Recent provocative evidence suggests that a board’s decision to remove a CEO from office is influenced by components of firm performance that have little to do with the CEO’s efforts or abilities. In particular, several authors have suggested that boards do not appropriately filter out industry performance measures when making CEO dismissal decisions. This evidence raises the alarming possibility that a substantial number of highly talented CEOs are removed from office for bad luck, while others are inefficiently retained because of good luck. The value loss from these types of inefficient removal decisions could be quite substantial if CEO talent is an important input into the value creation process.
A close look at this literature reveals substantial heterogeneity in research findings depending on important modeling choices including time period studied, how performance is measured, and how CEO turnover is identified. We argue that these variations in modeling choices can lead to systematic biases in research findings. In particular, press characterizations of CEO dismissals may reveal more about reporters’ beliefs regarding how CEOs are credited or blamed for firm performance than they do about the actual board decision-making process. In addition, extreme performance observations may be weighted more heavily in statistical models of turnover than is appropriate if extreme observations are often driven by random events that are largely outside the CEO’s control.
We re-examine this evidence in an unusually rich and comprehensive dataset that includes almost all CEO turnover events of U.S. publicly listed firms from 1990 onwards. The richness of this data allows us to nest many prior empirical modeling choices in a unified framework. We then examine the robustness of the most prominent findings in the CEO turnover literature related to the role of performance metrics in predicting CEO turnover. We first show that some of the classic findings in this literature are extremely robust, namely the significant sensitivity of CEO turnover to a firm’s stock performance. In particular, in any model in which CEO turnover is predicted as a function of relative-to-industry firm stock performance, there is always a highly significant coefficient indicating an elevated likelihood of turnover when firm performance is poor. This result holds independent of: (a) the sample period, (b) the definition of forced turnover, (c) the precise estimation procedure used, (d) the inclusion of various control variables, and (e) the precise way in which performance is measured.
With regard to the issue of the widely-discussed negative relation between turnover and industry performance, in almost all of the models that we examine, there is no convincing evidence of an independent role of stock returns in predicting CEO turnover. Only a very specific combination of modeling choices leads to this result. In particular, this result appears to be driven by (a) reliance on (potentially biased) press characterizations of CEO dismissals, (b) utilizing performance metrics that overweight extreme performance realization relative to what is suggested by the underlying theory, and (c) reliance on particular nonlinear functional forms in the estimation procedure.
This evidence is highly inconsistent with the notion that there are widespread inefficiencies in the CEO turnover process arising from ability inference errors and/or a corporate governance failure. The notion that CEOs are widely blamed by boards for bad luck, or credited for good luck, in the context of CEO removal decisions is simply not supported by the data. Thus, much of the recent discussion on the efficiency implications of the presence of industry performance factors in CEO turnover is misguided and attempting to address rare or non-existent behavior. Our evidence suggests that a simple efficient learning perspective with full industry relative performance evaluation (RPE) is a reasonable description of the CEO turnover process with respect to stock-based performance metrics. Thus, for future investigations of characteristics that affect the CEO turnover-performance relation, measuring stock performance on a relative-to-industry basis is an informative and reasonable baseline choice.
Our work also offers guidance into modeling choices that play an important role in learning about the CEO turnover process. One key choice is the turnover event categorization procedure, with choices that rely heavily on press characterizations almost surely incorporating systematic biases. We show that many CEO turnover events that do not appear overtly forced have all of the characteristics of a forced departure, including: (a) poor firm performance preceding the event, and (b) poor labor market opportunities for the CEO after the event. Thus, we suggest that the vast majority of CEO turnover events should be viewed as involuntary, and models of turnover should therefore incorporate a fairly broad definition of a CEO dismissal.
On a more technical note, we find that seemingly innocuous assumptions regarding how performance is measured can have a large effect on model inferences. For example, if performance is measured on a percentile basis rather than on an unadjusted basis, the statistical picture that emerges regarding turnover behavior changes in substantive ways. In addition, adding control variables to models that are statistically unrelated to the key explanatory variables of interest can also change the statistical inferences in important ways. Given this evidence, we recommend a broad set of model checks, including defining turnover and performance in multiple ways and using several different types of statistical models to assure that researchers’ conclusions regarding CEO turnover are robust and reflect true underlying economic behavior.
This post comes to us from C. Edward Fee the Morton A. Aldrich Professor of Business and Professor of Finance at Tulane University, Charles J. Hadlock, the A.J. Pasant Chair in Finance at Michigan State University, Jung Huang, Senior Consultant at Ernst & Young, and Joshua R. Pierce Associate Professor at the University of Kentucky, Department of Finance and Quantitative Methods. The post is based on the authors’ article, which is entitled “Robust Models of CEO Turnover: New Evidence on Relative Performance Evaluation” and available here.