Corporate Performance and CEO Turnover: The Role of Performance Expectations This paper proposes that the inconsistent findings in previous studies of the relationship between corporate performance and CEO turnover may be due to insufficient attention to the type of performance indicator used by the individuals responsible for making CEO turnover decisions, namely, the board of directors. We argue that the board develops expectations of corporate performance, which it then uses to judge the CEO's performance. The study reported here analyzes financial analysts' forecasts of corporate performance, as a surrogate for the expectations board members could be expected to have, and then examines the relationship of forecasts to turnover. The principal finding is that turnover occurs when reported annual earnings per share fall short of expectations. For a sample of 408 CEOs under the age of retirmenet, this measure of corporate performance is a predictor of CEO turnover, whereas mechanical algorithms of abnormal security returns and historical accounting ratios are not.
Over the past two decades numerous studies have been published on the relation between corporate performance and chief officer (CEO) turnover (see Furtado and Karan, 1990, for an extensive review). Although several theories have predicted a negative relationship (Gamson and Scotch, 1964; Salancik and Pfeffer, 1980; Salancik and Meindl, 1984; Tushman and Romanelli, 1985), empirical findings have been mixed. For example, change in return on equity was significant in studies by Allen and Panian (1982) and Lubatkin and Chung (1985) but not in studies by Robinson and Brief (1985) or Harrison, Torres, and Kukalis (1988).
We content that the inconsistent findings may be due to insufficient attention to the performance indicators used by the individuals responsible for CEO turnover decisions, namely, the board of directors. Our thesis is based on three principal assumptions. First, a pivotal role of the management compensation contract is to ensure that actions taken by management are in the best interests of the shareholders. Second, the types of organizational performance indicators of interest to the board of directors are stipulated in CEO compensation contracts. Third, the board develops expectations of these performance criteria and evaluates actual performance in relation to whether expectations are met. Failure to meet expectations may contribute to the dismissal of the CEO. Therefore, to the extent that a performance measure reflects the board's heuristic of differences from expectations it should be an effective predictor of CEO turnover.
This study tests a model of CEO turnover that includes three performance criteria frequently stipulated in CEO compensation contracts: stock price performance, earnings targets, and selected accounting ratios. Stock price performance and accounting ratios were measured by mechanically derived algorithms, used in other studies. In contrast, earnings targets were formulated using financial analysts' forecasts of earnings per share as a surrogate for the board of directors' expectations of earnings targets. We assume that analysts' forecasts reflect the board of directors' expectations about future performance because much of the information analysts work with comes from executive officers of the firm who are members of the board of directors or advisors to the board. In addition, the board's concerns are assumed to be for the interests of the owner-shareholders, who would also be the target audience for the financial analysts' forecasts.
CEO Compensation Contracts
Relationship between the board of directors and the CEO. A fundamental concern of the shareholders in the modern corporation arises from the separation of ownership from control. Because of the complexities of corporate operations, the owners (the shareholders and their representatives, the board of directors) are not able to manage all aspects of the corporation, and so they are forced to delegate control of operations to hired professionals (the CEO and the management team). Agency theory postulates that a conflict of interest can arise because the goals of the board of directors and the CEO do not automatically coincide. The board is primarily interested in increasing shareholder wealth by maximizing stock prices, whereas the CEO is motivated by self-interest and an increase in personal wealth through compensation and nonpecuniary benefits. The board is then faced with finding ways of ensuring that the CEO will act in the shareholders' interests. Compensation contracts and dismissals are mechanisms for controlling the CEO's actions and aligning the CEO's and shareholders' interests. These mechnisms serve to reduce agency costs (Jensen and Meckling, 1976; Fama, 1980; Lambert and Larcker, 1985; Eisenhardt, 1989; Walsh and Seward, 1990).
Performance measures in compensation contracts. CEO compensation contracts typically include multiple bonus schemes that can take three forms: stock option plans based on future stock prices, performance plans based on the attainment of corporate earnings targets, and, to a lesser extent, target accounting ratios selectively determined by the board of directors to quantify their specific objectives. Compensation contracts often contain bonus schemes based on more than one indicator of corporate performance. There are at least two reasons this practice occurs that can be derived from an agency theory framework. The first is grounded in a theoretical contract model that demonstrates that as the number of performance criteria increases, the evaluation of CEO effectiveness becomes more optimal (Holmstrom, 1979). Since each criterion included in the compensation contract can measure performance differently, combining them helps remove some of the noise contained in each individual measure, thereby providing a clearer assessment of the CEO's contribution to organizational performance. A second reason for multiple indicators is that performance measures must encourage the CEO to act in the shareholders' interests as well as safeguard some of the CEO's interests. Otherwise, the CEO may feel threatened or exploited an may not act in the organization's best interests (Lambert and Larcker, 1985). The three types of bonus plans and their relationship to the evaluation of CEOs are described below.
Stock plans. A primary measure of corporate performance is the change in stock price. Stock plans frequently appear as incentives in CEO compensation contracts in the form of stock options, stock appreciation rights, phantom stock, dividend units, and restricted stock (Larcker, 1983). Stock plans are typically viewed as a form of long-term compensation and are structured accordingly. For example, stock options are fixed at a level above the current selling price. If the stock price rises above the option price, the CEO is able to buy shares at the fixed price, while is below their prevailing market price.
Stock plans have potentially mixed success in aligning the shareholders' and the CEO's interests. On the one hand, by having a stake in the firm's performance in the stock market, the CEO has an incentive to undertake projects that improve the price of the company's shares. Any increase in share price makes the CEO better off. On the other hand, strictly relying on compensation via a stock plan may impose too much risk on the CEO, because the stock price includes the effects of factors beyond management's control. As a result, the CEO may adopt a more conservative investment strategy to protect his or her personal interests and forego risky ventures with potentially high returns to shareholders (Lambert and Larcker, 1985).
Earnings targets. A second method of evaluating firm performance is with annual earnings targets. Although stock prices and earnings are related, we know of no evidence that suggests that stock prices are fixed multiples of earnings. We interpret this to mean that security prices and earnings provide different indications of performnce. The additional information provided by earnings helps differentiate the effect of the CEO's actions from uncontrollable exogenous factors. Earnings targets also help balance the effects of risk that CEOs would incur if they were compensated solely by shares (Lambert and Larcker, 1985).
Bonus plans based on annual earnings targets are a common feature of American compensation plans. In 1980 they were used by 90 percent of the 1000 largest U.S. manufacturing corporations (Healy, 1985). Earnings-based bonus plans constitute a substantial portion of short-term executive compensation. For instance, in 1978 the median ratio of accounting bonus to base salary for senior executives was 52 percent (Fox, 1980). In recent discussions with the authors, Mercer Inc., a leading compensation firm, confirmed that the 52-percent figure remains current. Although earnings-based bonus schemes vary widely, they typically define a measure of reported earnings and an earnings target or lower bound. If reported earnings exceed the target, the contract defines the maximum percentage of the difference that can be allocated to a bonus pool. Typically, no funds are allocated to the pool if earnings are less than the target.
Selected accounting ratios. Accounting ratios are measures of profitability and efficiency that are tracked by internal and external evaluators of the firm to assess the firm's health (Weiner and Mahoney, 1981). Measures tested in empirical studies include return on assets (Virany, Tushman, and Romanelli, 1985; Harrison, Torres, and Kukalis, 1988), return on equity (James and Soref, 1981; Allen and Panian, 1982; Lubatkin and Chung, 1985; Robinson and Brief, 1985; Harrison, Torres, and Kukalis, 1988), and profit margin on sales (Salancik and Pfeffer, 1980; Harrison, Torres, and Kukalis, 1988). Accounting ratios are used much less frequently than share price and earnings targets. They generally take the form of current accounting performance measured against past performance. This tends to be done in an idiosyncratic way by each firm. For example, the CEO compensation plan at Toro Corporation in 1976 included a measure of the firm's average return on capital compared to the average for selected competitors over five years.
Expectations of Performance
The ideal approach to learning about the board of directors' expectations of earnings would be to measure them directly. However, it is difficult for outsiders to obtain such information, for several reasons. First, few firms release forecasts to the public. Only about 10 percent of New York Stock Exchange firms do so, and many of these do it sporadically (Less, 1981). Second, forecasts issued by corporations tend to be vauge, rather than stating performance as point estimates, such as "earnings are expected to show adverse comparisons with the prior year" (Foster, 1986: 280). Third, firm that release forecasts directly to the public are not a random sample of all firms (Foster, 1986: 280-282): they have a less variable earnings series (Imhoff, 1978), larger net assets (Ruland, 1979), and tend to release forecasts when the firm is experiencing good market security returns rather than poor returns (Penman, 1980). Other infrequent occasions for direct public disclosures by firms are to correct for unrealistic (overly optimistic or pessimistic) forecasts made by analysts, to reduce unequal access to private corporate information, to curry favor with analysts by lending credibility to their forecasts, and to raise new capital (Lees, 1981; Ajinkya and Gift, 1984).
Because it is difficult to get data directly from boards of directors about their performance expectations for their firms, we studied the forecasts made by independent financial analysts for firms' performance. For several reasons, we believe they can serve as a surrogate for the measure of the boards' expectations. Financial analysts' forecasts generally are regarded as a good proxy for the board of directors' beliefs about the firm's future performance (Imhoff and Lobo, 1984). Executive officers of the firm, who may themselves be members of the board, provide information to analysts and to the board. Financial analysts, like board members, obtain information from a variety of sources, such as past performance data, macroeconomic forecasts, industry trade-association reports, and annual and interim corporate reports (Foster, 1986: 264). However, analysts cite interviews with company executives as their most important source (Lees, 1981). In many firms the chief financial officer is responsible for providing information to analysts through meetings and special reports (Amling, 1984: 450), and some firms put a lot of time into talking to analysts on a formal, as well as an informal basis. Barrons reported that 625 firms scheduled approximately 900 meetings with financial analysts in the year ending June 1983. This was the first year of systematic reporting by Barrons and the list of reported meetings has expanded over time. A specific example is J.C. Penney Company, whose representatives conducted over 1,000 interviews with analysts in 1974 (Axelson, 1975).
Corporate disclosure of information to financial analysts is a more common practice than direct public announcements, and approximately 65 percent of firms registered on the New York Stock Exchange do so (Lees, 1981). Several reasons have been suggested for this practice. First, firms may believe that providing information prevents analysts from becoming disgruntled and discrediting the firm and that such cooperation pays off with greater visibility for the firm among analysts who will promote the sale of the firm's stock to investors (Ajinkya and Gift, 1984). A second reason is that the release of information to analysts is believed to perform a powerful signalling function for the firm. Firms may release their true forecasts to suggest that other information they report is also accurate. Third, the act of releasing forecasts is thought to convey a positive signal to the market about the firm's value. Forecasts show that corporations have the ability to recognize changes in their economic environment and adjust their production plans accordingly (Trueman, 1986).
It is in a firm's best interests to communicate accurately with analysts rather than try to manipulate them (Foster, 1986: 278). Firms might be tempted to announce a prediction lower than what they actually expect in order to inflate actual performance against a low criterion. Such a practice could benefit firms in the short run, but if a consistent pattern developed, financial analysts would simply discount corporate estimates. Firms also normally have incentives not to overestimate poor performance because predictions of poor performance erode investor confidence and depress stock prices, although they might deliberately predict poor performance to try to thwart an unfriendly takeover.
Another issue concerns whether forecasts remain stable over time. Brown, Foster, and Noreen (1985) found that approximately 80 percent of analysts' forecasts remain unchanged after they are issued. This suggests that the originally forecast performance number is likely to be the benchmark the board of directors use to evaluate actual performance.
Hypothesis. Our hypothesis is derived from the following premise of this paper: that indicators of corporate performance employed by the board of directors in CEO compensation contracts are used to align the interests of the two parties. We hypothesize that performance measured as the difference between actual performance and the board of directors' expectations will be a better predictor of CEO turnover than measures derived from merchanical algorithms.
Model. The model of CEO turnover to be tested consists of corporate performance measures representing each of the three bonus plans described above: long-term stock option plans, short-term earnings targets, and selected accounting ratios. The model also includes the following control variables used in previous studies: corporate performance growth, organizational size, and tenure of the departing (i.e., incumbent) CEO.
Two other measures in the model are time lags of corporate performance and controls for industry performance. Time lags are tested to determine the time periods of performance data that are used in making turnover decisions. The performance measures are also computed relative to their industry counterparts, according to conventions in the accounting literature. The industry often serves as a benchmark for evaluation of corporate performance (Lambert and Larcker, 1985; Foster, 1986: 225). In addition, when industry performance is controlled, the CEO's impact on corporate performance is more readily identified.
We apply the model to the incidence of CEO turnover in which the departing (i.e., incumbent) CEO is under the age of 64, thus avoiding the potentially confounding effects of retirement (Coughlan and Schmidt, 1985; Schwartz and Menon, 1985; Harrison, Torres, and Kukalis, 1988). In some companies it is customary for CEOs to retire once they reach a certain age, and corporate performance is not a factor in the turnover decision. In companies that do not have this practice, the CEO can try to schedule retirment following good corporate performance to improve his or her retirement benefits or reputation. Retirement and corporate performance have been found to be positively related (Sonnenfeld, 1988: 74). This suggests that turnovers due to retirement may be affected by the interests of the CEO rather than merely the interests of the board of directors.
Firms included in the sample had to meet the following selection criteria. The CEO had to be identified in the June 1982, 1983, and 1984 Forbes magazine survey of the 800 highest paid CEOs in America. The firm had to be followed by at least three investment analysts surveyed by the investment firm of Lynch, Jones and Ryan and be reported in their Institutional Brokers Estimate System (IBES) data base. The mean of the analysts' forecasts was used as the estimate of earnings per share. Actual earnings per share was also drawn from IBES. In addition, the firm had to be included on both the University of Chicago Center for Research on Stock Price (CRSP) tapes and the Primary-Supplementary-Tertiary data base from Standard and Poor's Compustat Industrial Data Base. Firms having corporate year-ends other than the traditional calendar year-end were included. This ensured representation of sectors such as retailing firms, which traditionally have January 31 year-ends. The integration of these data sources yielded a sample of 480 large publicly owned U.S. corporations traded on the New York and American Stock exchanges. Forbes provided the CEO's identity, age, and length of service with the organization. Security returns data were obtained from CRSP. All remaining financial data were extracted from Compustat.
Security analysts' forecasts can be obtained from both (1) primary sources--the reports issued by individual analysts--and (2) secondary sources--the reports of services that collect and distribute earnings forecasts made by analysts at many financial institutions. IBES is an example of a secondary source. Each months, IBES surveys individual financial analysts from the research departments of leading Wall Street and regional brokerage firms. The contributing brokers are selected because of the superior quality of their research, professional reputation, and client demand. IBES collects details of their earnings-per-share forecasts for one year ahead, two years ahead, and their estimate of a five-year earnings-per-share growth rate. In mid-1982, IBES surveyed over 130 brokerage houses that covered over 1,500 firms. Subscribers to the service, primarily institutional investors, can access data relating to individual firms.
The turnover rate for the 480 firms meeting the sample selection criteria was 11 percent (53 turnovers). The 72 firms in which the departing (i.e., incumbent) CEO was over age 63 had 31 turnovers (43 percent). These were considered to have CEOs who had reached retirment age and were eliminated from the analyses. The 408 firms in which the departing CEO was under 64 had 22 turnovers (5.3 percent). A Pearson chi-square test rejects, at the 1-percent leve, that there is no relationship between age of the departing CEO and the probability of turnover.
TURNOVER. The dependent variable was the incidence of turnover of CEOs in 1983. This year was selected for two reasons. First, the economy in the prior year was relatively stable, i.e., not marked by excessive inflation or recession. Second, by 1983, the number of firms included in the IBES data base had grown substantially since its inception in 1975, thereby providing a relatively large sample of analysts' forecasts of earnings and long-term growth estimates. We interpreted the Forbes report of a CEO having one year of tenure in 1984 as evidence that a turnover occurred in 1983. To ensure this was the CEO's first year, we checked the prior year's issue of Forbes for the name of the CEO. Proxy statements filed with the Securities and Exchange Commission were used to fill in missing turnover data. The variable was assigned a value of 0 if no turnover occurred, 1 if turnover occurred.
Unexpected earnings per share (UNEXPECTED). This performance measure was used to represent earnings-based bonus plans. Unexpected earnings per share (EPS) were measured as the difference between actual EPS and the mean of financial analysts' expectations of EPS as reported by IBES.
The formula for unexpected EPS was adapted from Brown and Rozeff (1978) as follows:
UNEXPECTED.sub.it = (ACTEPS.sub.it - FEPS.sub.it.)/
where UNEXPECTED.sub.it equals the unexpected earnings per share for firm i in period t; ACTEPS.sub.it equals the actual earnings per share for firm i in period t; and FEPS.sub.it equals analysts' forecasts of earnings per share six months perior to the corporate year-end for firm i in period t. Unexpected earnings per share in 1982 are represented by UNEXPECTED82 and unexpected earnings per share in 1981 by UNEXPECTED81.
We used expected rather than actual EPS in the donominator because we were interested in actual results in relation to predictions. Brown and Rozeff studied the accuracy of predictions versus actual results, hence the denominator was actual performance. A positive value denotes performance greater than analysts' predictions, a negative value represents performance less than predictions, and zero denotes performance identical to predictions. The estimate made six months prior to the earnings announcement (e.g., June estimates for firms having calendar year-ends, May estimates for November year-ends) was selected. This estimate would likely have been made following the prior year's earnings announcement and before the firm's first-quarter announcement in the following year. In addition, it was sufficiently distant from the turnover announcement to reduce the possibility that analysts would incorporate it into their estimate. A one-year lag of unexpected EPS performance was constructed from 1981 data.
This performance measure is susceptible to bias due to extreme outlying values. Situations can arise in which small changes in estimates can translate into large performance differences expressed as percentages. For example, if IBES analysts make an EPS estimate of $.10 and the firm reports an actual EPS of $.01, there is a 90-percent difference from expected performance. It is unlikely that capital markets would treat this as a grave error, given the small absolute amount involved. To prevent outliers from driving the results, we truncated all values beyond positive and negative 300 percent (Brown et al., 1987). Negative truncations numbered 22 in 1982 and 4 in 1981. No positive values required truncation.
Unexpected industry earnings per share (UNEXPIND). This measure represented the average difference between actual EPS and financial analysts' expectations of EPS for firms within each industry, grouped according to 2-digit Standard Industry Classification (SIC) code. The data came from IBES.
Cumulative abnormal security returns (CAR). This performance measure was used to represent the CEO's stock plan. Firm share price is available on a conntinuous basis from various media sources. However, the board of directors' expectations of share price are not as readily apparent. As a surrogate for the board's surprise, we used the abnormal stock returns of the firm in the year prior to the turnover. An abnormal return to a stock is "that part of the return that is unanticipated by a statistical or economic model of anticipated, normal returns" (Reinganum, 1985: 51). It represents the actual return minus the normal or predicted return. A positive abnormal return reflects performance that exceeds the market's expectations, and a negative abnormal return reflects performance that fails to meet expectations. According to the semi-strong form of the efficient markets hypothesis (Fama, 1980), security, prices reflect all publicly available information about the firm's future cash flows. Changes in security prices (i.e., security returns) represent changes in the capital market value of the firm and are considered to be unbiased estimates of the present value of all future cash flows of the firm (Fama et al., 1969).
Cumulatives abnormal returns were calculated for the 250 trading days in 1982. The procedure for constructing the measure was similar to that described by Reinganum (1985), and the data were obtained from the CRSP tapes. The one-year lag was created using data for the 250 trading days in 1981. Abnormal security returns are typically not adjusted for industry because they differentiate firm-specific performance from overall movements in the stock market (Reinganum, 1985; Coughlan and Schmidt, 1985).
Accounting ratios. These performance measures were used to represent selectively determined accounting-based bonus plans. We know of no source that systematically provides data on the specific formulae used by each firm. Therefore, we created various combinations of historical data to serve as surrogates. For each of the four accounting ratios used, we evaluated the firm's performance relative to its industry counterparts. Most recent performance was calculated as the change in the ratio from 1981 to 1982 compared with the industry. The one-year lag compared 1981 performance with 1980. A five-year average was calculated from 1978 to 1982. Return on assets (ROA), return on equity (ROE), and profit margin on sales (PROFIT) were obtained from Compustat. Earnings per share (EPS) was obtained from IBES.
Growth. Two measures of growth were used. Data on change in market share from 1981 to 1982 (MKTSHARE) were obtained from Compustat and were controlled for industry changes based on two-digit SIC codes. A five-year historical average growth rate from 1978 to 1982 (GROWTH78-82) was available from IBES. It represents the least-squares calculation of moving four quarters' actual earnings for the last twenty quarters. Variables representing growth reflect whether the firm's performance improved or declined relative to a previous period. It is particularly important to include measures of growth in order to ensure that results for the measure of unexpected differences from analysts' forecasts apply in conditions of decline as well as growth. Robinson and Brief (1985) found that weak sales growth over a five-year period was related to increased turnover; however, Salancik and Meindl (1984) found no relation.
Organizational size (ASSETS82), measured as the natural log of the firm's assets in 1982, was expected to be positively related to turnover. Turnover is less disruptive in larger organizations because they tend to be more formalized and decentralized, there may be more competition for the CEO's position, and CEOs in large firms may be more visible and hence be recruited by other firms (Harrison, Torres, and Kukalis, 1988). Of the many turnover studies that have included organizational size, three have found a significant positive relationship (Salancik and Pfeffer, 1980; Allen and Panian, 1982; Harrison, Torres, and Kukalis, 1988) The log was used to prevent this variable from dominating the analyses, given that the other variables were expressed as percentages. The data were obtained from Compustat. The recommended procedure for measuring organizational size is to use its log without controlling for industry (Harrison, Torres, and Kukalis, 1988).
Incumbent CEO's tenure (TENURE) was measured by the log of the number of years the CEO had held the position as of 1983. The information was taken from Forbes magazine. Arguments have been proposed for this variable being either positively or negatively associated with turnover. A negative relationship may occur if the CEO is able to establish a power base over time with a well-established group. Salancik and Meindl (91984) found that longer tenure in the CEO position was associated with fewer personnel changes in top-management teams. Conversely, tenure may be positively associated with turnover because long tenure means the individual is closer to retirement.
Logistic regression analysis was used because the dependent variable, incidence of CEO turnover, was dichotomous. Analyses were performed using three time periods of performance data. The first set of analyses consisted of measures of latest performance. The second set consisted of latest performance as well as a variable representing a one-year lag of each performance measure. The third set substituted five-year averages of performance for the accounting ratios. Each of the four accounting ratios was entered into the equations separately to compare with previous studies. The measures adopted are operationalized and described in the Appendix.
Descriptive Statistics and Correlations
Means, standard deviations, and correlations for all the variables are reported in Table 1. The third column of Table 1 shows that CEO tenure is the only variable significantly correlated with turnover. None of the performance measures or other control variables are related to turnover.
The model of CEO turnover consists of three categories of corporate performance relevant to executive compensation contracts: differences between actual and expected earnings per share, cumulative abnormal security returns, and selected accounting ratios. Results of analyses using measures of corporate performance in the last full year of the CEO's tenure are reported in Table 2. Equation 1 contains two mechanically derived performance measures--share price performance, represented by cumulative abnormal security returns in 1982 (CAR82), and an accounting ratio denoting the change in return on assets from 1981 to 1982 (ROA82)--and the control variables for growth (MKTSHARE), organizational size (ASSETS82), and CEO tenure (TENURE). Equations 2 through 4 have the same format, substituting a different accounting ratio each time. In equations 5 through 8, measures of differences between analysts' expectations and actual earnings per share are added to the set of variables included in equations 1 through 4, respectively. This two-step procedure enables us to determine whether the expectations variable contributes explained variance not accounted for by the mechanical algorithms for share prices and accounting ratios. The results are based on truncated values for the measure of differences from analysts' expectations of EPS. Analyses using nontruncated data produced a similar pattern of results.
Results of the first analysis are reported in column 1 of Table 2. Cumulative abnormal security returns (CAR82) do not have a statistically significant relationship with CEO turnover. The change in return on assets from 1981 to 1982 (ROA82) also is not significant. Results for the control variables in the first analysis are as follows. The coefficient for change in market share from 1981 to 1982 (MKTSHARE) was negative and significant, which suggests that CEO turnover is more likely to occur when organizational growth is negative, i.e.,when the firm's market share is shrinking. The coefficient for organizational assets in 1982 (ASSETS) was not significant. This may be due to the homogeneity of the sample, since all the firms were very large (mean assets in 1982 = $5.5 billion). The coefficient for length of the CEO's tenure in the organization in 1982 (TENURE) was positive and significant, which suggests that there is a greater likelihood of CEO turnover as the length of the incumbent's tenure in the position increases. The TENURE term was also squared, and the results remained substantially the same in all cases.
Using equation 1 as a base, we extended the analysis along several dimensions. First we reran equation 1, substituting other accounting ratios for return on assets. Equations 2, 3, and 4 include measures for changes from 1981 to 1982 in return on equity (ROE82), profits (PROFIT82), and earnings per share (EPS82), respectively. In each case, the accounting ratio was not statistically significant. The results for CAR82 and the control variables exhibited the same pattern as in equation 1. Thus we found no evidence to support the hypothesis that mechanically derived algorithms representing most recent stock price performance and accounting performance are predictors of CEO turnover.
The next step was to study the impact of a performance measure based on expectations. Equations 5 through 8 incorporate the differences between actual and expected earnings per share for the firm in 1982 (UNEXPECTED82) and its industry counterpart (UNEXPIND82) into the models tested in the first four equations. A comparison of equation 5 with equation 1 shows that the variables they have in common yielded essentially the same results: in both equations CAR82, ROA82, and ASSETS82 were not significant, MKTSHARE was significant and negative, and TENURE was significant and positive. Of greatest interest is the finding in equation 5 that differences between actual and expected earnings per share for the firm (UNEXPECTED82) were negative and significant. The control variable for industry (UNEXPIND) was not significant. This model is repeated in equations 6 through 8, substituting ROE82, PROFIT82, and EPS82 for ROA82. Results were similar to equation 5. These analyses provide support for the hypothesis that measures of performance based on differences from expectations are better predictors of CEO turnover than mechanically derived measures. An illustration of this conclusion can be found by comparing equations 4 and 8. In both equations the mechanical measure, EPS82, which compares current EPS with the prior year, was not significant. However, in equation 8, UNEXPECTED82, which is based on expectations of EPS, was significant.
The next set of analyses addressed the timeliness of the performance data considered in CEO turnover decisions. Table 3 is analogous to Table 2 except for a one-year lag of all the performance measures. As in Table 2, columns 1 through 4 of Table 3 contain all variables except the expectations variables. In equation 1, results for CAR82, ROA82, and the three control variables were similar to those of equation 1 in Table 2. The one-year lag for stock performance (CAR81) was not significant, and the one-year lag for return on assets (ROA81) was significant and positive. Results for equations 2 through 4 are comparable to their counterparts in Table 2. None of the one-year lagged performance measures was significant for the security price variables or the accounting variables.
Columns 5 through 8 of Table 3 include performance expectations measures in 1981 and 1982 and their industry controls. A comparison of column 5 in tables 2 and 3 reveals that results are unchanged with the addition of the one-year performance lags: as before, the significant variables were UNEXPECTED82, MKTSHARE, and TENURE. None of the performance lags was significant. Results were similar when other accounting ratios were substituted in equations 6 through 8. These analyses suggest the CEO turnover decisions are based on the most recent performance data and not on performance in the prior year.
The last set of tests extended the time horizon by measuring the accounting ratios as an average over the preceding five-year period. Abnormal security returns and unexpected earnings performance were measured in the most recent year, since five-year averages would entail assumptions of model stability over a greater period than currently accepted in the financial literature. The results are consistent with the previous analyses. In all cases, the five-year accounting ratios were not statistically significant (except ROA78-82), and the pattern of results for the other variables was unchanged. In particular, the earnings expectation measure (UNEXPECTED82) remained statistically significant in all equations.
The results thus show that three variables have a systematic relation with turnover in all three of our models: earnings per share below analysts' expectations in the most recent year, long tenure of the CEO, and declining market share. The first two variables were significant in all analyses, and market share was significant in all but two. In the first set of models, depicting most recent performance (Table 2, columns 5 through 8), neither cumulative abnormal returns nor any of the four accounting ratios was significant. In the second set of models, which included most recent performance as well as a one-year lag of the performance measures (Table 3, columns 5 through 8), results were the same except for a positive effect for the one-year lag of return on assets. In a final set of models (results not shown), which included five-year averages for the accounting ratios, the results paralleled those in the second model: most recent negative unexpected EPS and a positive five-year ROA were associated with turn-over. Results for the control variables were similar in the three models: tenure was always positive and significant, assets was always nonsignificant, and market share was always negative and was significant in all but two cases. To verify the relationship between unexpected EPS and turnover, a streamlined model was run containing most recent unexpected EPS for the firm and the industry and the two nonperformance control variables, assets and tenure. Once again unexpected earnings in the most recent year was negatively related to turnover (-.48, p. < .10), and tenure was positively related (1.27, p. < .05).
DISCUSSION AND CONCLUSION
The results of this study support the hypothesis that the negative relationship between corporate performance and CEO turnover will grow stronger the more a performance measure reflects the board of directors' expectations. The measure of differences between actual EPS and financial analysts' expectations, which served as a proxy for the board's expectations, was significant, whereas measures based on mechanical algorithms--changes in accounting ratios over time and cumulative abnormal security returns--were not significant.
One way to account for the greater explanatory power of performance criteria based on expectations, rather than mechanical algorithms, is suggested by human information processing theory. This theory asserts that people use implicit models consisting of hunches or intuition to make predictions. People may use quantitative information based on statistical or mechanical techniques but make subjective changes to such data (Tversky and Kahneman, 1974). They may do this in the belief that quantitative models are too static or insensitive to subtle factors (Einhorn and Hogarth, 1982).
Retirement Age and the
The results reported in this study apply to CEOs who presumably have not reached retirement age. The dynamics of the corporate performance--CEO turnover relationship may be different for CEOs who have reached the conventional retirement age. Poor performance may still be a factor, but two other scenarios are also possible. Turnover may occur because of age, regardless of corporate performance. Alternately, the CEO may decide to retire following good performance to increase his or her retirement benefits. We conducted exploratory analyses for the 72 CEOs in the data set who were over age 63. The data were analyzed in the same manner as for the under-age-64 sample. The only significant variable was CEO tenure, which was positively related to turnover in all equations. To determine whether any performance variables were related to turnover, we reran the analyses with tenure omitted. In the first set of models that included the most recent measures of performance, positive abnormal security returns (CAR82) were significant in three of the four equations. In the fourth case a negative change in profits (PROFIT82) was the only significant variable. In the second set of models, which included most recent performance and a one-year lag, there was a positive effect for the most recent cumulative abnormal returns (CAR82) and the one-year lag of unexpected EPS (UNEXPECTED81). One accounting ratio, most recent profit (PROFIT82), was positively related to turnover. In the third set of models, which included five-year accounting ratios, CAR82 was significantly positive in four of the five equations. No other performance variables were significant. Results for the control variables were similar in the three sets of models: market share was always non-significant and assets was nonsignificant in all but one equation, where it was significant and positive.
The results thus show that turnover of CEOs of retirement age tends to be associated with positive corporate performance rather than negative performance, as found in the pre-retirement group. One explanation is that the CEO is influencing the timing of retirement (Sonnenfeld, 1988: 74), and thus the board of directors does not have exclusive control of the turnover decision.
The main question the results of this study raise is why financial analysts' forecasts of firm performance are a better predictor of CEO turnover than mechanical algorithms. We have argued that analysts' forecasts are a proxy for the board of directors' forecasts. We presented evidence that corporate executives provide information to their boards as well as to financial analysts. We further argued that boards and analysts incorporate this information, along with other considerations, into their forecasts of firm performance.
A useful research agenda would be to examine compensation contracts and observe the various performance criteria stipulated in them. Given the exact terms of the contract, one could calculate the extent to which the firm's management had performed in accordance with the board's stated targets.
An alternative would involve examining more directly the process by which performance expectations are formulated by the board of directors. Instead of relying on proxy measures such as analysts' forecasts or formulae in compensation contracts, researchers could survey or interview board members to assess their knowledge of various measures of corporate performance and the extent to which they believe turnover decisions are based on objective performance or other factors.
Once we have a greater understanding of how expectations are formed, we could then study the factors affecting the board of directors' commitment to its performance expectations. This informant would enable analysts, investors, and CEOs to better predict whether the board will feel bound to its expectations and make turnover decisions based on them. Four factors that have been found to generate commitment are explicitness, revocability, volition, and publicity (Salancik, 1977). Researchers could investigate the degree to which the board's expectations reflects these factors and what the subsequent impact on CEO turnover is.
In conclusion, the relationship between corporate performance and CEO turnover is of considerable practical and conceptual interest. In the corporate world the phenomenon of
"The CEO Factor" exists, whereby investors, analysts, and practitioners view chief executive officers as responsible for making decisions that will affect corporate performance (Financial World, 1981). Agency theory suggest that an important function of the compensation contract is to align the interests of the manager with those of the shareholder's representatives, i.e., the board of directors. Since the board of directors cannot directly observe all of the CEO's actions, the board must rely on various outcomes of corporate performance to evaluate the CEO's effectiveness in what is a complex process. Further study of this process could make sense of the disparate findings from past studies of the performance-turnover relationship and clarify the board of directors' expectations for the CEO's role in firm performance.
Ajinkya, Bipin P., and Michael J. Gift
1984 "Corporate managers" earnings forecasts and symmetrical adjustments of market expectations." Journal of Accounting Research, 22: 425-444.
Allen, Michael Patrick, and Sharon K. Panian
1982 "Power, performance, and succession in the large corporation." Administrative Science Quarterly, 27: 538-547.
1984 Investments: An Introduction to Analysis and Management, 5th ed.: chap. 17. Englewood Cliffs, NJ: Prentice-Hall.
Axelson, Kenneth S.
1975 "A businessman's views on disclosure." Journal of Accountancy, 140: 42-46.
Brown, Lawrence D., Paul A. Griffen, Robert L. Hagerman, and Mark E. Zmijewski
1987 "Security analyst superiority relative to univariate time-series models in forecasting quarterly earnings." Journal of Accounting and Economics, 17: 61-88.
Brown, Lawrence D., and Michael S. Rozeff
1978 "The superiority of analyst forecasts as measures of expectations: Evidence from earnings." Journal of Finance, 33: 1-16.
Brown, Philip, George Foster, and Eric Noreen
1985 Security Analyst Multi-year Earnings Forecasts and the Capital Market, Sarasota, FL: American Accounting Association.
Coughlan, Anne T., and Ronald M. Schmidt
1985 "Executive compensation, management turnover, and firm performance." Journal of Accounting and Economics, 7: 43-66.
Einhorn, Hillel J., and Robin M. Hogarth
1982 "Prediction, diagnosis, and causal thinking in forecasting." Journal of Forecasting, 1: 23-36.
Eisenhardt, Kathleen M.
1989 "Agency theory: An assessment and review." Academy of Management Review, 14: 57-74.
Fama, Eugene F.
1980 "Agency problems and the theory of the firm." Journal of Political Economy, 88: 288-307.
Fama, Eugene F., L. Fisher, Michael C. Jensen, and Richard Roll
1969 "The adjustment of stock prices to new information." International Economic Review, 10: 1-21.
1981 "The CEO factor." Vol. 150(12): 21-23.
1986 Financial Statement Analysis, 2d ed. Englewood Cliffs, NJ: Prentice-Hall.
1980 "Top executive bonus plans." New York: The Conference Board.
Furtado, Eugene T. H., and Vijay Karan
1990 "Causes, consequences and shareholder wealth effects of management turnover: A review of the empirical evidence." Financial Management, 19(2): 60-75.
Gamson, William A., and Norman A. Scotch
1964 "Scapegoating in baseball." American Journal of Sociology, 70: 69-72.
Harrison, J. Richard, David L. Torres, and Sal Kukalis
1988 "The changing of the guard: Turnover and structural change in top-management positions." Administrative Science Quarterly, 33: 211-232.
Healy, Paul M.
1985 "The effect of bonus schemes on accounting decisions." Journal of Accounting and Economics, 7: 85-107.
1979 "Moral hazard and observability." Bell Jounal of Economics, 10: 74-91.
Imhoff, Eugene A., Jr.
1978 "The representativeness of management earnings forecasts." Accounting Review, 53: 836-850.
Imhoff, Eugene A., Jr., and Gerald J. Lobo
1984 "Information content of analysts' composite forecast revisions." Journal of Accounting Research, 22: 541-554.
James, David R., and Michael Soref
1981 "Profit constraints on managerial autonomy: Managerial theory and the unmaking of the corporation president." American Sociological Review, 46: 1-18.
Jensen, Michael C., and William H. Meckling
1976 "Theory of the firm: Managerial behavior, agency costs, and capital structure." Journal of Financial Economics, 3: 305-360.
Lambert, Richard A., and David F. Larcker
1985 "Executive compensation, corporate decisions-making and shareholder wealth: A review of the evidence." Midland Corporate Finance Journal, 2(4): 6-22.
Larcker, David F.
1983 "The association between performance plan adoption and corporate capital investment." Jounal of Accounting and Economics, 5: 3-30.
Lees, Francis A.
1981 "Public disclosure of corporate earnings forecasts." New York: The Conference Board.
Lubatkin, Michael, and Kae Chung
1985 "Leadership origing and organizational performance in prosperous and decline firms." Academy of Management Proceedings' 85: 25-29.
Penman, S. H.
1980 "An empirical investigation of the voluntary disclosure of corporate earnings forecasts." Journal of Accounting Research, 18(1): 132-160.
Reinganum, Marc R.
1985 "The effect of executive succession on stockholder wealth." Administrative Science Quarterly, 30: 46-60.
Robinson, Brian S., and Arthur P. Brief
1985 "CEO succession among America's largest firms." Paper presented at the national meeting of the Academy of Management, San Diego.
1979 "The time series of earnings for forecast reporting and nonreporting firms." Journal of Business Finance and Accounting, 6: 187-201.
Salancik, Gerald R.
1977 "Commitment and the control of organizational behavior and belief." In Barry M. Staw and Gerald R. Salancik eds.), New Directions in Organizational Behavior: 1-54. Chicago: St. Clair Press.
Salancik, Gerald R., and James R. Meindl
1984 "Corporate attributions as strategic illusions of management control." Administrative Science Quarterly, 29: 238-254.
Salancik, Gerald R., and Jeffrey Pfeffer
1980 "Effects of ownership and performance on executive tenure in U.S. corporations." Academy of Management Journal, 23: 653-664.
Schall, Lawrence D., and Charles W. Haley
1986 Introduction to Financial Management, 4th ed. New York: McGraw-Hill.
Schwartz, Kenneth B., and Krishnagopal Menon
1985 "Executive succession in failing firms." Academy of Management Journal, 28: 680-686.
1988 The Hero's Farewell: What Happens When CEOs Retire. New York: Oxford University Press.
1986 "Why do managers voluntarily release earnings forecasts?" Journal of Accounting and Economics, 8: 53-71.
Tushman, Michael, and Elaine Romanelli
1985 "Organization evolution: A metamorphic model of inertia and reorientation." In L. L. Cummings and Barry M. Staw (eds.), Research in Organizational Behavior, 7: 171-222. Greenwich, CT: JAI Press.
Tversky, Amos, and Daniel Kahneman
1974 "Judgment under uncertaity: Heuristics and biases." Science, 185: 1124-1131.
Virany, Beverly, Michael L. Tushman, and Elaine Romanelli
1985 "A longtudinal study of the determinants and effects of executive succession." Academy of Management Proceedings' 85: 186-190.
Walsh, James P., and James K. Seward
1990 "On the efficiency of internal and external corporate control mechanisms." Academy of Management Review, 15: 421-458.
Weiner, Nan, and Thomas A. Mahoney
1981 "A model of corporate performance as a function of environmental, organizational, and leadership influences." Academy of Management Journal, 24: 453-470.
APPENDIX: Measures of Firm Performance and Control Variables
Unexpected performance (UNEXPECTED), using a formula adapted from Brown and Rozeff (1978), was calculated as follows:
[Mathematical Expression Omitted],
where [UNEXPECTED.sub.it] equals the unexpected earnings per share for firm i in period t; [ACTEPS.sub.it] equals the actual earnings per share for firm i in period t; [FEPS.sub.it] equals analysts' forecasts of earnings per share six months prior to corporate year-end for firm i in period. t. UNEXPECTED82 equals the unexpected earnings per share in 1982, and UNEXPECTED81 equals unexpected earnings per share in 1981.
Unexpected industry performance (UNEXPIND) was calculated as above, substituting industry data for firm dats. UNEXPIND82 equals the unexpected industry earnings per share in 1982, and UNEXPIND81 equals the unexpected industry earnings per share in 1981.
Cumulative abnormal security returns (CAR) was measured using formulae from Fama et al's. (1969) market model:
[R.sub.jt] = [a.sub.j] + [b.sub.j.R.sub.mt] + [e.sub.jt']
[R.sub.jt] = continuously compounded rate of return on security j over period t (i.e., normal return for security j over period t);
[a.sub.j] = E([R.sub.jt]) - [b.sub.j.E]([R.sub.mt]);
[b.sub.j] = cov ([R.sub.j],[R.sub.mt])/var([R.sub.mt]);
[R.sub.mt] = continuously compounded rate of return on Standard and Poor's Composite Index (market index) over period t; [e.sub.jt] = disturbance term of security j over period t; and E([e.sub.]) = O
Abnormal return (prediction error) was calculated as follows:
[AR.sub.jt] = [R.sub.jt] - [a.sub.j] - [b.sub.j][R.sub.mt],
where a and b are estimated over the 250 trading days prior to corporate year end.
Cumulative abnormal return was calculated as follows:
CAR = [Mathematical Expression Ommitted]
where N equals 250 trading days. CAR82 represents the cumulative abnormal return in 1982, and CAR81 the cumulative abnormal return in 1981.
Accounting ratios were measured using formulae from Schall and Haley (1986):
Relative accounting ratio =
Firm Performance - Average Industry Performance / Average industry performance
where firm and industry performance are measured as follows : ROA (x) equals (income + interest expense)/assets; ROE (x) equals (income + interest expense)/(common + preferred shares outstanding); PROFIT (x) equals income/sales; EPS (x) equals income per common shares outstanding; using x to represent 1982 (current year), 1981 (prior year), and 1978 -1982 (5-year average), respectively.
Earnings growth (GROWTH78-82) was computed from IBES, using the five-year historical average growth rate of earnings for 1978-1982 that represents a least-squares calculation of moving four quarters' actual earnings for the last twenty quarters.
Firm sales 1982 Firm sales 1981 / Mean industry sales 1982 mean industry sales 1981 / Firm sales 1981/Mean industry sales 1981
Organizational size (ASSETS82) was measured as the natural log of the firm's assets in 1982.
CEO tenure (TENURE) was measured as the natural log of the number of years the CEO had held the position as of 1982.…