

0.0071(0.0244)

-0.1426***

(0.0242)

Industry-adjusted

stockreturn

-0.0641***

(0.0

106)

-0.0204(0.0200)

-0.0818***

(0.0157)

-0.0818***

(0.0198)

-0.0387***

(0.0147)

-0.0917***

(0.0148)

CEOagedummy

0.1369***

(0.0

081)

0.1255***

(0.0132)

0.1679***

(0.0139)

0.1112***

(0.0143)

0.1332***

(0.0106)

0.1415***

(0.0121)

Numberofobs

9982

2947

3903

3132

4937

5045

PseudoR2

0.0588

0.0600

0.0712

0.0591

0.0583

0.0712

PanelB

ReturnonS&P

500

-0.1396***

(0.0

215)

0.0454(0.0499)

-0.1537***

(0.0325)

-0.3713***

(0.0745)

0.0160(0.0429)

-0.1891***

(0.0299)

Industryreturn

returnonS&P

500

-0.0820***

(0.0

173)

0.0495(0.0455)

-0.0948***

(0.0264)

-0.0426(0.0405)

0.0082(0.0267)

-0.1473***

(0.0253)

Industry-adjusted

stockreturn

-0.0586***

(0.0

113)

-0.0157(0.0230)

-0.0709***

(0.0166)

-0.0828***

(0.0207)

-0.0303*

(0.0162)

-0.0881***

(0.0157)

ChangeinROA

-0.1594**

(0.07

10)

-0.0974(0.1399)

-0.3413***

(0.1034)

0.0287(0.1335)

-0.2651**

(0.1044)

-0.0413

(0.0954)

CEOagedummy

0.1422***

(0.0

081)

0.1326***

(0.0144)

0.1732***

(0.0142)

0.1137***

(0.0147)

0.1392***

(0.0113)

0.1452***

(0.0124)

n

9375

2624

3714

3037

4523

4852

PseudoR2

0.0612

0.0537

0.0770

0.0590

0.0587

0.0714

Pro

bitregressionestimatesofthelikelihoodofinternalCEOturnoverdurin

gtheperiodfrom

1992to2007.Internalturnoverexcludesturnoverduetomergers

andacquisitionsanddelistingsfroma

majorstockexchange.Occurrence

swheretheCEO

diesinofficearedefinedasnon-turnoverevents.The

dependent

variableequalsoneiftheCEOturnoversandzerootherwise.

DProbmeasuresthechangeintheprobabilityo

fCEOturnoverperunitchangeintherelevant

explanatoryvariables.Forindicatorva

riables,thecoefficientrepresentsth

echangeintheprobabilityassociatedwithmovingtheindicatorfrom

0to1.CEO

agedummyequals1iflaggedCEOageisgreaterthanorequalto60andzerootherwise.Robuststandarderro

rstocontrolforheteroskedasticityarereported

in

parentheses.

*,**,and***

indicatesignificanceatthe10%,5%,and1%

levels,respectively.




12/31

one-standard deviation (15%) decline in the S&P 500 index corresponds to anincrease of 2.7% in mean probability of CEO turnover.

Surprisingly, the relations between CEO turnover and overall market per-

formance and industry performance, respectively are positive and significant inthe 1992 to 1996 subperiod. One possible interpretation for these positivemarginal probabilities is that CEOs left office after good performance. Thiswould be the case if they were voluntary turnovers. While forced turnoversincreased from the 1992 to 1996 subperiod to the 1997 to 2002 subperiod (fromabout 12% to about 17%), this explanation is implausible because the greatmajority of turnovers in both subperiods are classified as voluntary.

Overall, the results in panel A indicate that the relation between internalCEO turnover and (poor) stock market performance appears to have intensified.In the period from 2000 to 2007, all three sensitivities to stock performance

3.2%, 2.7%, and 2.7%, respectively for one-standard deviation changes inindustry-adjusted, industry, and market performance are economically mean-ingful relative to average internal turnover of 12.55%.1

The second set of regressions in panel B of Table 4 includes the change inreturn on assets (ROA) in addition to the stock return variables. The results arequalitatively unchanged. The change in ROA is significant during the fullsample period from 1992 to 2007. While the estimate is statistically significant,a one-standard deviation increase in the change in ROA is associated with onlya 0.8% increase in the mean likelihood of CEO turnover.

In Table 5, we add 1 year of lagged stock performance variables to the regres-

sions. Overall, the results in Table 5 are consistent with those in Table 4. Forthe entire period, internal CEO turnover is significantly negatively relatedto industry-adjusted, industry, and overall market stock performance in thecurrent year. Turnover also is significantly negatively related to industry-adjusted and industry performance stock performance in the previous year.Strangely, turnover is significantly positively related to the lagged return on theS&P 500 (although the positive coefficient on the lagged return is smaller inmagnitude than the negative coefficient on the contemporaneous return).2 Asin the previous results, the regressions in Table 5 indicate that the turnover-performance relations are driven by the later subperiod.

To examine the change in the predicted probability associated with a changein the current and lagged industry-adjusted performance, we first estimate thepredicted probability of turnover evaluated at the mean values of all the regres-sors. Next, we re-estimate the predicted probability of turnover for current andlagged industry-adjusted stock returns evaluated at one-standard deviationbelow their means and for all other regressors evaluated at their means. For the

1 The predicted likelihood calculated at the mean of the independent variables is 10.97% for the

200 to 2007 subperiod.

2 In unreported tests, we test whether the sum of the current and lagged market returns are

statistically different from zero. For the full-sample period and the 2000 to 2007 subperiod,

the sum is statistically different from zero. It is not statistically different from zero for the 1992

to 1999 subperiod.




13/31

Ta

ble5


CEOturnoveronperformanceandlaggedperformance

Variable

19922007DProb

(s.e.)

19921996DProb

(s.e.)

1997

2002DProb

(s.e.)

20032007DProb

(s.e.)

19921999DProb

(s.e.)

20002007DProb

(s.e.)

ReturnonS&P500

-0.2038***

(0.0278)

0.0229(0.0664)

-0.3204***

(0.0495)

-0.3202***

(0.1144)

-0.0371(0.0501)

-0.2484***

(0.0359)

Laggedreturnon

S&P500

0.1103***

(0.0235)

0.0437(0.0449)

0.2609***

(0.0530)

0.1221(0.0750)

0.1137***

(0.0363)

0.1584***

(0.0332)

Industry-return-

returnonS&P

500

-0.0500***

(0.0172)

0.0853**

(0.0401)

-0.0687***

(0.0248)

-0.0321(0.0396)

0.0347(0.0244)

-0.1079***

(0.0246)

Laggedindustry

return-returnon

S&P500

-0.0630***

(0.0159)

-0.0683*

(0.0373)

-0.0269(0.0247)

-0.0616*

(0.0371)

-0.1029***

(0.0256)

-0.0471**

(0.0207)

Industry-adjusted

return

-0.0646***

(0.0104)

-0.0202(0.0201)

-0.0822***

(0.0153)

-0.0802***

(0.0199)

-0.0319**

(0.0144)

-0.0983**

(0.0147)

Laggedindustry-

adjustedstock

return

-0.0618***

(0.0102)

-0.0698***

(0.0210)

-0.0597***

(0.0149)

-0.0593***

(0.0193)

-0.0660***

(0.0155)

-0.0601***

(0.0134)

CE

Oagedummy

0.1366***

(0.0081)

0.1252***

(0.0133)

0.1663***

(0.0139)

0.1125***

(0.0144)

0.1331***

(0.0107)

0.1395***

(0.0121)

n

9850

2886

3861

3103

4859

4991

PseudoR2

0.0686

0.0692

0.0866

0.0652

0.0697

0.0838

Pro


gtheperiodfrom




swheretheCEO


dependent







0to1.CEO



in

parentheses.

*,**,and***






14/31

full sample period, the result is an increase in the predicted probability ofturnover of 5.18%. Repeating the same procedure, the 1992 to 1999 and 2000to 2007 subperiods result in increases in the probability of turnover of 0.1% and

6.6%, respectively. Again, the sensitivity to industry-adjusted performance forthe 2000 to 2007 period appears to be greater than any of the sensitivitiesreported in Murphy (1999). Similar exercises for market performance show thatsimilar to the results for industry-adjusted returns, the relations between marketperformance and turnover are greater in the 2000 to 2007 subperiod than in the1992 to 1999 subperiod.

In contrast, when we evaluate the change in current and lagged industryperformance, we find that one-standard deviation declines in industry per-formance are associated with larger increases in the 1992 to 1999 subperiodthan in the 2000 to 2007 subperiod. Specifically, a one-standard deviation

decline in current and lagged industry performance during the 1992 to 1999subperiod is associated with a 5.00% increase in turnover. In contrast, a one-standard deviation decline in current and lagged industry performance duringthe 2000 to 2007 subperiod is associated with a 3.8% increase in turnover.Finally, one-standard deviation declines in current and lagged market returnsare associated with a 2.5% increase in turnover for using the full-sampleperiod.

Overall, the results in Tables 4 and 5 suggest that since 2000, boards havebeen more sensitive to poor stock performance. It is also worth noting that theeconomic magnitudes of the effect are large. For the 2000 to 2007 period, the

regressions in Table 4 imply that a CEO whose firm performs one-standarddeviation better than the industry has a 5.2% lower likelihood of turnover whilea CEO whose firm performs one-standard deviation worse than the industry hasa cumulative 5.2% increase in the likelihood of turnover. From a base turnoverlevel of 12.5%, these imply likelihoods of 7.4% for the strong performer versus17.8% for the poor performer. These are economically meaningful differenceswith 7.4% implying a tenure of 13.5 years and 17.8% implying a tenure of5.6 years.

Table 6 repeats the Table 5 probit regressions for the full-sample period andthe 1992 to 1999 and 2000 to 2007 subperiods for two sets of firms: firms in theS&P 500 index and all other Fortune 500 firms. We do this for two reasons. First,many paper on executive compensation and corporate governance use theExecuComp dataset that includes only firms in various S&P indices. It is possiblethere is a selection bias in these firms. Second, investors may be more likely topay attention to firms in the S&P 500 index, and, if so, these firms would bemore likely to be monitored by the press and institutional investors. Thus, theeffect of stock market performance might be different for these firms (Bertrandand Mullainathan, 2001).

The coefficient patterns in Table 6 are qualitatively similar for the two sets offirms except for the 2000 to 2007 subperiod for which turnover for firms not inthe S&P 500 index is not statistically related to market performance. Turnoverin both sets of firms is significantly related to industry-adjusted and market




15/31

Ta

ble6


CEOturnoveronperformance

Variable

Firmsin

theS&P500

Firmsnotin

theS&P500

19922007DP

rob

(s.e)

19921999DProb

(s.e)

20002007DProb

(s.e)

19922007DProb

(s.e.)

19921999DProb

(s.e.)

20002007DProb

(s.e.)

ReturnonS&P500

-0.2328***

(0.0364)

-0.0486(0.0644)

-0.30

55***

(0.0470)

-0.1602**

(0.0429

)

-0.0067(0.0791)

-0.1732

(0.0549)

Laggedreturnon

S&P500

0.1347***

(0.0303)

0.0849*

(0.0478)

0.23

38***

(0.0425)

0.0656*

(0.0372)

0.1298**

(0.0548)

0.0555

(0.0533)

Industryreturn

returnonS&P

500

-0.0625***

(0.0221)

0.0298(0.0314)

-0.13

59***

(0.0309)

-0.0314(0.0276)

0.0424(0.0385)

-0.0838**

(0.0404)

Laggedindustry

returnreturnon

S&P500

-0.0579***

(0.0209)

-0.0881***

(0.0338)

-0.05

34**

(0.0273)

-0.0672***

(0.024

4)

-0.1034***

(0.0392)

-0.0430

(0.0319)

Industry-adjusted

stockreturn

-0.0484***

(0.0149)

0.0028(0.0195)

-0.11

70***

(0.0212)

-0.0812**

(0.0146

)

-0.0797**

(0.0212)

-0.0824***

(0.0198)

Laggedindustry-

adjustedstock

return

-0.0559***

(0.0147)

-0.0430**

(0.0215)

-0.07

14***

(0.0198)

-0.0696***

(0.014

1)

-0.0906***

(0.0224)

-0.0506***

(0.0179)

CE

Oagedummy

0.1625***

(0.0108)

0.1560***

(0.0141)

0.16

90***

(0.0163)

0.0967***

(0.012

0)

0.0974***

(0.0160)

0.0951***

(0.0178)

n

6015

2954

3061

3835

1905

1930

PseudoR2

0.0809

0.0760

0.11

04

0.0574

0.0772

0.0532

Pro


gtheperiodfrom




swheretheCEO


dependent







0to1.CEO



in

parentheses.

*,**,and***






16/31

stock performance over the entire sample period. As in the overall sample, therelations for firm in the S&P 500 index are stronger in the more recent 2000 to2007 period.

B. Internal turnover and governance variables

The previous sections document an increase in CEO turnover and turnover-performance sensitivity for large public companies in the United States. In thissection, we consider four possible sources of those increases corporate gov-ernance (or shareholder rights), shareholder blockholdings, board independ-ence, and the SOX legislation.

Recent work has suggested that differences in corporate governance andshareholder rights may have real effects. Gompers et al. (2003) find that differ-

ences in corporate governance and shareholder rights are related to stockreturns. Masulis et al. (2007) find that those differences in corporate governanceare related to acquisition behavior. In both papers, greater shareholder rights areassociated with higher stock returns.

In this section, we examine the relation between turnover, stock perform-ance, and governance. To do so, we use the Gompers Ishi Metrick (GIM)index developed by Gompers et al. (2003). They categorize 24 charter provi-sions, bylaw provisions, and other firm-level rules associated with corporategovernance into five types: (1) tactics for delaying hostile bidders; (2) votingrights; (3) director/officer protection; (4) other takeover defenses; and (5) state

laws. Their overall GIM index and the five component indices generally scoreone point for each provision that restricts shareholder rights or increasesmanagerial power. Thus, a higher GIM index score represents greater mana-gerial power (weaker shareholder rights).

We estimate turnover regressions that interact stock performance with themeasure of governance. We used both a continuous measure of the GIM indexas well as an indicator variable equal to one if the firms GIM index is in thehighest quintile that year and zero otherwise. To the extent that the GIM indexmeasures poor governance, the GIM index should have a negative effect on thelevel of turnover (i.e., poorly governed firms should have less turnover), whilethe interaction of the GIM index and stock performance should have a positiveeffect on turnover (i.e., turnover at poorly governed firms should be less sensi-tive to poor performance).

Table 7 presents results for probit regressions using the indicator variable forthe highest GIM index quintile. For the full-sample period, the marginal prob-ability associated with the GIM index is positive and marginally significant.That is, a high GIM index (fewer shareholder rights) is associated with slightlyhigher CEO turnover, not less.

For the sample overall and for each subperiod, most of the interaction termsare not statistically different from zero. For two coefficients, the GIM indexinteraction is significantly negative. Turnover is significantly more sensitive topoor lagged industry-adjusted performance for the high GIM index firms during




17/31

Table 7 Probit regressions of internal CEO turnover for Fortune 500 firms onperformance and governance

Variable Full sample 19921999 20002007

DProb (s.e.) DProb (s.e.) DProb (s.e.)

Return on S&P 500 -0.2146*** (0.0315) -0.0499 (0.0563) -0.2621*** (0.0408)Industry return

return on S&P 500-0.0485** (0.0196) 0.0346 (0.0275) -0.1071*** (0.0281)

Industry-adjusted stockreturn

-0.0683*** (0.0119) -0.0401** (0.0161) -0.0934*** (0.0171)

Lagged return on S&P500

0.1314*** (0.0267) 0.1316** (0.0418) 0.1803*** (0.0372)

Lagged industry return return on S&P 500

-0.0656*** (0.0180) -0.1077*** (0.0161) -0.0492** (0.0232)

Lagged

industry-adjustedstock return

-0.0529*** (0.0117) -0.0632*** (0.0179) -0.0457*** (0.0151)

G index 0.0029* (0.0016) 0.0018 (0.0021) 0.0041 (0.003)CEO age dummy 0.1358*** (0.0093) 0.1254*** (0.0120) 0.1455*** (0.0142)High G index dummy 0.0906 (0.1579) 0.1233 (0.2793) 0.0189 (0.1643)High G index dummy

x(Return on S&P500)

0.1057 (0.0771) 0.0207 (0.1412) 0.1273 (0.0985)

x(Industry return return on S&P500)

0.0321 (0.0468) 0.0413 (0.0684) 0.0161 (0.0674)

x(Industry-adjusted

stock return)

0.0429 (0.0308) 0.0660 (0.0454) 0.0252 (0.0430)

x(Lagged return onS&P 500)

-0.0915 (0.0636) -0.0652 (0.0959) -0.1158 (0.0906)

x(Lagged industryreturn return onS&P 500)

0.0153 (0.0468) 0.0063 (0.0699) 0.0087 (0.0624)

x(Lagged industry-adjusted stockreturn)

-0.0607* (0.0326) -0.0170 (0.0517) -0.0956** (0.0425)

x(G index) -0.0066 (0.0086) -0.0083 (0.0135) -0.0017 (0.0114)x(CEO age dummy) 0.0263 (0.0195) 0.0549** (0.0317) 0.0056 (0.0248)

n 9213 4557 4656

Pseudo R2 0.0722 0.0728 0.0885

Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firmsduring the period 1992 to 2007. Internal turnover excludes turnover due to mergers and acqui-sitions and delistings from a major stock exchange. Occurrences where the CEO dies in office aredefined as non-turnover events. The dependent variable equals one if the CEO turnovers and zerootherwise. DProb measures the change in the probability of CEO turnover per unit change in therelevant explanatory variables. For indicator variables, the coefficient represents the change inthe probability associated with moving the indicator from 0 to 1. Models are estimated withrobust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEOage is greater than or equal to 60 and zero otherwise.*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standarderrors to control for heteroskedasticity are reported in parentheses.




18/31

the overall sample period and in the 2000 to 2007 subperiod. These are theopposite signs one would expect if the GIM index measured poor governance. 3

Overall, then, we interpret these results as finding that the GIM measures of

governance or shareholder rights do not have an appreciable relation to CEOturnover. The most one can say is that the GIM measure is possibly associatedwith a somewhat faster response to poor industry-adjusted performance. Ourresults are consistent with Bhagat and Bolton (2006) who also fail to find asignificant effect on turnover when they interact governance and performance.

Next, we examine the relation between CEO turnover and blockholderownership. Institutional and blockholder ownership increase over the sampleperiod. We focus on blockholder ownership (where an institution owns at least5% of a firms outstanding shares) because blockholders have both the incentiveand the ability to monitor. Cremers and Nair (2005) find that blockholder

ownership and governance affect corporate valuations in certain circumstances.We follow Cremers and Nair (2005) and use the percentage of shares held ineach firm by the firms largest institutional blockholder where blockholders areshareholders with greater than 5% ownership of the firms outstanding shares.Data on institutional ownership are from Thomson-Reuters Institutional Hold-ings (13F) database.

Table 8 reports the mean and median block ownership and the presence of ablockholder for our sample firms. The average holding of blockholders andfraction of firms with a blockholder present increase over the sample period.During the 1992 to 1999 subperiod, the average blockholder ownership is 9.6%,

while 57% of the firms have a blockholder present. During the 2000 to 2007subperiod, blockholder ownership averages 14.98%, while almost 75% of thefirms have a blockholder present. The differences across subperiods are statisti-cally significant.

Table 9 reports the results of probit turnover regressions that include thepercentage of total blockholder ownership (at the end of the previous year) andthe interactions of the continuous measure and our three stock performancemeasures. As Table 9 shows, the marginal probability associated with the per-centage of total blockholder ownership while positive is not statistically differ-ent from zero for the full sample or for either subperiod. The only interactionterm that is statistically significant is the interaction of block ownership andindustry-adjusted stock returns in the 2000 to 2007 subperiod. One interpre-tation of this result is that blockholders are particularly active in firms thatunderperform their industries but only so in the latter time period.

This is somewhat suggestive of blockholders playing a role in the increase inindustry-adjusted turnover-performance sensitivity. These results are consistentwith those in Brav et al. (2008) who find increased CEO turnover associatedwith hedge fund activism and Del Guercio et al. (2008) who find increased CEOturnover associated with institutional investor voting.

3 Although not reported in a table, our results also are qualitatively similar when we interact

stock performance with the continuous measure of the GIM index.




19/31

Third, we consider the role of independent directors. Weisbach (1988) findsthat turnover-performance sensitivities are greater for firms with more inde-pendent boards. And over time, boards in the United States have becomemore independent. It is possible that this increased independence has playeda role in the changes in turnover. We obtain data on director independencefrom the IRRC Directors database on WRDS. We use the variable director typeto classify directors as independent. The IRRC data are available only from1996 onward.

Table 8 reports the percentage of independent directors and the fractionof firms that do not have a majority independent board. Similar to therise in blockholders, the percentage of independent directors on a firmsboard increases over time. The percentage increases from 63% to almost72% from the 1992 to 1999 subperiod to the 2000 to 2007 subperiod. Moreo-ver, the fraction of firms without a majority independent board decreasesfrom 24% during the 1992 to 1999 subperiod to 12% in the 2000 to 2007subperiod.

Table 10 reports the results of probit turnover regressions that account forboard independence. We include a dummy variable that equals one if the firmdoes not have a majority independent board. (The baseline, therefore, is a firmthat does have a majority independent board.) The results are qualitatively

Table 8 Block ownership and board of directors

Year Blockownership

Blockholderpresent

Percent (%) ofindependent

directors

Not anindependent

boardMean Median % of firms Mean Median % of firms

1992 8.12 5.02 50.121993 8.12 5.51 52.881994 8.56 5.25 52.821995 9.05 5.80 54.711996 9.42 6.08 56.97 61.5 63.64 28.221997 10.08 6.37 57.75 63.38 66.67 25.421998 11.01 7.55 75.12 63.94 66.67 22.421999 11.48 8.35 62.74 65.09 66.67 21.572000 12.53 9.54 63.41 66.15 70.00 20.732001 12.55 10.54 66.29 67.39 70.00 18.992002 13.72 11.53 72.93 69.63 72.73 15.182003 13.32 11.33 73.00 71.45 75.00 12.062004 16.42 14.36 80.12 72.82 75.00 9.422005 16.54 14.83 79.23 73.90 77.78 7.852006 17.10 14.74 81.00 74.32 77.78 8.142007 17.92 15.38 78.68 80.45 81.82 1.42

Block ownership and board independence for sample firms during the period 1992 to 2007.Institutional ownership data are from 13-filings. A block is defined as an institutional owner withgreater than a 5% ownership stake. Independent director data are taken from IRRC Directorsdatabase on WRDS and proxy filings.




20/31

Table 9 Probit regressions of internal CEO turnover for Fortune 500 firms onperformance and block ownership

Variable Full sample 19921999 20002007


Return on S&P 500 -0.2183*** (0.0394) -0.0389 (0.6573) -0.2672*** (0.0523)Industry return

return on S&P 500-0.0593** (0.0244) 0.0288 (0.0327) -0.1410*** (0.0359)


-0.0450*** (0.0147) -0.0522*** (0.0195) -0.0865*** (0.0231)

Lagged return on S&P500

0.1018*** (0.0329) 0.1436*** (0.0486) 0.1358*** (0.0472)


-0.0851*** (0.0226) -0.1162*** (0.0341) -0.0597** (0.0304)

Lagged industry-

adjusted stock return

-0.0450*** (0.0150) -0.0650*** (0.0213) -0.0297 (0.0197)

Percent of total blockownership

0.0116 (0.0300) 0.0847 (0.0903) -0.0303 (0.0355)

CEO age dummy 0.1370*** (0.0081) 0.1339*** (0.0107) 0.1395*** (0.0121)Percent of total block

ownershipx(Return on S&P

500)0.0998 (0.2212) -0.0270 (0.4471) 0.1133 (0.2844)


0.0792 (0.1311) 0.0443 (0.2080) 0.2303 (0.1769)

x(Industry-adjusted

stock return)

-0.0057 (0.0759) 0.1845 (0.1135) -0.0721 (0.1072)


0.0998 (0.2212) -0.2880 (0.3154) 0.1365 (0.2590)

x(Lagged industryreturn return onS&P 500)

0.1616 (0.1211) 0.1224 (0.2328) 0.0753 (0.1515)


-0.1255 (0.0771) -0.0055 (0.1377) -0.2016** (0.0958)

n 9848 4857 4991Pseudo R2 0.0698 0.0711 0.0859

Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firms duringthe period 1992 to 2007. Internal turnover excludes turnover due to mergers and acquisitions anddelistings from a major stock exchange. Occurrences where the CEO dies in office are defined asnon-turnover events. The dependent variable equals one if the CEO turnovers and zero other-wise. DProb measures the change in the probability of CEO turnover per unit change in therelevant explanatory variables. For indicator variables, the coefficient represents the change inthe probability associated with moving the indicator from 0 to 1. Models are estimated withrobust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEOage is greater than or equal to 60 and zero otherwise.*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standarderrors to control for heteroskedasticity are reported in parentheses.




21/31

similar but have a less natural interpretation when we use the percentage ofindependent directors. We also interact this indicator variable with the stockperformance variables. We estimate the probit for the 1996 to 2007 period, theperiod over which we have director data.

Table 10 indicates that the marginal probability associated with thenon-independent board indicator variable, while negative, is not statisticallysignificant. At the same time, two of the six interaction terms are positiveand statistically significant suggesting that non-independent boards areassociated with less turnover-performance sensitivity. Said another way,the results of the second regression suggest that more independent boardsare associated with more turnover and more turnover sensitivity to poorperformance.

Table 10 Probit regressions of internal CEO turnover for Fortune 500 firms onperformance and independent directors

Variable 19962007 19962007

DProb (s.e.) DProb (s.e.)

Return on S&P 500 -0.2638*** (0.0311) -0.2764*** (0.0336)Industry return return on S&P 500 -0.0716*** (0.0189) -0.0647* (0.0206)Industry-adjusted stock return -0.0750*** (0.0116) -0.0913*** (0.0130)Lagged return on S&P 500 0.1356*** (0.0272) 0.1542*** (0.0293)Lagged industry return return on S&P

500-0.0498*** (0.0172) -0.0623*** (0.0189)

Lagged industry-adjusted stock return -0.0600*** (0.0114) -0.0724*** (0.0127)CEO age dummy 0.1418*** (0.0097) 0.1418*** (0.0097)No independent board indicator variable -0.0120 (0.0095) -0.0147 (0.0125)No independent board indicator variable

x(Return on S&P 500) 0.0572 (0.0871)x(Industry return return on S&P 500) -0.0442 (0.0485)x(Industry-adjusted stock return) 0.0812*** (0.0295)x(Lagged return on S&P 500) -0.0877 (0.0778)x(Lagged industry return return on S&P

500) 0.0562 (0.0439)

x(Lagged industry-adjusted stock return) 0.0506* (0.0281)n 7374 7374Pseudo R2 0.0777 0.0807

Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firms duringthe period 1996 to 2007. Internal turnover excludes turnover due to mergers and acquisitions and

delistings from a major stock exchange. Occurrences where the CEO dies in office are defined asnon-turnover events. The dependent variable equals one if the CEO turnovers and zero other-wise. Independent director data are taken from IRRC Directors database on WRDS. No independ-ent board equals one if a majority of directors are not independent. DProb measures the changein the probability of CEO turnover per unit change in the relevant explanatory variables. Forindicator variables, the coefficient represents the change in the probability associated withmoving the indicator from 0 to 1. Models are estimated with robust standard errors to control forheteroskedasticity. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 andzero otherwise.*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standarderrors are in parentheses.




22/31

In Table 11, we examine whether the probability of CEO turnover is greaterin the years under the SOX legislation. We create an indicator variable equal toone for years 2003 and after and zero otherwise. In the first regression, we

include this indicator variable, current year measures of performance, andinteractions between the indicator variable and performance variables. In thesecond regression, we add the lagged performance measures and correspondinginteraction terms. In the last column of the table, we repeat the second regres-sion for the 2000 to 2007 period.

In the first regression, CEO turnover is significantly higher during yearsunder the SOX legislation. The marginal positive marginal probability impliesabout a 3% increase in turnover during these years. Additionally, the interactionterm between the SOX indicator variable and market performance is signifi-cantly negative implying a decrease in market performance is associated with

higher turnover during the years under the SOX legislation. When we add thelagged variables, the marginal probability associated with the SOX indicatorvariable is positive but no longer statistically significant in the full sample or the2000 to 2007 subperiod. For the 2000 to 2007 subperiod, the interaction termbetween the indicator variable and industry performance is positive and sig-nificant, but the sum of the industry performance and the interaction termis not statistically different from zero. We conclude that these results areinconclusive.

Taken together, the governance variables are of mixed help in explainingthe increase in turnover and turnover-performance sensitivity in the latter

half of our sample. While the GIM measure and the SOX legislation do notreliably affect turnover, the presence of large blockholders and independentboards are associated with greater turnover-performance sensitivity. The find-ings in Jenter and Lewellen (2010) for a different sample are consistent withthis latter result.

C. Forced turnover

Thus far, we have not distinguished between forced turnover and all otherturnovers. Jenter and Kanaan (forthcoming) focus exclusively on forcedturnover. As we do for all turnovers, they find that forced turnover isrelated to the three different measures of performance. It is possible that ourresults are driven by the forced turnover in our sample. To examine whetherthis is the case and whether performance is related differently to forced turno-ver and standard internal turnover, we estimate multinomial logit (MNL)regressions.

We follow Huson et al. (2001) in classifying turnover as forced. (Jenter andKanaan forthcoming) also use this classification scheme.) If an article in thebusiness press indicates that the CEO was fired, forced, or left following a policydisagreement or some other equivalent, then turnover is defined as forced. Forthe remaining announcements, succession is classified as forced when the CEOis under 60, and the first article reporting the announcement does not report




23/31

Table 11 Probit regressions of internal CEO turnover for Fortune 500 firms onperformance and SOX legislation

Variable Full sample Full sample 2000 - 2007


Return on S&P 500 -0.0917*** (0.234) -0.1899*** (0.0335) -0.4782*** (0.1944)Lagged return on S&P

500 0.0801*** (0.0301) 0.2833*** (0.0750)

Industry return return on S&P 500

-0.0585*** (0.0200) -0.0474** (0.0200) -0.1386*** (0.0320)


-0.0747*** (0.0186) -0.0245 (0.0286)


-0.0581*** (0.0122) -0.0586** (0.0120) -0.1090*** (0.0200)

Lagged industry-

adjusted stock return

-0.0620*** (0.0119) -0.0557*** (0.0173)

CEO age dummy 0.1354*** (0.0081) 0.1362*** (0.0081) 0.1395*** (0.0121)SOX dummy 0.0295*** (0.0109) 0.0174 (0.0181) 0.0448 (0.0327)SOX dummy

x(Return on S&P500)

-0.2834*** (0.0820) -0.1520 (0.1275) 0.1308 (0.2315)


0.0515 (0.0862) -0.1496 (0.1113)


0.0080 (0.0461) 0.0126 (0.0471) 0.1033* (0.0539)

x(Lagged industry

return return onS&P 500)

0.0069 (0.0440) -0.0442 (0.0495)

x(Industry-adjustedstock return)

-0.0291 (0.0249) -0.0266 (0.0248) 0.0224 (0.0298)


-0.0018 (0.0240) -0.0092 (0.0273)

n 9982 9850 4991Pseudo R2 0.0606 0.0698 0.0866

Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firms duringthe period 1992 to 2005. Internal turnover excludes turnover due to mergers and acquisitions and

delistings from a major stock exchange. Occurrences where the CEO is promoted to another CEOposition, remains CEO of the delisting firm, is selected for a government position, dies in office,or leaves office due to illness are defined as non-turnover events. The dependent variable equalsone if the CEO turnovers and zero otherwise. The SOX dummy equals one for years after 2002.DProb measures the change in the probability of CEO turnover per unit change in the relevantexplanatory variables. For indicator variables, the coefficient represents the change in the prob-ability associated with moving the indicator from 0 to 1. Models are estimated with robuststandard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age isgreater than or equal to 60 and zero otherwise.*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. p-values are inparentheses.




24/31

the reason for the departure as involving death, poor health, or the acceptanceof another position elsewhere.

We present the regression results in Table 12. The dependent variable catego-ries in the MNL estimations are unforced turnover, forced turnover, or noturnover. The Table reports the marginal effect of each regressor. The probabilityof forced turnover is significantly negatively related to the three components offirm stock performance firm performance relative to the industry, the industryrelative to the market, and the overall market in the current year and toindustry-adjusted return in the previous year. The probability of unforcedturnover is significantly negatively related to firm stock performance relative tothe industry and the overall market in the current year, and to industry stockreturn and industry-adjusted stock return in the previous year. As in the previ-

ous results, both types of turnover are positively related to lagged marketperformance.Overall, then, both forced turnover and unforced turnover are sensitive to all

three types of poor stock performance. This strongly suggests that a number ofunforced turnovers are not voluntary. In work that was partially motivated byour results, Jenter and Lewellen (2010) find strong evidence consistent with thisconclusion.

D. External turnover

As discussed earlier, in addition to internal turnover, we examine externalturnover. Recall, nonstandard or external turnover is turnover due to a merger

Table 12 Multinominal logit regression estimates of the likelihood of no CEOturnover, unforced CEO turnover, and forced CEO turnover for Fortune 500 firms

Variable Unforced CEO turnover Forced CEO turnover

DProb (s.e.) DProb (s.e.)

Return on S&P 500 -0.1402*** (0.0237) -0.0341*** (0.0084)Lagged return on S&P 500 0.0761*** (0.0204) 0.0182*** (0.0069)Industry return return on S&P 500 -0.0202 (0.0144) -0.0229*** (0.0052)Lagged industry return return on

S&P 500-0.0454*** (0.0134) -0.0037 (0.0048)

Industry-adjusted stock return -0.0285*** (0.0086) -0.0288*** (0.0033)Lagged industry-adjusted stock return -0.0399*** (0.0088) -0.0130*** (0.0037)CEO age dummy 0.1517*** (0.0079) -0.0087*** (0.0018)Pseudo R2 = 0.0878

Multinominal logit regression estimates of the likelihood of no CEO turnover, non-forced CEOturnover, and forced CEO turnover for Fortune 500 firms during the period 1992 to 2007. DProbmeasures the change in the probability of the particular choice per unit change in the relevantexplanatory variables. For indicator variables, the coefficient represents the change in the prob-ability associated with moving the indicator from 0 to 1. Models are estimated with robuststandard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age isgreater than or equal to 60 and zero otherwise.*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standarderrors are in parentheses.




25/31

or bankruptcy/delisting. We consider the CEO to have been turned over in amerger if his or her company is taken over by another company, and he or sheis not CEO of the combined company. We consider the CEO to have been

turned over in a bankruptcy if he or she is no longer CEO of the bankruptcompany. The incidence of external turnover is 4.7% per year over the sampleperiod.

Table 13 reports probit regressions of the probability of external turnover asa function of current stock market performance (columns 1 through 3) andcurrent and lagged performance (columns 4 through 6). Unlike the results forinternal turnover, in most specifications, external turnover is unrelated toperformance. When performance is significant, the marginal probabilities areeconomically small and in most cases positive. For example, when we includeonly current performance in the regressions, external turnover is positively

related to market performance in the earlier subperiod and to industry-adjustedstock performance in the full sample period and later subperiod. One interpre-tation of the lack of results for external turnover and performance is that, onaverage, mergers and acquisitions during this period were not disciplinary innature.

V. SUMMARY AND IMPLICATIONS

In this paper, we examine the extent and determinants of internal and exter-nal CEO turnover for a sample of large US companies from 1992 to 2007.

Total turnover, the sum of internal and external turnover, is 15.84%from 1992 to 2007, implying an average CEO tenure of less than 7 years.In the more recent period from 2000 to 2007, total CEO turnover increasesto 16.78%, implying an average tenure of just under than 6 years. Internalor board-driven turnover also rises substantially in the latter part of thesample.

We then look at how turnover varies with firm stock performance. Previouswork suggests a modest relation between internal (board initiated) turnoverand firm stock performance. We find a stronger and significant relationbetween internal turnover and three different components of firm stockperformance performance relative to the industry, performance of the indus-try relative to the stock market, and the performance of the overall stockmarket (Jenter and Kanaan forthcoming) obtain similar results for forcedturnover.) The sensitivities are economically meaningful. Both types of inter-nal turnover forced and unforced are sensitive to all three types of poorstock performance.

Internal turnover after 2000 is more strongly related to all three measures ofstock performance. In fact, the sensitivity to stock performance appears to begreater than that in any of the periods between 1970 and 1995 studied inMurphy (1999).

We next consider four possible explanations for or factors that drive thechanges in turnover and turnover-performance sensitivity. There is some




26/31

Ta

ble13

Probitregressionsoftheprobabilityofexterna

lCEOturnoveronperformance

Variable

19922007DProb

(s.e.)

19921999DProb

(s.e.)

2000

2007DProb

(s.e.)

19922007DProb

(s.e.)

19921999DProb

(s.e.)

20002007DProb

(s.e.)

ReturnonS&P500

-0.0007(0.00

48)

0.0177**

(0.0073)

-0.00

26(0.0058)

-0.0095(0.0059)

0.0160(0.0099)

-0.0105

(0.0067)

Laggedreturnon

S&P500

0.0125**

(0.0052)

0.0162**

(0.0075)

0.0168***

(0.0062)

Industryreturn

returnonS&P

500

-0.0010(0.00

45)

0.0027(0.0066)

-0.00

72(0.0058)

0.0015(0.0044)

0.0051(0.0052)

-0.0052

(0.0053)

Laggedindustry

returnreturnon

S&P500

-0.0068(0.0029)

-0.0034(0.0042)

-0.0094**

(0.0038)

Industry-adjusted

stockreturn

0.0047**

(0.022)

0.0037(0.0030)

0.00

44*

(0.0026)

0.0042(0.0021)

0.0034(0.0027)

0.0033

(0.0024)

Laggedindustry-

adjustedstock

return

-0.0034(0.0025)

-0.0023(0.0033)

-0.0041

(0.0027)

CE

Oagedummy

0.0027(0.00

81)

-0.000(0.0020)

0.06

6(0.0032)

0.0025(0.0018)

0.0001(0.0018)

0.0057**

(0.0029)

n

8842

4400

4442

8719

4325

4391

PseudoR2

0.0135

0.0268

0.02

74

0.0250

0.0458

0.0449

Pro

bitregressionestimatesofthelikelihoodofexternalCEOturnoverforFortune500firmsduringtheperiod1

992to2005.Thedependentvariableequalsone

ift

heCEOturnoversandzerootherwise.

DProbmeasuresthechangeintheprobabilityofCEOturnoverperun

itchangeintherelevantexplanatoryvariables.

Forindicatorvariables,thecoefficient

representsthechangeintheprobabilityassociatedwithmovingthein

dicatorfrom

0to1.Modelsareestimatedwith

rob

uststandarderrorstocontrolforh

eteroskedasticity.CEOagedummyequals1iflaggedCEOageisgreate

rthanorequalto60andzeroothe

rwise.

*,**,and***


levels,respectively.Robuststandarderrorsto

controlforheteroskedasticityare

reportedin

parentheses.




27/31

evidence that the increases in turnover and turnover-performance sensitivityare related to increases in blockholdings and board independence. Turnover andturnover-performance sensitivity are not reliably related to the Gompers et al.

(2003) governance index.External turnover turnover primarily related to acquisitions also is sig-

nificantly related to all three measures of performance. As with internal turn-over, the results are stronger in the later subperiod.

Our results have several implications. First, they suggest that the CEO job ismore precarious than in the past. When external takeovers are included, theaverage tenure of a CEO has declined to less than 6 years for the recent 1998 to2005 period. The recent tenures are substantially shorter than those reported inthe previous work for the 1970s, 1980s, and 1990s. The shorter tenures appearto have continued out of sample.

For individual CEOs, the shorter expected tenure likely offsets some of thebenefits of the increase in CEO pay since the mid-1990s. For example, theannual pay of S&P 500 CEOs roughly doubled in real terms from the 19921997period to the 19982005 period. Our estimates suggest that the total pay of anindividual CEO over his or her entire expected term increased by less than thisbecause the expected tenure at the higher pay declined by one quarter to onethird.

This calculation would be inaccurate if severance agreements around internalturnover and takeovers are both large and have increased over time. If, instead,the severance agreements are small, then they do not have much of an effect on

a CEOs total pay. Yermack (2006) and Hartzell et al. (2004) study severanceagreements, respectively, around internal and external takeover events. In fact,the average and median magnitudes they report are modest. (They do not studywhether these payments have changed over time.) Severance agreements, there-fore, are unlikely to alter the conclusion that the job of CEO has become riskierand that the shorter expected tenures of CEOs partially offset increases in CEOpay.

The coincidence of shorter CEO tenures and higher CEO pay is consistentwith Hermalin (2005) who presents a model that predicts increased boardvigilance will be associated with both outcomes. The relations with moreindependent boards and blockholders also are consistent with the increasedvigilance in Hermalins model.

Second, the similar results for the turnover-performance sensitivities offorced and unforced turnover suggest that many unforced turnovers are notvoluntary. This greater incidence of involuntary turnover, in turn, implies boththat boards do a better job of disciplining poor management and that the CEOsjob is more precarious than is commonly believed.

Third, our results suggest an evolving role for boards. In a sample fromthe 1980s, Morck et al. (1989) find that internal turnover is related toindustry-adjusted performance while external turnover from hostile take-overs is related to industry performance. They interpret this as indicatingboards respond well to poor performance relative to the industry but do




28/31

not respond well to poor industry performance. The external takeover marketbecomes active in reaction to poor industry performance and a need forrestructuring.

Our results suggest that boards respond not only to poor performance relativeto the industry but also to poor industry performance and to poor marketperformance. To the extent that internal turnover has increased, boards alsoappear to monitor more frequently. One interpretation of these results is thatboards possibly encouraged by large shareholders perform both the role theyperformed in the 1980s and the role that hostile takeovers played then. Theincreased turnover associated with blockholdings and board independence isconsistent with this interpretation.

The result that boards do not index CEO turnover to the industry or themarket is noteworthy in light of criticisms of boards for not indexing CEO

pay to the industry or the market. Bebchuk and Fried (2002) interpret the lackof indexing of pay as a failure of governance. Our results on turnover inconjunction with those in Morck et al. (1989) for the earlier period provide analternative explanation. When an industry or the overall economy performspoorly, it is sometimes efficient for the board to bring on a new CEO torespond to the new industry or market conditions. The recent (and out ofsample) high turnover in the financial services industry is consistent withthis.

Fourth, the shorter expected CEO tenures and sensitivity of those tenuresto stock performance have implications for the measurement of CEO pay.

The shorter expected tenures suggest that the estimates of CEO pay basedStandard and Poors ExecuComp data may be overstated. While optiongrants typically have a 10-year life, ExecuComp uses a 7-year life becauseexecutives rarely wait until the expiration date to exercise their options.This adjustment assumes that CEOs will remain with the company for atleast 7 years. If a CEO has an expected initial tenure of 6 years, the Execu-Comp assumption will tend to overstate the value of option grants every yearof the CEOs tenure with the overstatement increasing each year. Thisassumes that CEOs forfeit unvested options and/or must exercise vestedoptions when they leave the company. For internal turnover, Yermack (2006)finds that this tends to be the default policy for most companies, and com-panies deviate from those policies in only 16% of the internal turnovers hestudies. ExecuComp also values restricted stock grants as fully vested when, infact, they usually vest over a period of time. This also will tend to overstateexecutive compensation.

The sensitivity of turnover to performance implies that the vesting andeffective life of stock options are not independent of performance. The optionsof CEOs of companies that perform poorly will both have a shorter effective lifeand will be worth less. The Black-Scholes methodology does not take thesecorrelations into account. This, in turn, implies that a proper valuation of stockoptions such as, using a binomial tree approach would incorporate thesecorrelations.




29/31

Finally, shorter CEO tenures, the greater sensitivity to stock performance, aswell as higher CEO pay may have created a greater incentive for CEOs to engagein earnings management or manipulation.

Bernadette MintonFisher College of BusinessThe Ohio State University834 Fisher HallColumbusOH [email protected]

REFERENCES

Bebchuk, L., and J. Fried (2002), Managerial Power and Rent Extraction in the Designof Executive Compensation, University of Chicago Law Review, 69, 751846.

Bebchuk, L., and Y. Grinstein (2005), The Growth of U.S. Executive Pay, Oxford Reviewof Economic Policy, 21, 283303.

Bertrand, M., and S. Mullainathan (2001), Are CEOs Rewarded for Luck? The Oneswithout Principles Are, Quarterly Journal of Economics, 116, 90132.

Bhagat, S., and B. Bolton (2006), Corporate Governance and Firm P

Kaplan and Minton_How Has CEO Turnover Changed

Documents

Transcript of Kaplan and Minton_How Has CEO Turnover Changed