Loss aversion and managerial decisions: evidence from Major League Baseball.

Pedace, Roberto ; Smith, Janet Kiholm

I. INTRODUCTION AND MOTIVATION

Previous research indicates that management changes are important events for organizations, partly because they lead to reversals of poor prior decisions, (1) However, this research begs the question of why it may be necessary to replace the manager in order to bring about such changes. In this article, we test for evidence of loss aversion and distinguish between loss aversion and agency costs as explanations for why managers "hang on" to poorly performing assets or employees they were instrumental in acquiring or hiring. The study contributes to the literature on managerial decisions and to the sports economics literature.

It may be necessary to replace managers if their investment decisions are affected by loss aversion. As Aronson, Wilson, and Akert (2010) point out, avoiding admission of mistakes is a common human trait. Once we have committed time or energy to a cause, it seems to be nearly impossible to convince us that the cause is unworthy. Psychological aversion to admitting mistakes is related to loss aversion and reflects a cognitive bias. The manager fails to see the problem or discounts negative experience as non-representative.

Agency costs provide an alternative explanation for why it may be necessary to replace managers to rectify prior mistakes. Agency costs arise when an agent (e.g., a manager) acts in his/her own interest but not in the interest of the principal (e.g., the firm owner). In the presence of asymmetric information, managers may have incentives to disguise or conceal poor decisions to avoid negative consequences such as reduced marketability as a manager, demotion, or termination. This concern with career and reputation can lead to overlooking inefficiencies, including asset underperformance and overpayment for the underperforming assets.

Using three decades of data on managerial and player turnover in Major League Baseball (MLB), we: (1) estimate the tendency of incumbent managers to hold on to low-performing players rather than divest them; (2)test the hypothesis that new managers unwind the mistakes of previous managers; and (3) distinguish between loss aversion and agency costs as alternative explanations for retention of poor-performing players.

Our empirical analysis focuses on poor-performing players who were hired by previous managers. This is because either loss aversion or agency cost may have induced the prior manager to hang onto poor-performing players. New managers can be expected to move quickly to address these types of problems; hence, we can use new manager retention decisions to determine if the decisions are affected by whether or not the poor performer was hired by a previous manager. Failure of the acquiring manager to divest poor-performing players could be consistent with either the manager's aversion to loss (a cognitive bias) or the manager's unwillingness to admit mistakes (agency costs in the form of career concerns). To distinguish between these, we follow the career concerns literature and reason that experienced managers are less likely to suffer reputational harm if they terminate poor performers whom they hired. Loss aversion, in contrast, does not suggest a relationship between managerial experience and failure to terminate poor performers.

There are several advantages to using baseball data for this study. The empirical design holds industry and competitive conditions constant and allows us to directly observe the performance of acquired assets (players) and how managers react to their performance once it is revealed. We are able to identify the manager who made the acquisition decision and can measure the performance of individual players. We also observe the player retention decisions of the subsequent manager (the manager who did not acquire the player). Hence, we can evaluate how the acquiring manager's behavior differs from that of the manager who inherits poor-performing players, some who were acquired by the prior manager and others who were hired by even earlier managers. This distinction between players acquired by the previous manager and those hired by other managers enables us to test whether acquiring managers are averse to admitting/recognizing mistakes compared to managers who were not involved with the initial acquisition.

Regarding key assets, managers in all industries possess hidden information about their productivity and synergies. For personnel, this includes information about collegiality, synergies and conflicts among team members, potential impairments, and so on. Because much of the information about performance of key assets is hidden from outsiders, it is difficult to evaluate whether a manager made a mistake by acquiring them. In professional baseball, this limitation is diminished because much of the player performance information is public. Objective measures of absolute and relative performance, such as batting success and fielding ability, are available to all interested parties. Hence, in contrast to more general business settings, observable performance measures are less noisy indicators of productivity.

The empirical results for the full sample of players show that poor-performing players are significantly more likely to be divested by new managers than by continuing managers. This result confirms the importance of new managers reversing the poor decisions made by their predecessors. Moreover, we can trace the aversion to divesting poor performers to the managers who were responsible for acquiring the players who turn out to be low performers. New managers tend to reverse the decisions of these acquiring managers. If efficiency guides retention decisions and if managers do not suffer from behavioral bias related to loss aversion or agency cost in the form of career concerns, there is no reason to expect that the choice to divest would vary systematically with the identity of the manager who hired the poor-performing players. In a test designed to distinguish between loss aversion and agency cost, we find that the experience of the acquiring manager does not affect the probability that the new manager divests the poor-performing player. We conclude that our findings are most consistent with loss aversion on the part of the immediate past manager.

II. BACKGROUND LITERATURE

The economics and psychology literature suggest two primary explanations as to why managers might be resistant to recognizing losses and divesting underperforming assets--cognitive bias and agency cost. Loss aversion, a form of cognitive bias, refers to the tendency of people to fail to perceive their past mistakes and has been convincingly demonstrated by Kahneman and Tversky (1979) and Tversky and Kahneman (1991). According to the theory, the possibility of being wrong is dissonance-arousing, so people will change their perceptions to make their decisions seem better. (2) In contrast, the agency cost explanation for non-profit maximizing decisions argues that a manager maximizes his or her own utility function, which may reflect career concerns, an interest in being surrounded by employees he of she likes, and so on. The manager's failure to rid an organization of inefficiencies that redound to the manager arises because acknowledging the inefficiencies (or errors in judgment) can have negative consequences for the manager's current job and subsequent marketability.

Several well-known studies are consistent with the agency cost interpretation of "holding onto losers." Boot (1992) studies corporate divestitures and the question of why firms delay selling underperforming divisions. He points out that asymmetric information allows managers to hide their prior bad investment choices from investors. Cho and Cohen (1997) suggest that holding onto losers allows managers to "blur" their poor performance under cover of the remaining operating units. Firms tend not to divest poorly performing business units, unless the firm experiences significant underperformance relative to industry peers. (3)

In their analyses of relevant literature, Baker and Ruback (2008) and Camerer and Malmendier (2005) note that the predictions of the behavioral models typically look very much like those of rational agency cost. To test the behavioral theories, it is necessary to distinguish between outcomes that are related to cognitive bias and outcomes that are related to agency or information problems. Hence, a primary contribution of this article is to distinguish empirically between agency cost and behavioral explanations for the managerial failure to divest poor performers. In addition, our research design uses a large database of baseball player acquisitions and divestitures that overcomes common obstacles that face researchers testing for managerial loss aversion. In particular, we are able to observe and measure performance of specific managerial decisions and to tie acquisitions and divestitures of particular assets to specific managers. Because we know the identities of the acquiring managers, we can observe whether new managers reverse the poor decisions of the acquiring managers.

III. ANALYTICAL FRAMEWORK

A. Hypotheses

We model the choice to retain/divest a player in any year as being dependent on individual player statistics, experience, and contractual constraints, as well as managerial characteristics and team performance. The unit of analysis is a player-team-year. In the tests below, we classify players hired by the current manager, the immediate prior manager, or an even earlier manager. For each observation, we also classify the manager as new or continuing.

The outcome we examine is that a player in a given year either leaves the team or is retained. That is, we observe whether, in the subsequent year, a player is no longer playing for his current team. We use the term "divestiture" to mean that the player is not observed on the same team in the subsequent year. (4) We examine the impact of manager changes on the probability of player divestiture, controlling for player and team performance, manager career success, and other variables that may affect the retention decision.

Below, we develop empirical tests to identify loss aversion in decision making and to distinguish between the loss aversion hypothesis and the agency cost hypothesis:

H1: Loss Aversion. Due to the previous manager's behavioral bias (loss aversion), new managers are more likely to rid the team of low performers hired by the previous manager than are continuing managers. The tendency for loss aversion applies, without regard to experience, to the previous manager who acquired the players and who failed to divest the low performers.

H2: Agency Cost. Agency cost manifested as career concerns suggests that the manager who acquired the poor performing players will be reluctant to recognize/acknowledge his mistakes, and instead will retain the players, out of concern for his reputation as a manager. This concern is more likely to affect retention choices for managers with less experience/reputation and suggests that a new manager, who is not accountable for the acquiring manager's mistakes, will be more likely to divest poor performers who were acquired by an inexperienced manager.

We employ a three-step empirical approach. First, we model the probability of player retention, using all player-team-year observations, including players on teams with new or continuing managers. Loss aversion suggests that a new manager is hired when a previous manager has failed to allocate resources optimally, and importantly for this study, has failed to rid the team of low-performing players. Both loss aversion and agency cost imply that the likelihood of divestiture goes up when the manager is new and when the new manager is confronted with low-performing players. While the first step is useful for examining the possible presence of a behavioral bias, it does not fully address the question of whether we can attribute the acquiring manager's divestiture decisions to the prior manager's loss aversion. The second step, therefore, is to estimate a model to test the hypothesis that new managers, who inherit low performers, reverse the decisions of the acquiring managers. Specifically, we test whether the new manager is more likely to divest low-performing players who were hired by the immediate past manager compared to low-performing players hired by an even earlier manager.

At the end of each season, managers make decisions on which players to retain. They do so based on their assessment of the player's ability, reflected, in part, on past performance, experience, tenure, and so on. However, it the manager's decision reflects loss aversion, then when determining whether to keep a player he hired himself, the manager will exhibit a bias toward retention. Hence, it follows that a new manager will be more likely to dismiss a poor-performing player hired by the immediate past manager (who acquired the player), compared to a poor-performing player who was acquired by an even earlier manager. This is because the new manager will partly be undoing the effect of the previous manager's bias (the bias will not exist for players hired by earlier managers since any bias will already have been undone by a subsequent manager). Thus, we focus on poor-performing players and distinguish between those hired by the immediate past manager versus an even earlier manager. (5) If we find that new managers reverse the poor acquisition decisions of their predecessor, then the third step is to evaluate possible explanations for why the acquiring manager failed to divest the poor performers. This could be the result of behavioral bias of the manager may have been fully aware of his mistake but, because of career concerns, did not want to admit it by divesting the player. We distinguish between these two possibilities by assessing whether the acquiring manager had an established reputation earned through his years of experience as a manager, while simultaneously controlling for managerial success (measured by career winning percentage). Experienced managers are less likely to have career concerns related to admitting mistakes than are inexperienced managers. In contrast, cognitive bias does not imply a relationship between loss aversion and the acquiring manager's experience.

B. Empirical Models

Equation (1) provides the first step model to examine the probability of player divestiture, using all player-team-year observations, including players on teams with either new or continuing managers:

(1) Pr([y.sub.1ij(t+1)] = 1) = F([[alpha].sub.1] + [x.sub.ijt][[beta].sub.1] + [w.sub.jt] [[gamma].sub.1] + [[theta].sub.1][m.sub.jt] + [[delta].sub.1][q.sub.ijt] + [[phi].sub.1][z.sub.ijt])

where the subscripts i, j, and t denote player, team, and year, respectively. The value of [y.sub.1] is equal to 1 if the player is not on the same team in the subsequent year (divested), 0 otherwise, and F is the standard cumulative normal distribution. The x vector captures player characteristics, including player performance (measured by At bats and Slugging percentage), a dummy variable indicating if a player's contract included a no-trade clause, and a dummy variable indicating if the player is a "Low performer" (in the bottom quartile of both performance statistics of all players in a given year). At bats and Slugging percentage are averaged over (up to) three prior seasons. The results are not sensitive to using other measures of performance, such as batting average or on-base percentage.

The w vector includes the team's winning percentage, averaged over (up to) three prior seasons, and the manager's experience and career winning percentage. The variable, m, is a dummy variable indicating that the manager is new to the team, and q is the interaction of the New Manager and Low performer dummy variables. We also include a dummy variable, z, to control for whether the new manager and the player were previously matched on a team.

In MLB, two types of managers may influence retention and hiring choices. The general manager (GM) is comparable to the CEO and oversees player transactions and contract negotiations. The GM normally is the person who hires and tires the coaching staff, including the team manager and field coaches. However, in baseball parlance, the term "manager" almost always refers to the field or team manager. The team manager (TM) controls team tactics, sets the line-up, and determines substitutions throughout the game. While we expect that the GM normally will be responsible for player acquisition and retention decisions, these decisions may be informed by the TM's experience with the player. Hence, we include data on both types of managers.

To test our hypotheses, we include two sets of variables--one identifies whether the management is "new" and the other identifies the interaction of Low performer with New Manager (GM or TM). We do not observe when, during the season, the new manager comes on board, but instead observe the team's managers at the beginning of each season. It is reasonable that it would take a season to make the changes in personnel that the new manager deems necessary; hence, we define a New Manager as one in his first or second year with the team. (6)

Referring to Equation (1), we expect individual performance to be a primary determinant of divestiture, as good performers are expected to create more wins for the team, thereby attracting more attendance. Similarly, players with longer tenure on a team are likely to become symbolic ambassadors for their organizations and may have accumulated specific capital (e.g., familiarity with the organization's goals and strategies), so they are also expected to have lower exit propensities. Holding individual performance and tenure constant, it is not clear, on net, how MLB experience, which is correlated with player age, will affect divestiture probability.

Another factor expected to affect player divestiture is organizational success. If players find successful teams more desirable, they will be less likely to leave voluntarily. In addition, managers who have been successful with a particular portfolio of players may be reluctant to make personnel changes. Therefore, we expect better team performance to be associated with lower divestiture probability.

Contractual commitments can restrict the ability of teams to terminate players. The most recent and increasingly popular innovation is the "no-trade" clause, which gives players the contractual option to restrict their trade to a preferred group of teams or to reject any trade altogether. Consequently, a player with a no-trade clause in his contract is less likely to be divested.

As a first step in the analysis, estimating Equation (1) allows us to establish whether the decisions of new managers regarding poor-performing players are different from the decisions of continuing managers. The predicted sign for the interaction between Low performer and a New Manager (GM or TM) is positive for both the loss aversion and agency cost hypotheses. As a second step, estimating Equation (2) allows us to examine the choices of new managers who inherit underperforming players who were acquired by the previous manager or an even earlier manager:

(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where r is a dummy variable indicating the player was acquired by the previous manager (immediate predecessor of the current manager), s is a dummy variable indicating the player was acquired by an earlier manager (the reference group consists of all other players), and u and v are the interactions, respectively, of r and s with q (where q interacts New Manager and the Low performer dummy variables). As before, the dependent variable is equal to 1 if the player is not observed on the same team (divested) in year t + 1, and 0 otherwise.

With this specification, we use Wald tests to evaluate (1) whether the likelihood of retention is different for players acquired by previous managers compared to players acquired by earlier managers and (2) whether the interacted variable, New Manager x Low performer x Acq. previous Manager, is statistically different from the interacted variable, New Manager x Low performer x Acq. earlier Manager. The latter test evaluates whether new managers come on board and remove the poor performers who were acquired by the new manager's immediate predecessor (relative to those acquired by an earlier manager). If the likelihood of divestiture is higher for those poor performers acquired by previous managers, it suggests that the previous managers exhibited a positive bias toward those players by retaining them.

The third step of the analysis allows us to distinguish between career concerns and loss aversion as explanations for why an acquiring manager would hold on to a low-performing player. The career concerns literature as developed by Holmstrom (1999) and Holmstrom and Ricart i Costa (1986), suggests that managers develop reputations over time based on their investment and personnel decisions. Once a manager makes an investment decision, the manager's career success is tied to its success or failure, and this creates a bias toward continuation of the investment. Even if the manager has information that suggests that abandonment of the project is more profitable for the firm, concerns with reputation creates incentives to continue the investment in an effort to hide the mistake. As Baker (2000) explains, the agency cost associated with this bias is stronger for less experienced managers because the impact of failure to a manager with no track record is larger than for a seasoned manager. In their study of career concerns in the mutual fund industry, Chevalier and Ellison (1999) proxy for experience using managerial age. We are able to measure experience directly (years in MLB), and to evaluate how experience affects player retention choices.

To test for career concerns, we estimate Equation (3) using a sample limited to players who were not acquired by the current manager (the default group is players acquired by earlier managers): (7)

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where q is, as in Equations (1) and (2), an interaction of new manager and low performer, r is, as in Equation (2), a dummy variable indicating that the player was acquired by the previous manager (the default is the players acquired by earlier managers), c is a dummy variable indicating that the previous manager who acquired the player has low experience, measured as years in a MLB manager role, and d and g are interacted variables. The variable d is a three-way interaction (Low performer x New Manager x Acq. previous Manager) and g is a four-way interaction (interacts d with c, or Low performer x New Manager x Acq. previous Manager x Acq. previous low experience). Hence, the specification allows us to ascertain whether the experience of the acquiring manager has an effect on the new manager's decision to retain a low-performing player. That is, does an experienced manager, who acquired a player who proves to be a poor performer, behave differently than inexperienced managers when making retention decisions? We evaluate this question by observing how new managers react when they inherit low performers. The career concerns hypothesis predicts that new managers are more likely to divest poor performers who were previously acquired by a manager with low experience. The acquiring manager may have retained the player out of concern that terminating the player would imply that the manager made a hiring mistake and would reflect poorly on his reputation and career. Such a concern is less likely to motivate a manager who has experience, and an established reputation in the industry. In contrast, the cognitive bias of loss aversion transcends experience and does not imply any relationship between the new manager's retention decision and the acquiring manager's experience.

IV. DATA AND RESULTS

Our full sample consists of 15,880 player-team-year observations. These are based on MLB team roster appearances from 1976 through 2005 for individuals with no missing information on batting performance. We selected these years to include all available data since "free agency" was instituted in December, 1975. (8) Player performance information is acquired from Lahman's Archive and the USA Today Web site, while no-trade clause data is gathered from the MLB.com Web site. Since the source data begins in 1949, we are able to accurately calculate player experience and tenure, even for the early years of our sample period. This information also allows us to match a team's acquisition of a player to a particular GM and TM, even if it occurred prior to 1976. Teams are uniquely identified in the source data according to their name and location. For analysis purposes, teams are coded as new franchises only if a franchise location change is observed.

We use batting statistics because we are interested in measuring individual performance rather than joint performance. In contrast to pitching and defensive statistics (fielding performance), posting of hitting statistics is largely independent of teammates. In addition, there is considerable judgment involved in measuring defense (e.g., fielding errors). Bradbury (2007) documents that although some spillovers may occur from batter to batter, the spillovers on defense are much more problematic.

Table 1 contains definitions and descriptive statistics of the primary variables used in the models. As shown, on average, 42% of the observations are associated with player divestiture. Players are separated into quartiles of performance. A quartile is defined using all player observations in a given year. To be classified in the bottom 25%, the player must be in the bottom 25% of all players in terms of both Slugging average and At bats at time t. We measure the Slugging average and the At bat statistics as rolling 3-year averages (t - 2 through t). If t - 2 does not exist for a player at a given point in time, then it is the average of t and t - 1. If the player is a rookie, then we use performance at time t. Players, on average, have 6.3 years of experience in MLB, compared to 5.7 for GMs and 9.0 for TMs. Players on average have 2.7 years of tenure with a specific team. As shown, 33% of the observations are associated with new GMs (defined as in their first or second year with a team), 53% are associated with new TMs, and 22% with new TMs and GMs. There are few cases in our sample where the player and manager have previously been matched on a team (less than 1% for the GM or TM).

The first three columns of Table 2 show results of estimating Equation (1) using three specifications: Model 1 includes the New GM indicator and the interaction of New GM with Low performer; Model 2 includes the New TM indicator and the interaction of New TM with Low performer; Model 3 includes an indicator variable (GMTM) that equals 1 if the observation is associated with a New GM and a New TM, and an interaction of GMTM with Low performer. The results are based on data for all players and all managers. We report the linear probability model (LPM) estimates, following Ai and Norton (2003), who argue that the marginal effects on probit interaction terms can be misinterpreted because estimated coefficients do not fully account for the interactive nature of the explanatory variables. (9)

Player performance statistics have the expected signs, as low performers are more likely to be divested. The presence of a New GM does not, by itself, affect the choice to divest a player, nor does the presence of a New TM or an entirely new management team (New GMTM). Loss aversion suggests that when new managers are confronted with low performers acquired by a previous GM or TM, they are more likely to divest those players than better-performing players. This is because new managers will not internalize the same dissonance in divesting an inherited poor-performing player as would a continuing manager who had hired the player. The result, however, could also be consistent with agency cost if the acquiring manager held onto the poor performers because of career concerns. As shown, the interaction effects (bolded) are all positive and significant. On average, when a manager is new, the divestiture probability for a poor-performing player increases by 4-6 percentage points.

Models 4-6 of Table 2 show the results of estimating Equation (2). The models test the hypothesis that the new manager is more likely to reverse poor decisions made by the previous acquiring manager (compared to earlier managers). When the previous manager retains the poor performer, the implication is that the new manager will "undo" the inefficiency. To evaluate the hypothesis, we conduct a Wald test of significance of the difference in coefficients for the interacted variables: New Manager x Low performer x Acq. previous Manager and New Manager x Low performer x Acq. earlier Manager. As shown, in the final rows of the table, the Wald tests reveal that retention choices are significantly different when the acquiring GM was the immediate predecessor of the current GM, and more importantly, that new GMs are significantly more likely to divest poor performing players hired by the previous, as opposed to earlier, GMs. The difference is significant at the .03 level in a one-tailed test. The difference is not significant for the TMs. These results indicate that retention choices are primarily driven by the GM, which is consistent with job descriptions as the GM normally oversees player acquisition and retention.

Hence, Table 2 indicates that poor-performing players are significantly more likely to be divested by the GM when the players were acquired by the previous GM. This result is expected if the acquiring manager displays loss aversion, but it arguably could be due to agency cost if the acquiring manager held onto the poor performers because of career concerns. To consider the potential for career concerns to affect retention choices, we examine the impact of the previous manager's experience on the retention choice of new managers using a subsample of observations where the player was not acquired by the current GM. The sample excludes players acquired by the current manager so that we can distinguish between the cases where a player was acquired by the previous GM versus an even earlier GM (the reference group) and reduces the number of interacted variables in the regression so that interpretation is more tractable. Using this approach, we can evaluate whether new manager choices can be traced to the acquiring manager's failure to recognize a loss, whether intentional (career concerns) or not (loss aversion).

We measure the experience of the acquiring GM as of the time the player was acquired (mean of 5.37 years for the full sample). We create a dummy variable for acquiring managers with low experience (Acq. GM low experience), defined as fewer than 5 years in MLB; 0 otherwise. Table 3 shows the results of estimating Equation (3), which includes the indicator variable for experience and two interaction terms: Acq. previous GM x New GM x Low performer and Acq. previous GM x New GM x Low performer x Acq. GM low experience. As shown, the incremental effect of experience on the likelihood that the new GM will divest a poor performing player acquired by the previous GM is not significant. That is, the new manager's retention choices are not significantly different when a low performer is inherited from an inexperienced versus an experienced acquiring manager. This finding is not consistent with the career concerns hypothesis. Loss aversion, in contrast, does not imply that the acquiring manager's experience will have a significant impact on the retention decision of the new manager.

A. Additional Robustness Checks

As noted above, the data do not indicate why players leave, so we cannot directly differentiate between involuntary separations (e.g., management-initiated terminations and trades) and voluntary separations (e.g., retirements and player-initiated trades). The inclusion of voluntary separations in the data reduces the likelihood of finding a relationship between performance, acquisition status, and player terminations. In order to assess the importance of this factor, we identified a subsample consisting of players who are least likely to have player--initiated career terminations and tradesiplayers who are not eligible for free agency. (10) We then re-estimated the regressions on this subsample. The results (not reported) are similar to those in Tables 2 and 3; all coefficients have the same sign, approximately equivalent magnitudes, and similar levels of statistical significance. (11)

As an additional check, we re-estimated the models in Tables 2 and 3 on two different time periods, where one might plausibly argue that performance statistics and hiring decisions could differ--the period prior to the introduction of interleague play and the period after (1976-1996 and 1997-2005). Because the two leagues have different rules regarding designated hitters (DH), MLB determined that interleague games would be played based on the home ballpark's rules so that DH are used in AL parks but not NL parks. The results (not reported) are not significantly affected by dividing the sample.

V. DISCUSSION

There are many reasons for managerial changes and varied expectations for what will occur after the changes. Among possible explanations for why new managers are brought on board is that previous managers are unable to recognize their mistakes, or unwilling to admit them, and hence do not terminate poorly performing projects, assets, or employees. While many studies address cognitive bias associated with loss, and there are many studies that document agency costs in managerial decisions, we are unaware of any study that distinguishes between these potential explanations for the failure of managers to rid the organization of losers. In this article, we use data from MLB to evaluate the alternative hypotheses. Important advantages of sports data are that manager and player changes are regularly observable, individual productivity is transparent and measurable, and hiring decisions can be traced to a specific manager.

Our key finding is that general managers (the baseball equivalent of CEOs) display behavior consistent with loss aversion in that acquiring managers are reluctant to divest poor performers. Specifically, new general managers are more likely to divest players when the poor performing player was acquired by the previous manager, as opposed to having been acquired by an even earlier manager. This behavior is not affected by the experience/reputation of the acquiring manager, pointing to loss aversion as the explanation rather than agency cost in the form of career concerns. Thus, it appears that team owners may hire new managers, in part to reverse player acquisition decisions of the new manager's immediate predecessor. Importantly, it is not that new managers rid the team of all poor performers or simply divest players who were hired by a previous manager. Instead, new managers appear to address a specific behavioral problem associated with the immediate past manager who acquired the poor-performer--namely, the tendency to "hold on to losers" who were hired by that manager.

While we use baseball data to evaluate the hypotheses, the findings have implications for the broader market of CEOs and high-level managers. If loss aversion is suspected, there may be ways for owners or boards of directors to address the specific behavioral issue directly rather than resorting to replacing the manager. As examples, incentive contacts could specifically contemplate the manager's possible loss aversion bias, and may include the requirement to specify performance benchmarks, ex ante, against which the realized performance of the asset can be assessed.

doi: 10.1111/j.1465-7295.2012.00463.x

ABBREVIATIONS

DH: Designated Hitters

GM: General Manager

LPM: Linear Probability Model

MLB: Major League Baseball

TM: Team Manager

REFERENCES

Ai, C., and E.C. Norton. "Interaction Terms in Logit and Probit Models." Economics Letters, 80(1), 2003, 123-29.

Aronson, E., T. D. Wilson, and R. M. Akert. Social Psychology. 7th ed. Englewood Cliffs, NJ: Prentice Hall, 2010.

Baker, M. "Career Concerns and Staged Investment: Evidence from the Venture Capital Industry." Harvard Business School, Working Paper, 2000.

Baker, M., and R. Ruback. "Behavioral Corporate Finance: A Survey," in Handbook of Corporate Finance: Empirical Corporate Finance, Vol. 2, edited by B. Espen Eckbo. Handbook of Finance Series. Amsterdam, The Netherlands: Elsevier/North-Holland, 2008.

Bollinger, C., and J. L. Hotchkiss. "The Upside Potential of Hiring Risky Workers: Evidence from the Baseball Industry." Journal of Labor Economics, 21(4), 2003, 924-44.

Boot, A. "Why Hang on to Losers? Divestitures and Takeovers." Journal of Finance, 47(4), 1992, 1401-24.

Bradbury, J. C. The Baseball Economist, New York City: Dutton, Penguin Group, 2007.

Camerer, C., and U. Malmendier. "'Behavioral Economics of Organizations," in Economic Institutions and Behavioral Economics, edited by P. Diamond and H. Vartiainen. Yrjo Jahnsson Foundation, Princeton, NJ: Princeton University Press, 2005.

Chevalier, J., and G. Ellison. "Career Concerns of Mutual Fund Managers." Quarterly Journal of Economics, 114, 1999, 389-432.

Cho, M.-H., and M. A. Cohen. "The Economic Causes and Consequences of Corporate Divestiture." Managerial and Decision Economics, 18(5), 1997, 367-74.

Holmstrom, B. "Managerial Incentive Problems: A Dynamic Perspective." Review of Economic Studies, 66(1), 1999, 169-82.

Holmstrom, B., and J. Ricart i Costa. "Managerial Incentives and Capital Management." Quarterly Journal of Economics, 101(4), 1986, 835-60.

Jin, L., and A. Scherbina. "Inheriting Losers." Review of Financial Studies, 24(3), 2011, 786-820.

Kahneman, D., J. L. Knetsch, and R. H. Thaler. "Experimental Test of the Endowment Effect and the Coase Theorem." Journal of Political Economy, 98(6), 1990, 1325-48.

Kahneman, D., and A. Tversky. "Prospect Theory: An Analysis of Decision under Risk." Econometrica, 47(2), 1979, 263-91.

Knox, R. E., and J. A. Inkster. "Post Decision Dissonance at Post Time." Journal of Personality and Social Psychology, 8(4), 1968, 319-23.

Lehn, K., and M. L. Mitchell. "Do Bad Bidders Become Good Targets?" Journal of Political Economy, 98(2), 1990, 372-98.

Tversky, A., and D. Kahneman. "Loss Aversion in Riskless Choice: A Reference Dependent Model." Quarterly Journal of Economics, 106(4), 1991, 1039-61.

Weisbach, M. S. "CEO Turnover and the Firm's Investment Decisions." Journal of Financial Economics, 37(2), 1995, 159-88.

(1.) See Lehn and Mitchell (1990) and Weisbach (1995) as examples.

(2.) In an early experiment, for example, Knox and Inkster (1968) found that bettors at a horse track believed bets were more likely to succeed immediately after being placed. Kahneman, Knetsch, and Thaler (1990) offer loss aversion to explain the sunk cost fallacy and for the endowment effect.

(3.) Also, see Jin and Scherbina (2011) who study the phenomenon of holding onto losers in the context of mutual fund management.

(4.) As the data do not indicate why players leave, we cannot directly differentiate between involuntary and voluntary separations, which could bias our analysis against finding an impact of managerial change on divestiture. Below, we provide an approach to deal with this potential issue.

(5.) The empirical distinction between players hired by the immediate past manager versus an earlier manager, along with the control for years of player experience, allows us to rule out the possibility that a new manager is more likely to terminate poor performers because he has more time than the prior manager to observe the players who were hired only a few years prior to the start of the new manager's tenure. If, controlling for player experience we do observe that the identity of the acquiring manager is important to retention decision (which we do), it cannot be because of this additional observation time, as the extra time occurs whether or not the acquiring manager was the previous manager or an earlier manager.

(6.) The window is supported by a report in The Economist, 2001, "To Cut or Not to Cut." February 10. 67-68, which indicates that, in most industries, newly hired managers make personnel changes within their first 2 years.

(7.) The restricted sample allows us to focus on possible motivations behind the acquiring manager's failure to divest a poor performer without having to rely on an overly complicated model with numerous interactions.

(8.) Prior to a legal ruling in 1975, all players were restricted agents who were "owned" by the team who hired them unless they were traded or released.

(9.) Probit results are available upon request and are not substantially different than those reported here.

(10.) Players with less than 6 years of experience are not eligible for free agency. Bollinger and Hotchkiss (2003) similarly identify these "reserve clause" players. With 6 years of service, the player is eligible for open contract negotiations.

(11.) Results for all robustness checks are available from the authors upon request.

ROBERTO PEDACE and JANET KIHOLM SMITH *

* We thank Dan Bernhardt, Jeff Borland (the editor), Richard Burdekin, Rick Smith, and seminar participants at the University of California Riverside for insightful comments on earlier drafts. Catherine Powers provided excellent research assistance. The Financial Economics Institute at CMC provided research support.

Pedace: Department of Economics, Scripps College, Claremont, CA 91711. Phone 951-318-2439, Fax 909-607-7142, E-mail rpedace@scrippscollege.edu

Smith: Robert Day School of Economics and Finance, Claremont McKenna College, Claremont, CA 91711. Phone 909-994-5757, Fax 909-607-6955, E-mail jsmith@cmc.edu

TABLE 1
Variable Definitions and Summary Statistics

Variable Name Definition

Player turnover, performance, and experience

Player 1 if player is not observed on the team in year t
separation + 1; 0 otherwise

Slugging Slugging average = (total bases-at bats), excludes
 walks; averaged over (up to) three seasons (t,
 t - 1, and t - 2)

At bats Total plate appearances; averaged over (up to)
 three seasons (t, t - 1, and t - 2)

Experience Player experience, measured as years in MLB

Player tenure Player tenure, measured as years with current team

No-trade 1 if player contract includes a no-trade clause; 0
clause otherwise

Low performer 1 if player performance is in lowest quartile for
 all players, measured by Slugging and At bats; 0
 otherwise

Team and manager characteristics

Win percent Team winning percentage, averaged over t, t - l,
 and t - 2

American 1 if team is in American League; 0 otherwise
League

New GM 1 if the GM associated with the player-year-
 observation is in his first or second year with
 team; 0 otherwise

New TM 1 if TM associated with the player-year-
 observation is in his first or second year with
 team; 0 otherwise

New GMTM 1 if both GM and TM are new (in first or second
 year with team); 0 otherwise

GM experience Experience of current GM, measured as years
 in MLB

TM experience Experience of current TM, measured as TM years in
 MLB

GM career win Current GM's winning percentage, averaged over
percent entire career

TM career win Current TM's winning percentage, averaged over
percent entire career

Player-manager matching characteristics

New GM x Low 1 if GM is new and player is a low performer; 0
performer otherwise

New TM x Low If TM is new and player is a low performer; 0
performer otherwise

New GMTM x Low 1 if both GM and TM are new and player is a low
performer performer; 0 otherwise

New GM x 1 if GM is new and player-GM combination was
Player match observed on a previous team; 0 otherwise

New TM x 1 if TM is new and player-TM combination was
Player match observed on a previous team; 0 otherwise

New GMTM x l if GM and TM are new and player-GM or player-TM
Player match combination was observed on a previous team; 0
 otherwise

Variable Name M Statistic SD
 Median

Player turnover, performance, and experience

Player 0.4244 0.0000 0.4943
separation

Slugging 0.3659 0.3737 0.1356

At bats 229.73 188.33 183.67

Experience 6.2920 5.0000 4.4509

Player tenure 2.7176 2.0000 2.4799

No-trade 0.0052 0.0000 0.0717
clause

Low performer 0.1491 0.0000 0.3562

Team and manager characteristics

Win percent 0.4974 0.5000 0.0699

American 0.5545 1.0000 0.4970
League

New GM 0.3288 0.0000 0.4698

New TM 0.5283 1.0000 0.4992

New GMTM 0.2236 0.0000 0.4166

GM experience 5.7116 4.0000 4.6472

TM experience 8.9808 7.0000 7.2334

GM career win 0.4959 0.4931 0.0545
percent

TM career win 0.4975 0.5003 0.0521
percent

Player-manager matching characteristics

New GM x Low 0.0486 0.0000 0.2151
performer

New TM x Low 0.0808 0.0000 0.2725
performer

New GMTM x Low 0.0339 0.0000 0.1811
performer

New GM x 0.0024 0.0000 0.0489
Player match

New TM x 0.0084 0.0000 0.0915
Player match

New GMTM x 0.0108 0.0000 0.1032
Player match

Notes: The table shows variable definitions and summary statistics
for 15,880 player-year observations for MLB batters, 1976-2005. All
variables are defined for player i, team j, year t, unless otherwise
noted.

TABLE 2
Regression Models of Player Divestitures, All Players

 Base Models

 1 New GM 2 New TM 3 New GM/TM
Independent Variables Model Model Model

Slugging -0.3983 *** -0.3961 *** -0.3946 ***
 0.0536 0.0534 0.0533
At bats -0.0008 *** -0.0008 *** -0.0008 ***
 0.0000 0.0000 0.0000
Experience 0.0320 *** 0.0321 *** 0.0320 ***
 0.0010 0.0010 0.0010
Player tenure -0.0031 -0.0029 -0.0029
 0.0022 0.0022 0.0022
No-trade clause -0.2435 *** -0.2454 *** -0.2459 ***
 0.0377 0.0379 0.0380
Win percent -0.4883 *** -0.4205 *** -0.4341 ***
 0.0657 0.0609 0.0673
GM experience 0.0010 0.0015
 0.0010 0.0010
TM experience 0.0009 0.0007
 0.0006 0.0006
GM career win percent -0.0225 0.0153
 0.0921 0.9442
TM career win percent -0.2242 *** -0.2603 ***
 0.0870 0.0882
Low performer 0.0602 *** 0.0550 *** 0.0625 ***
 0.0176 0.0203 0.0169
New GM -0.0059
 0.0094
New GM x Player match 0.0447
 0.0669
New TM x Player match -0.0216
 0.0413
New GMTM x Player -0.0071
 match 0.0360
New GM x Low performer 0.0446 **
 0.0214
New TM 0.0126
 0.0087
New TM x Low performer 0.0386 *
 0.0207
New GMTM 0.0049
 0.0105
New GMTM x Low 0.0594 ***
 performer 0.0233
Acq. previous GM

Acq. earlier GM

New GM x Low performer
 x Acq. previous GM
New GM x Low performer
 x Acq. earlier GM
Acq. previous TM

Acq. earlier TM

New TM x Low performer
 x Acq. previous TM

New TM x Low performer
 x Acq. earlier TM
Acq. previous GMTM

Acq. earlier GMTM

New GMTM x Low
 performer x Acq.
 previous GMTM
New GMTM x Low
 performer x Acq.
 earlier GMTM
Year fixed effects Yes Yes Yes
Team fixed effects Yes Yes Yes
League dummy Yes Yes Yes
[R.sup.2] 0.1857 0.1863 0.1864
Wald tests for value)
 differences in
 coefficients
 (F-stat and p value)
Test 1: Acq. Previous
 Manager--Acq.
 earlier Manager
Test 2: New Manager
 x Low performer x
 Acq. previous
 Manager--New Manager
 x Low performer x
 Acq. earlier Manager

 Models with Acquiring Manager
 Variables and Interactions

 4 New GM 5 New TM 6 New GM/TM
Independent Variables Model Model Model

Slugging -0.3968 *** -0.3916 *** -0.3888 ***
 0.0536 0.0534 0.0530
At bats -0.0008 *** -0.0008 *** -0.0008 ***
 0.0000 0.0000 0.0000
Experience 0.0323 *** 0.0321 *** 0.0318 ***
 0.0010 0.0010 0.0010
Player tenure -0.0035 -0.0053 ** -0.0084 ***
 0.0026 0.0027 0.0027
No-trade clause -0.2451 *** -0.2432 *** -0.2420 ***
 0.0383 0.0375 0.0374
Win percent -0.4875 *** -0.4217 *** -0.4278 ***
 0.0654 0.0610 0.0670
GM experience 0.0011 0.0018 *
 0.0010 0.0010
TM experience 0.0009 0.0009
 0.0006 0.0006
GM career win percent 0.0393 0.0150
 0.0917 0.0938
TM career win percent -0.2249 *** -0.2483 ***
 0.0871 0.0880
Low performer 0.0613 *** 0.0559 *** 0.0644 ***
 0.0176 0.0203 0.0169
New GM -0.0119
 0.0103
New GM x Player match 0.0498
 0.0671
New TM x Player match -0.0181
 0.0417
New GMTM x Player -0.0051
 match 0.0364
New GM x Low performer 0.0403 *

 0.0232
New TM 0.0460
 0.0098
New TM x Low performer 0.0423 **
 0.0220
New GMTM -0.0015
 0.0105
New GMTM x Low 0.0560 ***
 performer 0.0242
Acq. previous GM 0.0277 **
 0.0128
Acq. earlier GM -0.0185
 0.0217
New GM x Low performer 0.0435
 x Acq. previous GM 0.0443
New GM x Low performer -0.1983
 x Acq. earlier GM 0.1230
Acq. previous TM 0.0233 **
 0.0112
Acq. earlier TM 0.0249
 0.0172
New TM x Low performer 0.0137
 x Acq. previous TM 0.0330
New TM x Low performer -0.0842
 x Acq. earlier TM 0.0827
Acq. previous GMTM 0.0209 **
 0.0101
Acq. earlier GMTM 0.0567 ***
 0.0152
New GMTM x Low 0.0588
 performer x Acq. 0.0610
 previous GMTM
New GMTM x Low -0.1916
 performer x Acq. 0.2330
 earlier GMTM
Year fixed effects Yes Yes Yes
Team fixed effects Yes Yes Yes
League dummy Yes Yes Yes
[R.sup.2] 0.1865 0.1867 0.1874
Wald tests for
 differences in
 coefficients
 (F-stat and p value)
Test 1: Acq. Previous 5.58 ** 0.01 6.88 ***
 Manager--Acq. p = .0182 p = .9164 p = .0087
 earlier Manager
Test 2: New Manager 3.44 * 1.32 1.23
 x Low Performer x
 Acq. previous p = .0639 p = .2499 p = .2667
 Manager--New Manager
 x Low performer x
 Acq. earlier Manager

Notes: The table shows results of linear probability models, where the
dependent variable equals 1 if the player was divested from team in the
subsequent year and 0 otherwise. Regressions include all 15,880
observations. Each column reports the coefficient and robust standard
error, adjusted for clustering by player.

Statistical significance at the 1%, 5%, and 10% levels in two-tailed
tests is indicated by ***, **, and *, respectively.

TABLE 3

Regression Models of Player Divestitures, Players Not Acquired by
Current GM

Independent Variables 1 M SD

Slugging -0.6851 *** 0.3888 0.0967
 0.1002
At bats -0.0006 *** 318.04 176.88
 0.0001
Experience 0.02595 *** 7.9997 4.3478
 0.0021
Player tenure -0.0032 4.9314 3.3394
 0.0032
Win percent -0.5421 *** 0.4993 0.0697
 0.1493
No trade clause -0.2168 *** 0.0089 0.0941
 0.0684
GM experience 0.0012 4.1572 4.4381
 0.0020
GM career win percent -0.1380 0.4994 0.0613
 0.1720
Low performer 0.1386 ** 0.0513 0.2206
 0.0614
New GM 0.0051 0.6264 0.4838
 0.0185
New GM x Player match -0.0989 0.0010 0.0319
 0.1574
New GM x Low performer -0.2399 * 0.0393 0.1943
 0.1370
Acq. previous GM 0.0337 * 0.7517 0.4321
 0.0204
Acq. GM low exp. 0.0062 0.4856 0.4999
 0.0735
New GM x Low performer x Acq. 0.1811 0.0367 0.1882
 prev. GM 0.1428
New GM x Low performer x Acq. 0.0793 0.0204 0.1414
 prev. GM x Acq. GM low exp. 0.0786
League dummy Yes
Year and team fixed effects Yes
[R.sup.2] 0.1724

Notes: The table shows results of linear probability models, where
the dependent variable equals I if the player was divested from the
team in the subsequent year and 0 otherwise. The sample is limited
to players who were not hired by the current general manager; 3,919
observations. The variable "Acq. GM low exp" is equal to I if the GM
who acquired the player had less than 5 years of MLB experience and
0 otherwise. Each column reports the coefficient and robust standard
error adjusted for clustering by player. Mean and standard deviation
statistics for each independent variable are shown in the final two
columns.

Statistical significance at the 1%, 5%, and 10% levels in two-
tailed tests is indicated by ***, **, and *, respectively.