Is a CEO Turnover Good or Bad News?...Is a CEO Turnover Good or Bad News? Abstract We investigate...

Is a CEO Turnover Good or Bad News?∗

Axel Kind† Yves Schläpfer‡

April 2010

Abstract

We investigate the information content of CEO turnovers by analyzing abnormal stockreturns and abnormal trading volumes in the surrounding of the announcement date. Thesample consists of 208 CEO turnovers between January 1998 and June 2009 for companiesbelonging to the Swiss Performance Index. The single most important variable in assessingthe value of a turnover news is found to be the quality of the departing CEO as proxied by theprior stock-price performance relative to the market. In line with economic intuition and inaccordance with previous studies, the departure of outperforming (underperforming) managersrepresents bad (good) news for shareholders. Outside successions and forced turnovers yieldsignificant positive abnormal returns. However, a forced turnover does not per se represent apositive signal to shareholders. On the contrary, investors seem to critically assess the qualityof the board’s firing decision by considering the quality of the departing manager. When atalented CEO is dismissed or forced to leave, shareholders appear to disesteem the board’sdecision. This finding is confirmed in multivariate cross-sectional regressions and is robustto time subperiods and alternative test statistics. Trading volume is found to consistentlyincrease for all types of CEO turnovers. However, the size of the reaction crucially depends onthe characteristics of the turnover event, with forced turnovers generating the largest impact onthe turnover announcement day (+196.37%). Finally, the operating performance significantlyincreases (decreases) in the years following (preceding) CEO turnovers and reflects on averagethe short-term stock-price reaction around the announcement date.

Keywords: Corporate governance; CEO turnover; Firm performance; Trading volumeJEL codes: G14; G30; G34; M51

∗Contact information: Email A. Kind: [email protected]; Email Y. Schläpfer: [email protected]. Aspecial thanks goes to Marco Poltera for his excellent research assistance.†Department of Finance, University of Basel and Swiss Institute of Banking and Finance, University of St. Gallen.‡Department of Finance, University of Basel.

Is a CEO Turnover Good or Bad News?

Abstract

We investigate the information content of CEO turnovers by analyzing abnormal stockreturns and abnormal trading volumes in the surrounding of the announcement date. Thesample consists of 208 CEO turnovers between January 1998 and June 2009 for companiesbelonging to the Swiss Performance Index. The single most important variable in assessingthe value of a turnover news is found to be the quality of the departing CEO as proxied by theprior stock-price performance relative to the market. In line with economic intuition and inaccordance with previous studies, the departure of outperforming (underperforming) managersrepresents bad (good) news for shareholders. Outside successions and forced turnovers yieldsignificant positive abnormal returns. However, a forced turnover does not per se represent apositive signal to shareholders. On the contrary, investors seem to critically assess the qualityof the board’s firing decision by considering the quality of the departing manager. When atalented CEO is dismissed or forced to leave, shareholders appear to disesteem the board’sdecision. This finding is confirmed in multivariate cross-sectional regressions and is robustto time subperiods and alternative test statistics. Trading volume is found to consistentlyincrease for all types of CEO turnovers. However, the size of the reaction crucially depends onthe characteristics of the turnover event, with forced turnovers generating the largest impact onthe turnover announcement day (+196.37%). Finally, the operating performance significantlyincreases (decreases) in the years following (preceding) CEO turnovers and reflects on averagethe short-term stock-price reaction around the announcement date.

1 Introduction

The duty of a Chief Executive Officer is to maximize shareholders’ wealth by tak-ing sensible management decisions. Given the scope and importance of this mission, it is evidentthat a CEO turnover represents a major event in the history of any corporation, with possibly farreaching consequences for the company and its shareholders. The aim of this paper is to assess theinformation content of CEO turnovers from the stockholders’ perspective. In particular, we aim atexplaining the cross-sectional variation of abnormal returns by considering crucial characteristicsof the turnover, such as the departure type (forced or voluntary), the successor origin (insiders oroutsiders), the prior performance of the incumbent manager, and combinations thereof.

A real-world example should best illustrate how investors interpret the information related to aCEO turnover for determining the value of a stock. The example refers to Mr. Fred Kindle, formerCEO of ABB Ltd, a global leader in power and automation technologies.1 Under his leadership,

1Despite the fact that the story of Mr. Kindle motivates very well this study and is a prime example of aCEO turnover that achieved broad media attention and had a shocking impact on investors, we should remark that

1

ABB recovered from financial distress and achieved a record result in 2007. In January 2005, thedate of Kindle’s appointment as CEO, ABB’s stock price was around 6 Swiss francs. Under hisleadership the stock price rose to around 25 Swiss francs in February 2008. According to the Fi-nancial Times, ABB’s recovery qualifies as a prime case study of successful company restructuring.2

Nevertheless, on February 13, 2008, ABB surprisingly announced in an official statement that FredKindle would leave ABB “due to irreconcilable differences about how to lead the company”.3 TheCFO of ABB, Michel Demaré, was appointed interim CEO, but at that time was also considered asa potential candidate for the position of permanent CEO. Several news agencies4 speculated thatthis decision was mainly caused by a power struggle with the president of the board, Hubertusvon Gruenberg, regarding ABB’s acquisition strategy. The departure of Fred Kindle was a bigsurprise to the financial community as everybody agreed that he was doing an outstanding job atABB. Strikingly, the Financial Times referred to him as the “wunderkind chief executive”. Thefact that investors shared the same perception about Fred Kindle’s work performance is cruciallycaptured by the stock-price reaction on the announcement date of his departure. In spite of thesimultaneous announcements of a record result for the year 2007, a doubling of the dividend, andupcoming share repurchases, the stock price dropped sharply by 5.14% on that very same day.According to Bloomberg, ABB’s stock price lost interim about 10%, which was the worst drop ofthe previous three years. This example highlightens two key insights that go along with this paper.First, investors may attribute a great importance to the information of a CEO departure. Second,the valuation consequences of a CEO turnover strongly depend on the characteristics and circum-stances of this event. While the academic literature has already stressed the beneficial effects offorced turnovers, the example of Mr. Kindle suggests that the prior performance of the departingmanager may convey equally important information. These insights motivate us in studying thestock-price impact of CEO turnovers by considering a variety of characteristics associated withthese events.

A large part of prior research on CEO turnovers focuses on the relation between the CEO’sperformance and the turnover probability (Coughlan and Schmidt, 1985; Weisbach, 1988; Warner,Watts, and Wruck, 1988; Parrino, 1997; Suchard, Singh, and Barr, 2001). This strand of literaturecomes to the conclusion that there is a negative relation between performance and the probabilityof a (forced) turnover. However, while statistically significant, this negative relation is found to beeconomically weak.

A second strand of literature which is of immediate relevance for the hypotheses studied in thispaper investigates the impact of CEO turnovers on the company’s performance. Contributionsdiffer with respect to the performance measures employed (stock price reactions vs. accountingperformance measures) and the characteristics of the turnovers considered: (i) outside vs. insidesuccessions, (ii) forced vs. voluntary turnovers, (iii) governance differences, (iv) gender differences,

this event could not be included in the empirical analysis due to a “no-confounding-event” criterion applied in theconstruction of the sample. More precisely, the release of other valuation-relevant news (dividends and earningsannouncement) on the date of the dismissal of Mr. Kindle prevents us from using this data point.

2http://www.ft.com/cms/s/0/5a5db304-da69-11dc-9bb9-0000779fd2ac.html3ABB Press Release: “ABB CEO Fred Kindle leaves company”, available at http://www.abb.com.4Amongst others Reuters, Bloomberg, Timesonline, Financial Times, Financial Times Deutschland, Handelsblatt

and Handelszeitung covered the news regarding the departure of Fred Kindle at ABB.

2

etc.. Table 1 presents an overview of the most important contributions in this field together with asummary of the main empirical findings.

The majority of studies (Reinganum, 1985; Furtado and Rozeff, 1987; Warner, Watts, and Wruck,1988; Bonnier and Bruner, 1989; Borokhovich, Parrino, and Trapani, 1996; Dherment-Ferere andRenneboog, 2002; Huson, Parrino, and Starks, 2001; Dahya and McConnell, 2005; Adams and Mansi,2009) detect significant positive abnormal returns following turnovers with company-outsiders assuccessors. Notable exceptions are Worrell, Davidson, and Glascock (1993) and Khanna and Poulsen(1995). The former find mixed results and the latter do not report significant abnormal returnsby examining companies eventually filing for Chapter 11 protection, neither for outside nor insidesuccessions.

The majority of studies that also investigate the subsample of forced turnovers find that ab-normal returns following firings are higher than those following voluntary turnovers (Furtado andRozeff, 1987; Denis and Denis, 1995; Huson, Parrino, and Starks, 2001; Dherment-Ferere and Ren-neboog, 2002; Adams and Mansi, 2009). Borokhovich, Parrino, and Trapani (1996) report significantpositive abnormal performance for forced turnovers in combination with outside successions. OnlyDedman and Lin (2002) detect significant negative abnormal returns on average after a CEO’sdismissal.

Weisbach (1988) investigates the relation between the composition of the board of directors(insider- vs. outsider-dominated boards) and CEO turnovers. He reports positive announcementeffects but no difference in the impact between the firing decision of outsider- and insider-dominatedboards. Similarly, Fisman, Khurana, and Rhodes-Kropf (2005) examine the role of managerial en-trenchment for the board’s firing decision and its consequence on the stock-price, finding a weaklysignificant relation between entrenchment and abnormal returns.

Lee and James (2007) and Coxbill, Sanning, and Shaffer (2009) study gender effects by compar-ing the price impact of male vs. female CEO appointments. While the former detect significantlylower (and negative) abnormal returns for women CEOs, the latter report a negative but insignif-icant impact on the stock price. Finally, Johnson, Magee, Nagarajan, and Newman (1985) andWorrell, Davidson, Chandy, and Garrison (1986) measure the stock-price reaction in the aftermathof key-executives’ deaths. The latter report negative and significant effects in association with sud-den deaths and the former significantly positive (negative) reactions following the death of companyfounders (non-founders).

As mentioned, accounting performance measures can be used as an alternative way to assessthe value of a CEO turnover. Several papers (e.g. Denis and Denis, 1995; Khurana, 2001; Dedmanand Lin, 2002; Huson, Malatesta, and Parrino, 2004; Fisman, Khurana, and Rhodes-Kropf, 2005;Hillier and McColgan, 2005; Dezso, 2007) investigate the impact of CEO changes on key account-ing performance measures. Typically, the papers evidence that a CEO-turnover is preceded bydeteriorating accounting figures, which improve after the new CEO takes up his/her position. Thedescribed pattern is generally more pronounced for forced than for voluntary turnovers.

CEO turnover in relation to manager quality (proxied by prior company or stock performance)is investigated in Weisbach (1988) and Bonnier and Bruner (1989). Weisbach (1988) finds a strongand negative relation between the intersection of prior stock performance with outside boards andabnormal returns around the CEO turnover. Bonnier and Bruner (1989) investigate underperform-

3

ing firms jointly identified by the criteria of negative earnings and dividend omission prior to themanagement turnover. The reason to select this particular sample of management turnover is thatthe abnormal return at the announcement of a (forced) management turnover comprises two effectsas described by Warner, Watts, and Wruck (1988). First, if the turnover signals managerial qualitythat is worse than anticipated then the information effect is negative. Second, the real effect ispositive if the (board-initiated) turnover is appreciated by the shareholders. Therefore, positiveabnormal returns are observed only if the real effect outweighs the information effect. By using asample of companies for which the bad performance has already recognized they try to reduce thenegative information effect and estimate a more accurate real effect.

In this paper, we investigate stock returns, operating return on assets, as well as trading volumesin the surrounding of CEO turnover events. The sample consists of CEO turnovers at companiesin the Swiss Performance Index during the period between January 1998 and June 2009. The fi-nal sample comprises 208 CEO turnovers. It is survivorship-bias free since it also includes CEOturnovers at companies that exited the index. The CEO-turnover variables considered in the inves-tigation are the successor origin, the departure type, and the prior stock performance.

This paper contributes in three ways to the existing literature. First, it corroborates the findingsof the existing literature on CEO turnovers by employing a new hand-collected data sample. Inparticular, we are able to show that the most important results reported for the US also apply tothe Swiss market. This is interesting because - as shown in LaPorta, de Silanes, and Shleifer (1999)and Faccio and Lang (2002) - the Swiss market is characterized by a less atomistic ownership struc-ture with more family-controlled companies and fewer active investors. Second, the paper classifiesturnovers with respect to (i) the departure type of the CEO (voluntary vs. forced), (ii) the successororigin (internal or external), (iii) the prior performance (outperformance or underperformance), andall the interactions of those variables. Most importantly, in contrast to previous research, the paperemphasizes that forced turnovers do not always constitute a positive signal to shareholders: Forcedturnovers of underperforming managers trigger significantly positive abnormal returns, while forcedturnovers of overperforming managers are associated with negative abnormal returns. This suggeststhat shareholders assess the quality of the board’s firing decision by considering the performanceof the departing CEO. Third, instead of solely focusing on abnormal returns, the paper analyzesthe impact of CEO turnover announcements on trading volumes5 and long-term accounting perfor-mance measures.

The results obtained in this paper indicate that outside successors, forced turnovers, and priorunderperformance of the company under the departing manager lead to significant positive abnor-mal returns: +1.85%, +2.74%, and +1.94% on average for the [-2 0] event window. The largestaverage abnormal returns are detected for the following two categories of CEO turnovers: (i) forceddepartures in conjunction with an outside successor (+6.71%)6 and (ii) forced departures of under-performing managers (+5.73%). Interestingly, while forced turnovers of underperforming managerstrigger positive and significant abnormal returns (+5.73%), forced turnovers of outperforming CEOsare associated with negative abnormal returns (−2.24%). This suggests that shareholders assess

5To the best of our knowledge, the only study that investigates abnormal returns and trading volumes aroundmanagement-turnover announcement is Cools and van Praag (2007).

6When not otherwise stated, the abnormal returns reported in parethesis refer to [-3 0] event windows

4

the quality of the firing decision by the board of directors by considering the prior performance ofthe company under the departing CEO. The results are robust with respect to time subperiods anda wide range of parametric and non-parametric test specifications.

The trading volume of the average company affected by a CEO turnover is found to increase byalmost 130% on the announcement day. Particularly large increases in trading volume are observedfor forced-turnover announcements (mean: +196.37%) and forced turnovers of underperformingCEOs (mean: +259.64%). In accordance with theoretical papers on the release of public informa-tion, absolute abnormal stock returns are found to be positively correlated with abnormal tradingvolumes.

The long-term relation between CEO turnovers and operating performance, which is measuredas the ratio of operating income and book value of total assets, shows that a CEO turnover istypically preceded by deteriorating operating performance and followed by a steady increase inoperating performance. Besides for the total sample, this pattern is particularly evident for thesubsamples of outside successions, forced departures, departures of underperforming CEOs, andthe intersection of forced departures with underperforming CEOs.

The remainder of this paper is structured as follows. Section 2 develops hypotheses regardingthe expected stock-price reaction following different types of CEO turnovers. Section 3 deals withthe sample construction and describes the final sample of CEO turnovers. Section 4 presents theempirical analysis of this paper. After addressing the setup of the event-study and presenting gen-eral results related to the impact of CEO turnovers on abnormal stock returns (Subsection 4.1),we perform a cross-sectional analysis of the results by regressing abnormal returns against selectedCEO-turnover variables (Subsection 4.2) and provide extensive robustness checks (Subsection 4.3).Subsection 4.4 focuses on the impact of CEO turnovers on the trading volume and Subsection 4.5investigates the long-term impact of turnovers on the companies’ operating performance. Section 5provides a summary of the paper and concludes.

2 Theory, Prior Empirical Results, and Hypotheses

In this section we formulate hypotheses regarding the impact of CEO turnovers on the corre-sponding stock prices. In particular, we are interested in differentiating between various charac-teristics of the turnover, such as (i) the successor origin (inside vs. outside successions), (ii) thedeparture type (forced vs. voluntary departures) and, (iii) the relative performance of the companyunder the departing CEO. Such hypotheses are meant to reflect explicit theories proposed in theCEO-turnover literature, plausible extensions thereof, and empirical results from previous studies.

2.1 Outside Succession

It is a well established empirical fact that in the majority of CEO turnovers the successor is acompany insider. The corporate-finance literature has proposed a number of plausible reasons for

5

the Board’s preference for inside over outside CEOs. These are (i) the company-specific human cap-ital accumulation theory of Dherment-Ferere and Renneboog (2002), (ii) the quality-measurementtheory, (iii) the tournament theory of Chan (1996).

Dherment-Ferere and Renneboog (2002) argues that inside candidates have two main advantagesover outsiders. First, over the years they have the opportunity to accumulate valuable company-specific knowledge about processes and technologies. Second, they can exploit an already existingsocial network to acquire specific internal information. Thus, the accumulation of company-specifichuman capital naturally makes insiders more attractive than outsiders for a CEO position.

Another explanation for the reluctance to appoint outside candidates to the CEO position arisesfrom the less accurate estimation of their quality. The history of an insider in the company auto-matically generates a performance track record that can be easily used by the Board of Directors toassess the insider’s quality. Conversely, the information basis to estimate the quality of an outsideris much smaller, which makes this choice intrinsically more risky.

Finally, Chan (1996) argues that considering outsiders for the CEO position can reduce theincentives and hence the motivation of lower-level executives. Clearly, if outsiders are includedinto the circle of potential successors the chance for insiders to become CEOs diminishes. Thus,Chan (1996) suggests that the preference for inside successors may represent a natural attempt tomotivate employees by strengthening the link between performance and reward. By considering thegeneral preference for insiders as CEO successors, we argue that the board of directors will appointan outsider as CEO only if his/her quality exceeds by far that of the best available insider. Inthis light, we might expect the stock price to react positively once the information about a CEOturnover with an outside succession is released, which leads us to the formulation of the followinghypothesis.

Hypothesis 1: The announcement of an outside succession yields positive abnormalreturns.

By considering the empirical evidence reported in Table 1, column 6, the above hypothesis seemsto be backed by the majority of previous studies.

2.2 Forced Turnovers

The board of directors has the non-transferable and indefeasible duty of nomination and dis-missal of the management of the company. Ideally, we would expect the board to act in the bestinterest of shareholders when deciding about a CEO’s dismissal.

The improved management hypothesis presented by Huson, Malatesta, and Parrino (2004) statesthat forced management turnovers induce a higher expected company performance through in-creased managerial quality. Since the quality of the CEO is not directly observable, companydirectors will infer the quality of a CEO from his/her past performance. A CEO will be replaced

6

if the realized performance is sufficiently low and the expected benefit of a turnover exceeds theexpected cost. More precisely, the resulting costs from the turnover have to be more than offsetby the quality differential separating the new and the incumbent CEO. Following this argument,investors should interpret the firing decision by the board as a positive signal about the quality ofthe appointed CEO and the value of the firm.

Under the scapegoat hypothesis based on Holmström (1979), Shavell (1979), and Mirrlees (1976)firings of CEOs occur even though all managers are assumed to be identical in terms of quality.The threat of a dismissal merely serves to ensure adequate effort by the incumbent CEO. In equilib-rium, all CEOs provide the same effort and low performance arises just by chance. In case of poorperformance, the board of directors will fire a CEO just to maintain the threat of dismissal therebycreating the incentive to supply the optimal level of effort. Since in this model the poor performanceleading to the dismissal of a CEO is simply the result of luck and not poor managerial quality orlack of effort, the fired CEO can be viewed as a scapegoat. Also in this case, it is conceivable thatinvestors will interpret the dismissal of the CEO as a positive signal that testifies the responsibilityof the board in providing adequate effort incentives for CEOs.

Since the majority of empirical studies (cf. Table 1, column 9) associates a positive stock-pricereaction with the announcement of a CEO dismissal, thus providing supportive evidence for thetheoretical predictions, we formulate our second hypothesis as follows.

Hypothesis 2: The announcement of a forced CEO turnover yields positive abnormalreturns.

In case of a voluntary retirement, a responsible board of directors should appoint the managerwith the highest quality as CEO successor. However, this does not imply a specific quality differ-ential of the new CEO over the departing one. Further, voluntary departures due to the age ofthe CEO can be anticipated, which should suggest small price reactions. Negative returns couldoccur if the departure of the incumbent CEO is associated with a loss of valuable company-specifichuman capital or if the voluntary decision to leave the company is motivated by superior (negative)information of the departing CEO about the future development of the company. The empirical lit-erature offers a very mixed picture concerning the impact of voluntary CEO turnovers on abnormalreturns. While the majority of papers do not find significant abnormal returns (cf. Table 1, column10), Adams and Mansi (2009) and Hillier and McColgan (2005) report positive and significant ARsand (Neumann and Voetmann, 2005) negative and significant ARs. While in view of the theoryand previous empirical evidence, we expect voluntary retirements to cause smaller stock-price reac-tions than forced departures, we do not see compelling theoretical arguments to formulate a specifichypothesis.

2.3 Performance

Previous studies show an inverse relation between the stock performance and the probability ofa CEO turnover (e.g. Warner, Watts, and Wruck, 1988; Weisbach, 1988, among others). In this

7

paper, we investigate the impact of the company’s prior performance under the departing CEO(which can be seen as a proxy for his/her skills) on the abnormal stock returns in the surroundingsof the announcement of the CEO departure. When an underperforming CEO leaves the company,we expect shareholders to benefit from the turnover.

Hypothesis 3: In case of prior underperformance of the stock relative to the marketindex, the announcement of a CEO turnover yields positive abnormal returns.

In case of prior overperformance, the turnover announcement is expected to cause negativeabnormal returns because the value of all future projects should be reduced to take into accountthe departure of a talented and successful CEO.

Hypothesis 4: In case of prior overperformance of the stock relative to the market index,the announcement of a CEO turnover yields negative abnormal returns.

2.4 Combinations of Turnover Characteristics

In this paper, we consider the three above variables (successor type, departure type, and com-pany’s prior relative stock performance) not only as stand-alone characteristics but also in con-nection with each other. More precisely, we investigate whether the interaction of those variablesconveys additional valuation-relevant information. While we refrain from discussing all possiblecombinations to be constructed by pairs and triples of those variables, some considerations regard-ing specific intersections are useful.

First, based on the previously mentioned theoretical considerations and the available empiricalfindings, the combination of forced turnovers with outside successions - the two characteristics thatin previous studies are found to deliver the highest abnormal returns - is expected to yield positiveabnormal returns.

Hypothesis 5: The announcement of a forced turnover with an outside successor yieldspositive abnormal returns.

Second, we challenge the notion that forced turnovers per se represent positive news to share-holders on average. We believe that forced turnovers have to be examined in connection with theskills of the departing CEO (proxied by the company’s relative stock performance). In particular,we expect to observe positive abnormal returns for forced turnovers of underperforming managersand negative abnormal returns for forced departures of overperforming managers. While the firstevent reflects a wise decision by the company’s board of directors, the latter does not.

8

Hypothesis 6: The announcement of a forced turnover of an underperforming manageryields positive abnormal returns.

Hypothesis 7: The announcement of a forced turnover of an overperforming manageryields negative abnormal returns.

2.5 Trading Volume

Analyzing trading volumes in addition to abnormal returns can likely add an important dimen-sion to the analysis of CEO turnovers. As noted by Beaver (1968), trading volume of a given securityindicates a lack of consensus among investors regarding the price of that security. Along these linesHolthausen and Verrecchia (1990) conclude: “If one defines information content as the ability of aninformation signal to alter investors’ beliefs, evidence on volume reactions is as relevant for assess-ing information content as evidence on unexpected price changes.” To derive testable hypothesesrelated to the trading volume, we follow Kim and Verrecchia (1991b). In their theoretical study7

on the relation between abnormal returns and abnormal trading volumes after the release of publicinformation, they argue as follows: “Volume reflects the sum of differences in traders’ reaction; thechange in price measures only the average reaction. As a result, volume is proportional to both theabsolute price change and the measure of differential precision”. The first part of their conclusionmotivates us to state the following hypothesis:

Hypothesis 8: The higher the absolute abnormal returns caused by the announcementof a CEO turnover, the higher the abnormal trading volume.

Further, we conjecture that out of the different subsamples of CEO turnovers, the one thatconveys the highest degree of surprise will likely relate to forced turnovers of overperforming CEOs.

Hypothesis 9: The announcement of a forced turnover of an overperforming managertriggers the highest abnormal trading volume.

2.6 Operating Performance

If we assume markets to be efficient, the abnormal returns in the surrounding of a CEO-turnoverannouncement date reflect investors’ perceptions about fundamental changes in the value of thecompany. From finance 101, we know that the fundamental value of a company results from thesum of all the expected future free-cash flows discounted by the appropriate risk-adjusted interest

7Other papers that address in a theoretical framework the trading volume around news releases include Holthausenand Verrecchia (1990), Kim and Verrecchia (1991a, 1994), and Demski and Feltham (1994).

9

rate. Thus, changes in the value of a company either reflect changes in the relevant discount rateor changes in the expected future cash flows (or both). If we rule out that a CEO turnover has animpact on the relevant discount rate,8 the different company valuation before and after the CEO-turnover announcement will solely reflect changes in the expected cash flows and thus in the futureoperating performance of a company. Based on this line of reasoning, we formulate the followinghypothesis.

Hypothesis 10: The higher the abnormal returns caused by the announcement of a CEOturnover, the greater the increase in the future operating performance.

3 Data

3.1 Sample Construction

The data investigated in this paper comprises CEO turnovers of companies in the Swiss Perfor-mance Index (SPI) between 1998 and June 2009. The sample is hand-collected and initially consistsof 347 turnovers at 184 companies.

To cover all CEO turnovers, we apply the following procedure. First, the complete list of CEOchanges is obtained by collecting the annual reports of all companies included in the SPI sinceJanuary 1998. Second, to verify the event and obtain the exact CEO-turnover announcement date,the following sources are screened: (i) ad-hoc disclosures at the SIX Swiss Exchange, (ii) articles inleading Swiss financial and business newspapers (in particular, “NZZ”, “Finanz und Wirtschaft”,and “Handelszeitung”), (iii) company-specific news provided by Bloomberg, (iv) company Internetsites, and (v) selected Internet news sites.9

To be included in the final sample, we require CEO-turnover events to cumulatively satisfy thefollowing criteria: (i) The date of the announcement, i.e. the first day investors can trade on theCEO-turnover information, has to be identifiable (this also includes the information on whetherthe announcement was made before or after the closing of the market); (ii) The relevant detailsregarding the departing and the incoming CEO (age, succession type, and successor origin) have tobe known; (iii) There must be no confounding events, such as earnings or dividend announcementsor mergers and acquisitions, in a three-day time period around the turnover announcement; (iv)The CEO turnover may not be directly related to the takeover of the company; (v) Furthermore,there has to be a sufficiently long stock-price history before the CEO-turnover announcement dateand the stock must be traded at least at 100 days (40%) in the estimation period to guarantee areasonably accurate estimation of the market model.

8This would be the case if the average successor reduces the systematic risk of a company or has an impact on themarket premium. An analysis of betas before and after the event date rules out the first hypothesis and the latterappears highly implausible.

9Internet sites include www.news.ch, www.swissinfo.ch, and http://moneycab.presscab.com.

10

The number of events that had to be excluded due to the criteria (i) to (v) are reported in Table 2.

[Table 2 about here]

3.2 Explanatory Variables

For all CEO-turnover events, we gather detailed information regarding the company, the de-parting CEO, the new CEO, and the turnover characteristics. The main explanatory variables inthis study are the successor origin (inside vs. outside successions), the departure type (forced vs.voluntary departures), and the prior performance relative to a broad stock-market index.

In the related literature, the classification of the successor origin is measured by applying twoalternative rules. According to the first rule, the new CEO is classified as an outsider if the ap-pointment as CEO occurs on the same date as he/she joined the company; All other successors areclassified as insiders. According to the second rule, the new CEO is considered as an outsider ifhe/she has a working history in the relevant company of less than a year (see e.g. Parrino, 1997). Inthis paper, successors are classified as insiders or outsiders according to the former rule. However,applying the latter method does not qualitatively alter the major findings of the paper. In oursample, 75 CEO turnovers (36% of the total sample) are classified as outside successions and 133turnovers (64% of the total sample) are classified as inside successions.

The division into forced and voluntary departures is carried out following the methodology of De-nis and Denis (1995). Based on the information provided by diverse media reports, such as leadingSwiss financial and business newspapers, ad-hoc news, and company statements, a CEO turnoveris classified as forced if it is accompanied by an internal conflict with the Board. In those cases inwhich the turnover cannot be directly assigned to the sample of forced or voluntary turnovers onthe basis of the available data, we apply the following decision scheme: If the departing CEO is notover 64 years old and the new appointed CEO is an outsider, the CEO turnover is assigned to thesubsample of forced turnovers. By applying this procedure, the sample of forced turnovers consistsof 60 events (29%) and the sample of voluntary turnovers includes 148 events (71%).

The companies’ relative stock performance is calculated against the Swiss Performance Indexover the same time period that is used to estimate the market model, i.e. over a 250 trading-dayperiod from day −260 to day −11 prior to the CEO-turnover announcement. Depending on the signof the prior relative performance, the event is assigned either to the over- or the underperformingsubsample: 118 turnovers (57%) are preceded by prior underperformance and 90 turnovers (43%)prior overperformance.

Table 3 provides a breakdown of the final sample by year and turnover characteristics. Whilethere seem to be an overall increase in the total number of CEO turnovers over the years, thisdevelopment is far from being steady and smooth.


11

Figure 1 is a Venn diagram that splits the sample according to the main CEO-turnover char-acteristics (together with their intersections): outside successions, forced departures, and turnoversof underperforming managers. It is worth noting that the percentages of outside successions andforced departures are, with 36% and 29% of the sample respectively, higher than those reported inearlier studies. For instance, Adams and Mansi (2009) report for the period between 1990 and 200029.4% of outside replacements and 19.6% of forced turnovers; Parrino (1997) classifies only 15%of all turnovers in his sample as outside successions and only 13% as forced departures; Finally,Clayton, Hartzell, and Rosenberg (2005), examining the impact of CEO turnovers on stock-pricevolatility, classify 20.6% as outside successions and 17.4% as forced departures.

[Figure 1 about here]

3.3 Control Variables

Abnormal stock returns induced by a turnover announcement might be influenced by a numberof variables besides the origin of the new CEO, the performance of the departing CEO, and theturnover type. For instance, a blue-chip stock might react stronger to a CEO-turnover announce-ment because investors follow the company news more attentively or the media offer a strongercoverage. Conversely, small companies might get less attention because they are often owned byfamilies, weaker media and analysts’ coverage or are simply too risky for investors to take actionafter a turnover announcement. To account for the existence of such effects related to the size of acompany, we include as a control variable in our regressions the logarithm of company’s total assets(SIZE) which we obtain from Datastream.

Other variables that will likely play an important role in determining the magnitude of thestock-price impact are the age of the departing and the appointed CEO. If the incumbent CEOis close to retirement age, his/her departure might be anticipated and thus lack a strong surpriseeffect. In this case, the age of the departing CEO will have a dampening effect on the news impact.The appointment of a young CEO which is relatively unknown and potentially less experiencedcould also have a material effect on the stock-price reaction. To control for age-related effects, weinclude in the cross-sectional regressions the variables AGEDEP and AGEINC, i.e. the logarithmof age of the departing and incoming CEO, respectively.

12

4 Empirical Analysis

4.1 Abnormal Returns

In order to investigate the impact of CEO-turnover announcements on stock prices, we applystandard event-study methodology. As usual, the tests rely on the assumption of market efficiency,i.e. that stock prices reflects all relevant information and thus quickly incorporate the effect relatedto CEO-turnover news. Consequently, we choose a short-term event window to measure the impacton stock prices. Following Brown and Warner (1985) and McWilliams and Siegel (1997) this eventwindow should be long enough to capture the impact of the event, but short enough to minimize theinfluence of confounding effects unrelated to the event. Consequently, we choose to employ differentevent windows ranging from one to five days. To account for potential information leakage, we letsome event windows start before the CEO-turnover announcement date.

In accordance with the bulk of the literature on short-term event studies, we calculate abnormalstock returns in the event window by subtracting from realized stock returns the normal returnsobtained by the market model. The parameters of the market model are estimated over a 250trading-day period ending 11 days before the CEO-turnover announcement date. To make surethat the findings of the paper are not driven by inaccurate parameter estimates, the estimation ofthe market model is performed both by simple OLS and robust linear regressions (Huber, 1973).The latter approach is an iteratively re-weighted least squares algorithm (Holland and Welsch,1977). In particular, in each iteration the weights are calculated by applying the bisquare functionto the residuals from the previous iteration. Since the results of the two estimation procedures donot lead to qualitatively different results, for the sake of brevity, only the robust regression resultsare reported in the paper. In addition to measuring the magnitude of mean and median abnormalreturns, it is critical to determine their statistical significance. For this purpose we employ theStandardized Cross-Sectional test by Boehmer, Musumeci, and Poulsen (1991) (henceforth simplydenoted by Boehmer test). In addition, we also consider the Wilcoxon Signed Rank test by Wilcoxon(1945) (henceforth simply denoted by Wilcoxon test) which is a non-parametric test and thus doesnot rely on specific assumptions about the distribution of stock returns.

We start the empirical investigation of this paper by measuring abnormal returns (ARs) forthe whole sample and for selected subsamples defined by the key turnover characteristics: successororigin (insider or outsider), departure type (forced or voluntary), and CEO’s prior performance (un-derperformance or overperformance). Table 4 reports both average and median abnormal returnstogether with test statistics obtained by the Boehmer, Musumeci, and Poulsen (1991) test and thenon-parametric Wilcoxon Signed Rank test by Wilcoxon (1945).10

When considering the results related to the whole sample (Panel A), we observe that for allevent windows considered, the average and median abnormal returns are positive (the only excep-tion is the median AR two days before the event). In the period [-2 0] both the Boehmer and the

10While standardized residuals from the market model - and not abnormal returns - are used in the calculationof the test statistics of Boehmer, Musumeci, and Poulsen (1991), in Table 4 we still report mean abnormal returnsbecause of their easier interpretation.

13

Wilcoxon test detect significant ARs at the 10% level with a mean ARs of +0.74% and a medianARs of +0.32%. The period [-1 1] is significant at the 5% level for the Boehmer test with a mean(median) AR of 0.65% (+0.35%). The mean AR of +0.42% on the announcement day is significantat the 10% level for the Boehmer test. Overall, CEO turnovers seem to convey good news forinvestors on average. This result finds grafical support in the development of the cumulative ARsdisplayed in Figure 2 (a).

If we consider the subsample of turnovers with outside successors (cf. Table 4, Panel B), abnor-mal returns become larger and more significant. For instance, mean ARs for the periods [-2 0], [-10], and [-1 1] are all significant at the 5% level with values of +1.85%, +1.41% and +1.58%, respec-tively. By comparing in Figure 2 (b) the ARs of outside successions to the ARs of inside successionsit becomes evident that the former convey positive news about the value of the company whereasthe latter do not.11 Furthermore, the difference in the cumulated AR of 1.96% for the two samplesover the window [-3 0] is significant at the 10% level (see Table 5). The findings validate Hypothesis1 of this paper and conform to the theory that an outside candidate for the CEO position mustoffer distinguishable better qualities than an inside candidate to be appointed CEO.

According to Hypothesis 2, we expect the stock price to rise following the decision to fire aCEO. Also this hypothesis finds support in the event-study results reported in Table 4, Panel Cand depicted in the ARs evolution in Figure 2 (c). The mean AR for the subsample of forcedturnovers are significant at the announcement day and for all event windows longer than one day.We observe that the mean abnormal return is largest for the window [-2 0] with a value of +2.74%.By comparing in Figure 2 (c) the development of ARs for forced vs. voluntary turnovers one canvisually capture the striking difference in the pattern of the two lines: increasing in correspondenceof forced turnovers and very close to zero for unforced turnovers. For all tested event windows,ARs following unforced turnovers are found to be insignificant at all conventional confidence levels(results reported in Table B.1 in the Appendix). Table 5 shows that the difference in means of3.12% between forced and voluntary turnover over the event window [-3 0] is significant at the 5%level.

The impact of the departure of underperforming CEOs is shown in Table 4, Panel D. Strikingly,all the mean ARs for the event windows [-2 0], [-1 0], [-1 1], and 0 are statistically significant atthe 1% level with values as high as +1.94%, +1.56%, +1.62%, and +1.25%, respectively. A verysimilar picture arises when focusing on median ARs and considering the Wilcoxon test. Accordingly,Hypothesis 3 stating that the information of the departure of underperforming CEOs will triggersignificantly positive abnormal returns finds strong empirical support in our sample. Conversely,ARs related to departing CEOs with positive prior performance are found to be negative but in-significant (results reported in Table B.1 in the Appendix). Furthermore, we observe in Table 5 thatthe difference in means of 2.96% over the window [-3 0] is significant at the 1% level. Consequently,also Hypothesis 4 is confirmed in our sample. A direct comparison of the impact of the turnoverof under- and overperforming CEOs is depicted in Figure 2 (d) with a clear upward trend for theformer and a downward trend for the latter.

The results obtained for outside successions and forced turnovers are in line with the findingsreported by other authors in in the related literature (cf. Table 1). For both subsamples, positive

11As reported in Table B.1 in the Appendix, ARs of inside successions are very close to zero and insignificant.

14

and significant abnormal returns could be detected. Conversely, voluntary turnovers and insidesuccessions induce minor and insignificant stock-price reactions. Not reported in previous work,the impact of a CEO turnover on stock returns depends on the performance of the departing CEO.When underpeforming (overperforming) CEOs leave the company, the stock experiences a signifi-cantly positive (negative) price shock.

The results for the samples restricted to one turnover characteristic are displayed in a compactform in Panel A of Table 7, where the ranking of ARs are based on the return period [-3 0]. Weobserve that forced turnovers, underperforming CEOs, and outside successors generate the largestand most significant ARs with respective values of +2.94%, +2.00%, and +1.98%. Conversely, thesmallest AR is associated with overperforming CEOs. It amounts to −0.96% but is not statisticallysignificant.

At this point, it is natural to extend the current analysis and define further subsamples bycombining different turnover characteristics. The goal of this exercise is to identify certain CEO-turnover constellations that convey either particularly positive or negative news to shareholders. Ina first step we calculate ARs for all 12

((

32

)

· (2 · 2))

possible subsamples (by using a [-3 0] event

window) that can be constructed by pairwise interrelating the three selection criteria (turnover type,successor origin, and prior CEO performance) and rank them accordingly (cf. Panel B of Table7). Some results deserve our attention. First, the subsamples obtained by the pairwise intersectionof Outsider, Forced, and Underperformance generate the largest ARs: +6.71% (t-value = 2.68),+5.73% (t-value = 3.74), and +3.84% (t-value = 2.90) for Outsider & Forced; Forced & Under-performance; and Outsider & Underperformance, respectively. Even when considering alternativeevent windows (cf. Table 4, Panels E and H) those pairs of characteristics generate the strongest(and most significant) price reactions. According to Hypothesis 5 of this paper, the combinationOutsider & Forced should generate positive ARs. Since this subsample is associated with the largestand most significant ARs, the data support Hypothesis 5. Second, it is of special interest to observethat forced turnovers rank both 2nd and 12th (last) in this list. In particular, when consideringforced turnovers of underperforming CEOs a mean AR as large as +5.73% is achieved (associatedwith a t-value of 3.74). Conversely, when forced turnovers of overperforming CEOs are considered,ARs become negative! This finding is depicted in Figure 2 (f) and confirmed by alternative eventwindows in Table 4, Panels G and H (although the negative ARs associated with forced turnoversof overperforming CEOs are not statistically significant at low confidence levels). However, the re-ported results are further supported by the large and at the 1% level significant difference of 7.97%in the mean AR over the window [-3 0] reported in Table 6. This finding suggests that shareholdersassess the quality of the firing decision by the board of directors by considering the quality andskills of the departing CEO. If the relative stock performance under the departing CEO is positive,shareholders seem to disfavour the decision and adjust downward their estimates about the valueof the company. The double edged impact of forced turnovers is also apparent in Table 6, wherewe observe on the one hand mostly significant (9 out of 11 comparisons) negative differences inmeans of the other subsamples against the sample of forced turnovers of underperforming managers(FOR*UND) and, on the other hand exclusively positive differences against the sample of forcedturnovers of overperforming managers (FOR*OVE), whereof four are significant at least at the 10%level. The sole exception of a larger cumulated AR than the one calculated for forced turnovers

15

of underperforming managers is the sample of outside successors combined with a forced turnover.The just described empirical results seem to support Hypothesis 6 and Hypothesis 7 of this paper.Finally, it is interesting to note that outside CEOs trigger positive abnormal returns only if theprevious CEO was fired (Figure 2 (e)), not if he/she voluntary left the company.

If we define subsamples based on the interaction of all three turnover characteristics, we obtaina total of eight subsamples (23). Table 7, Panel C ranks those subsamples by their ARs in the eventwindow [-3 0]. Not all too surprisingly, the subsample with the most significant ARs (Boehmert-value = 4.29, Wilcoxon z-value = 3.05) is characterized by forced departures of underperformingCEOs substituted by firm outsiders. The mean (median) ARs over the window [-3 0] of the 14events falling in this category amounts to +10.61% (+9.41%). The most negative AR (−2.82%)corresponds to the intersection of forced departures with inside successions at overperforming com-panies. The columns R1, R2, and R3 in Table 7, Panel C evidence an interesting pattern which canbe interpreted as follows. First, the prior performance (R1) is the single most important variable indeciding whether a turnover represents good or bad news. Turnovers at underperforming compa-nies are always associated with positive ARs, while turnovers at overperforming companies triggernegative ARs. Second, column R2 impressively shows the amplifying effect of forced turnovers incombination with under-, and overperforming companies. Third, in column R3 subsamples withoutside successors rank above the corresponding subsamples with inside successors. Thus, whileoutside successors are ceteris paribus viewed more positively than inside successors, the other vari-ables (departure type and prior performance) seem to be more relevant in judging the value of aCEO-turnover event.




4.2 Cross-Sectional Regressions

In this subsection we evaluate the cross-sectional information content of CEO turnovers byregressing ARs against a set of explanatory variables. More precisely, we regress ARs in the eventwindow [-2 0] on an outside-succession dummy (OUT) a forced-turnover dummy (FOR), priorperformance under the departing CEO (OVE)12, size of the company measured as the logarithm oftotal assets (SIZE), the age of the departing CEO (AGEDEP), as well as the age of the incomingCEO (AGEINC). The regressions results are reported in Table 8. Regressions 1 to 3 are univariateregressions of ARs against a single turnover characteristic and thus simply replicate some of theresults obtained by analyzing ARs of selected subsamples. For instance, the sum of the constant,

12The variable PERF denotes both a continuous variable and a dummy variable, taking on the value of one if thecompany experienced prior relative overperformance and zero otherwise.

16

0.0011, and the coefficient of the variable OUT, 0.0173, in Table 8 is equivalent (except for roundingerrors) to the mean [0 2]-AR for the subsample of turnovers with outside successors, 0.0185, in Table4, Panel B. Nonetheless, performing cross-sectional regressions offers two decisive advantages: First,it allows to measure the impact of continuous variables (and not only dummy variables); Second,it allows to combine the effect of multiple variables on ARs and to isolate the impact of each them.Not surprisingly, the coefficients related to outside successions (OUT), forced departure (FOR), andoverperformance (OVE) are all found to be significant in the univariate regressions. The size ofthe company has a slightly negative but insignificant effect on ARs. The coefficient related to theage of the departing manager (AGEDEP) is negative and weakly significant and the one relatedto the age of the incoming CEO (AGEINC) is positive but not significant. Importantly, in themultivariate regression with all six explanatory variables (Table 8, Regression 7) the coefficientsand significances of the variables OUT, FOR, and PERF experience remarkably small changes. Inthe multivariate regression, almost six percent of the cross-sectional variance can be traced back tothe six explanatory variables.


In a next step, we investigate whether these results are consistent over different return periods.For that purpose, we regress in Table 9 ARs of six event windows against all variables found to besignificant in Regression 7 of Table 8. By looking at the results, it stands out that the signs of thecoefficients remain unchanged and the significances are reasonably consistent across the performedregressions. The variables OUT, FOR, and OVE are always significant, except for OUT in theperiods [-1 1] and [-2 1]. The coefficient of SIZE is always negative but in solely two cases ([-3 1]and [-1 1]) a weak significance at the 10% level is detected. The results from Table 9 reinforce ourview that the successor origin, the turnover type, and the prior CEO performances are key variablesin explaining cross-sectional variations in abnormal returns and determining whether a turnover isgood news or bad news to shareholders.


Motivated by the previous results about the impact of forced turnovers in dependence of theCEO’s prior performance, we add an additional dummy variable, FOR×OVE, to capture thisinteraction. As can be seen in Panel A of Table 10, for both event windows considered FOR×OVEis significant at the 1% level and the adj. R2 rise to 0.1363 for the [-3 0] event window and 0.0959for the [-2 0] event window. The results reported in Panel B of Table 10 are obtained by definingOVE as the realized relative performance during the estimation period (and avoid to transform itinto a dummy variable). Also under this specification FOR×OVE remains significantly negative atthe 1% level and the adj. R2 increases to 0.2184 for the period [-3 0] and 0.2114 for the period [-20].


17

It is of particular interest to examine the impact on shareholder value of those turnovers inwhich the CEOs is forced to leave the company by a deliberate decision of the board of directors.Table 11 presents the results of cross-sectional regressions on the subsample of forced turnovers.In all regressions, the coefficients of the variables OUT and OVE have the expected sign and arehighly significant. In particular, knowing that the performance of a fired CEO was positive inthe year preceding the dismissal induces an average drop in shareholder value of approximately0.8%. The results of Table 11 emphasize once more the importance of the prior performance ofthe departing CEO in assesing the news of a firing. Conversely, they challenge the oversimplifyingview that the dismissal of a CEO represents per se good news. Similarly, the results suggest thatshareholders systematically mistrust the board of directors to act in their best interest when theyfire an outperforming CEO.


4.3 Robustness

In this subsection we intend to check the robustness of our results with respect to (i) econometrictest specifications and (ii) time subperiods. To further ensure the robustness of the findings, we re-examine the event-study results by implementing, in addition to the already employed StandardizedCross-Sectional test by Boehmer, Musumeci, and Poulsen (1991) and Wilcoxon Signed Rank testby Wilcoxon (1945), an additional battery of parametric and non-parametric event-study tests.The parametric methods include the Traditional test by Brown and Warner (1980), the Portfoliotest by Brown and Warner (1980), a test that accounts for first-order autocorrelation in abnormalstock returns by Ruback (1982), and the Standardized-Residual test by Patell (1976). The non-parametric test statistics have the advantage of avoiding to assume a specific distribution for theabnormal returns. The non-parametric test employed in this study include a Bootstrap test, theCorrado Rank test by Corrado (1989), and the Generalized Rank test by Kolari and Pynnonen(2008). Table 12 provide an overview on all test used in this paper and Table 13 illustrates thecomputation of the corresponding test statistics.



One major worry in event studies concerns the risk of detecting spurious abnormal returns. Ineconometrics, the risk of erroneously rejecting the null hypothesis is called error of type I. Theprobability of incurring into this error should be determined by the confidence level, α. However,misspecified test statistics may lead to higher or lower rejection rates of true null hypotheses. To shedlight on this issue and gain insights about the reliability of the statistical inferences in this paper,we perform empirical experiments that evaluate the performance of three very different event-study

18

tests: the Bootstrap method, the Standardized Cross-Sectional test by Boehmer, Musumeci, andPoulsen (1991), and the Wilcoxon Signed Rank test by Wilcoxon (1945). The historical-simulationexperiment uses the daily stock-price histories of 160 SPI companies in the time period from January1990 to June 2009. It consists of two steps. In the first one, we generate 250 random samplesof N pseudo or “fake” CEO turnovers, each one representing a possible empirical sample of CEOturnovers. Clearly, since the turnovers did not actually take place, we know that the null hypothesisof zero abnormal returns is indeed true. Consequently, we expect the event-study tests to rejectthe null hypothesis with frequency α. Each of the 250 samples is constructed as follows: First, werandomly draw with replacement N (N ∈ {10; 20; 50; 100; 200}) securities from the overall sampleof 160 securities. Second, each sampled security gets a randomly drawn pseudo-event day which alsodefines both the estimation and event windows. In particular, in accordance with the event-studydesign employed in the paper, the days −260 till day −11 refer to the estimation window with dayzero referring to the event date.13 In a second step, we conduct for each of the 250 pseudo-turnoversamples a complete event-study test, i.e. we (i) estimate the market model by OLS, (ii) calculateabnormal returns in the event window, and (iii) determine the rejection of the null hypothesis basedon a particular test and confidence level, α. The rejection rates reported in Table 16 represent theaverage number of samples for which abnormal returns are found to be significantly different fromzero divided by the sample size (N). Since for all three models the test statistics reject the nullhypothesis at approximately the significance levels, the Type I error rate is acceptable and the testsdo not seem prone to detect spurious abnormal returns. Interestingly, the rather involved Bootstrapevent-study test does not seem to deliver better results than the other two test statistics.


To asses the power of the tests, i.e. the capability of rejecting falls null hypotheses, we repeatthe above simulation experiment by artificially introducing in the event window of each pseudoCEO turnover abnormal returns in the range of −3% to +3%. In Figure 3 the empirical rejectionrates are plotted against the level of abnormal returns used to contaminate the stock-price returnsin the event window. As expected, the larger the absolute value of the contamination, the higherthe rejection rates. For sufficiently large abnormal returns the probability of rejecting the null ofzero abnormal returns reaches 100%. Figures 3 (a)-(d) differ with respect to the size of the CEO-turnover sample and highlight the fact that the sample size is critical for obtaining a high testpower. By comparing the shapes of the rejection curves, the bootstrap test appears to be slightlyoutperformed by the other two test statistics.


13To estimate the market model for the event study, we require the companies to be a member of the SwissPerformance Index during all 261 days prior to the randomly drawn event date. If this condition is not fulfilled anew pseudo event day is randomly drawn for this security. This cycle continues until a complete history of securityreturns is generated.

19

It is now interesting to investigate the robustness of the basic results on CEO-turnover subsam-ples by employing all event-study tests previously introduced. Table 14 reports t-values associatedwith the different tests, event windows, and CEO-turnover subsamples. It is reassuring to observethat the significant results obtained earlier for outside successions, forced departures, and priorunderperformance are strongly supported by the additional test methods. Furthermore, we obtainsignificant negative ARs for the subsample of overperforming companies for five out of six parametrictest methods (Table 14, Panel E). Similar conclusions can be drawn for the subsample constructedby combining outside successions with forced and voluntary departures as well as forced departureswith the performance of the departing CEO. In particular, we emphasize that the finding of signif-icant negative abnormal returns for forced turnovers at overperforming companies (Hypothesis 7)and significant positive returns at underperforming companies (Hypothesis 6) is robust with respectto alternative test specifications.


Finally, for testing robustness over time, we divide the sample into two subperiods of equallength (1998-2003 and 2004-2009) and perform cross-sectional regressions on both set of abnormalreturns. It is reassuring to observe that regardless of the time subsamples and event windows allthe coefficients in Table 15 have the expected sign. Having said that, the results related to the sub-period 1998-2003 show a higher number of significant coefficients and are characterized by higherR2 and thus a better fit. However, for the event window [-3 0], even in the latter period the dummyvariables FOR and FOR×OVE are statistically significant at the 5% confidence level. Therefore,the dependency of the announcement impact of forced turnovers on prior performance is a featurethat is present in both subperiods.


4.4 Trading Volume

As mentioned earlier in the paper, trading volume could convey even more information aboutthe relevance of a CEO turnover to investors than abnormal returns. For this reason, we conductanother event-study based on the stock trading volume (TV ), i.e. the number of traded shares(data item “Turnover by Volume” in Datastream). To provide an immediate interpretation of theimpact of CEO-turnover news on the trading volume, we first consider the following standardizedmeasure of abnormal trading volume:

STV Ait =TVitTV Nit

− 1, (1)

20

where TVit is the realized trading volume of company i at time t and TV Ni is the normal, i.e.the expected, trading volume which is measured in accordance with a constant-mean model asthe average trading volume between day −60 and day −11: TV Ni = 1/50 · ∑−11

t=−60 TVit. STV isparticularly easy to interpret because it simply represents the percentage increase in trading volumerelative to a 50-day benchmark period before the event date.

Figure 4 and Figure 5 depict the daily average and cumulative average standardized abnormaltrading volumes. All graphs show a distinct increase of the average STV A on the day of theCEO turnover announcement and the following one. In Figure 4 (a), we see that for the totalsample the average trading volume on the event day is +129.97% larger than the average tradingvolume in the estimation period. By examining Figures 4 (b)-(f), it is apparent that the changes intrading volumes strongly depend on the characteristics of the turnover event. For instance, forceddepartures (graph (c)) have a much larger impact on trading volume than voluntary departures(+196.37% vs. +103.04%). Surprisingly, the largest average STV A is caused by forced turnoversat underperforming companies (+259.64%) and not - as conjectured in Hypothesis 9 of this paper- by forced turnovers of overperforming CEOs (+78.87%).



By following the previous literature on trading-volumes in correspondence of news releases(Ajinkya and Jain, 1989; Cready and Ramanan, 1991; Campbell and Wasley, 1996), we performan event study based on a constant-mean model applied to the logarithm of the trading volume(LTV = ln(TV )). As mentioned in the above articles, working with log-transformed trading vol-umes is preferable because its empirical distribution is closer to the normal and allows for moreaccurate statistical inferences.14 Similar to abnormal stock returns, abnormal trading volumes ofcompany i on day t, LTV Ait , must be calculated as the difference between realized and normal LTV .In accordance with the constant-mean model, the normal daily LTV in the event window is cal-culated as the average LTV from day −60 to day −11 prior to the turnover announcement date:LTV Ni = 1/50 ·∑−11

t=−60 LTVit.15 Finally, abnormal LTV for each event are obtained by subtracting

normal LTV s from the realized ones: LTV Ait = LTVit − LTV Nit .Table 17 presents the test statistics of the event study. In the total sample (Panel A) the increase

in trading volume before the CEO-turnover announcement indicates the existence of informationleakage. Significant abnormal trading volumes are detected for the periods [-2 2], [1 3], as wellas day 0 and day +1. The trading-volume effect is particularly pronounced for the subsamples of

14As an alternative trading-volume metrics, we also employ the relative trading volume (RTV ), defined as the ratioof the logarithm of traded shares and shares outstanding (data item “Number of Shares in Issue” in Datastream). Ina formula: RTV = ln ((100 · V/NOSH) + c), where TV denotes the number of traded shares, NOSH is the numberof outstanding shares, and c = 0.000255 denotes a constant which precludes taking a logarithm of zero when noshares are traded (Cready and Ramanan, 1991). In general, the results obtained with this alternative methodologydo not materially differ from those described and reported in the paper.

15Days in which no trading was reported are not considered when calculating this average.

21

forced departures in Table 17, Panel C (similar to Cools and van Praag (2007)), outside successorswith forced departures (Panel E), and forced turnovers of underperforming managers (Panel H).


In Table B.2 of the Appendix we show that an increase in trading volume is also observed forinside successions, voluntary departures, and departures of overperforming managers. The resultsof those subsamples show how trading volume can capture the trading relevance of news in cases inwhich we do not observe significant abnormal returns (cf. Table B.1 in the Appendix).

By regressing in Table 18 the abnormal trading volume on turnover characteristics only thedeparture type (forced vs. voluntary) is found to be significant.


To address Hypothesis 8 of this paper, we regress absolute abnormal returns against abnormaltrading volumes, LTV A. For all event windows considered, we find a highly significant slope coef-ficients (cf. Table 19). The strongest relation refers to the event day with a t-value of 6.4273 andan adj. R2 of 0.1589. These results support the hypothesis that belief updates by investors, asmeasured by absolute price changes, generate trading.


4.5 Operating Performance

To round off the analysis on CEO turnovers we test whether they have a longer-term impacton the operating performance of a company. This longer-term effect is measured by the operatingreturn on asset (OROA). We calculate the operating return on total assets in period t, OROAt, asthe ratio of operating income (Datastream item 137) and book value of total assets (Datastreamitem 392). We do not consider the return on assets in the financial year of the CEO turnover becauseit is affected by both the old and the new CEO.16 Therefore, OROA0 indicates the operating returnon assets of the financial year ending before the CEO turnover and OROA+1 denotes the returncorresponding to the first full-time year under the lead of the new CEO. We follow the methodologyproposed by Barber and Lyon (1996) and calculate normal, or expected, operating performance,

16It is often argued that the departing CEO has incentives to artificially increase the reported earnings in a lastattempt to keep his/her position and that the newly appointed CEO has incentives to reduce reported earnings tocredit poor performance to the predecessors and augments the chance that subsequent good performance will beattributed to him/her. The second mentioned discretionary behavior that a new CEO has the incentive to decreaseearnings in the year of his appointment is known as taking a so-called “earnings bath” (Elliott and Shaw, 1988;DeAngelo, 1988; Murphy and Zimmerman, 1993; Pourciau, 1993; Wells, 2002).

22

OROANit , as the sum of the lagged company’s OROA, OROAi,t−1, and the change in the median

industry OROA from year t − 1 to year t, ∆OROAI

it (without considering company i). Thus, theabnormal operating performance, OROAAit, accounts for industry effects and is calculated as

OROAAit = OROAit −OROANit , with (2)

OROANit = OROAi,t−1 + ∆OROAIit. (3)

The sample used in this analysis (N = 136) is smaller than the one used for the event studies onabnormal returns and trading volumes (N = 208) because we need three (two) years of accountingdata before (after) the financial year of the CEO turnover to perform the analysis and the lack ofdata for some companies restricts our sample. Again, year 0 and year +1 refer to the last full yearunder the departing CEO and the first full year under the new CEO, respectively.

Median OROAs for an industry are calculated starting from a sample of 246 companies in theSPI index (both present and past constituencies of the index). First, we consider all companiesin the same “Sector 4” industry code of Datastream reporting in the same calendar year as thecompany of interest. If less than three companies with available data belong to the same Sector 4,we consider all companies which belong to the larger “Sector 3” industry code of Datastream. Ifeven with this broader industry definition we cannot find at least three companies with availableOROAs, we employ the median change over all companies which report OROAs for the years t− 1and t.

Figure 6 (a) depicts the abnormal OROAs in the financial years around the CEO-turnover an-nouncement date. The inverted tent shape of the graph indicates that CEO turnovers are generallypreceded by deteriorating operating performance and followed by a steady increase in profitability.Hence, CEO turnovers seem to have an overall positive impact on the real operating performanceof a company. This pattern is more pronounced than in the earlier studies conducted by Denis andDenis (1995), Dedman and Lin (2002), and Huson, Malatesta, and Parrino (2004).17 The graphs inFigure 6 tend to mirror the results of the analysis on abnormal stock returns: The subsamples withthe more pronounced changes in OROAA coincide with the subsamples with the largest abnormalstock returns. Particularly large changes in abnormal operating performance are associated withoutside successions (Figure (b)), forced departures (Figure (c)), prior underperformance (Figure(d)), outside successions in connection with forced departures (Figure (e)), and the intersectionof forced turnovers with prior underperformance (Figure (f)). On the contrary, the subsamples ofvoluntary turnovers (Figure (c)), prior overperformance (Figure (d)), outside successions in com-bination with voluntary turnovers (Figure (e)), and forced turnovers with prior overperformance(Figure (f)) do not trigger large changes of the abnormal operating performance.


17Other studies (Dezso, 2007; Fisman, Khurana, and Rhodes-Kropf, 2005; Hillier and McColgan, 2005; Khurana,2001) that examine the impact of CEO turnovers on companies’ operating performance arrive to similar conclusions.

23

Table 20 provides formal econometric tests about the significances of the abnormal OROA overspecified periods.18 Panel A of this table confirms that OROAA significantly worsen in the run-upto the CEO turnover. In the year following the CEO change, OROAAs tend to rise but are notstatistically different from zero. Only over a two-year period following the turnover, the positiveimpact on the operating performance kicks in and leads to high levels of statistical significance.Surprisingly, this operating-performance pattern holds for the majority of subsamples yielding de-creasing OROAAs prior the CEO-turnover year and increasing OROAAs after the turnover. Whenconsidering median changes, the subsamples in which CEO turnovers have a particularly strong andsignificant impact on OROAs are those related to outside successors (Panel B), forced departures(Panel D), prior stock underperformance (Panel F), and the intersection of forced departures withunderperformance (Panel K). For example, the subsample of forced departures yield a mean changein OROAA of +2.92% (t-value = 2.4106) and a median change in OROAA of +1.98% (z-value= 2.6375). In the subsample of forced departures with underperforming companies we measure amean change of +3.56% (t-value = 2.1730) and a median change of +2.61% (z-value = 2.2824).


To test whether different CEO-turnover characteristics have a significantly different impact onthe operating performance, we regress cross-sectional OROAA changes over the time period [1 2]on a set of explanatory variables. While most of the coefficients have the same sign as in thecross-sectional regressions based on abnormal returns (cf. Table 8), only the coefficients related tothe outsiders dummy (OUT) and the company’s relative overperformance (OVE) are statisticallydifferent from zero at the 10% confidence level.19


Finally, we are interested in finding out whether the stock price reaction in the surrounding of theannouncement date correlates with the future development of the company’s operating performance(Hypothesis 10). For this purpose, we regress using robust linear regressions20 OROAA changes overthe period [0 2] (financial years) on the ARs from the period [-3 0] (trading days). Table 22, PanelA reports for the total sample a significant relation between ARs and OROAA changes with acoefficient of 0.1112 (t-value = 2.4758). This positive relation also holds for most subsamples (cf.

18The same analysis was repeated by employing OROA changes as a measure of abnormal operating performance,which implies that the normal OROA for the next year is proxied by the OROA from the previous year, hence wecalculate OROAAit = OROAit − OROAi,t−1. This alternative model specification did not qualitatively alter theresults reported in Table 20.

19When performing the same cross-sectional regressions based on a [0 2] event window, the significances of OVEand OUT vanish and SIZE becomes weakly significant. The latter result could reflect the higher degree of sluggishnessin the restructuring of larger companies.

20For the robust regression we apply the bisquare weighting function as before in Section 4.1 for estimating themarket model for the expected security returns.

24

Table 22, Panel B). A particularly strong connection between ARs and OROAs can be observedfor the subsample of outside successors (β = 0.3991, t-value = 3.88). Overall, it is reassuring (fromthe view point of market efficiency) that the information content of CEO turnovers appears to bereflected both in the short-term stock-market reaction and in the long-term operating performanceof the company.


5 Summary and Conclusions

In this paper we investigate the information content of CEO turnovers in the Swiss stock mar-ket by analyzing abnormal stock returns and trading volumes around the turnover announcementdate. The sample consists of 208 CEO turnovers between January 1998 and June 2009 in compa-nies belonging to the Swiss Performance Index. The results provide strong evidence that selectedcharacteristics of CEO turnovers are determinant in explaining the stock-market reaction.

The relative stock-price performance under the departing manager is found to be the singlemost important variable in assessing the shareholder-value content of a CEO-turnover. In line witheconomic intuition, the departure of outperforming (underperforming) managers represents bad(good) news. Outside successions and forced turnovers yield, in accordance with previous studies,significant positive abnormal returns. Most interestingly, while forced turnovers of managers withpoor prior performance trigger positive and significant abnormal returns, forced turnovers of out-performing managers are associated with negative abnormal returns. Therefore, a forced turnoverdoes not - as often argued in the literature - per se constitute a positive signal to shareholders.On the contrary, shareholders seem to critically assess the quality of the board’s firing decision byconsidering the skills of the departing manager as proxied by the prior relative stock performance.When an outperforming CEO is dismissed or forced to leave, shareholders seem to disesteem theboard’s decision. This finding is confirmed in multivariate cross-sectional regressions and is robustto time subperiods and alternative test statistics.

Trading volume is found to consistently increase for all types of CEO-turnovers. However, thesize of the impact crucially depends on the characteristics of the turnover event, with forced depar-tures causing the largest effect on trading-volume (+196.37%). Furthermore, in line with theoreticalmodels on news releases (e.g. Kim and Verrecchia, 1991b), we observe a statistically significant re-lation between absolute abnormal returns and abnormal trading volumes.

By examining the long-term relation of CEO-turnovers and companies’ operating performance,we report a significant deterioration of the operating performance in the forefront of the turnoverand a significant improvement afterward. The relation between abnormal stock returns aroundthe announcement day and subsequent changes in operating performance is positive and statisti-cally significant. This result indicates that the short-term investors’ reaction to a CEO-turnoverannouncement reflects on average the new long-term operating prospects of that company.

25

References

Adams, J. C., and S. A. Mansi (2009): “CEO turnover and bondholder wealth,” Journal ofBanking and Finance, 33, 522–533.

Ajinkya, B. B., and P. C. Jain (1989): “The behavior of daily stock market trading volume,”Journal of Accounting and Economics, 11, 331–359.

Barber, B. M., and J. D. Lyon (1996): “Detecting abnormal operating performance: Theempirical power and specification of test statistics,” Journal of Financial Economics, 41, 359–399.

Beatty, R. P., and E. J. Zajac (1987): “CEO change and firm performance in large corpora-tions: Succession effects and manager effects,” Strategic Management Journal, 8(4), 305–317.

Beaver, W. H. (1968): “The information content of annual earnings announcements,” Journal ofAccounting Research, 6, 67–92.

Boehmer, E., J. Musumeci, and A. B. Poulsen (1991): “Event-study methodology underconditions of event-induced variance,” Journal of Financial Economics, 30, 253–272.

Bonnier, K. A., and R. F. Bruner (1989): “An analysis of stock price reaction to managementchange in distressed firms,” Journal of Accounting and Economics, 11, 95–106.

Borokhovich, K., R. Parrino, and T. Trapani (1996): “Outside directors and CEO selec-tion,” Journal of Financial and Quantitative Analysis, 31, 337–355.

Brown, S., and J. Warner (1980): “Measuring security price performance,” Journal of FinancialEconomics, 8, 205–258.

(1985): “Using daily stock returns: The case of event studies,” Journal of FinancialEconomics, 14, 3–31.

Campbell, C. J., and C. E. Wasley (1996): “Measuring abnormal daily trading volume forsamples of NYSE/ASE and NASDAQ securities using parametric and nonparametric test statis-tics,” Review of Quantitative Finance and Accounting, 6, 309–326.

Chan, W. (1996): “External recruitment versus internal promotion,” Journal of Labor Economics,14, 555–570.

Clayton, M. C., J. C. Hartzell, and J. Rosenberg (2005): “The impact of CEO turnoveron equity volatility,” Journal of Business, 78, 1779–1808.

Cools, K., and C. M. van Praag (2007): “The value relevance of top executive departures:Evidence from the Netherlands,” Journal of Corporate Finance, 13, 721–742.

26

Corrado, C. J. (1989): “A nonparametric test for abnormal security price performance in eventstudies,” Journal of Financial Economics, 23, 385–395.

Coughlan, A., and R. Schmidt (1985): “Executive compensation, management turnover, andfirm performance: An empirical investigation,” Journal of Accounting and Economics, 7, 43–66.

Coxbill, A. L., L. W. Sanning, and S. Shaffer (2009): “Market reaction to the announcementof a male-to-female CEO turnover,” CAMA Working Paper Series.

Cready, W. M., and R. Ramanan (1991): “The power of tests employing log-transformedvolume in detecting abnormal trading,” Journal of Accounting and Economics, 14, 203–214.

Dahya, J., A. A. Lonie, and D. M. Power (1998): “Ownership structure, firm performanceand top executive change: An analysis of UK firms,” Journal of Business Finance & Accounting,25, 1089–1118.

Dahya, J., and J. J. McConnell (2005): “Outside directors and corporate board decisions,”Journal of Corporate Finance, 11, 37–60.

Danisevska, P., A. de Jong, and M. Rosellon (2003): “Disciplining managers: Evidencefrom two-tier boards,” Working Paper.

DeAngelo, L. E. (1988): “Managerial competition, information costs, and corporate governance:The use of accounting performance measures in proxy contests,” Journal of Accounting andEconomics, 10, 3–36.

Dedman, E., and S.-J. Lin (2002): “Shareholder wealth effects of CEO departures: Evidencefrom the UK,” Journal of Corporate Finance, 8, 81–104.

Demski, J. S., and G. A. Feltham (1994): “Market response to financial reports,” Journal ofAccounting and Economics, 17, 3–40.

Denis, D. J., and D. K. Denis (1995): “Performance changes following top management dis-missals,” The Journal of Finance, 50, 1029–1057.

Dezso, C. L. (2007): “Entrenchment and changes in performance following CEO turnover,” Work-ing Paper.

Dherment-Ferere, I., and L. Renneboog (2002): “Share price reactions to CEO resignationsand large shareholder monitoring in listed French companies,” In: McCahery, J., et al. (Eds.),Convergence and Diversity of Corporate Governance Regimes and Capital Markets. Oxford Uni-versity Press, pp. 297–324.

Elliott, J. A., and W. H. Shaw (1988): “Write-Offs as accounting procedures to manageperceptions,” Journal of Accounting Research, 26, 91–119.

27

Faccio, M., and L. Lang (2002): “The ultimate ownership of western european corporations,”Journal of Financial Economics, 65, 365–395.

Fisman, R., R. Khurana, and M. Rhodes-Kropf (2005): “Governance and CEO turnover:Do something or do the right thing?,” Working Paper.

Furtado, E. P., and M. S. Rozeff (1987): “The wealth effects of company initiated manage-ment changes,” Journal of Financial Economics, 18, 147–160.

Hillier, D., and P. McColgan (2005): “Firm performance, entrenchment and CEO successionin family-managed firms,” Working Paper.

Holland, P. W., and R. E. Welsch (1977): “Robust regression using iteratively reweightedleastsquares,” Communications in Statistics - Theory and Methods, A6, 813–827.

Holmström, B. (1979): “Moral hazard and observability,” The Bell Journal of Economics, 10(1),74–91.

Holthausen, R. W., and R. E. Verrecchia (1990): “The Effect of informedness and consensuson price and volume behavior,” The Accounting Review, 65(1), 191–208.

Huber, P. J. (1973): “Robust regression: Asymptotics, conjectures, and monte carlo,” Annals ofStatistics, 1(5), 799–821.

Huson, M. R., P. H. Malatesta, and R. Parrino (2004): “Managerial succession and firmperformance,” Journal of Financial Economics, 74, 237–275.

Huson, M. R., R. Parrino, and L. T. Starks (2001): “Internal monitoring mechanisms andCEO turnover: A long-term perspective,” Journal of Finance, 56, 2265–2297.

Johnson, W. B., R. P. Magee, N. J. Nagarajan, and H. A. Newman (1985): “An analysisof the stock price reaction to sudden executive deaths: Implications for the managerial labormarket,” Journal of Accounting and Economics, 7, 151–174.

Kang, J. K., and A. Shivdasani (1996): “Does the Japanese governance system enhance share-holder wealth? Evidence from the stock-price effects of top management turnover,” The Reviewof Financial Studies, 9, 1061–1095.

Khanna, N., and A. B. Poulsen (1995): “Managers of financially distressed firms: Villains orscapegoats?,” The Journal of Finance, 50, 919–940.

Khurana, R. (2001): “The performance consequences of CEO turnover,” Working Paper, MITand Harvard University.

Kim, O., and R. E. Verrecchia (1991a): “Market reaction to anticipated announcements,”Journal Financial Economics, 30, 273–309.

28

(1991b): “Trading volume and price reactions to public announcements,” Journal ofAccounting Research, 29(2), 302–321.

(1994): “Market liquidity and volume around earnings announcements,” Journal of Ac-counting and Economics, 17, 41–67.

Kolari, J., and S. Pynnonen (2008): “Generalized rank test for testing cumulative abnormalreturns in event studies,” Working Paper.

LaPorta, R., F. L. de Silanes, and A. Shleifer (1999): “Corporate ownership around theworld,” Journal of Finance, 54(2), 471–517.

Lee, P., and E. H. James (2007): “She’-E-Os: Gender effects and stock price reactions to theannouncements of top executive appointments,” Strategic Management Journal, 28(3), 227–241.

Mahajan, A., and S. Lummer (1993): “Shareholder wealth effects of management changes,”Journal of Business Finance and Accounting, 20(3), 393–410.

McWilliams, A., and D. Siegel (1997): “Event studies in management research: Theoreticaland empirical issues,” The Academy of Management Journal, 40, 626–657.

Mirrlees, J. A. (1976): “The optimal structure of incentives and authority within an organiza-tion,” The Bell Journal of Economics, 7(1), 105–131.

Murphy, K. J., and J. L. Zimmerman (1993): “Financial performance surrounding CEOturnover,” Journal of Accounting and Economics, 16, 273–315.

Neumann, R., and T. Voetmann (2005): “Top executive turnovers: Separating decision andcontrol rights,” Managerial and Decision Economics, 26, 25–37.

Park, S., and M. S. Rozeff (1994): “The role of outside shareholders, outside boards, andmanagement entrenchment in CEO selection,” Working Paper.

Parrino, R. (1997): “CEO turnover and outside succession: A cross-sectional analysis,” Journalof Financial Economics, 46, 165–197.

Patell, J. M. (1976): “Corporate forecasts of earnings per share and stock price behavior: Em-pirical tests,” Journal of Accounting Research, 14, 246–276.

Penman, S. (1982): “Insider trading and the dissemination of firms forecast information,” TheJournal of Business, 55, 479–503.

Pourciau, S. (1993): “Earnings management and nonroutine executive changes,” Journal of Ac-counting and Economics, 16, 317–336.

Reinganum, M. R. (1985): “The effect of executive succession on stockholder wealth,” Adminis-trative Science Quarterly, 30, 46–60.

29

Ruback, R. (1982): “The effect of discretionary price control decisions on equity values,” Journalof Financial Economics, 10, 83–106.

Setiawan, D. (2008): “An analysis of market reaction to CEO turnover announcement: The casein indonesia,” International Business & Economics Research Journal, 7, 119–128.

Shavell, S. (1979): “Risk sharing and incentives in the principal and agent relationship,” Journalof Business, 10(1), 55–73.

Shen, W., and A. A. Cannella (2003): “Will succession planning increase shareholder wealth?Evidence from investor reactions to relay CEO successions,” Strategic Management Journal, 24(2),191–198.

Suchard, J., M. Singh, and R. Barr (2001): “The market effects of CEO turnover in Australianfirms,” Pacific-Basin Finance Journal, 9, 1–27.

Warner, J. B., R. L. Watts, and K. H. Wruck (1988): “Stock prices and top managementchanges,” Journal of Financial Economics, 20, 461–492.

Weisbach, M. (1988): “Outside directors and CEO turnover,” Journal of Financial Economics,20, 431–460.

Wells, P. (2002): “Earnings management surrounding CEO changes,” Accounting and Finance,42(2), 169–193.

Wilcoxon, F. (1945): “Individual comparisons by ranking methods,” Biometrics Bulletin, 1(6),80–83.

Worrell, D. L., W. N. Davidson, P. Chandy, and S. L. Garrison (1986): “Manage-ment turnover through deaths of key executives: Effects on investor wealth,” The Academy ofManagement Journal, 29, 674–694.

Worrell, D. L., W. N. Davidson, and J. L. Glascock (1993): “Stockholder reactions todepartures and appointments of key executives attributable to firings,” The Academy of Manage-ment Journal, 36, 387–401.

30

A Bootstrap Event-Study Test

The bootstrap event-study test is implemented in three steps. First, we calculate an ordinaryt-value for the sample of abnormal returns on day t: t-valuet = ARt/σ(ARt)

√N , where ARt is the

mean abnormal return for day t and σ(ARt) is the cross-sectional standard deviation of abnormalreturns. Second, we generate a distribution of t-values under the null-hypothesis, i.e. we re-sample10, 000 times from the initial sample of abnormal returns by randomly drawing for each event oneabnormal return from the estimation window of 250 days. For each random sample we calculate themean and the standard deviation of the abnormal returns and subsequently the t-valuer, where rstands for random. Third, we calculate the p-valuet, depending on our hypothesis, as the percentageof t-valuer above or below t-valuet from the initial sample, i.e. we calculate either

p-valuet = (#t-valuer ≥ t-valuet)/10, 000

orp-valuet = (#t-valuer ≤ t-valuet)/10, 000.

Finally, we convert the obtained p-value to the corresponding t-value of a t-distribution to allow acomparison with the other test statistics.

For multi-day abnormal returns we randomly draw for each event a number of abnormal returnsequal to the number of days in the period of interest, i.e., if we want to test the period [-2 0] onsignificance, we draw three abnormal returns from the estimation period. Afterward, we calculatethe randomly sampled cumulative abnormal return for each event. That followed, we calculatethe mean cumulative abnormal returns over the sample as well as the standard deviation and thecorresponding t-value. Then, we proceed in the same way as already described above.

Furthermore, we apply the same procedure to mean values for single and multi-day returns.This implies that we calculated the average over the sample of abnormal returns on day t, ARt.Afterward, we generated a distribution of the mean abnormal return under the null-hypothesissimilar to the procedure for the t-values by calculating random samples of mean abnormal returns,ARr. By applying this procedure, we obtained a slightly higher significance for testing on abnormalreturns by applying the procedure to mean abnormal returns instead of t-values, however, only ina irrelevant way for the interpretation of the results.

31

B Additional Results for Abnormal Returns and Trading

Volume

[Table B.1 about here]

[Table B.2 about here]

32

Table 1: Related Empirical Literature on CEO Turnovers

This table provides an overview of important empirical contributions related to this paper. The table provides information about the market investigated (Market),the time period covered by the empirical sample (Years), the executive position under investigation (Pos.), the number of management turnovers considered in thewidest sample (Sample), the event window used for calculating abnormal returns (Window), and test results related to Hypothesis 1 (H1: Succession Type) andHypothesis 2 (H2: Departure Type) of this paper. OUT and INS indicate samples considering only company outsiders and company insiders as newly appointedmanagers. In spite of the clear-cut classification provided in this table, the studies can differ in the specific mechanisms used to classify turnovers. Furthermore, inthe last column we indicate whether the study also examines the impact of manager turnovers on return on assets (ROA) and trading volumes (TV ). ***, **, and *denote significances at the 1%, 5%, and 10% confidence level, respectively, in a two-tailed test.

Authors Market Years Pos. Sample Window H1: Succession Type H2: Departure Type ROA/TVOUT INS FOR VOL

Reinganum (1985) USA 1978–1979 Top 353 [0] 1.17%** -0.13% – – −/−Beatty and Zajac (1987) USA 1979–1980 CEO 209 [0] 0.10% 0.00% – – −/−Furtado and Rozeff (1987) USA 1975–1982 Top 323 [0 1] 0.72% 1.05%*** 1.03%** – −/−Warner, Watts, and Wruck (1988) USA 1963–1978 Top 230 [-1 0] 0.34%** – 0.14% – −/−Mahajan and Lummer (1993) USA 1972–1983 Top 498 [-1 0] – – -0.73%* 0.21% −/−Worrell, Davidson, and Glascock (1993) USA 1963–1987 Top 62 [-1 0] -1.17% 0.83% 0.38% – −/−Park and Rozeff (1994) USA 1979–1986 CEO 385 [-1 0] 0.61% -0.34% – – X/−Denis and Denis (1995) USA 1985–1988 CEO 328 [-1 0] – – 2.50%*** 0.61% X/−Khanna and Poulsen (1995) USA 1980–1990 Top 121 [-1 0] -0.26% 0.00% – – −/−Huson, Parrino, and Starks (2001) USA 1971–1995 CEO 854 [-2 2] 2.49%*** – 2.02%*** – −/−Shen and Cannella (2003) USA 1988–1997 CEO 177 [-1 1] 1.95%*** – – – −/−Adams and Mansi (2009) USA 1973–2000 CEO 674 [-1 1] 2.42%*** 0.15% 2.43%*** 0.27%** −/−Dahya, Lonie, and Power (1998) UK 1989–1992 Top 105 [-1 0] – – 0.12%** -0.02% −/−Dedman and Lin (2002) UK 1990–1995 CEO 251 [-1 1] – – -3.40%*** 0.13% X/−Dahya and McConnell (2005) UK 1988–1999 CEO 523 [-1 0] 0.79%*** 0.20% – – −/−Hillier and McColgan (2005) UK 1993–1998 CEO 462 [-1 1] – – 11.82% 0.92%** X/−Dherment-Ferere and Renneboog (2002) FR 1988–1992 CEO 92 [-1 0] – – 0.50% 0.40% −/−Kang and Shivdasani (1996) JP 1985–1990 CEO 432 [-1 0] 0.95%** 0.38%** 1.02%** 0.40%** −/−Setiawan (2008) ID 1992–2003 CEO 59 [0] 0.90% -2.30% -1.20% 0.00% −/−Neumann and Voetmann (2005) DK 1994–1998 CEO 81 [-1 1] – – 1.10%** -1.00%** −/−Danisevska, de Jong, and Rosellon (2003) NL 1993–1999 CEO 84 [0 1] – – -0.54% – −/−Cools and van Praag (2007) NL 1991–1999 Top 227 [0 1] – – 0.97% – −/X

33

Table 2: Sample Construction

This table gives an overview about the sample construction. In particular, the table reports the number of events excluded from the

sample due to the selection criteria presented in Section 3.1.

Step CEO-Turnover Sample Number of CEO Turnovers1. Total CEO Turnovers for the period 1998 to 2009 3472. Missing exact announcement time and date -563. Confounding events -67

· Simultaneous earnings and dividend announcements 62· Organizational change (e.g. merger, acquisition) 5

4. Insufficient stock price data -16· Less than 40% of trading days in estimation period 11· Too short stock price history to estimate market model 5

5. Final sample of CEO turnovers 208

34

Table 3: Statistics on CEO Turnovers in the Sample

This table reports the number and characteristics of CEO turnovers between January 1998 and June 2009 that comply with the sample selection criteria presented

in Section 3.1. The sample includes 208 CEO turnovers at 127 companies. Total Turnover indicates the number of all CEO turnovers during a particular year.

Successor Origin is subdivided into the categories Insider and Outsider, depending on whether the newly appointed CEO is already a company employee or not.

Departure is subdivided into the categories Forced and Voluntary, depending on whether the departing CEO was forced to leave the company by a boards decision

or not. Performance is subdivided into the categories Underperformance and Overperformance, depending on the prior performance of the stock price relative to a

broad market index (SPI) in the one-year period before the CEO-turnover announcement date.

Year Total Turnovers Successor Origin Departure PerformanceInsider Outsider Voluntary Forced Underperf. Overperf.

1998 5 3 2 2 3 4 11999 10 10 0 9 1 8 22000 13 8 5 12 1 8 52001 16 8 8 11 5 11 52002 24 14 10 12 12 21 32003 17 13 4 10 7 8 92004 26 16 10 22 4 11 152005 11 6 5 7 4 6 52006 25 17 8 20 5 13 122007 29 22 7 21 8 10 192008 19 9 10 15 4 8 112009 13 7 6 7 6 10 3Total 208 133 75 148 60 118 90Total % 100% 64% 36% 71% 29% 57% 43%

35

Table 4: Abnormal Returns

This table reports the mean and median abnormal stock returns obtained from an event study of CEO turnover announcements. As

indicated in column 1 of the table the abnormal returns refer to different event windows. Mean and median abnormal returns are shown in

column 2 and column 4, respectively. The table presents results both for the entire sample (Panel A) and for selected subsamples (Panels

B-H). The parameters for the market model are estimated over a period of 250 trading days ending 11 days prior the CEO turnover

announcement. The one-sided event-study test statistics are the Standardized cross-sectional test by Boehmer, Musumeci, and Poulsen

(1991) (t-value) and the Wilcoxon signed rank test (z-value). ***, **, and * denote significances at the 1%, 5%, and 10% confidence

level, respectively, in a two-tailed test.

Mean abnormal Median abnormalDays return t-value return z-value

Panel A: Total Sample (n = 208)[-2 0] 0.74%* 1.4381 0.32%* 1.4705[-1 0] 0.53% 1.2453 0.01% 0.5017[-1 1] 0.65%** 1.7612 0.35% 1.0770

-2 0.21% 0.8265 0.07%** 1.7835-1 0.11% -0.0887 -0.05% 0.45910 0.42%* 1.4919 0.01% 1.0310

Panel B: Outside Successor (n = 75)[-2 0] 1.85%** 2.2346 0.51%** 1.8852[-1 0] 1.41%** 1.8146 0.00% 1.1248[-1 1] 1.58%** 1.8962 0.71%* 1.5630

-2 0.44%* 1.5270 0.07%* 1.5683-1 0.33% 0.6197 -0.06% 0.24290 1.08%* 1.5760 0.09%* 1.4733

Panel C: Forced Departure (n = 60)[-2 0] 2.74%* 1.6192 0.81%** 1.8551[-1 0] 2.27%** 1.7771 0.78% 1.1705[-1 1] 1.94%* 1.3021 0.39% 0.7067

-2 0.47% 0.2008 0.00% 0.9791-1 0.55% 0.6733 -0.03% 0.41960 1.72%** 1.6898 0.21%* 1.3472

Panel D: Prior Underperformance (n = 118)[-2 0] 1.94%*** 3.1035 0.84%*** 2.6250[-1 0] 1.56%*** 2.7211 0.20%** 1.8758[-1 1] 1.62%*** 2.9018 0.77%** 2.1094

-2 0.38%* 1.4225 0.10%** 1.6824-1 0.31% 0.9166 -0.03% 0.20280 1.25%*** 2.6401 0.19%** 2.0450

36

Table 4: – Continued


Panel A: Prior Overperformance – H4 (n = 90)[-2 0] -0.84% -1.1554 -0.01% 0.9999[-1 0] -0.82% -1.2317 -0.13%* 1.6276[-1 1] -0.62% -0.6568 -0.42% 0.9878

-2 -0.01% -0.3144 0.01% 0.8188-1 -0.17% -1.0855 -0.06% 0.99590 -0.66% -0.7752 0.00% 0.9838

Panel F: Outside Successor with Forced Departure – H5 (n = 21)[-2 0] 5.94%** 2.1617 2.21%** 2.0681[-1 0] 4.72%* 1.6637 0.87%* 1.6162[-1 1] 5.28%* 1.4928 1.72% 1.1644

-2 1.21% 1.2635 0.00%* 1.4772-1 0.63% 0.2651 -0.08% 0.15640 4.10%* 1.5260 0.60%** 1.9986

Panel G: Forced Departure with Underperformance – H6 (n = 39)[-2 0] 5.07%*** 2.7397 3.98%*** 2.7212[-1 0] 4.00%** 2.3002 2.10%** 1.9956[-1 1] 3.71%** 2.0555 2.97%* 1.4374

-2 1.07%* 1.4885 0.31%** 1.7723-1 1.07%* 1.5108 0.14% 1.11640 2.93%** 1.9069 1.81%** 1.8421

Panel H: Forced Departure with Overperformance – H7 (n = 21)[-2 0] -1.58% -1.1688 -0.15% 1.0254[-1 0] -0.94% -1.1093 -1.32%* 1.5815[-1 1] -1.35% -1.2798 -0.86%* 1.3382

-2 -0.63% -0.8742 -0.06% 1.0254-1 -0.42% -1.2395 -0.11% 1.16440 -0.52% -0.2220 0.00% 0.8168

37

Table 5: Test on Differences in Means (Restriction on One Turnover Characteristics)

This table shows the differences in mean cumulative abnormal returns between all pairwise combinations of samples restricted on one

turnover characteristics over the window [-3 0], CAR1−CAR2, where CARi is the mean of the sample i. The t-statistic is calculated by

t =(

CAR1 − CAR2

)

/√

σ̂2(CAR1)/n1 + σ̂2(CAR2/n1), where σ̂2(CARi) is the variance, and ni the size of sample i. The construction

of the subsamples is based on the turnover characteristics successor origin (OUT denotes outside and INS inside successions), departure

type (FOR denotes forced and VOL voluntary turnovers), and prior performance (OVE denotes prior overperformance and UND prior

underperformance). ***, **, and * denote significance at the 1%, 5%, and 10% confidence level, respectively.

OUT VOL FOR UND OVEINS -1.96%* 0.19% -2.93%** -1.99%** 0.97%

OUT 2.16%** -0.96% -0.03% 2.94%***VOL -3.12%** -2.18%*** 0.78%FOR 0.94% 3.90%***UND 2.96%***

38

Table 6: Test on Differences in Means (Restriction on Two Turnover Characteristics)

This table shows the differences in mean cumulative abnormal returns between all pairwise combinations of samples restricted on two turnover characteristics over the

window [-3 0], CAR1 − CAR2, where CARi is the mean of the sample i. The t-statistic is calculated by t =(

CAR1 − CAR2

)

/√

σ̂2(CAR1)/n1 + σ̂2(CAR2/n1),

where σ̂2(CARi) is the variance, and ni the size of sample i. The construction of the subsamples is based on the intersection of two turnover characteristics, which

are successor origin (OUT denotes outside and INS inside successions), departure type (FOR denotes forced and VOL voluntary turnovers), and prior performance

(OVE denotes prior overperformance and UND prior underperformance). ***, **, and * denote significance at the 1%, 5%, and 10% confidence level, respectively.

INS*FOR INS*UND INS*OVE OUT*VOL OUT*FOR OUT*UND OUT*OVE VOL*UND VOL*OVE FOR*UND FOR*OVEINS*VOL -1.26% -1.35% 0.93% -0.49% -7.07%*** -4.20%*** 0.03% -0.52% 0.21% -6.08%*** 1.88%INS*FOR -0.08% 2.19% 0.77% -5.81%** -2.93% 1.30% 0.74% 1.48% -4.82%** 3.14%*INS*UND 2.28%** 0.85% -5.73%** -2.85%* 1.38% 0.82% 1.56% -4.74%*** 3.23%**INS*OVE -1.42% -8.00%*** -5.13%*** -0.90% -1.45% -0.72% -7.01%*** 0.95%

OUT*VOL -6.58%*** -3.70%** 0.53% -0.03% 0.70% -5.59%*** 2.37%OUT*FOR 2.88% 7.11%*** 6.55%*** 7.28%*** 0.99% 8.95%***OUT*UND 4.23%*** 3.67%** 4.41%*** -1.89% 6.08%***OUT*OVE -0.56% 0.18% -6.12%*** 1.85%VOL*UND 0.73% -5.56%*** 2.40%*VOL*OVE -6.30%*** 1.67%FOR*UND 7.97%***

39

Table 7: Abnormal-Return Ranking for Subsamples

This table shows the abnormal-return rankings for subsamples based on the turnover characteristics successor origin (OUT denotes

outside and INS inside successions), departure type (FOR denotes forced and VOL voluntary turnovers), and prior performance (OVE

denotes prior overperformance and UND prior underperformance). The subsamples are constructed based on one (Panel A), two (Panel

B), and three (Panel C) turnover characteristics. Rank indicates the position of the subsamples with respect to the mean AR in the

event window [-3 0] (Mean AR [-3 0]). Sample Restriction identifies the sample construction based on one, two, or three sample criteria

as listed under R1, R2, and R3. Events reports the number of CEO turnovers included in each subsample. Finally, t-value indicates the

test statistics obtained by a two-tailed Boehmer test (Boehmer, Musumeci, and Poulsen, 1991). ***, **, and * denote significances at

the 1%, 5%, and 10% confidence level, respectively.

Rank Sample Restriction Events Mean t-valueR1 R2 R3 AR [-3 0]

Panel A: Restriction on One Turnover Characteristics1 FOR - - 60 2.94%* 1.93892 UND - - 118 2.00%*** 3.40863 OUT - - 75 1.98%** 2.36884 INS - - 133 0.01% 0.40085 VOL - - 148 -0.18% 0.30806 OVE - - 90 -0.96% -1.2637

Panel B: Restriction on Two Turnover Characteristics1 OUT FOR - 21 6.71%** 2.68312 FOR UND - 39 5.73%*** 3.73523 OUT UND - 42 3.84%*** 2.89574 INS UND - 76 0.99%** 2.17345 INS FOR - 39 0.91% 0.89816 VOL UND - 79 0.16% 1.27097 OUT VOL - 54 0.13% 1.06048 INS VOL - 94 -0.36% -0.26949 OUT OVE - 33 -0.39% 0.128710 VOL OVE - 69 -0.57% -0.635811 INS OVE - 57 -1.29% -1.438112 FOR OVE - 21 -2.24%* -1.3576

Panel C: Restriction on Three Turnover Characteristics1 UND FOR OUT 14 10.61%*** 4.28682 UND FOR INS 25 2.99%** 2.13983 UND VOL OUT 28 0.45% 0.88444 UND VOL INS 51 0.01% 0.91495 OVE VOL OUT 26 -0.21% 0.58866 OVE VOL INS 43 -0.79% -1.00477 OVE FOR OUT 7 -1.08% -1.40178 OVE FOR INS 14 -2.82% -1.0534

40

Table 8: Cross-Sectional Regressions of Abnormal Returns

This table reports the results of regressing the [-2 0]-abnormal returns on a set of explanatory variables: succession type, departure

type, relative performance, logarithm of total assets, and the age of the departing and the appointed CEO. The sample consists of 208

CEO turnovers at 127 companies in the Swiss Performance Index over the time period between December 1998 and June 2009. Outside

successions and forced departures are denoted by OUT and FOR, respectively. OVE is a dummy variables and denotes companies whose

stock experienced an overperformance against a broad market index of Swiss stocks in the estimation period of the market model. SIZE

denotes the logarithm of total assets, AGEDEP the age of the departing CEO and AGEINC of the appointed CEO. The notations

***, **, and * denote significance at the 1%, 5%, and 10% confidence level, respectively, and the corresponding t-values are depicted in

parentheses.

Parameter Regr. 1 Regr. 2 Regr. 3 Regr. 4 Regr. 5 Regr. 6 Regr. 7CONST 0.0011 -0.0007 0.0194*** 0.0178 0.1847* -0.0816 0.1012

(0.1999) (-0.1354) (3.2240) (0.6394) (1.3897) (-0.6513) (0.5705)OUT 0.0173** 0.0172**

(1.8075) (1.8185)FOR 0.0282*** 0.0240**

(2.8041) (2.2124)OVE -0.0278*** -0.0259***

(-3.0349) (-2.8342)SIZE -0.0007 -0.0008

(-0.3795) (-0.4058)AGEDEP -0.0445* -0.0199

(-1.3349) (-0.5772)AGEINC 0.0229 -0.0014

(0.7108) (-0.0427)adj. R2 0.0060 0.0274 0.0335 -0.0091 -0.0011 -0.0073 0.0595

41

Table 9: Cross-Sectional Regressions for Different Event Windows

This table reports the results of regressing the abnormal returns of different event windows on succession type, departure type, relative

performance and logarithm of total assets. The sample consists of 208 CEO turnovers at 127 companies in the Swiss Performance

Index over the time period between 1998 and 2009. Outside successions and forced departures are denoted by OUT denotes and FOR,

respectively. OVE is a dummy variables and denotes companies whose stock experienced an overperformance against a broad market

index of Swiss stocks in the estimation period of the market model. The fourth variable, SIZE, denotes the logarithm of total assets.

The notations ***, **, and * denote significance at the 1%, 5%, and 10% confidence level, respectively, and the corresponding t-values

are depicted in parentheses.

Days CONST OUT FOR OVE SIZE adj. R2

[-3 0] 0.0253 0.0191** 0.0291*** -0.0273*** -0.0015 0.0868(0.9001) (2.0521) (2.9599) (-3.0462) (-0.8069)

[-3 1] 0.0592** 0.0181* 0.0241** -0.0268*** -0.0037* 0.0522(1.7416) (1.6073) (2.0241) (-2.4684) (-1.6299)

[-2 0] 0.0178 0.0172** 0.0259*** -0.0256*** -0.0009 0.0672(0.6294) (1.8295) (2.6105) (-2.8325) (-0.4841)

[-2 -1] 0.0043 0.0071 0.0092* -0.0080* -0.0002 0.0077(0.2558) (1.2542) (1.5417) (-1.4679) (-0.1826)

[-1 0] 0.0139 0.0137* 0.0225*** -0.0220*** -0.0007 0.0538(0.5303) (1.5711) (2.4431) (-2.6195) (-0.4230)

[-1 1] 0.0478* 0.0127 0.0175* -0.0215** -0.0029* 0.0289(1.5229) (1.2189) (1.5878) (-2.1387) (-1.3957)

42

Table 10: Extended Cross-Sectional Regressions


performance, logarithm of total assets and a cross-term of forced turnovers with prior performance. The coefficient estimates are for a

sample of 208 CEO turnovers at 127 companies in the Swiss Performance Index in the time period between 1998 and 2009. Panel A reports

the results with OVE as a dummy variable where stocks overperforming the broad market index in the estimation period of the market

model take on the value of one and zero otherwise. In Panel B the performance dummy is replaced by the realized relative performance

during the estimation period. OUT denotes outside successions. FOR denotes forced departures. SIZE denotes the logarithm of total

assets. In Panel A FOR*OVE is a dummy variable denoting forced turnovers in combination with prior overperformance and in Panel

B FOR*OVE is the interaction term of forced turnovers with prior performance. The notations ***, **, and * denote significance at the

1%, 5%, and 10% confidence level, respectively, and the corresponding t-values are depicted in parentheses.

Days CONST OUT FOR OVE SIZE FOR*OVE adj. R2

Panel A: Prior Performance as Dummy Variable[-3 0] -0.0059 0.0199** -0.0159 -0.0077 0.0029 -0.0708*** 0.1363

(-0.7433) (2.2109) (-1.0316) (-0.7576) (0.2740) (3.5857)[-2 0] -0.0024 0.0175** -0.0089 -0.0105 0.0010 -0.0553*** 0.0959

(-0.2954) (1.9023) (-0.5659) (-1.0033) (0.0950) (2.7416)Panel B: Prior Performance as Continuous Variable

[-3 0] 0.0058 0.0128* 0.0055 -0.0013 -0.0009 -0.0816*** 0.2184(0.2262) (1.4738) (0.5598) (-0.0929) (-0.5021) (-4.4574)

[-2 0] -0.0018 0.0107 0.0019 -0.0074 -0.0002 -0.0770*** 0.2114(-0.0698) (1.2303) (0.1883) (-0.5353) (-0.1270) (-4.1933)

43

Table 11: Cross-Sectional Regressions for Forced Turnovers

This table reports the results of regressing the abnormal returns of forced turnovers from the period [−3 0] on succession type, departure

type, relative performance and logarithm of total assets. The coefficient estimates are for a sample of 60 CEO turnovers at 45 companies

in the Swiss Performance Index in the time period between 1998 and 2009. OUT denotes outside successions. OVE denotes prior realized

relative performance as continuous variable. SIZE denotes the logarithm of total assets. The notations ***, **, and * denote significance

at the 1%, 5%, and 10% confidence level, respectively, and the corresponding t-values are depicted in parentheses.

Parameter Regr. 1 Regr. 2 Regr. 3 Regr. 4CONST 0.0091 0.0023 0.0574 -0.0252

(0.6467) (0.2101) (0.9015) (-0.4630)OUT 0.0581*** 0.0373**

(2.4512) (1.7962)OVE -0.0853*** -0.0797***

(-5.5773) (-5.1427)SIZE -0.0019 0.0011

(-0.4478) (0.3150)adj. R2 0.0621 0.3263 -0.0315 0.3398

44

Table 12: Overview on Tests

Method Literature Characteristics

Panel A: Parametric Tests

Traditional Brown and Warner (1980) t-values are calculated as the average of abnormal returns dividedby the average of standard deviation of abnormal returns. Theunderlying assumption of the test is cross sectional independenceacross the securities.

Portfolio Brown and Warner (1980) This test statistic takes into account the cross-sectional depen-dence of the security returns.

IncorporatingAutocorrelation

Ruback (1982) This test adjusts for first-order autocorrelation in abnormal re-turns.

Standardized Residual Patell (1976) The event-period residuals are standardized by the standard de-viation calculated over the estimation period. This standardiza-tion diminishes the problem of heteroskedastic event-day resid-uals. Therefore, stocks with large variances are prevented fromdominating the test statistics.

StandardizedCross-Sectional

Boehmer, Musumeci, andPoulsen (1991)

This test is a combination of the standardized-residual method-ology by Patell (1976) and the ordinary cross-sectional approachapplied amongst others by Penman (1982). The test works wellin case of event-induced variance increases and eliminates theheteroskedasticity problem associated with the ordinary cross-sectional approach .

Panel B: Non-Parametric Tests

Wilcoxon Signed Rank Wilcoxon (1945) Both the sign and the magnitude of the abnormal returns areincorporated for calculating the test statistic.

Corrado Rank Corrado (1989) There is no requirement of a symmetric security return distribu-tion and the test is resistant to event-induced variance.

Generalized Rank Kolari and Pynnonen(2008)

The test does not lose power in testing for cumulative abnormalperformance. Test is robust against serial abnormal return corre-lation, clustering effects and event-induced variance increases.

Bootstrap - The distribution under H0 is generated with random draws fromthe estimation period residuals from each event in the sample. Fora detailed description of the Bootstrap method see Section 4.3.

45

Table 13: Overview on Test Statistics

Panel A shows the test statistic, t = AR∗

(τ1, τ2)/σ̂∗(

AR∗

(τ2, τ2))

, for the parametric methods. The corresponding enumerator is shown

in the second column, denoted by AR∗

(τ1, τ2), and the denominator in the third column, denoted by σ̂∗(

AR∗

(τ2, τ2))

. The variable

ARit represents the abnormal return of company i on day t, the variable AR∗it denotes the variable employed in the test statistic, and

σ̂(ARi) is the daily standard deviation of the abnormal returns over the estimation period. The parameter N denotes the number of

events, the estimation period covers the days -260 to day -11 prior the event, i.e. consists of a time span of 250 days. The variables τ1

and τ2 denote the first and the last day, respectively, of the event window. The nonparametric test statistics are shown in Panel B.

Panel A: Parametric Tests

Method AR∗

(τ1, τ2) σ̂∗(

AR∗

(τ2, τ2))

Traditional (Brown andWarner, 1980)

∑τ2

t=τ1

(

1N

∑N

i=1ARit) [

(τ2 − τ1 + 1) 1N2

∑N

i=1

(∑

−11

t=−260(AR∗it −AR

∗

i )2)]1/2

Portfolio (Brown andWarner, 1980)

∑τ2

t=τ1

(

1N

∑N

i=1ARit)

[

(τ2 − τ1 + 1) 1250−1

∑

−11

t=−260(AR

∗

t −AR∗

)2

]1/2

,

where AR∗

= 1N

∑

−11

t=−260ARit

IncorporatingAutocorrelation (Ruback,

1982)

∑τ2

t=τ1

(

1N

∑N

i=1ARit)

[

T · 1250−1

∑

−11

t=−260(AR

∗

t −AR∗

)2 + 2 · (T − 1) + Cov(

AR∗

t , AR∗

t−1

)

]1/2

,

where T = τ2 − τ1 + 1

Standardized Residual(Patell, 1976)

∑τ2

t=τ1

(

1N

∑N

i=1ARitσ̂(ARi)

) [

(τ2 − τ1 + 1)(

N(250−2)250−4

)]1/2

StandardizedCross-Sectional (Boehmer,Musumeci, and Poulsen,

1991)

∑τ2

t=τ1

(

1N

∑N

i=1ARitσ̂(ARi)

)

[

1N(N−1)

∑N

i=1

(

AR∗it −AR∗

t

)2]1/2

Panel B: Nonparametric TestsMethod Test Statistic

Wilcoxon Signed Rank(Wilcoxon, 1945)

Z =W−

n(n+1)4

√

n(n+1)(2n+1)24

, where W = min(∑N

i=1B+i rank (ARi(τ1, τ2)) ,

∑N

i=1B−i rank (ARi(τ1, τ2))

)

, where

B+i =

{

1 if ARi(τ1, τ2) > 0

0 if ARi(τ1, τ2) < 0and B−i =

{

1 if ARi(τ1, τ2) < 0

0 if ARi(τ1, τ2) > 0

Corrado Rank (Corrado,1989)

t =

1N

∑N

i=1

(

∑τ2

t=τ1Kit−(τ2−τ1+1)E(Ki)

)

σ̂(K)(τ2−τ1+1)1/2 , where Kit = rank(ARit), t = −260,−259, ...,+5,

E(Ki) = 0.5 + 0.5 · 266 and σ̂(K) =

[

1266

∑5

t=−2601N

∑N

i=1

(

Kit − E(Ki)

)2]1/2

Generalized Rank (Kolariand Pynnonen, 2008)

t = K0−0.5σ̂(K)

, where K0 =

rank

(

∑τ2

t=τ1Kt

)

T ′+1, σ̂(K) =

[

1T ′

∑5−(τ1−τ2)

t=−260

(

Kt − 0.5)2]1/2

,

Kt = 1N

∑N

i=1Kit, Kit =

rank(GSARit)T ′+1

, and T ′ = T − (τ2 − τ1 + 1)

GSARit =

{

AR∗itσ̂(AR∗)

, for t = τ1, ..., τ2

AR∗it, for t = −260, ..., τ1 − 1, τ2 + 1, ..., 5, where AR∗it = ARit

σ̂(ARi)

Bootstrap See Appendix A for a detailed description.

46

Table 14: Robustness of Abnormal Returns

To test the abnormal returns on significance nine event-study methodologies were employed in this paper. This table shows the test results for all the employed event

study-methodologies. The applied test statistics are the Traditional and the Portfolio test (Brown and Warner, 1980), the Standardized-residual test (Patell, 1976), the

Ruback (1982) test, the Standardized cross-sectional test (Boehmer, Musumeci, and Poulsen, 1991), a Bootstrap test, the Corrado (1989) rank test, the Generalized

rank test (Kolari and Pynnonen, 2008) and the Wilcoxon signed rank test. Tests have been conducted one-sided. The notations ***, **, and * denote significance at

the 1%, 5%, and 10% confidence level, respectively.

Days Traditional Portfolio Patell Ruback Boehmer Bootstrap Corrado Rank Gen. Rank WilcoxonPanel A: Total Sample (n = 208)

[-3 0] 1.7852** 1.7730** 2.2085** 1.8874** 1.4862* 1.3732* 2.2928** 1.8144** 1.5510*[-2 0] 2.1118** 2.0973** 2.3628*** 2.2162** 1.4381* 1.4357* 2.2578** 1.8323** 1.4705*[-1 1] 1.8562** 1.8435** 2.9882*** 1.9480** 1.7612** 1.1600 1.9195** 1.7142** 1.0770

-1 0.5214 0.5179 -0.1063 0.5179 -0.0887 0.3752 -0.2172 -0.1286 0.45910 2.0973** 2.0829** 3.2096*** 2.0829** 1.4919* 1.0787 2.0285** 2.1001** 1.03101 0.5964 0.5923 2.0934** 0.5923 1.5900* 0.4214 1.5134* 1.5888* 0.9642

Panel B: Outside Successor – H1 (n = 75)[-3 0] 2.4959*** 2.4648*** 2.8092*** 2.5180*** 2.3688** 2.3539** 2.6633*** 2.5577*** 2.2231**[-2 0] 2.6903*** 2.6567*** 2.9239*** 2.7075*** 2.2346** 1.9695** 2.3460*** 2.0244** 1.8852**[-1 1] 2.3086** 2.2798** 2.9923*** 2.3234** 1.8962** 1.4268* 2.4302*** 1.9951** 1.5630*

-1 0.8261 0.8158 0.7079 0.8158 0.6197 0.6999 0.4771 0.5265 0.24290 2.7289*** 2.6949*** 3.0266*** 2.6949*** 1.5760* 1.3190* 1.8660** 1.9083** 1.4733*1 0.4436 0.4380 1.4693* 0.4380 1.0554 0.3289 1.8660** 1.9083** 1.0984

Panel C: Forced Departure – H2 (n = 60)[-3 0] 3.1979*** 3.2682*** 3.4673*** 3.2666*** 1.9389** 2.5846*** 2.9645*** 2.6479*** 2.1938**[-2 0] 3.4447*** 3.5204*** 3.2791*** 3.5189*** 1.6192* 2.3024** 2.5431*** 2.3023** 1.8551**[-1 1] 2.4378*** 2.4914*** 2.7454*** 2.4904*** 1.3021* 1.3598* 1.4503* 1.0999 0.7067

-1 1.1875 1.2136 0.9177 1.2136 0.6733 0.9748 0.5294 0.5783 0.41960 3.7480*** 3.8304*** 4.4747*** 3.8304*** 1.6898** 1.7233** 2.2387** 2.2789** 1.3472*1 -0.7130 -0.7287 -0.6180 -0.7287 -0.4188 -0.4281 -0.2560 -0.2031 0.6331

Panel D: Prior Underperformance – H3 (n = 118)[-3 0] 3.3993*** 3.4374*** 4.6588*** 3.6436*** 3.4086*** 3.3195*** 3.6047*** 3.5371*** 2.8050***[-2 0] 3.8034*** 3.8461*** 4.8378*** 4.0491*** 3.1035*** 2.9196*** 3.2572*** 3.0859*** 2.6250***[-1 1] 3.1750*** 3.2106*** 4.9164*** 3.3801*** 2.9018*** 2.2395** 2.8483*** 2.9905*** 2.1094**

-1 1.0666 1.0786 1.0508 1.0786 0.9166 1.0463 0.5375 0.6051 0.20280 4.2335*** 4.2811*** 5.6815*** 4.2811*** 2.6401*** 2.4394*** 2.9721*** 3.0244*** 2.0450**1 0.1990 0.2013 1.8177** 0.2013 1.3016* 0.2621 1.4237* 1.4857* 0.6673

47


Days Traditional Portfolio Patell Ruback Boehmer Bootstrap Corrado Rank Gen. Rank WilcoxonPanel E: Prior Overperformance – H4 (n = 90)

[-3 0] -1.8269** -1.9216** -1.9770** -1.9818** -1.2637 -2.3096** -0.3301 -0.8553 0.9878[-2 0] -1.8408** -1.9362** -1.9475** -1.9898** -1.1554 -2.3201** -0.0105 -0.4493 0.9999[-1 1] -1.3704* -1.4415* -1.0867 -1.4814* -0.6568 -1.8682** -0.0940 -0.5651 0.9878

-1 -0.6422 -0.6755 -1.3648* -0.6755 -1.0855 -0.9392 -0.9142 -0.8509 0.99590 -2.4998*** -2.6293*** -1.6261* -2.6293*** -0.7752 -2.2276** -0.0586 -0.0000 0.98381 0.7683 0.8081 1.1010 0.8081 0.9093 0.5833 0.8100 0.8639 0.6699

Panel F: Outside Successor with Forced Departure – H5 (n = 21)[-3 0] 3.5388*** 3.5430*** 3.3983*** 3.6069*** 2.6831*** 3.3548*** 3.5368*** 3.1169*** 2.5894***[-2 0] 3.6129*** 3.6172*** 3.2186*** 3.6750*** 2.1617** 2.3883** 3.0242*** 2.4911*** 2.0681**[-1 1] 3.2107*** 3.2145*** 2.7494*** 3.2659*** 1.4928* 1.7295** 2.5152*** 1.4114* 1.1644

-1 0.6591 0.6599 0.3650 0.6599 0.2651 0.4806 0.2812 0.3096 0.15640 4.3189*** 4.3240*** 3.9980*** 4.3240*** 1.5260* 1.7143* 3.1157*** 3.1322*** 1.9986**1 0.5831 0.5838 0.4184 0.5838 0.2868 0.3987 0.9595 0.9850 0.2954

Panel G: Forced Departure with Underperformance – H6 (n = 39)[-3 0] 4.3526*** 4.4115*** 6.0257*** 4.4681*** 3.7352*** 4.1047*** 4.2748*** 4.0425*** 3.3492***[-2 0] 4.4468*** 4.5069*** 5.5638*** 4.5583*** 2.7397*** 3.3128*** 3.3867*** 3.1659*** 2.7212***[-1 1] 3.2578*** 3.3019*** 4.7497*** 3.3395*** 2.0555** 1.9516** 2.3079** 1.9267** 1.4374*

-1 1.6194* 1.6413* 2.1671** 1.6413* 1.5108* 1.3763* 1.3676* 1.4023* 1.11640 4.4565*** 4.5168*** 5.8027*** 4.5168*** 1.9069** 2.1181** 2.5191*** 2.5484*** 1.8421**1 -0.4332 -0.4390 0.2903 -0.4390 0.1830 -0.1391 0.1107 0.1511 0.3349

Panel H: Forced Departure with Overperformance – H7 (n = 21)[-3 0] -2.3271** -2.4030** -2.3510** -2.4071** -1.3576* -2.1104** -0.8998 -1.0214 1.3034*[-2 0] -1.8926** -1.9544** -2.0394** -1.9573** -1.1688 -1.7857** -0.3903 -0.5175 1.0254[-1 1] -1.6209* -1.6737* -1.8323** -1.6763* -1.2798 -1.9410** -0.7349 -1.0840 1.3382*

-1 -0.8721 -0.9005 -1.4021* -0.9005 -1.2395 -1.0212 -0.9831 -0.9489 1.16440 -1.0891 -1.1246 -0.3442 -1.1246 -0.2220 -1.0596 0.2856 0.3144 0.81681 -0.8463 -0.8739 -1.4401* -0.8739 -1.1557 -1.0217 -0.5755 -0.5431 0.8516

48

Table 15: Cross-Sectional Regressions for Subperiods


performance, logarithm of total assets and a cross-term of forced turnovers with prior overperformance. Panel A shows the regression

results for the time period 1998 - 2003 and Panel B the results for the period 2004 - 2009. OUT denotes outside successions. FOR

denotes forced departures. OVE assumes a value of one if the company’s stock experienced an overperformance against broad market

index of Swiss stocks over the estimation period of the market model and zero otherwise. SIZE denotes the logarithm of total assets.

FOR*OVE is a dummy variable representing forced turnovers in combination with prior overperformance. The notations ***, **, and *

denote significance at the 1%, 5%, and 10% confidence level, respectively, and the corresponding t-values are depicted in parentheses.

Days CONST OUT FOR OVE SIZE FOR*OVE adj. R2

Panel A: 1998 - 2003 (N = 85)[-3 0] 0.0215 0.0240* 0.0790*** -0.0017 -0.0024 -0.1057*** 0.1850

(0.4102) (1.4349) (4.0981) (-0.0809) (-0.6874) (-2.9371)[-2 0] 0.0515 0.0236 0.0799*** -0.0068 -0.0044 -0.0906** 0.1498

(0.8748) (1.2580) (3.6979) (-0.2929) (-1.1398) (-2.2457)Panel B: 2004 - 2009 (N = 123)

[-3 0] 0.0170 0.0145* 0.0313** -0.0138 -0.0010 -0.0400** 0.0543(0.5583) (1.3980) (1.9892) (-1.1967) (-0.5057) (-1.7242)

[-2 0] -0.0110 0.0112 0.0097 -0.0171** 0.0014 -0.0156 0.0273(-0.4178) (1.2526) (0.7111) (-1.7154) (0.7824) (-0.7781)

49

Table 16: Rejection Rates for Different Event-Study Tests

This table shows the empirical rejection rates for the Bootstrap, the Standardized cross-sectional (Boehmer, Musumeci, and Poulsen,

1991) and the Wilcoxon signed rank test when there is no abnormal return present in the data. The underlying sample for this historical

simulation are the 160 stock price histories in the period from January 1990 to June 2009 from the sample with a complete data set where

we refrained to eliminate data points due to confounding events. This initial sample is employed to generate 250 random samples by

choosing each time randomly a number of securities, N (N ∈ {10; 20; 50; 100}) with replacement. The Bootstrap test was conducted by

resampling 10, 000 times from our initial sample of abnormal returns. The rejection rates of the simulations are obtained in a two-tailed

test at the 5% significance level.

Method Number of Events10 20 50 100 200

Panel A: Significance Level 10%Bootstrap 10.80% 9.60% 10.00% 10.00% 11.60%Boehmer 7.60% 8.00% 10.80% 8.40% 12.00%Wilcoxon 8.80% 9.20% 10.00% 9.20% 15.20%

Panel B: Significance Level 5%Bootstrap 6.40% 6.00% 4.00% 5.60% 6.80%Boehmer 5.20% 5.20% 4.80% 4.80% 7.20%Wilcoxon 2.00% 5.60% 6.00% 4.80% 4.80%

Panel C: Significance Level 1%Bootstrap 1.60% 1.60% 1.60% 1.20% 1.60%Boehmer 0.00% 0.80% 1.20% 0.80% 1.60%Wilcoxon 0.00% 0.00% 0.80% 0.40% 1.60%

50

Table 17: Impact of CEO Turnovers on Trading Volume

This table reports the test statistics for the trading volume around the CEO turnover announcement. The mean and median abormal trading volume are calculated

with non-log-transformed trading volumes as in Equation 1. The applied test statistics are the Traditional and the Portfolio test (Brown and Warner, 1980), the

Standardized-residual test (Patell, 1976), the Ruback (1982) test, the Standardized cross-sectional test (Boehmer, Musumeci, and Poulsen, 1991), a Bootstrap test,

the Corrado (1989) rank test, the Generalized rank test (Kolari and Pynnonen, 2008) and the Wilcoxon signed rank test. The mean abnormal trading volume is shown

in the second column and the median abnormal trading volume in the seventh column. Tests have been conducted one-sided. The notations ***, **, and * denote

significance at the 1%, 5%, and 10% confidence level, respectively.

Days Mean Portfolio Patell Boehmer Bootstrap Median Corrado Rank Gen. Rank WilcoxonPanel A: Total Sample (n = 208)

[-3 -1] 34.05% 1.7750** -0.7321 -0.4228 2.8649*** -41.94% 1.7227** 1.5263* 0.2531[-2 2] 240.93% 10.2419*** 7.3697*** 3.1013*** 3.7864*** 43.06% 5.0670*** 3.7656*** 4.6761***[1 3] 110.30% 6.5323*** 4.3169*** 2.1706** 3.7864*** 6.82% 3.6639*** 2.8348*** 3.4277***

-1 10.65% 2.0788** 0.1190 0.0969 3.2382*** -17.03% 1.4244* 1.5277* 1.13800 129.97% 10.6347*** 10.0972*** 5.9646*** 3.7864*** 17.65% 4.0434*** 4.0844*** 6.6057***1 55.66% 6.2036*** 5.3683*** 3.7534*** 3.7864*** -2.40% 3.0663*** 3.1306*** 4.6059***

Panel B: Outside Successor (n = 75)[-3 -1] 39.40% 1.7899** -0.3498 -0.2009 2.6901*** -38.90% 1.7371** 1.4622* 0.5175[-2 2] 233.39% 6.0089*** 3.4585*** 1.7136** 3.9113*** 39.06% 4.1500*** 2.7370*** 3.4218***[1 3] 88.00% 3.8223*** 2.0051** 1.2119 3.9113*** 28.38% 2.9223*** 1.8768** 2.4449***

-1 13.38% 1.7667** -0.0086 -0.0072 3.1158*** -24.15% 1.3475* 1.4265* 1.2832*0 131.73% 5.3504*** 4.4253*** 2.9159*** 3.9113*** 20.16% 2.9972*** 3.0435*** 3.8918***1 59.21% 3.3154*** 3.3265*** 2.4292*** 3.9113*** 0.00% 2.4329*** 2.4904*** 3.0627***

Panel C: Forced Departure (n = 60)[-3 -1] -4.87% -0.0667 -0.4720 -0.3310 0.7282 -47.02% 0.5557 0.1632 0.3681[-2 2] 331.44% 6.0040*** 7.5014*** 3.7412*** 3.9621*** 100.56% 4.6931*** 3.4792*** 3.9826***[1 3] 141.12% 3.3428*** 4.7670*** 2.5986*** 3.4602*** 50.59% 3.3408*** 2.4722*** 2.5545***

-1 11.71% 0.8879 0.5960 0.5460 1.6355* -12.01% 1.0352 1.1213 0.86130 196.37% 6.8708*** 9.2885*** 5.5586*** 3.9621*** 66.96% 4.3189*** 4.3394*** 5.2635***1 101.87% 3.9077*** 5.5451*** 3.8861*** 3.9621*** 28.34% 3.1529*** 3.1967*** 3.6734***

Panel D: Prior Underperformance (n = 118)[-3 -1] 57.52% 2.8406*** 1.1912 0.6639 3.1305*** -29.66% 1.7465** 1.6776** 1.0863[-2 2] 257.13% 9.9621*** 7.5451*** 3.2646*** 3.8393*** 65.55% 5.2588*** 4.0283*** 4.8754***[1 3] 126.62% 6.9453*** 4.7257*** 2.5042*** 3.8393*** 31.14% 3.9939*** 3.0258*** 3.8872***

-1 19.25% 2.5033*** 0.9669 0.7751 3.1608*** -9.85% 1.5181* 1.6129* 1.9080**0 132.07% 9.1759*** 8.7676*** 5.0737*** 3.8393*** 21.87% 4.0452*** 4.0827*** 5.5978***1 58.94% 6.0338*** 4.8974*** 3.6558*** 3.8393*** 4.12% 3.1819*** 3.2390*** 4.6579***

51


Days Mean Portfolio Patell Boehmer Bootstrap Median Corrado Rank Gen. Rank WilcoxonPanel E: Prior Overperformance (n = 90)

[-3 -1] 3.27% -0.3021 -2.4768*** -1.5178* 0.9621 -71.42% 1.4383* 0.6445 0.9596[-2 2] 219.70% 4.9439*** 2.5643*** 1.0461 3.6778*** -5.25% 4.0726*** 2.7963*** 1.4868*[1 3] 88.89% 2.5380*** 1.1516 0.5452 2.6316*** -40.63% 2.6975*** 2.3029** 0.8550

-1 -0.63% 0.5028 -0.9261 -0.7702 1.1859 -34.54% 1.0934 1.1964 0.45270 127.21% 6.3488*** 5.3109*** 3.2293*** 3.8781*** -1.00% 3.4454*** 3.4956*** 3.5952***1 51.35% 2.9949*** 2.5533*** 1.6552* 3.1241*** -15.59% 2.4652*** 2.5373*** 1.7684**

Panel F: Outside Successor with Forced Departure (n = 21)[-3 -1] 34.09% 0.2083 0.0294 0.0192 0.8632 -38.90% 1.2023 0.8157 0.1564[-2 2] 292.17% 3.3321*** 3.2357*** 1.6723* 4.4929*** 64.62% 3.6097*** 2.3860*** 2.1724**[1 3] 55.49% 0.7268 0.7922 0.5038 1.3830* -74.18% 1.5486* 1.1573 0.2259

-1 24.38% 0.7930 0.4596 0.4097 1.4575* -11.99% 1.2162 1.2802 0.67780 178.20% 3.9666*** 4.1613*** 2.3491** 3.8193*** 38.23% 3.0071*** 3.0390*** 2.6242***1 76.66% 1.5587* 2.5454*** 1.6021* 1.7833** 17.61% 1.9742** 2.0247** 1.1991

Panel G: Forced Departure with Underperformance (n = 39)[-3 -1] 41.33% 1.3094* 1.3269* 0.8761 1.5995* -20.63% 0.8910 0.5960 0.8792[-2 2] 478.75% 7.5214*** 9.1301*** 4.6366*** 4.1047*** 194.92% 5.2586*** 4.2049*** 4.1586***[1 3] 201.82% 4.7667*** 5.8474*** 3.2996*** 4.1047*** 74.84% 3.5490*** 2.8145*** 3.1678***

-1 30.05% 1.6691* 1.6731* 1.5343* 2.1574** -0.87% 1.3963* 1.4698* 1.8421**0 259.64% 7.0978*** 9.5123*** 5.6078*** 4.1047*** 72.79% 4.5197*** 4.5328*** 4.6051***1 137.85% 4.9204*** 6.0309*** 4.4266*** 4.1047*** 30.50% 3.4383*** 3.4723*** 4.1028***

Panel H: Forced Departure with Overperformance (n = 21)[-3 -1] -90.66% -2.2056** -2.6060*** -2.3547** -1.5154* -77.63% -0.1421 -1.2961* 2.2419**[-2 2] 57.88% 1.0719 0.2374 0.1341 1.9780** 12.98% 2.4625*** 1.3600* 0.7473[1 3] 28.41% -0.3151 0.0890 0.0484 0.5571 -47.24% 2.0672** 1.2384 0.1217

-1 -22.34% -0.7183 -1.2726 -1.2399 -0.4109 -28.57% 0.1612 0.2458 1.23390 78.87% 3.6037*** 2.7373*** 1.8518** 3.8193*** 38.23% 2.7829*** 2.8201*** 2.3114**1 35.06% 0.6580 1.1542 0.7905 0.9690 -19.21% 1.8072** 1.8620** 0.5040

52

Table 18: Cross-Sectional Regressions Abnormal Trading Volume

This table reports the results of regressing the abnormal trading volume of the event day on succession type, departure type, relative

performance, logarithm of total assets and the age of the departing and the appointed CEO. The sample consists of 179 CEO turnovers

in companies of the Swiss Performance Index. Outside successions and forced departures are denoted by OUT and FOR, respectively.

OVE is a dummy variables and denotes for companies whose stock experienced an overperformance against a broad market index of

Swiss stocks in the estimation period of the market model. SIZE denotes the logarithm of total assets, AGEDEP the age of the departing

CEO and AGEINC of the appointed CEO. The notations ***, **, and * denote significance at the 1%, 5%, and 10% confidence level,

respectively, and the corresponding t-values are depicted in parentheses.

Parameter Regr. 1 Regr. 2 Regr. 3 Regr. 4 Regr. 5 Regr. 6 Regr. 7CONST 0.4829*** 0.3988*** 0.5273*** -0.0200 1.1722 -1.6383 -1.4850

(5.9520) (5.2585) (6.1245) (-0.0514) (0.6274) (-0.9362) (-0.5809)OUT 0.0422 0.0772

(0.3126) (0.5662)FOR 0.3442** 0.3164**

(2.4380) (2.0205)OVE -0.0675 -0.0203

(-0.5154) (-0.1541)SIZE 0.0361 0.0298

(1.3520) (1.0869)AGEDEP -0.1690 0.1066

(-0.3610) (0.2149)AGEINC 0.5509 0.2630

(1.2216) (0.5588)adj. R2 -0.0093 0.0186 -0.0085 -0.0009 -0.0091 -0.0025 0.0036

53

Table 19: Regression of Absolute Abnormal Returns on Trading Volume

This table reports the results for regressing absolute abnormal returns on the log-transformed abnormal trading volume for 208 CEO-

turnover announcements. The notations ***, **, and * denote significance at the 1%, 5%, and 10% confidence level, respectively, in a

two-tailed test. The corresponding t-values are depicted in parentheses.

Parameter [-1] [0] [-1 1] [-2 2]CONST 0.0187*** 0.0195*** 0.0350*** 0.0431***

(11.0858) (6.4010) (8.3243) (9.6323)LTV A 0.0071*** 0.0185*** 0.0097*** 0.0063***

(2.8586) (6.4273) (4.8977) (4.4779)adj. R2 0.0288 0.1589 0.0956 0.0798

54

Table 20: Operating Performance

This table reports the results for the companies’ operating performance (in %) adjusted by lagged company’s performance and controlled

for industry effects, whereby from each company’s operating performance we subtract the lagged company’s OROA and the change in

the median operating performance of the companies in the same industry which in total represents the expected OROA for year t. After

obtaining the adjusted OROA values for the years -2 to +2 we calculate first differences from year t to year t− 1 which are subsequently

used for obtaining the test results in the table below. The notations ***, **, and * denote significance at the 1%, 5%, and 10% confidence

level, respectively, for two-tailed tests.

Years Mean Change T-Test Median Change WilcoxonPanel A: Total Sample (n = 136)

[-2 0] -0.54% -0.7078 -0.46%* 1.6747[-1 0] -1.27%* -1.7145 -0.62%** 2.4068[0 1] 1.14% 0.9669 0.64% 1.3685[0 2] 2.19%*** 3.0344 0.86%*** 3.2930

Panel B: Outside Successor (n = 51)[-2 0] 0.06% 0.0476 -0.37% 0.5999[-1 0] -1.01% -0.8332 -0.46% 1.2279[0 1] -1.42% -0.7759 0.53% 0.0281[0 2] 2.04%** 2.0991 1.28%*** 2.6902

Panel C: Inside Successor (n = 85)[-2 0] -0.89% -0.9162 -0.56%* 1.6629[-1 0] -1.43% -1.5149 -0.76%** 2.0879[0 1] 2.69%* 1.7625 0.78%* 1.7900[0 2] 2.28%** 2.2782 0.86%** 1.9784

Panel D: Forced Departures (n = 39)[-2 0] -0.83% -0.6342 -0.84% 1.2559[-1 0] -0.75% -0.6183 -1.12% 1.0327[0 1] -0.20% -0.1116 1.08% 0.7257[0 2] 2.92%** 2.4106 1.98%*** 2.6375

Panel E: Voluntary Departures (n = 97)[-2 0] -0.42% -0.4502 -0.31% 1.1712[-1 0] -1.48% -1.6075 -0.44%** 2.1895[0 1] 1.68% 1.1236 0.68% 1.1460[0 2] 1.89%** 2.1332 0.57%** 2.2327

Panel F: Prior Underperformance (n = 83)[-2 0] -0.42% -0.4486 -1.49%* 1.8615[-1 0] -0.75% -0.9256 -1.01%* 1.6889[0 1] 0.15% 0.1193 1.70% 0.8172[0 2] 2.62%*** 2.8159 2.18%*** 3.3007

55


Years Mean Change T-Test Median Change WilcoxonPanel G: Prior Overperformance (n = 53)

[-2 0] -0.72% -0.5574 0.15% 0.2523[-1 0] -2.09% -1.4695 -0.24%* 1.6687[0 1] 2.70% 1.1617 0.13% 1.0225[0 2] 1.51% 1.3185 0.11% 0.7215

Panel H: Outside Successor with Forced Departure (n = 15)[-2 0] 0.46% 0.1740 -0.73% 0.2840[-1 0] -2.12% -1.4621 -1.39% 1.0223[0 1] -1.33% -0.3198 1.33% 0.2840[0 2] 2.59% 1.2415 2.59% 1.4767

Panel I: Outside Successor with Voluntary Departure (n = 36)[-2 0] -0.11% -0.0823 -0.41% 0.6127[-1 0] -0.55% -0.3390 -0.20% 0.9112[0 1] -1.47% -0.7372 0.37% 0.1885[0 2] 1.81% 1.6664 0.79%** 2.1838

Panel J: Forced Departure with Overperformance (n = 12)[-2 0] -0.22% -0.2547 0.15% 0.1569[-1 0] -1.66% -1.1843 -0.47% 1.0983[0 1] -2.36% -0.7284 0.06% 0.4707[0 2] 1.46% 1.0733 0.57% 1.0983

Panel K: Forced Departure with Underperformance(n = 27)[-2 0] -1.10% -0.5911 -1.35% 1.3694[-1 0] -0.34% -0.2083 -0.97% 0.6967[0 1] 0.76% 0.3553 1.54% 0.6246[0 2] 3.56%** 2.1730 2.61%** 2.2824

56

Table 21: Cross-Sectional Regressions Operating Performance

This table reports the results of regressing the change in adjusted OROA after the CEO turnover on succession type, departure type,

relative performance, logarithm of total assets and the age of the departing and the appointed CEO. The sample consists of 136 CEO

turnovers. The dependent variable is the change in adjusted OROA over the years [1 2] after the turnover. Outside successions and forced

departures are denoted by OUT and FOR, respectively. OVE is a dummy variables and denotes companies whose stock experienced an

overperformance against a broad market index of Swiss stocks in the estimation period of the market model. SIZE denotes the logarithm

of total assets, AGEDEP the age of the departing CEO and AGEINC of the appointed CEO. The notations ***, **, and * denote

significance at the 1%, 5%, and 10% confidence level, respectively, in a two-tailed test and t-values are depicted in parentheses.

Parameter Regr. 1 Regr. 2 Regr. 3 Regr. 4 Regr. 5 Regr. 6 Regr. 7CONST -0.0041 0.0021 0.0247* 0.0508 -0.0375 -0.1818 -0.3593

(-0.3257) (0.1766) (1.9433) (0.7346) (-0.1217) (-0.6975) (-0.9023)OUT 0.0387* 0.0408*

(1.8871) (1.9533)FOR 0.0290 0.0279

(1.3148) (1.2099)OVE -0.0367* -0.0369*

(-1.8007) (-1.8026)SIZE -0.0028 -0.0022

(-0.5901) (-0.4635)AGEDEP 0.0120 0.0559

(0.1556) (0.7044)AGEINC 0.0496 0.0439

(0.7381) (0.6542)adj. R2 0.0112 -0.0021 0.0089 -0.0124 -0.0149 -0.0109 0.0188

57

Table 22: Regressing Change in Adjusted OROAs [0 2] on CARs [-3 0]

This table shows the results from a robust linear regression of the changes in adjusted OROAs from the years [0 2] on the CARs from the

event period [-3 0]. The notations ***, **, and * denote significance at the 1%, 5%, and 10% confidence level, respectively, in a two-tailed

test. The corresponding t-values are depicted in parentheses.

Parameter Underlying SamplePanel A: Total Sample (N = 145)

CONST 0.0017(0.4290)

CAR [-3 0] 0.1112**(2.4758)

Panel B: Single RestrictionINS OUT VOL FOR UND OVE

(N = 93) (N = 52) (N = 103) (N = 42) (N = 88) (N = 57)CONST -0.0018 0.0057 0.0023 0.0103 0.0041 -0.0001

(-0.4567) (0.7428) (0.5934) (0.8104) (0.6281) (-0.0313)CAR [-3 0] -0.0088 0.3991*** 0.1137** 0.0982 0.1012 0.0278

(-0.2006) (3.8800) (2.5387) (0.6346) (1.3715) (0.5175)Panel C: Double Restriction

INS*VOL INS*FOR OUT*VOL OUT*FOR INS*UND INS*OVE(N = 67) (N = 26) (N = 36) (N = 16) (N = 58) (N = 35)

CONST -0.0021 0.0148 0.0094 0.0073 -0.0003 -0.0038(-0.4791) (1.0694) (1.3614) (0.3343) (-0.0486) (-0.6875)

CAR [-3 0] -0.0090 -0.1775 0.3026*** 0.4983* -0.0083 -0.0309(-0.1962) (-0.9774) (2.9302) (2.0626) (-0.1213) (-0.5345)

Panel D: Double Restriction cont.OUT*UND OUT*OVE VOL*UND VOL*OVE FOR*UND FOR*OVE(N = 30) (N = 22) (N = 59) (N = 44) (N = 29) (N = 13)

CONST 0.0001 0.0070 0.0026 0.0027 0.0515*** -0.0081(0.0078) (0.8782) (0.4439) (0.4996) (2.8237) (-1.0113)

CAR [-3 0] 0.5022*** 0.1991 0.0513 0.1742*** -0.1852 -0.0952(3.0884) (1.4696) (0.7475) (2.9454) (-0.9468) (-0.5727)

58

Table B.1: Abnormal Returns

This table reports test results on abnormal returns for different event windows. The return periods for calculating the abnormal returns

are shown in the first column. The mean and median abnormal returns are shown in column 2 and in column 4, respectively. Parameters

for the market model are estimated over 250 trading days ending 11 days prior the CEO turnover announcement. The applied event-study

test statistics are the Standardized cross-sectional test by Boehmer, Musumeci, and Poulsen (1991) for obtaining the t-values and the

Wilcoxon signed rank test for calculating the z-values. The notations ***, **, and * denote significance at the 1%, 5%, and 10% confidence

level, respectively, in a two-tailed test.


Panel A: Inside Successor (n = 133)[-2 0] 0.11% 0.4213 0.25% 0.3515[-1 0] 0.03% 0.3978 0.02% 0.3155[-1 1] 0.12% 0.8460 0.00% 0.1830

-2 0.08% 0.1778 0.07% 1.0477-1 -0.02% -0.5399 -0.05% 0.74220 0.05% 0.7653 0.00% 0.1763

Panel B: Voluntary Departure (n = 148)[-2 0] -0.07% 0.4903 0.24% 0.5054[-1 0] -0.18% 0.1108 -0.05% 0.2910[-1 1] 0.13% 1.1945 0.35% 0.9571

-2 0.10% 0.9372 0.11%* 1.5391-1 -0.07% -0.6307 -0.06% 0.87100 -0.10% 0.5017 0.00% 0.4097

Panel C: Prior Overperformance (n = 90)[-2 0] -0.84% -1.1554 -0.01% 0.9999[-1 0] -0.82% -1.2317 -0.13%* 1.6276[-1 1] -0.62% -0.6568 -0.42% 0.9878

-2 -0.01% -0.3144 0.01% 0.8188-1 -0.17% -1.0855 -0.06% 0.99590 -0.66% -0.7752 0.00% 0.9838

59

Table B.1: – Continued


Panel A: Inside Successor with Voluntary Departure (n = 94)[-2 0] -0.26% -0.1241 0.08% 0.0283[-1 0] -0.35% -0.3965 -0.00% 0.6015[-1 1] 0.11% 0.5759 0.00% 0.2960

-2 0.08% 0.5348 0.12% 1.1219-1 -0.24% -1.1601 -0.06% 1.18970 -0.11% 0.1847 0.00% 0.2998

Panel B: Inside Successor with Prior Overperformance (n = 57)[-2 0] -1.10% -1.2887 -0.02% 1.0766[-1 0] -0.95% -1.1503 -0.14%* 1.4738[-1 1] -0.88% -0.9048 -0.46%* 1.3388

-2 -0.16% -0.7641 0.12% 0.4568-1 -0.31%* -1.3162 -0.11% 1.13220 -0.64% -0.5862 0.08% 0.5999

Panel C: Voluntary Departure with Prior Overperformance (n = 69)[-2 0] -0.61% -0.6569 0.16% 0.6248[-1 0] -0.79% -0.9278 -0.13% 1.0613[-1 1] -0.40% -0.1341 -0.38% 0.4036

-2 0.18% 0.6134 0.19%* 1.3363-1 -0.09% -0.6054 -0.05% 0.57100 -0.70% -0.7420 0.00% 0.7025

60

Table B.2: Impact of CEO Turnovers on Trading Volume

This table reports the test statistics for the trading volume around the CEO turnover announcement. The mean and median abormal trading volume are calculated

with non-log-transformed trading volumes as in Equation 1. The applied test statistics are the Traditional and the Portfolio test (Brown and Warner, 1980), the

Standardized-residual test (Patell, 1976), the Ruback (1982) test, the Standardized cross-sectional test (Boehmer, Musumeci, and Poulsen, 1991), a Bootstrap test,

the Corrado (1989) rank test, the Generalized rank test (Kolari and Pynnonen, 2008) and the Wilcoxon signed rank test. The mean abnormal trading volume is shown

in the second column and the median abnormal trading volume in the seventh column. Tests have been conducted one-sided. The notations ***, **, and * denote

significance at the 1%, 5%, and 10% confidence level, respectively.

Days Mean Portfolio Patell Boehmer Bootstrap Median Corrado Rank Gen. Rank WilcoxonPanel A: Inside Successor (n = 133)

[-3 -1] 31.03% 0.6667 -0.6528 -0.3767 1.7015** -51.23% 1.4857* 1.3703* 0.0236[-2 2] 245.19% 8.9946*** 6.6192*** 2.5842*** 3.8254*** 59.74% 5.1070*** 3.8936*** 3.2800***[1 3] 122.87% 5.7488*** 3.8929*** 1.8025** 3.6323*** -1.01% 3.7533*** 2.8496*** 2.4895***

-1 9.11% 1.1737 0.1553 0.1239 1.7079** -16.11% 1.2966* 1.4081* 0.46600 128.97% 10.3990*** 9.3041*** 5.2248*** 3.8254*** 17.31% 4.3169*** 4.3539*** 5.3035***1 53.66% 5.8346*** 4.2154*** 2.8705*** 3.8254*** -3.23% 3.1509*** 3.2166*** 3.4866***

Panel B: Voluntary Departure (n = 148)[-3 -1] 49.83% 2.3928*** -0.5673 -0.3074 2.9772*** -40.54% 2.1321** 1.8126** 0.4939[-2 2] 204.24% 8.2105*** 3.9606*** 1.5883* 3.8143*** 11.29% 4.6707*** 3.3009*** 2.8829***[1 3] 97.80% 5.6628*** 2.0825** 1.0208 3.8143*** -2.15% 3.4050*** 2.6933*** 2.3737***

-1 10.22% 1.9563** -0.2384 -0.1859 2.7660*** -29.92% 1.4616* 1.5641* 0.78870 103.04% 7.9677*** 6.0561*** 3.6375*** 3.8143*** -5.46% 3.4268*** 3.4826*** 4.2076***1 36.92% 4.7358*** 2.8335*** 2.0058** 3.8143*** -11.68% 2.6628*** 2.7367*** 2.9748***

Panel C: Prior Overperformance (n = 90)[-3 -1] 3.27% -0.3021 -2.4768*** -1.5178* 0.9621 -71.42% 1.4383* 0.6445 0.9596[-2 2] 219.70% 4.9439*** 2.5643*** 1.0461 3.6778*** -5.25% 4.0726*** 2.7963*** 1.4868*[1 3] 88.89% 2.5380*** 1.1516 0.5452 2.6316*** -40.63% 2.6975*** 2.3029** 0.8550

-1 -0.63% 0.5028 -0.9261 -0.7702 1.1859 -34.54% 1.0934 1.1964 0.45270 127.21% 6.3488*** 5.3109*** 3.2293*** 3.8781*** -1.00% 3.4454*** 3.4956*** 3.5952***1 51.35% 2.9949*** 2.5533*** 1.6552* 3.1241*** -15.59% 2.4652*** 2.5373*** 1.7684**

61

Table B.2: – Continued

Days Mean Portfolio Patell Boehmer Bootstrap Median Corrado Rank Gen. Rank WilcoxonPanel A: Inside Successor with Voluntary Departure (n = 94)

[-3 -1] 54.63% 0.9659 -0.3855 -0.2067 1.8409** -44.80% 2.1677** 1.9651** 0.1339[-2 2] 200.63% 5.2113*** 3.4098*** 1.2518 3.6717*** 9.65% 4.7403*** 3.4424*** 1.6573**[1 3] 96.16% 2.9941*** 1.1965 0.5390 2.7575*** -47.45% 3.1476*** 2.6487*** 0.9258

-1 10.86% 0.7918 -0.0743 -0.0563 1.4543* -23.62% 1.4619* 1.5742* 0.13390 96.95% 6.1835*** 5.6131*** 3.1112*** 3.8711*** -5.46% 3.6727*** 3.7284*** 3.2147***1 28.02% 2.9691*** 1.7871** 1.2027 3.2504*** -18.96% 2.5935*** 2.6768*** 1.6950**

Panel B: Inside Successor with Prior Overperformance (n = 57)[-3 -1] 24.32% -0.8133 -1.4474* -0.8498 0.1607 -75.00% 1.0317 0.8704 1.0527[-2 2] 223.39% 2.9319*** 3.2823*** 1.2479 2.3366** 7.08% 3.9935*** 2.8392*** 0.9653[1 3] 105.46% 1.4694* 1.9447** 0.8440 1.4792* -45.88% 2.7579*** 2.4425*** 0.6714

-1 4.05% -0.0463 -0.1623 -0.1288 0.3421 -23.04% 0.8685 0.9865 0.50450 111.31% 4.5320*** 5.5797*** 3.3087*** 3.7619*** -2.24% 3.9332*** 3.9776*** 3.0073***1 43.64% 2.0291** 2.1644** 1.4280* 2.0304** -19.21% 2.4458*** 2.5259*** 1.2673

Panel C: Voluntary Departure with Prior Overperformance (n = 69)[-3 -1] 31.86% 0.8431 -1.3911* -0.7918 1.4854* -59.45% 1.8872** 1.3758* 0.0209[-2 2] 268.95% 5.2197*** 2.7976*** 1.0633 3.4372*** -17.58% 4.1006*** 2.7067*** 1.2825*[1 3] 107.30% 3.1494*** 1.2662 0.5756 2.6490*** -35.39% 2.5269*** 2.3078** 1.0015

-1 5.97% 0.9801 -0.3557 -0.2839 1.4839* -37.65% 1.3178* 1.4129* 0.12850 141.92% 5.4935*** 4.5553*** 2.6765*** 3.9289*** -4.93% 3.1656*** 3.2200*** 2.8011***1 56.31% 3.1572*** 2.2793** 1.4452* 2.9781*** -11.98% 2.3448*** 2.4173*** 1.6831**

62

Figure 1: Sample Composition

This figure shows the composition of the sample with respect to the departure type (“FORCED”), the prior relative stock performance

(“UNDERPERFORMANCE”), and the successor origin (“OUTSIDER”). The figure reports the number of events that fall in each

category (“N”) together with the percentage of the total sample (in parenthesis).

N=26(12.50%)

N=14(6.73%)

N=7(3.37%)

N=14(6.73%)

N=28(13.46%)

N=25(12.02%)

N=51(24.52%)

TOTAL SAMPLE: N=208 (100%)

OUTSIDERN=75,(36.06%)

FORCEDN=60,(28.85%)

UNDERPERFORMANCEN=118,(56.73%)

Insider & Voluntary& Overperformance

N=43(20.67%)

63

Figure 2: Abnormal Returns

These figures show the average cumulative abnormal returns over a 7-day event window around the CEO turnover announcement date

(t=0). While the cumulative abnormal returns for the total sample are shown in Figure (a), Figures (b)-(f) refer to cumulative abnormal

returns of selected subsamples. The grey areas depict 5% confidence intervals. If a specific hypothesis was set up for a subsample, the

intervals refer to 1.64 sigma bounds. In no hypothesis was formulated, the intervals refer to 1.96 sigma bounds around the average

cumulative abnormal returns.

(a) Total Sample (b) Successor Origin

−3 −2 −1 0 1 2 3−4

−2

0

2

4

6

8

10x 10

−3 Cumulative Abnormal Return

Day

Total Sample

−3 −2 −1 0 1 2 3−0.015

−0.01

−0.005

0

0.005

0.01

0.015

0.02

0.025Cumulative Abnormal Return

Day

InsiderOutsider

(c) Departure Type (d) Prior Relative Performance

−3 −2 −1 0 1 2 3−0.02

−0.01

0

0.01

0.02

0.03


Day

VoluntaryForced

−3 −2 −1 0 1 2 3−0.015

−0.01

−0.005

0

0.005

0.01

0.015

0.02


Day

UnderperformanceOverperformance

(e) Outside Successions and Departure Type (f) Forced Departures and Prior Performance

−3 −2 −1 0 1 2 3−0.04

−0.02

0

0.02

0.04

0.06

0.08


Day

OUT*FOROUT*VOL

−3 −2 −1 0 1 2 3−0.04

−0.02

0

0.02

0.04

0.06


Day

FOR*OVEFOR*UND

64

Figure 3: Rejection Rates of the Null Hypothesis of Zero Abnormal Performance

These figures show the empirical rejection rates (y-axis) for the Bootstrap, the Standardized cross-sectional (Boehmer, Musumeci, and

Poulsen, 1991), and the Wilcoxon signed rank test in dependence of the abnormal returns (x-axis). Abnormal returns are artificially

introduced on the event day of the the randomly drawn sample and range between -3% and +3%. The underlying sample for this

historical simulation comprises 160 realized stock-price histories in the period from January 1990 to June 2009. This original sample is

used to generate 250 event-study samples by randomly choosing with replacement a number of securities, N (N ∈ {10; 20; 50; 100}). The

Bootstrap test is conducted by resampling 1, 000 times from the initial sample of abnormal returns. The rejection rates of the simulations

are obtained in a two-tailed test at the 5% significance level. Figure (a), Figure (b), Figure (c), and Figure (d) show the empirical

rejection rate for a sample of 10, 20, 50, and 100 securities, respectively.

(a) Sample of 10 Events (b) Sample of 20 Events

−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 30

10

20

30

40

50

60

70

80

90

100Rejection Rates of the Null−Hypothesis with 10 Events

Induced Level of Abnormal Performance (in %)

Rej

ectio

n R

ate

(in %

)

BootstrapBoehmerWilcoxon

−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 30

10

20

30

40

50

60

70

80

90



Rej

ectio

n R

ate

(in %

)


(c) Sample of 50 Events (d) Sample of 100 Events

−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 30

10

20

30

40

50

60

70

80

90



Rej

ectio

n R

ate

(in %

)


−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 30

10

20

30

40

50

60

70

80

90



Rej

ectio

n R

ate

(in %

)


65

Figure 4: Abnormal Trading Volume

These figures show the daily standardized mean abnormal trading volume, STV A, in percentage points over a 7-trading-day window in

the surroundings of the CEO turnover announcement date (t = 0). While Figure (a) refers to the total sample, Figures (b)-(f) plot the

development of the cumulative standardized mean abnormal trading volume for selected subsamples. The dashed lines show the standard

error of the mean daily abnormal trading volumes.


−3 −2 −1 0 1 2 30

20

40

60

80

100

120

140Daily Standardized Abnormal Trading Volume in %

Day

Total Sample

−3 −2 −1 0 1 2 30

20

40

60

80

100

120


Day

InsiderOutsider


−3 −2 −1 0 1 2 3−50

0

50

100

150


Day

VoluntaryForced

−3 −2 −1 0 1 2 3−20

0

20

40

60

80

100

120


Day



−3 −2 −1 0 1 2 3−50

0

50

100

150


Day

OUT*FOROUT*VOL

−3 −2 −1 0 1 2 3−50

0

50

100

150

200

250


Day

FOR*OVEFOR*UND

66

Figure 5: Cumulative Abnormal Trading Volume

These figures show the cumulative standardized mean abnormal trading volume, STV A, in percentage points over a 7-trading-day window

in the surroundings of the CEO turnover announcement date (t = 0). While Figure (a) refers to the total sample, Figures (b)-(f) plot

the development of the cumulative standardized mean abnormal trading volume for selected subsamples. The dashed lines represent 1.96

sigma bounds around the cumulative abnormal trading volume.


−3 −2 −1 0 1 2 3−100

0

100

200

300

400

500Cumulative Standardized Abnormal Trading Volume in %

Day

Total Sample

−3 −2 −1 0 1 2 3−100

0

100

200

300

400


Day

InsiderOutsider


−3 −2 −1 0 1 2 3−100

0

100

200

300

400

500


Day

VoluntaryForced

−3 −2 −1 0 1 2 3−100

0

100

200

300

400


Day



−3 −2 −1 0 1 2 3−100

0

100

200

300

400

500


Day

OUT*FOROUT*VOL

−3 −2 −1 0 1 2 3−200

0

200

400

600

800


Day

FOR*OVEFOR*UND

67

Figure 6: Operating Performance

These figures show the development of the median abnormal operating performance in the years surrounding the CEO turnover. In

particular, abnormal operating performance is calculated by subtracting from the realized company’s operating performance the sum

of the lagged company’s OROA and the change in the median operating performance of the companies in the same industry. While

Figure (a) refers to the total sample, Figures (b)-(f) plot the development of the median abnormal operating performance for selected

subsamples.


−2 −1 0 1 2−0.6

−0.5

−0.4

−0.3

−0.2

−0.1

0

0.1

0.2

0.3Median Industry and Lagged Performance Adjusted OROA

Year

Total Sample

−2 −1 0 1 2−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1Median Industry and Lagged Performance Adjusted OROA

Year

InsiderOutsider


−2 −1 0 1 2−1.5

−1

−0.5

0

0.5


Year

VoluntaryForced

−2 −1 0 1 2−2

−1.5

−1

−0.5

0

0.5


Year

OverperformanceUnderperformance

(e) Departure Type with Outside Successions (f) Forced Departures with Prior Performance

−2 −1 0 1 2−1.5

−1

−0.5

0

0.5

1

1.5Median Industry and Lagged Performance Adjusted OROA

Year

OUT*FOROUT*VOL

−2 −1 0 1 2−2

−1.5

−1

−0.5

0

0.5


Year

FOR*OVEFOR*UND

68

Is a CEO Turnover Good or Bad News?...Is a CEO Turnover Good or Bad News? Abstract We investigate...

Documents

Transcript of Is a CEO Turnover Good or Bad News?...Is a CEO Turnover Good or Bad News? Abstract We investigate...