SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix

We study CEO turnover – both internal (board driven) and external (through takeover and bankruptcy) – from 1992 to 2007 for a sample of large US companies. Annual CEO turnover is higher than that estimated in previous studies over earlier periods. Turnover is 15.8% from 1992 to 2007, implying an average tenure as CEO of less than 7 years. In the more recent period since 2000, total CEO turnover increases to 16.8%, implying an average tenure of less than 6 years. Internal turnover is significantly related to three components of firm stock performance – performance relative to industry, industry performance relative to the overall market, and the performance of the overall stock market. The relations are stronger in the more recent period since 2000. We find similar patterns for both forced and unforced turnover, suggesting that some, if not most, turnover labeled as unforced is actually not voluntary. The turnover-performance sensitivity is modestly related to block shareholder ownership and board independence.

I. Introduction

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix

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

While a great deal of work has focused on changes in CEO pay, recent changes in CEO turnover and board behavior have received little attention. Earlier work and casual empiricism suggest that the CEO's job has become riskier over time. Khurana (2002) reports that CEO turnover increased in the 1990s relative to the 1970s and 1980s. Murphy and Zabojnik (2008) and Jensen et al. (2004) also report that turnover has increased in the 1990s, although the magnitude 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 the year 2000 so they are unable to consider the period in which corporate governance 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 large US companies. We consider turnover that occurs through takeover and bankruptcy as well as turnover in ongoing companies. When takeovers and bankruptcies are taken into consideration, the job of CEO in large US companies appears more precarious than before, particularly since 2000. Annual CEO turnover is 15.8% from 1992 to 2007, implying an average tenure as CEO of less than 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 CEO turnover increases to about 16.8%, implying an average CEO tenure of about 6 years. Internal or board-driven turnover also rises substantially, increasing from 10.9% in the first part of the sample to 12.4% in the latter part of the sample. Looked at another way, only 21.30% of CEOs in place in 1992 remained CEO in 1999, while only 16.35% of CEOs in place in 2000 remained CEO in 2007.

We then examine how turnover varies with firm performance. Previous work suggests a modest relation between internal (board-initiated) turnover and firm stock performance (see Murphy 1999 and Jensen et al. 2004). We find stronger (and more significant) relations. Internal turnover is related to three different components of total firm stock performance. Turnover is sensitive to the stock performance of the firm relative to the industry, the stock performance of the industry relative to the stock market (under certain specifications), and the performance of the overall stock market. (Jenter and Kanaan forthcoming obtain similar results for forced turnover.) The sensitivity to one-standard deviation differences in each measure is economically meaningful. We find similar results for both forced and unforced turnover.

We also find that internal turnover since 2000 is more strongly related to all three measures of stock performance. In fact, the sensitivity to stock performance appears to be greater than that in any of the periods between 1970 and 1995 studied in Murphy (1999). Ironically, it appears that during the period in which boards have been criticized, boards have become increasingly sensitive to firm stock performance.

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

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

As we discuss in more detail in the conclusion, our results have a number of implications. First, CEO tenures have declined, suggesting the CEO job is more 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. The recent tenures are substantially shorter than those reported in previous work for the 1970s, 1980s, and 1990s. For individual CEOs, the shorter expected tenure likely offsets some of the benefits of the increase in CEO pay over this period. This result is consistent with Hermalin (2005) who presents a model that predicts increased board vigilance will be associated with shorter CEO tenures (and higher CEO pay). The aforementioned relations with more independent boards and blockholders also are consistent with the increased vigilance in Hermalin's model.

Second, our similar results for the turnover-performance sensitivities of forced and unforced turnover suggest that a number of turnovers labeled as unforced are, in fact, not voluntary. In subsequent work, partially motivated by our results, Jenter and Lewellen (2010) find additional evidence consistent with this conclusion.

Third, the results suggest an evolving role for boards. In a sample from the 1980s, Morck et al. (1989) find that internal turnover is related to industry-adjusted performance while external turnover from hostile takeovers is related to industry performance. They interpret this as indicating boards respond well to poor performance relative to the industry, but do not respond well to poor industry 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 relative to the industry but also to poor industry performance and to poor market performance. One interpretation of these results is that boards (perhaps in concert with shareholders) perform both the role they performed in the 1980s and some of the role that hostile takeovers played then.

Fourth, the shorter expected CEO tenures and sensitivity of those tenures to stock performance have implications for the measurement of CEO pay. The shorter expected tenures suggest that the estimates of CEO pay used in most compensation studies may be overstated.

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

This paper was written contemporaneously with Jenter and Kanaan (forthcoming) who study related issues in a sample of CEO turnovers from 1993 to 2001. They focus on forced CEO turnover, rather than all CEO turnovers. Forced turnovers represent somewhere between 15% and 25% of total internal turnovers. As we do for forced and unforced turnover, they find that forced CEO turnover is significantly related to industry-adjusted, industry, and market returns. They focus most of their paper on verifying this effect for forced turnover and explaining why boards might behave this way. They also study a larger sample of firms but over a shorter time period. Unlike us, they do not focus on the level of total turnover, the annual variation in that turnover, and do not consider external turnovers. Given their shorter sample, they also do not consider how turnover behavior changes over time.

Our paper also is related to that of Mikkelson and Partch (1997) who compare complete management turnover in US companies in two 5-year periods – the active takeover market of 1984–1988 and the less active market of 1989–1993. In the active takeover period, they find that 39% of firms experience CEO turnover, and 23% of firms experience complete management turnover; in the less active period, 34% of firms experience CEO turnover, while 16% of firms experience complete management turnover. They find that the decline in turnover frequency is more pronounced among poorly performing firms. They argue that the activity of the external takeover market affects the ‘intensity of management discipline.’ Our results suggest that the intensity of management discipline has increased since the end of their sample period and likely exceeds the intensity of the active takeover period. Huson et al. (2001) also examine CEO turnover across subperiods to see if the relation between performance and turnover has changed over time. Using four 6-year subperiods from 1971 to 1994, they document that while CEO turnover 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 performance has changed.

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

II. Sample and Data

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix

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

We follow the sample firms from the first year they appear on a Fortune list until the end of the sample period or until the firm exits the sample because of a merger, acquisition, or delisting from a major stock exchange. Thus, we continue to follow a firm even if it falls out of the Fortune 500. We identify CEO turnovers using the Fortune 500 and Fortune 1000 lists, 10-K filings, proxy statements, Dun and Bradstreet's Million Dollar Directory, the Wall Street Journal, and Lexis/Nexis business news searches.

III. Turnover Levels

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix

Turnover in a given fiscal year, T, means that the CEO in the spring of year T is 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 turnover that is associated with a company's board of directors. For standard turnover, a company remains publicly listed over the course of the year, but the CEO in the spring is no longer the CEO the following spring. This is the turnover that is generally 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 her company 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 many mergers, the target CEO leaves the combined company. In those instances in which the CEO target remains, the target CEO generally experiences a reduction in pay and power. Total turnover is the sum of internal and external turnover.

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

figure

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

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

Download figure to PowerPoint

Table 1. CEO turnover
YearNumber of firmsTotal turnoverStandard (internal) turnover
(1)(2)(1)(2)
n%n%n%n%
  1. CEO turnovers in publicly traded Fortune 500 companies between 1992 and year-end 2007. Total turnover is all CEO turnovers including turnover due to mergers and acquisitions and delistings from a major stock exchange. Standard (internal) turnover excludes turnover due to mergers and acquisitions 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 is selected. (2) Defines occurrences where the CEO dies as a non-turnover event. Data are from annual Fortune 500 lists, 10-K filings, proxy statements, and the Wall Street Journal. Year denotes the fiscal year-end for the sales data on which Fortune ranks firms. (i.e., 1992 corresponds to the 1993 April/May Fortune list.)

19924605712.395712.395411.745411.74
19934865711.735711.754910.104910.10
1994727689.35669.08567.70547.43
199574111315.2511315.259212.429212.42
199673910313.9410013.537510.15729.74
199773611315.3511115.08689.24668.97
199872213018.0113018.018712.058712.05
199971615321.3715321.3710114.1110114.11
200070318526.3218326.0313419.0613218.78
20016839714.209714.20639.22639.22
200267613319.6713119.3810215.0910014.79
2003663466.94466.94345.13345.13
20046699414.059213.757811.667611.36
200567912718.7012618.569413.849313.70
200666311216.8911116.748312.528212.37
200765211016.8710716.418412.888112.42
Total10,715169815.84168015.67125411.70123611.54
1992–1996315339812.6239312.4632610.3432110.18
1997–2002423681119.1580519.0055513.1054912.96
2003–2007333148914.7048214.4937311.2136611.00
1992–1999532779414.9178714.7758210.9357510.79
2000–2007539390416.7889316.5767212.4766112.27

There are three noteworthy patterns across the panels. First, total turnover levels are substantially higher than those typically reported. Overall, total turnover under definition 1 is 15.8% over the entire sample period implying an average CEO tenure of 6.3 years. This is substantially higher than that reported in Jensen et al. (2004) and Murphy and Zabojnik (2008) who study a different sample of large firms (from the Forbes lists) over three decades from 1970 to 2000. They report turnover of 10.2% in the 1970s, 10.0% in the 1980s, and 11.3% in the 1990s. All of these measures, however, are for standard or internal turnover. For our sample period of 1992 to 2007, we obtain a standard turnover of 11.7%, similar to their results for the 1990s. At 11.7%, the estimated average CEO 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 the levels of both total and internal turnover. For example, using definition 1 total turnover is as low as 6.94% in 2003 (and only 9.35% in 1994), and as high as 26.32% in 2000 (and 21.37% in 1999).

Third, turnover increased significantly in the latter part of the sample. In the earlier period from 1992 to 1996, total CEO turnover using definition 1 is 12.62% per year implying an average tenure of 7.9 years. In the period from 1997 to 2002, total turnover increases to 19.15% per year, implying an average tenure of just 5.2 years. This period roughly coincides with the large increase in CEO pay described in Bebchuk and Grinstein (2005). The period in which CEO pay increased substantially coincides with a period in which CEO tenure decreases substantially. It is worth adding that the increased level of turnover began well before 2002 (when the Enron and Worldcom scandals became apparent and Sarbanes-Oxley was passed). In the more recent period from 2003 to 2007, total CEO turnover declines from the middle period to 14.70%, implying an average CEO turnover of 6.8 years. Internal or board- driven turnover follows a similar trend.

As stated previously, we continue to follow a firm once it is included in the Fortune 500, even if it drops from the Fortune 500. This may make our results harder 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 turnover separately for firms in the Fortune 500 and those not in the Fortune 500 in the particular year. The increase in turnover from the earlier part of the sample to the latter part of the sample is consistent across both Fortune 500 and non-Fortune 500 companies. While it appears that annual turnover is larger for non-Fortune 500 firms, annual tests of the differences in average turnover shows that the turnover across groups is not different in all but 3 years.

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

Table 2. Turnover across periods
VariableTotal turnoverStandard (internal) turnover
(1)(2)(1)(2)
ΔProb (p-value)ΔProb (p-value)ΔProb (p-value)ΔProb (p-value)
  1. Probit regression estimates of the likelihood of CEO turnover during the period from 1992 to 2007 to test whether the probability of turnover is higher in the 2000 to 2007 period. Total turnover is all CEO turnovers including turnover due to mergers and acquisitions and delistings from a major stock exchange. Standard (internal) turnover excludes turnover due to mergers and acquisitions 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 is selected. (2) Defines occurrences where the CEO dies in office as a non-turnover event. The dependent variable equals one if the CEO turnovers and zero otherwise. ΔProb represents the change in the probability associated with moving the indicator from 0 to 1. Robust standard errors to control for heteroskedasticity are reported in parentheses.

  2. ** and ***indicate significance at the 5% and 1% levels, respectively.

2000 or later indicator variable0.01873*** (0.0071)0.01800*** (0.0070)0.0165** (0.0065)0.0158** (0.0064)
n10,71510,71510,27210,272
Pseudo R20.00080.00070.00230.0008

Table 3 presents the turnover data in a different way. We compare the fraction of the CEOs in 1992 who are no longer CEOs in 1999 to the fraction of CEOs who are CEOs in 2000 and no longer CEOs in 2007. The Table shows that 78.70% of CEOs in 1992 were no longer CEOs in 1999, while over 83% of CEOs in 2000 were no longer CEOs by 2007. Again, this result suggests that the job of CEO has become increasingly precarious over the sample period.

Table 3. CEO turnover
  1. Number and percent of firms experiencing no turnover over a 5-year period. Turnover is measured using total turnover. Total turnover is all CEO turnovers including turnover due to mergers and acquisitions and delistings from a major stock exchange. Occurrences where the CEO dies in office is defined as a non-turnover event.

Year1992
Number of firms in the sample in 1992460
Number of firms experiencing no turnover between 1992 and 199998
Percent of firms experiencing no turnover between 1992 and 199921.30%
Year2000
Number of firms in the sample in 2000703
Number of firms experiencing no turnover between 2000 and 2007115
Percent of firms experiencing no turnover between 2000 and 200716.35%

The turnover also is substantially greater than that measured by Mikkelson and Partch (1997) over two earlier 5-year periods. In the active takeover period from 1983 to 1988, they find that 39% of firms experience CEO turnover (and 23% of firms experience complete management turnover); in the less active period from 1989 to 1993, 34% of firms experience CEO turnover (while 16% of firms experience complete management turnover). Unfortunately, these results are not directly comparable because the sample in Mikkelsen and Partch consists of smaller firms.

IV. The Relation between Turnover and Performance

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix

A. Internal turnover

We estimate pooled annual probit regressions to examine the likelihood of internal CEO turnover. In all of the probit regressions, the dependent variable is equal to one if a CEO turns over and zero otherwise. Turnover is measured using definition 2 (i.e., deaths are coded as non-turnover events). The Tables report the marginal changes in the probability of internal CEO turnover implied by the probit coefficient estimates that result from a unit change in the explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. These marginal sensitivities, labeled ‘ΔProb,’ are economically equivalent to coefficient estimates from ordinary least squares estimation. In the discussion below, we focus on the significance of these marginal effects. Robust standard errors to control for heteroskedasticity are reported in parentheses.

In all of the probit regressions, three proxies are used to measure stock market performance. First, we measure market performance using the annual return on the S&P 500 index. Second, relative industry performance is measured at the two-digit SIC code level and equals the difference between the return on the median firm in the industry and the return on the S&P 500 index. Third, relative firm performance is measured as the industry-adjusted firm stock return, which is equal to the firm stock return minus the return for the median firm in the same two-digit SIC code. Returns are measured over the calendar year period. Lagged returns are measured over the previous calendar year. For example, for companies in Fortune's April 2006 issue, we measure stock returns for calendar year 2005 and lagged stock returns for calendar year 2004.

Table 4 reports the results for two sets of probit regressions. In panel A, we include measures of stock market performance and an indicator variable equal to one if lagged CEO age is greater than or equal to 60 (CEO age dummy). In the regressions in panel B, we add the change in ROA as a measure of operating performance where ROA equals the sum of income before extraordinary income plus interest and related expenses divided by total assets. We use change in ROA because 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 longer periods, 1992 to 1999 and 2000 to 2007.

Table 4. Probit regressions of the probability of internal CEO turnover on performance
Variable1992–2007 ΔProb (s.e)1992–1996 ΔProb (s.e)1997–2002 ΔProb (s.e)2003–2007 ΔProb (s.e.)1992–1999 ΔProb (s.e.)2000–2007 ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of internal CEO turnover during the period from 1992 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 as non-turnover events. The dependent variable equals one if the CEO turnovers and zero otherwise. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise. Robust standard errors to control for heteroskedasticity are reported in parentheses.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

Panel A      
Return on S&P 500−0.1188*** (0.0203)0.1121*** (0.0406)−0.1477*** (0.0311)−0.3533*** (0.0722)0.0891** (0.0366)−0.1823*** (0.0286)
Industry return – return on S&P 500−0.0772*** (0.0163)0.0700* (0.0396)−0.0963*** (0.0252)−0.0463 (0.0385)0.0071 (0.0244)−0.1426*** (0.0242)
Industry-adjusted stock return−0.0641*** (0.0106)−0.0204 (0.0200)−0.0818*** (0.0157)−0.0818*** (0.0198)−0.0387*** (0.0147)−0.0917*** (0.0148)
CEO age dummy0.1369*** (0.0081)0.1255*** (0.0132)0.1679*** (0.0139)0.1112*** (0.0143)0.1332*** (0.0106)0.1415*** (0.0121)
Number of obs998229473903313249375045
Pseudo R20.05880.06000.07120.05910.05830.0712
Panel B      
Return on S&P 500−0.1396*** (0.0215)0.0454 (0.0499)−0.1537*** (0.0325)−0.3713*** (0.0745)0.0160 (0.0429)−0.1891*** (0.0299)
Industry return – return on S&P 500−0.0820*** (0.0173)0.0495 (0.0455)−0.0948*** (0.0264)−0.0426 (0.0405)0.0082 (0.0267)−0.1473*** (0.0253)
Industry-adjusted stock return−0.0586*** (0.0113)−0.0157 (0.0230)−0.0709*** (0.0166)−0.0828*** (0.0207)−0.0303* (0.0162)−0.0881*** (0.0157)
Change in ROA−0.1594** (0.0710)−0.0974 (0.1399)−0.3413*** (0.1034)0.0287 (0.1335)−0.2651** (0.1044)−0.0413 (0.0954)
CEO age dummy0.1422*** (0.0081)0.1326*** (0.0144)0.1732*** (0.0142)0.1137*** (0.0147)0.1392*** (0.0113)0.1452*** (0.0124)
n937526243714303745234852
Pseudo R20.06120.05370.07700.05900.05870.0714

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

Table 4 documents turnover increases with poor industry-adjusted stock performance over the entire sample period, and particularly, in the latter subperiods. For example, for the 2000 to 2007 subperiod, a one-standard deviation (35.4%) decline in a firm's industry adjusted stock return is associated with an increase of 3.2% in the mean likelihood of CEO turnover. From a base turnover level of 12.55%, this change implies a likelihood of 9.35% for a CEO whose firm performs one-standard deviation above the industry versus 15.75% for a CEO whose firm performs one-standard deviation below the industry. These are economically meaningful differences with 9.35% implying a tenure of 10.7 years and 15.75% implying a tenure of 6.3 years.

The reported marginal probabilities appear to be greater than the sensitivities reported 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 that turnover is not statistically related to industry-adjusted performance during the 1992 to 1996 period.

CEO turnover also is related to poor industry performance. Again, the negative relation for the full sample period appears to be driven by the subperiod 1997 to 2002 if we split the time period in three or by the 2000 to 2007 subperiod if we divide the period in half. For the 2000 to 2007 subperiod, a one-standard deviation (18.7%) decline in industry performance is associated with a 2.7% increase in the mean likelihood of CEO turnover.

Lower overall market performance, as measured by the return on the S&P 500 index, is also associated with a higher likelihood of internal CEO turnover for the full sample and the later subperiods. For the 2000 to 2007 subperiod a one-standard deviation (15%) decline in the S&P 500 index corresponds to an increase of 2.7% in mean probability of CEO turnover.

Surprisingly, the relations between CEO turnover and overall market performance and industry performance, respectively are positive and significant in the 1992 to 1996 subperiod. One possible interpretation for these positive marginal probabilities is that CEOs left office after good performance. This would be the case if they were voluntary turnovers. While forced turnovers increased from the 1992 to 1996 subperiod to the 1997 to 2002 subperiod (from about 12% to about 17%), this explanation is implausible because the great majority of turnovers in both subperiods are classified as voluntary.

Overall, the results in panel A indicate that the relation between internal CEO 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 in industry-adjusted, industry, and market performance – are economically meaningful relative to average internal turnover of 12.55%.1

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

In Table 5, we add 1 year of lagged stock performance variables to the regressions. Overall, the results in Table 5 are consistent with those in Table 4. For the entire period, internal CEO turnover is significantly negatively related to industry-adjusted, industry, and overall market stock performance in the current 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 the S&P 500 (although the positive coefficient on the lagged return is smaller in magnitude than the negative coefficient on the contemporaneous return).2 As in the previous results, the regressions in Table 5 indicate that the turnover-performance relations are driven by the later subperiod.

Table 5. Probit regressions of the probability of internal CEO turnover on performance and lagged performance
Variable1992–2007 ΔProb (s.e.)1992–1996 ΔProb (s.e.)1997–2002 ΔProb (s.e.)2003–2007 ΔProb (s.e.)1992–1999 ΔProb (s.e.)2000–2007 ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of internal CEO turnover during the period from 1992 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 as non-turnover events. The dependent variable equals one if the CEO turnovers and zero otherwise. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise. Robust standard errors to control for heteroskedasticity are reported in parentheses.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

Return on S&P 500−0.2038*** (0.0278)0.0229 (0.0664)−0.3204*** (0.0495)−0.3202*** (0.1144)−0.0371 (0.0501)−0.2484*** (0.0359)
Lagged return on S&P 5000.1103*** (0.0235)0.0437 (0.0449)0.2609*** (0.0530)0.1221 (0.0750)0.1137*** (0.0363)0.1584*** (0.0332)
Industry-return-return on S&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)
Lagged industry return-return on S&P 500−0.0630*** (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)
Lagged industry-adjusted stock 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)
CEO age dummy0.1366*** (0.0081)0.1252*** (0.0133)0.1663*** (0.0139)0.1125*** (0.0144)0.1331*** (0.0107)0.1395*** (0.0121)
n985028863861310348594991
Pseudo R20.06860.06920.08660.06520.06970.0838

To examine the change in the predicted probability associated with a change in the current and lagged industry-adjusted performance, we first estimate the predicted probability of turnover evaluated at the mean values of all the regressors. Next, we re-estimate the predicted probability of turnover for current and lagged industry-adjusted stock returns evaluated at one-standard deviation below their means and for all other regressors evaluated at their means. For the full sample period, the result is an increase in the predicted probability of turnover of 5.18%. Repeating the same procedure, the 1992 to 1999 and 2000 to 2007 subperiods result in increases in the probability of turnover of 0.1% and 6.6%, respectively. Again, the sensitivity to industry-adjusted performance for the 2000 to 2007 period appears to be greater than any of the sensitivities reported in Murphy (1999). Similar exercises for market performance show that similar to the results for industry-adjusted returns, the relations between market performance and turnover are greater in the 2000 to 2007 subperiod than in the 1992 to 1999 subperiod.

In contrast, when we evaluate the change in current and lagged industry performance, we find that one-standard deviation declines in industry performance are associated with larger increases in the 1992 to 1999 subperiod than in the 2000 to 2007 subperiod. Specifically, a one-standard deviation decline in current and lagged industry performance during the 1992 to 1999 subperiod is associated with a 5.00% increase in turnover. In contrast, a one-standard deviation decline in current and lagged industry performance during the 2000 to 2007 subperiod is associated with a 3.8% increase in turnover. Finally, one-standard deviation declines in current and lagged market returns are associated with a 2.5% increase in turnover for using the full-sample period.

Overall, the results in Tables 4 and 5 suggest that since 2000, boards have been more sensitive to poor stock performance. It is also worth noting that the economic 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-standard deviation better than the industry has a 5.2% lower likelihood of turnover while a CEO whose firm performs one-standard deviation worse than the industry has a cumulative 5.2% increase in the likelihood of turnover. From a base turnover level of 12.5%, these imply likelihoods of 7.4% for the strong performer versus 17.8% for the poor performer. These are economically meaningful differences with 7.4% implying a tenure of 13.5 years and 17.8% implying a tenure of 5.6 years.

Table 6 repeats the Table 5 probit regressions for the full-sample period and the 1992 to 1999 and 2000 to 2007 subperiods for two sets of firms: firms in the S&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 the ExecuComp dataset that includes only firms in various S&P indices. It is possible there is a selection bias in these firms. Second, investors may be more likely to pay attention to firms in the S&P 500 index, and, if so, these firms would be more likely to be monitored by the press and institutional investors. Thus, the effect of stock market performance might be different for these firms (Bertrand and Mullainathan, 2001).

Table 6. Probit regressions of the probability of internal CEO turnover on performance
VariableFirms in the S&P 500Firms not in the S&P 500
19922007 ΔProb (s.e)1992–1999 ΔProb (s.e)2000–2007 ΔProb (s.e)19922007 ΔProb (s.e.)1992–1999 ΔProb (s.e.)2000–2007 ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of internal CEO turnover during the period from 1992 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 as non-turnover events. The dependent variable equals one if the CEO turnovers and zero otherwise. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise. Robust standard errors to control for heteroskedasticity are reported in parentheses.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

Return on S&P 500−0.2328*** (0.0364)−0.0486 (0.0644)−0.3055*** (0.0470)−0.1602** (0.0429)−0.0067 (0.0791)−0.1732 (0.0549)
Lagged return on S&P 5000.1347*** (0.0303)0.0849* (0.0478)0.2338*** (0.0425)0.0656* (0.0372)0.1298** (0.0548)0.0555 (0.0533)
Industry return – return on S&P 500−0.0625*** (0.0221)0.0298 (0.0314)−0.1359*** (0.0309)−0.0314 (0.0276)0.0424 (0.0385)−0.0838** (0.0404)
Lagged industry return – return on S&P 500−0.0579*** (0.0209)−0.0881*** (0.0338)−0.0534** (0.0273)−0.0672*** (0.0244)−0.1034*** (0.0392)−0.0430 (0.0319)
Industry-adjusted stock return−0.0484*** (0.0149)0.0028 (0.0195)−0.1170*** (0.0212)−0.0812** (0.0146)−0.0797** (0.0212)−0.0824*** (0.0198)
Lagged industry-adjusted stock return−0.0559*** (0.0147)−0.0430** (0.0215)−0.0714*** (0.0198)−0.0696*** (0.0141)−0.0906*** (0.0224)−0.0506*** (0.0179)
CEO age dummy0.1625*** (0.0108)0.1560*** (0.0141)0.1690*** (0.0163)0.0967*** (0.0120)0.0974*** (0.0160)0.0951*** (0.0178)
n601529543061383519051930
Pseudo R20.08090.07600.11040.05740.07720.0532

The coefficient patterns in Table 6 are qualitatively similar for the two sets of firms except for the 2000 to 2007 subperiod for which turnover for firms not in the S&P 500 index is not statistically related to market performance. Turnover in both sets of firms is significantly related to industry-adjusted and market stock performance over the entire sample period. As in the overall sample, the relations for firm in the S&P 500 index are stronger in the more recent 2000 to 2007 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 this section, we consider four possible sources of those increases – corporate governance (or shareholder rights), shareholder blockholdings, board independence, and the SOX legislation.

Recent work has suggested that differences in corporate governance and shareholder rights may have real effects. Gompers et al. (2003) find that differences in corporate governance and shareholder rights are related to stock returns. Masulis et al. (2007) find that those differences in corporate governance are related to acquisition behavior. In both papers, greater shareholder rights are associated with higher stock returns.

In this section, we examine the relation between turnover, stock performance, and governance. To do so, we use the Gompers Ishi Metrick (GIM) index developed by Gompers et al. (2003). They categorize 24 charter provisions, bylaw provisions, and other firm-level rules associated with corporate governance into five types: (1) tactics for delaying hostile bidders; (2) voting rights; (3) director/officer protection; (4) other takeover defenses; and (5) state laws. Their overall GIM index and the five component indices generally score one point for each provision that restricts shareholder rights or increases managerial power. Thus, a higher GIM index score represents greater managerial power (weaker shareholder rights).

We estimate turnover regressions that interact stock performance with the measure of governance. We used both a continuous measure of the GIM index as well as an indicator variable equal to one if the firm's GIM index is in the highest quintile that year and zero otherwise. To the extent that the GIM index measures poor governance, the GIM index should have a negative effect on the level of turnover (i.e., poorly governed firms should have less turnover), while the interaction of the GIM index and stock performance should have a positive effect on turnover (i.e., turnover at poorly governed firms should be less sensitive to poor performance).

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

Table 7. Probit regressions of internal CEO turnover for Fortune 500 firms on performance and governance
VariableFull sample1992–19992000–2007
ΔProb (s.e.)ΔProb (s.e.)ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firms during the period 1992 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 as non-turnover events. The dependent variable equals one if the CEO turnovers and zero otherwise. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. Models are estimated with robust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standard errors to control for heteroskedasticity are reported in parentheses.

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 stock return−0.0683*** (0.0119)−0.0401** (0.0161)−0.0934*** (0.0171)
Lagged return on S&P 5000.1314*** (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-adjusted stock return−0.0529*** (0.0117)−0.0632*** (0.0179)−0.0457*** (0.0151)
G index0.0029* (0.0016)0.0018 (0.0021)0.0041 (0.003)
CEO age dummy0.1358*** (0.0093)0.1254*** (0.0120)0.1455*** (0.0142)
High G index dummy0.0906 (0.1579)0.1233 (0.2793)0.0189 (0.1643)
High G index dummy   
x(Return on S&P 500)0.1057 (0.0771)0.0207 (0.1412)0.1273 (0.0985)
x(Industry return – return on S&P 500)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 on S&P 500)−0.0915 (0.0636)−0.0652 (0.0959)−0.1158 (0.0906)
x(Lagged industry return – return on S&P 500)0.0153 (0.0468)0.0063 (0.0699)0.0087 (0.0624)
x(Lagged industry-adjusted stock return)−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)
n921345574656
Pseudo R20.07220.07280.0885

For the sample overall and for each subperiod, most of the interaction terms are not statistically different from zero. For two coefficients, the GIM index interaction is significantly negative. Turnover is significantly more sensitive to poor lagged industry-adjusted performance for the high GIM index firms during the overall sample period and in the 2000 to 2007 subperiod. These are the opposite 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 CEO turnover. The most one can say is that the GIM measure is possibly associated with a somewhat faster response to poor industry-adjusted performance. Our results are consistent with Bhagat and Bolton (2006) who also fail to find a significant effect on turnover when they interact governance and performance.

Next, we examine the relation between CEO turnover and blockholder ownership. Institutional and blockholder ownership increase over the sample period. We focus on blockholder ownership (where an institution owns at least 5% of a firm's outstanding shares) because blockholders have both the incentive and 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 in each firm by the firm's largest institutional blockholder where blockholders are shareholders with greater than 5% ownership of the firm's outstanding shares. Data on institutional ownership are from Thomson-Reuters Institutional Holdings (13F) database.

Table 8 reports the mean and median block ownership and the presence of a blockholder for our sample firms. The average holding of blockholders and fraction 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 2007 subperiod, blockholder ownership averages 14.98%, while almost 75% of the firms have a blockholder present. The differences across subperiods are statistically significant.

Table 8. Block ownership and board of directors
YearBlock ownershipBlockholder presentPercent (%) of independent directorsNot an independent board
MeanMedian% of firmsMeanMedian% of firms
  1. 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 with greater than a 5% ownership stake. Independent director data are taken from IRRC Directors database on WRDS and proxy filings.

19928.125.0250.12   
19938.125.5152.88   
19948.565.2552.82   
19959.055.8054.71   
19969.426.0856.9761.563.6428.22
199710.086.3757.7563.3866.6725.42
199811.017.5575.1263.9466.6722.42
199911.488.3562.7465.0966.6721.57
200012.539.5463.4166.1570.0020.73
200112.5510.5466.2967.3970.0018.99
200213.7211.5372.9369.6372.7315.18
200313.3211.3373.0071.4575.0012.06
200416.4214.3680.1272.8275.009.42
200516.5414.8379.2373.9077.787.85
200617.1014.7481.0074.3277.788.14
200717.9215.3878.6880.4581.821.42

Table 9 reports the results of probit turnover regressions that include the percentage of total blockholder ownership (at the end of the previous year) and the interactions of the continuous measure and our three stock performance measures. As Table 9 shows, the marginal probability associated with the percentage of total blockholder ownership while positive is not statistically different from zero for the full sample or for either subperiod. The only interaction term that is statistically significant is the interaction of block ownership and industry-adjusted stock returns in the 2000 to 2007 subperiod. One interpretation of this result is that blockholders are particularly active in firms that underperform their industries but only so in the latter time period.

Table 9. Probit regressions of internal CEO turnover for Fortune 500 firms on performance and block ownership
VariableFull sample1992–19992000–2007
ΔProb (s.e.)ΔProb (s.e.)ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firms during the period 1992 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 as non-turnover events. The dependent variable equals one if the CEO turnovers and zero otherwise. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. Models are estimated with robust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standard errors to control for heteroskedasticity are reported in parentheses.

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)
Industry-adjusted stock return−0.0450*** (0.0147)−0.0522*** (0.0195)−0.0865*** (0.0231)
Lagged return on S&P 5000.1018*** (0.0329)0.1436*** (0.0486)0.1358*** (0.0472)
Lagged industry return – return on S&P 500−0.0851*** (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 block ownership0.0116 (0.0300)0.0847 (0.0903)−0.0303 (0.0355)
CEO age dummy0.1370*** (0.0081)0.1339*** (0.0107)0.1395*** (0.0121)
Percent of total block ownership   
x(Return on S&P 500)0.0998 (0.2212)−0.0270 (0.4471)0.1133 (0.2844)
x(Industry return – return on S&P 500)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)
x(Lagged return on S&P 500)0.0998 (0.2212)−0.2880 (0.3154)0.1365 (0.2590)
x(Lagged industry return – return on S&P 500)0.1616 (0.1211)0.1224 (0.2328)0.0753 (0.1515)
x(Lagged industry-adjusted stock return)−0.1255 (0.0771)−0.0055 (0.1377)−0.2016** (0.0958)
n984848574991
Pseudo R20.06980.07110.0859

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

Third, we consider the role of independent directors. Weisbach (1988) finds that turnover-performance sensitivities are greater for firms with more independent boards. And over time, boards in the United States have become more independent. It is possible that this increased independence has played a role in the changes in turnover. We obtain data on director independence from the IRRC Directors database on WRDS. We use the variable director type to classify directors as independent. The IRRC data are available only from 1996 onward.

Table 8 reports the percentage of independent directors and the fraction of firms that do not have a majority independent board. Similar to the rise in blockholders, the percentage of independent directors on a firm's board increases over time. The percentage increases from 63% to almost 72% from the 1992 to 1999 subperiod to the 2000 to 2007 subperiod. Moreover, the fraction of firms without a majority independent board decreases from 24% during the 1992 to 1999 subperiod to 12% in the 2000 to 2007 subperiod.

Table 10 reports the results of probit turnover regressions that account for board independence. We include a dummy variable that equals one if the firm does not have a majority independent board. (The baseline, therefore, is a firm that does have a majority independent board.) The results are qualitatively similar but have a less natural interpretation when we use the percentage of independent directors. We also interact this indicator variable with the stock performance variables. We estimate the probit for the 1996 to 2007 period, the period over which we have director data.

Table 10. Probit regressions of internal CEO turnover for Fortune 500 firms on performance and independent directors
Variable1996–20071996–2007
ΔProb (s.e.)ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firms during the 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 as non-turnover events. The dependent variable equals one if the CEO turnovers and zero otherwise. Independent director data are taken from IRRC Directors database on WRDS. No independent board equals one if a majority of directors are not independent. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. Models are estimated with robust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standard errors are in parentheses.

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 5000.1356*** (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 dummy0.1418*** (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)
n73747374
Pseudo R20.07770.0807

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

In Table 11, we examine whether the probability of CEO turnover is greater in the years under the SOX legislation. We create an indicator variable equal to one for years 2003 and after and zero otherwise. In the first regression, we include this indicator variable, current year measures of performance, and interactions between the indicator variable and performance variables. In the second regression, we add the lagged performance measures and corresponding interaction terms. In the last column of the table, we repeat the second regression for the 2000 to 2007 period.

Table 11. Probit regressions of internal CEO turnover for Fortune 500 firms on performance and SOX legislation
VariableFull sampleFull sample2000 - 2007
ΔProb (s.e.)ΔProb (s.e.)ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of internal CEO turnover for Fortune 500 firms during the 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 CEO position, 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 equals one if the CEO turnovers and zero otherwise. The SOX dummy equals one for years after 2002. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. Models are estimated with robust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. p-values are in parentheses.

Return on S&P 500−0.0917*** (0.234)−0.1899*** (0.0335)−0.4782*** (0.1944)
Lagged return on S&P 5000.0801*** (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)
Lagged industry return – return on S&P 500−0.0747*** (0.0186)−0.0245 (0.0286)
Industry-adjusted stock return−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 dummy0.1354*** (0.0081)0.1362*** (0.0081)0.1395*** (0.0121)
SOX dummy0.0295*** (0.0109)0.0174 (0.0181)0.0448 (0.0327)
SOX dummy   
x(Return on S&P 500)−0.2834*** (0.0820)−0.1520 (0.1275)0.1308 (0.2315)
x(Lagged return on S&P 500) 0.0515 (0.0862)−0.1496 (0.1113)
x(Industry return – return on S&P 500)0.0080 (0.0461)0.0126 (0.0471)0.1033* (0.0539)
x(Lagged industry return – return on S&P 500) 0.0069 (0.0440)−0.0442 (0.0495)
x(Industry-adjusted stock return)−0.0291 (0.0249)−0.0266 (0.0248)0.0224 (0.0298)
x(Lagged industry-adjusted stock return) −0.0018 (0.0240)−0.0092 (0.0273)
n998298504991
Pseudo R20.06060.06980.0866

In the first regression, CEO turnover is significantly higher during years under the SOX legislation. The marginal positive marginal probability implies about a 3% increase in turnover during these years. Additionally, the interaction term between the SOX indicator variable and market performance is significantly negative implying a decrease in market performance is associated with higher turnover during the years under the SOX legislation. When we add the lagged variables, the marginal probability associated with the SOX indicator variable is positive but no longer statistically significant in the full sample or the 2000 to 2007 subperiod. For the 2000 to 2007 subperiod, the interaction term between the indicator variable and industry performance is positive and significant, but the sum of the industry performance and the interaction term is not statistically different from zero. We conclude that these results are inconclusive.

Taken together, the governance variables are of mixed help in explaining the increase in turnover and turnover-performance sensitivity in the latter half of our sample. While the GIM measure and the SOX legislation do not reliably affect turnover, the presence of large blockholders and independent boards are associated with greater turnover-performance sensitivity. The findings in Jenter and Lewellen (2010) for a different sample are consistent with this latter result.

C. Forced turnover

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

We follow Huson et al. (2001) in classifying turnover as forced. (Jenter and Kanaan forthcoming) also use this classification scheme.) If an article in the business press indicates that the CEO was fired, forced, or left following a policy disagreement or some other equivalent, then turnover is defined as forced. For the remaining announcements, succession is classified as forced when the CEO is under 60, and the first article reporting the announcement does not report the reason for the departure as involving death, poor health, or the acceptance of another position elsewhere.

We present the regression results in Table 12. The dependent variable categories in the MNL estimations are unforced turnover, forced turnover, or no turnover. The Table reports the marginal effect of each regressor. The probability of forced turnover is significantly negatively related to the three components of firm stock performance – firm performance relative to the industry, the industry relative to the market, and the overall market – in the current year and to industry-adjusted return in the previous year. The probability of unforced turnover is significantly negatively related to firm stock performance relative to the industry and the overall market in the current year, and to industry stock return and industry-adjusted stock return in the previous year. As in the previous results, both types of turnover are positively related to lagged market performance.

Table 12. Multinominal logit regression estimates of the likelihood of no CEO turnover, unforced CEO turnover, and forced CEO turnover for Fortune 500 firms
VariableUnforced CEO turnoverForced CEO turnover
ΔProb (s.e.)ΔProb (s.e.)
  1. Multinominal logit regression estimates of the likelihood of no CEO turnover, non-forced CEO turnover, and forced CEO turnover for Fortune 500 firms during the period 1992 to 2007. ΔProb measures the change in the probability of the particular choice per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. Models are estimated with robust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standard errors are in parentheses.

Return on S&P 500−0.1402*** (0.0237)−0.0341*** (0.0084)
Lagged return on S&P 5000.0761*** (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 dummy0.1517*** (0.0079)−0.0087*** (0.0018)
Pseudo R2 = 0.0878  

Overall, then, both forced turnover and unforced turnover are sensitive to all three types of poor stock performance. This strongly suggests that a number of unforced turnovers are not voluntary. In work that was partially motivated by our results, Jenter and Lewellen (2010) find strong evidence consistent with this conclusion.

D. External turnover

As discussed earlier, in addition to internal turnover, we examine external turnover. Recall, nonstandard or external turnover is turnover due to a merger or bankruptcy/delisting. We consider the CEO to have been turned over in a merger if his or her company is taken over by another company, and he or she is 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 bankrupt company. The incidence of external turnover is 4.7% per year over the sample period.

Table 13 reports probit regressions of the probability of external turnover as a function of current stock market performance (columns 1 through 3) and current and lagged performance (columns 4 through 6). Unlike the results for internal turnover, in most specifications, external turnover is unrelated to performance. When performance is significant, the marginal probabilities are economically small and in most cases positive. For example, when we include only current performance in the regressions, external turnover is positively related to market performance in the earlier subperiod and to industry-adjusted stock performance in the full sample period and later subperiod. One interpretation of the lack of results for external turnover and performance is that, on average, mergers and acquisitions during this period were not disciplinary in nature.

Table 13. Probit regressions of the probability of external CEO turnover on performance
Variable1992–2007 ΔProb (s.e.)1992–1999 ΔProb (s.e.)2000–2007 ΔProb (s.e.)1992–2007 ΔProb (s.e.)1992–1999 ΔProb (s.e.)2000–2007 ΔProb (s.e.)
  1. Probit regression estimates of the likelihood of external CEO turnover for Fortune 500 firms during the period 1992 to 2005. The dependent variable equals one if the CEO turnovers and zero otherwise. ΔProb measures the change in the probability of CEO turnover per unit change in the relevant explanatory variables. For indicator variables, the coefficient represents the change in the probability associated with moving the indicator from 0 to 1. Models are estimated with robust standard errors to control for heteroskedasticity. CEO age dummy equals 1 if lagged CEO age is greater than or equal to 60 and zero otherwise.

  2. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standard errors to control for heteroskedasticity are reported in parentheses.

Return on S&P 500−0.0007 (0.0048)0.0177** (0.0073)−0.0026 (0.0058)−0.0095 (0.0059)0.0160 (0.0099)−0.0105 (0.0067)
Lagged return on S&P 5000.0125** (0.0052)0.0162** (0.0075)0.0168*** (0.0062)
Industry return – return on S&P 500−0.0010 (0.0045)0.0027 (0.0066)−0.0072 (0.0058)0.0015 (0.0044)0.0051 (0.0052)−0.0052 (0.0053)
Lagged industry return – return on S&P 500−0.0068 (0.0029)−0.0034 (0.0042)−0.0094** (0.0038)
Industry-adjusted stock return0.0047** (0.022)0.0037 (0.0030)0.0044* (0.0026)0.0042 (0.0021)0.0034 (0.0027)0.0033 (0.0024)
Lagged industry-adjusted stock return−0.0034 (0.0025)−0.0023 (0.0033)−0.0041 (0.0027)
CEO age dummy0.0027 (0.0081)−0.000 (0.0020)0.066 (0.0032)0.0025 (0.0018)0.0001 (0.0018)0.0057** (0.0029)
n884244004442871943254391
Pseudo R20.01350.02680.02740.02500.04580.0449

V. Summary and Implications

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix

In this paper, we examine the extent and determinants of internal and external 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 increases to 16.78%, implying an average tenure of just under than 6 years. Internal or board-driven turnover also rises substantially in the latter part of the sample.

We then look at how turnover varies with firm stock performance. Previous work suggests a modest relation between internal (board initiated) turnover and firm stock performance. We find a stronger and significant relation between internal turnover and three different components of firm stock performance – performance relative to the industry, performance of the industry relative to the stock market, and the performance of the overall stock market (Jenter and Kanaan forthcoming) obtain similar results for forced turnover.) The sensitivities are economically meaningful. Both types of internal turnover – forced and unforced are sensitive to all three types of poor stock performance.

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

We next consider four possible explanations for or factors that drive the changes in turnover and turnover-performance sensitivity. There is some evidence that the increases in turnover and turnover-performance sensitivity are related to increases in blockholdings and board independence. Turnover and turnover-performance sensitivity are not reliably related to the Gompers et al. (2003) governance index.

External turnover – turnover primarily related to acquisitions – also is significantly related to all three measures of performance. As with internal turnover, the results are stronger in the later subperiod.

Our results have several implications. First, they suggest that the CEO job is more precarious than in the past. When external takeovers are included, the average tenure of a CEO has declined to less than 6 years for the recent 1998 to 2005 period. The recent tenures are substantially shorter than those reported in the previous work for the 1970s, 1980s, and 1990s. The shorter tenures appear to have continued out of sample.

For individual CEOs, the shorter expected tenure likely offsets some of the benefits of the increase in CEO pay since the mid-1990s. For example, the annual pay of S&P 500 CEOs roughly doubled in real terms from the 1992–1997 period to the 1998–2005 period. Our estimates suggest that the total pay of an individual CEO over his or her entire expected term increased by less than this because the expected tenure at the higher pay declined by one quarter to one third.

This calculation would be inaccurate if severance agreements around internal turnover 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 severance agreements, respectively, around internal and external takeover events. In fact, the average and median magnitudes they report are modest. (They do not study whether these payments have changed over time.) Severance agreements, therefore, are unlikely to alter the conclusion that the job of CEO has become riskier and that the shorter expected tenures of CEOs partially offset increases in CEO pay.

The coincidence of shorter CEO tenures and higher CEO pay is consistent with Hermalin (2005) who presents a model that predicts increased board vigilance will be associated with both outcomes. The relations with more independent boards and blockholders also are consistent with the increased vigilance in Hermalin's model.

Second, the similar results for the turnover-performance sensitivities of forced and unforced turnover suggest that many unforced turnovers are not voluntary. This greater incidence of involuntary turnover, in turn, implies both that boards do a better job of disciplining poor management and that the CEO's job is more precarious than is commonly believed.

Third, our results suggest an evolving role for boards. In a sample from the 1980s, Morck et al. (1989) find that internal turnover is related to industry-adjusted performance while external turnover from hostile takeovers is related to industry performance. They interpret this as indicating boards respond well to poor performance relative to the industry but do not respond well to poor industry 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 relative to the industry but also to poor industry performance and to poor market performance. To the extent that internal turnover has increased, boards also appear to monitor more frequently. One interpretation of these results is that boards – possibly encouraged by large shareholders – perform both the role they performed in the 1980s and the role that hostile takeovers played then. The increased turnover associated with blockholdings and board independence is consistent with this interpretation.

The result that boards do not index CEO turnover to the industry or the market 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 lack of indexing of pay as a failure of governance. Our results on turnover in conjunction with those in Morck et al. (1989) for the earlier period provide an alternative explanation. When an industry or the overall economy performs poorly, it is sometimes efficient for the board to bring on a new CEO to respond to the new industry or market conditions. The recent (and out of sample) high turnover in the financial services industry is consistent with this.

Fourth, the shorter expected CEO tenures and sensitivity of those tenures to stock performance have implications for the measurement of CEO pay. The shorter expected tenures suggest that the estimates of CEO pay based Standard and Poor's ExecuComp data may be overstated. While option grants typically have a 10-year life, ExecuComp uses a 7-year life because ‘executives rarely wait until the expiration date to exercise their options.’ This adjustment assumes that CEOs will remain with the company for at least 7 years. If a CEO has an expected initial tenure of 6 years, the ExecuComp assumption will tend to overstate the value of option grants every year of the CEO's tenure with the overstatement increasing each year. This assumes that CEOs forfeit unvested options and/or must exercise vested options when they leave the company. For internal turnover, Yermack (2006) finds that this tends to be the default policy for most companies, and companies deviate from those policies in only 16% of the internal turnovers he studies. ExecuComp also values restricted stock grants as fully vested when, in fact, they usually vest over a period of time. This also will tend to overstate executive compensation.

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

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

  1. 1

    The predicted likelihood calculated at the mean of the independent variables is 10.97% for the 200 to 2007 subperiod.

  2. 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.

  3. 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.

References

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix
  • Bebchuk, L. , and J. Fried (2002), ‘Managerial Power and Rent Extraction in the Design of Executive Compensation’, University of Chicago Law Review, 69, 751846.
  • Bebchuk, L. , and Y. Grinstein (2005), ‘The Growth of U.S. Executive Pay’, Oxford Review of Economic Policy, 21, 283303.
  • Bertrand, M. , and S. Mullainathan (2001), ‘Are CEOs Rewarded for Luck? The Ones without Principles Are’, Quarterly Journal of Economics, 116, 90132.
  • Bhagat, S. , and B. Bolton (2006), ‘Corporate Governance and Firm Performance’, Working paper, University of Colorado at Boulder.
  • Brav, A. , W. Jiang , F. Partnoy , and R. Thomas (2008), ‘Hedge Fund Activism, Corporate Governance, and Firm Performance’, Journal of Finance, 63, 172975.
  • Cremers, K. J. M. , and V. Nair (2005), ‘Governance Mechanisms and Equity Prices’, Journal of Finance, 60, 285994.
  • Del Guercio, D. , L. Wallis , and T. Woidtke (2008), ‘Do Boards Pay Attention when Institutional Investors ‘Just Vote No’?’, Jorunal of Financial Economics, 90, 84103.
  • Gompers, P. A. , J. L. Ishii , and A. Metrick (2003), ‘Corporate Governance and Equity Prices’, Quarterly Journal of Economics, 118, 10755.
  • Hartzell, J. , E. Ofek , and D. Yermack (2004), ‘What's in It for Me? CEOs Whose Firms Are Acquired’, Review of Financial Studies, 17, 3761.
  • Hermalin, B. (2005), ‘Trends in Corporate Governance’, Journal of Finance, 60, 235184.
  • Huson, M. , R. Parrino , and L. Starks (2001), ‘Internal Monitoring Mechanisms and CEO Turnover: A Long Term Perspective’, Journal of Finance, 56, 226597.
  • Jensen, M. , K. Murphy , and E. Wruck (2004), ‘CEO Pay.. and How to Fix It’, Working paper, Harvard Business School.
  • Jenter, D. , and F. Kanaan (forthcoming), ‘CEO Turnover and Relative Performance Evaluation’, Journal of Finance.
  • Jenter, D. , and K. Lewellen (2010), ‘Performance Induced CEO Turnover’, Working paper, Stanford University.
  • Khurana, R. (2002), Searching for a Corporate Savior: The Irrational Quest for Charismatic Ceos. Princeton, NJ: Princeton University Press.
  • Masulis, R. , C. Wang , and F. Xei (2007), ‘Corporate Governance and Acquirer Returns’, Journal of Finance, 62, 185189.
  • Mikkelson, W. , and M. Partch (1997), ‘ “The Decline of Takeovers and Disciplinary Managerial Turnover,” with Megan Partch’, Journal of Financial Economics, 44, 20528.
  • Morck, R. , A. Shleifer , and R. Vishny (1989), ‘Alternative Mechanisms for Corporate Control’, American Economic Review, 79, 842852.
  • Murphy, K. J. (1999), ‘Executive Compensation’, in O. Ashenfelter and D. Card (eds), Handbook of Labor Economics, Vol. 3, San Diego, CA: Elsevier Inc., pp. 2485525.
  • Murphy, K. J. , and J. Zabojnik (2008), ‘Managerial Capital and the Market for Ceos’, Working paper, USC.
  • Yermack, D. (2006), ‘Golden Handshakes: Separation Pay for Retired and Dismissed Ceos’, Journal of Accounting and Economics, 41, 23756.
  • Weisbach, M. S. (1988), ‘Outside Directors and CEO Turnover’, Journal of Financial Economics, 20, 431460.

Appendix

  1. Top of page
  2. Abstract
  3. I. Introduction
  4. II. Sample and Data
  5. III. Turnover Levels
  6. IV. The Relation between Turnover and Performance
  7. V. Summary and Implications
  8. References
  9. Appendix
Table A1. CEO turnover
YearNumber of firmsTotal turnoverStandard (internal) turnover
(1)(2)(1)(2)
n%n%n%n%
  1. CEO turnovers in publicly traded Fortune 500 companies between 1992 and year-end 2007. Total turnover is all CEO turnover including turnover due to mergers and acquisitions and delistings from a major stock exchange. Standard (internal) turnover excludes turnover due to mergers and acquisitions 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 is selected. (2) Defines occurrences where the CEO dies as a non-turnover event. Data are from annual Fortune 500 lists, 10-K filings, proxy statements, and the Wall Street Journal. Year denotes the fiscal year-end for the sales data on which Fortune ranks firms. (i.e., 1992 corresponds to the 1993 April/May Fortune list.)

Panel A: Firms in the Fortune 500
19924435311.965311.965311.965311.96
19934585211.355211.35449.61449.61
1994470316.60316.60224.68224.68
19954807215.007215.005711.885711.88
19964735410.995210.99418.67398.25
19974857515.267415.265110.525010.31
19984848818.978818.976313.026313.02
19994899820.049820.047214.727214.72
200050213426.4913326.499719.329619.12
20014906713.676713.67489.80489.80
20024829920.129720.128016.6%7816.18
2003469224.69224.69142.99142.99
20044757214.957114.956213.056112.84
20054858517.538517.536513.406513.40
20064787415.277315.275711.925611.72
20074687214.967014.965511.755311.32
Total76311,14815.04%1,13814.91%88111.55%87111.41%
1992–1996232426211.27%26011.19%2179.342159.25
1997–2002293256119.13%55719.00%41114.0240713.88
2003–2007237532513.68%32113.52%25310.6524910.48
1992–1999378352313.83%52013.75%40310.6640010.58%
2000–2007384962516.2461816.0647812.4247112.24
Panel B: Firms not in the Fortune 500
199217423.53423.5315.8815.88
199328517.86517.86517.86517.86
19942573714.403513.603413.233212.45
19952614115.714115.713513.413513.41
19962654918.494818.113412.833312.48
19972503714.803614.40176.80166.40
19982374117.304117.302410.132410.13
19992275524.235524.232912.782912.78
20002015125.375024.883718.413617.91
20011922915.102915.10157.81157.81
20021943417.533417.532211.342211.34
20031942412.372412.372010.312010.31
20041942211.342110.82168.25157.73
20051944221.654121.132914.952814.43
20061853820.543820.542614.052614.05
20071843820.653720.112915.762815.22
Total308054717.7653917.5037312.1136511.85
1992–199682813616.4313316.0610913.1610612.80
1997–2002130124718.9924518.8314411.0714210.91
2003–200795116417.2516116.9312012.6211712.30
1992–1999154226917.4426517.1917911.6117511.35
2000–2007153827818.0827417.8219412.6119012.35