Abstract
This study investigates the effect on stock return volatility of a significant event in the life of a firm, a change in its Chief Executive Officer (CEO). Citing weaknesses in the prior literature, we bring a new approach to re-examine the issue. Firstly, we use a relatively unbiased classification system using both company announcements and media reports. Secondly, we use short-term stock return volatility as a more accurate estimator to isolate the effect of a single disclosure. We find strong evidence that the level of stock return volatility increases following announcements of CEO departures, and that the increase is significantly higher following announcements of forced departures compared to voluntary departures. The results are consistent with signalling effect theory in that forced dismissals convey previously unknown information to the market. Signed cumulative abnormal returns are also more negative for a forced CEO departure.
1. Introduction
Chief Executive Officers (CEOs) are often considered to be the public face of a company, as they directly impact upon the allocation of a firm’s resources and ultimately firm value; a change in executive leadership is a significant event in the life of a firm. Investors revise expectations of firm value when firms release CEO departure announcements through the stock exchange. Many studies have shown that CEO changes can have a significant impact on shareholder wealth and on firm operations (Denis and Denis, 1995; Huson et al., 2004; Weisbach, 1995). Aivazian et al. (2009) argue that when CEOs are perfect substitutes for one another, managers are interchangeable inputs into the production process, and thus unlikely to affect firm value. In reality, however, managers tend to differ in their preferences, risk aversion, skill levels and opinions so that corporate policies, and therefore firm value, depend on who is in control. Theoretically, during the period that investors are revising their expectations of firm value, stock return volatility will be heightened (Jackson, 2011).
Specifically, the research question addressed in this paper is whether the level of market uncertainty, as proxied by stock return volatility, differs between voluntary and forced departures in the Australian context. The level of market uncertainty is a question of interest both to the company and market participants, as volatility affects various aspects of the firm. For the company, volatile stock prices influence not only firm value, but also investment and financing decisions through the cost of capital, and internal firm decisions such as whether to issue stock-based compensation (Baiman and Verrecchia, 1995). 1 Investors are concerned about the level of short-term stock return volatility, as unstable stock prices affect investment strategy and the value of their investment, and may also create valuable arbitrage opportunities (Fleming et al., 2001). Short-term stock return volatility is also important from a high-frequency trader’s perspective in the sense that they rely on volatility in order to profit from rapid price swings in the market.
Information effect theory suggests that stock return volatility increases when new information is released to the market, as market participants revise their assessments of firm value by accounting for the implications of the information. The longer the information is analysed by the market, indicated by prolonged, heightened volatility, the greater the uncertainty of the implications for firm value (Bookstaber and Pomerantz, 1989). This temporary spike in stock return volatility returns to normal levels once the revised firm value has been determined at an equilibrium point (Garman and Klass, 1980).
While many prior studies have found significant changes in stock prices at the announcement of a CEO turnover, few studies examine the stock return volatility effects arising from uncertainty relating to the CEO departure. Two studies, Clayton et al. (2005) and Intintoli (2010), examine the impact of CEO departure announcements on stock return volatility, finding a significant increase in long-run volatility following CEO departure announcements, particularly for forced departures. 2 However, as these studies categorised forced and voluntary departure announcements based only on information contained in the official departure announcement, instances of forced departures, where the firm deliberately avoided specifying the departure as a dismissal, are misclassified.
In addition, both Clayton et al. (2005) and Intintoli (2010) examine the long-run volatility effects of a change in CEO. We take a different approach, by examining the short-term shift in volatility and short-term abnormal returns created by an individual corporate disclosure. 3 Since long-run volatility primarily reflects the underlying economics of the firm, long-window studies could lack the power to document short-term shifts in volatility created by individual corporate disclosures. Further, long-run volatility is limited in its ability to capture the uncertainty surrounding a particular event. Where stock return volatility is calculated over a period of two to three years, there will be other information that will impact on the estimation, unrelated to the CEO departure.
Motivated by these two weaknesses in the existing literature, this study employs a more accurate method of classification between forced and voluntary departures than that used by either Clayton et al. (2005) or Intintoli (2010). As shareholders respond to CEO departure announcements based on their opinion of departure type formed from existing knowledge of the departure context, this study more closely aligns departure classification by examining media reports and other press releases prior to the CEO departure, and classifying based on this understanding of departure context. We argue that both these weaknesses will bias previous results upwards. In the first instance, a misclassification of departure reasons will result in the ‘forced’ sample containing the most extreme values, inflating the effects and distorting the difference between the two samples.
Secondly, long-term measures of volatility may be inflated by confounding events not linked to the announcement of the CEO departure, which may bias towards finding results that a change in CEO will impact volatility. By contrast, short-term measures of daily stock return volatility are able to isolate the effect of the CEO departure announcement without the contagion of other information being disclosed throughout the estimation period. As a result, we argue that short-term volatility is a more accurate estimator to isolate the effect of a single disclosure. We re-examine the issue, addressing both of these weaknesses.
To investigate the research question, a sample of companies in Australia releasing an announcement of a CEO departure between the years 1998 and 2009 was analysed. Even though there are differences in the Australian institutional environment (Lau et al., 2009; Suchard et al., 2001), 4 we do not expect these to produce different results to those studies conducted in the US. Consistent with prior studies in Australia, we study departures of both managing directors (MDs) and CEOs, which are used interchangeably in Australia (Wells, 2002). Both will be referred to henceforth as CEOs. To confirm that the Australian environment produces similar results, we first examine our research question using departure classification rules based solely on the reasons given in the company announcements. As expected, there is an under-representation of forced departures. Despite a lack of statistical power, using short-term volatility produces similar inferences to those documented in prior literature (Clayton et al., 2005; Intintoli, 2010).
We hypothesise, consistent with Clayton et al. (2005) and Intintoli (2010), that this increase in stock return volatility is greater for forced departure announcements compared to voluntary departure announcements. While both types of departure are associated with uncertainty relating to the succeeding CEO’s relative ability, forced departures are associated with an additional signalling effect. This effect relates to uncertainty about possible changes in the firm’s strategy and its future prospects, as CEO dismissals usually signal severe internal firm problems and a possible change in company direction in an attempt to revive firm performance (Clayton et al., 2005).
Results show that stock return volatility for firms with CEO departure announcement increases to nearly twice that of a control sample on day –1. More substantial differences are observed on the announcement date, where the volatility for CEO departure announcement firms reaches nearly three times the volatility of their matched counterparts. This heightened volatility gradually eases and returns to normal levels on day +4, signalling that the uncertainty stemming from CEO departure announcements lingers for a few days following the announcement. This result thus provides evidence consistent with heightened market uncertainty following CEO departure announcements, which is not resolved until at least four days after the announcement.
Multivariate analyses controlling for departure-specific and firm-specific variables similarly find that volatility following forced departure announcements is significantly higher than the volatility following voluntary departure announcements. Therefore, results are consistent with the signalling effect hypothesis that the stock return volatility for forced departures is greater than that for voluntary departure announcements.
Signed abnormal returns surrounding the announcement of a CEO departure are also examined and the results are consistent with our predictions. Firstly, compared to a matched-control group, abnormal returns for the sample of CEO departure firms are more negative. When extended to examining the difference in abnormal returns between forced and voluntary departures, abnormal losses are greater for forced departures. Importantly, results indicate that the cumulative abnormal returns surrounding the announcement of a CEO departure are less of a loss, all else being equal, when a successor is simultaneously announced.
Overall, we make a number of contributions to the literature. Firstly, consistent with prior literature (Engel et al., 2003; Farrell and Whidbee, 2003; Pourciau, 1993), we use a more accurate way to differentiate between forced and voluntary departures. By using media reports to classify CEO departures into comprehensive departure reasons, and linking these departure reasons to departure type, we provide an effective method of categorisation with a balance between subjectivity and accuracy. The importance of a classification system using corroborating evidence is demonstrated by comparing the results of the categorisation used in this paper with one based solely on the reasons given in the company announcements. Secondly, findings contribute to the limited stream studying volatility effects from CEO departure announcements.
Our results are significant for a number of reasons. Firstly, the improved classification system demonstrates the usefulness of the financial press in providing additional and, arguably, relatively unbiased information to markets, especially when firms are faced with incentives for framing their disclosures in a certain light. Secondly, results on the cumulative abnormal returns highlight the importance for firms to quickly resolve any uncertainties about the future direction of the firm by simultaneously announcing a CEO successor.
The rest of the paper is organised as follows. Prior literature is reviewed in the next section. The third section sets out the formal development of our testable hypotheses. Section 4 describes the data selection and research design. Our empirical results are presented in the fifth section, and the sixth section concludes.
2. Prior literature
2.1. Announcement of CEO departure
The literature surrounding CEO departures can be categorised into two main streams. The first examines the pre-announcement period, and documents which variables affect the likelihood of CEO turnover, while the second stream studies the post-announcement period and investigates the impact of CEO turnover on capital markets.
Many studies have documented an association between CEO turnover probability and firm performance. Since voluntary CEO turnover is likely to be attributable to reasons unrelated to internal firm performance, the focus of this literature is on predicting the incidence of forced CEO turnover. Empirical evidence shows a negative association between CEO turnover probability and firm performance as measured by current and lagged accounting performance or share market returns (Gregory-Smith et al., 2009; Leker and Salomo, 2000; Warner et al., 1988; Weisbach, 1988). Other indicators examined as measures of firm performance are the number of negative performance surprises (Dikolli et al., 2009) and deviation from expected performance (Farrell and Whidbee, 2003). However, a board’s decision to dismiss a CEO may not depend on any one single measure, but rather on a variety of measures considered collectively (Engel et al., 2003).
Within an Australian context, Suchard et al. (2001) do not find a significant relationship between current share market performance and CEO turnover. Their results suggest a relationship between CEO turnover and one-year lagged performance for large firms. They attribute this difference in findings for Australian firms to the less effective market for corporate control in Australia, where boards face less pressure to remove CEOs for poor short-term performance. Their study includes all types of CEO departure, so it is possible they did not find an association between firm performance and CEO turnover because of voluntary departures, which do not appear to have any association with firm performance.
Lau et al. (2009) revisit this question, limiting their sample to forced dismissals among the top 100 companies in Australia, where corporate governance mechanisms are considered to be more effective. They use both accounting and market-based performance measures, and find evidence of a negative relation between corporate performance and the likelihood of CEO turnover.
The second stream of literature relating to CEO departures focuses on the post-announcement period and examines how capital market participants react to announcements of CEO departures. The majority of these studies investigate abnormal returns associated with the announcements, while few study the effects on trading volume or stock return volatility.
Clayton et al. (2005) are the first to report on the effect of CEO turnover announcements on stock return volatility, which is seen as a measure of short-term uncertainty about firm value. Consistent with intuition, they find CEO departures on average lead to increased stock return volatility. The authors propose the strategy hypothesis and the ability hypothesis to explain this increase in volatility. The strategy hypothesis proposes that volatility following a CEO departure results from increased uncertainty about the corporate strategy that would be adopted by the new CEO, while the ability hypothesis proposes that volatility stems from uncertainty relating to the skill level of the incoming CEO. This ability hypothesis is in line with Dikolli et al. (2009), who suggest that others are initially uncertain about the ability of the new CEO, but their uncertainty is gradually resolved over time.
While both the strategy and ability hypotheses work simultaneously to heighten stock market uncertainty following all types of CEO departure, Clayton et al. (2005) suggest that the strategy effect is significantly higher for forced CEO departures. They consider this finding is due to dismissals being a means of indicating to investors a change in corporate strategy in an attempt to improve performance, while voluntary resignations are not accompanied by such signals. Consistent with this argument, they find that forced CEO turnovers lead to a more substantial increase in stock return volatility compared to voluntary turnovers.
Succession type has also been found to interact with departure type in determining the level of market uncertainty following CEO departures. Assuming that forced departures are accompanied by a change in corporate strategy, the ability of insiders to continue in the firm under a new direction is no more certain than the ability of an outsider. This argument is supported by Clayton et al. (2005) and Intintoli (2010) in their findings that inside and outside successions have no significant difference on volatility for forced departures. However, succession type directly influences the level of market uncertainty following voluntary resignations, because the market is more certain of the performance of insiders, compared to outsiders, where transfer of internal knowledge is expected (Clayton et al., 2005).
There is evidence that regardless of departure type, successions by insiders significantly reduce market uncertainty (Clayton et al., 2005). This can be explained by insiders being less likely to bring about dramatic change in strategy, and investors being better able to predict their ability. However, prior research also suggests that appointment of an outsider who comes from a well-known firm in the same industry can mediate uncertainty surrounding their ability (Clayton et al., 2005).
In summary, Clayton et al. (2005) and Intintoli (2010) find higher stock return volatility following forced departure announcements compared to voluntary departures. Both studies rely on simple categorisations, classifying a departure as forced when it is specifically stated as such in the announcement, or when the departing CEO is under the age of 60 and does not leave for health reasons or to assume a role in another firm, whereas we categorise departures as forced or voluntary by scrutinising the context of the turnover using additional media reports.
2.2. Classification of departure type
Existing literature has acknowledged the difficulty of distinguishing between forced and voluntary departures (Dissanaike and Papazian, 2005; Warner et al., 1988; Weisbach, 1988). This difficulty arises from the observation that companies often mask the true reasons behind a CEO turnover when it is initiated by the board, disguising it as an ‘early retirement’ or ‘for personal reasons’, or providing no reason at all. To limit subjectivity and for simplicity, early studies have based the classification between forced and voluntary departures entirely on information provided in the announcement. These studies identify departures as forced if it is explicitly stated using terms such as ‘dismissal’, ‘conflict’ or ‘asked to resign’, and classify the rest as voluntary (Engel et al., 2003; Mahajan and Lummer, 1993; Worrell et al., 1993). This method of classification is reliant on details provided by the firm in the announcement, thus failing to capture instances where firms deliberately avoid explicating the nature of the CEO departure.
However, market participants have access to additional information from a variety of sources, including analyst reports, previous announcements and the media. By drawing upon information contained in such sources, market participants can form an opinion about the true reason for departure. Dherment-Ferere and Renneboog (2002) and Cools and Van Praag (2007) categorise a departure as forced where it is either stated to be forced in the announcement, or where a forced departure is alluded to in media reports. This categorisation method has been complemented by simple rules, such as using CEO age to distinguish between genuine retirements and forced early retirements (Kind and Schlapfer, 2010; Parrino, 1997). While CEO age is a simple rule and easily applied, Dissanaike and Papazian (2005) find many instances of genuine early retirements, suggesting that the age criterion is an inaccurate discriminator.
Addressing this weakness, Dissanaike and Papazian (2005) use clinical investigation to scrutinise the context of each CEO turnover, using media and analyst reports. They form an opinion for the reason of departure based on these additional sources of information, and classify each individual departure as either voluntary or forced. While this method is more accurate in the absence of inside information, the lack of specific rules creates subjectivity and inconsistency.
Consequently, a number of studies have attempted to achieve a balance between objectivity and accuracy by using publicly available reports to determine departure reason, while classifying departures according to specific guidelines to limit subjectivity and provide for consistency. Lubatkin et al. (1989) and Cools and Van Praag (2007) use information from media reports to sort departures into a number of categories based on departure reason, such as dismissed, ousted, internal change, interim appointment, retirement, retired to part time, change of control, head-hunted and unclassified. Each departure reason is then categorised as either voluntary or forced. This method provides for the same level of accuracy as the clinical method, but offers greater clarity and consistency. Effectively, the subjectivity associated with determining the reason for departure is balanced with the objective association of the departure type with the departure reason. Since this method adopts a balance between subjectivity, efficiency and accuracy, our classification system adopts the same approach. A summary of the departure reasons and their categorisation as voluntary or forced is contained in Table 1.
Departure categories.
This table presents the departure categories in which each observation in the CEO departure announcement sample was classified. Each departure reason is associated with a category for departure reason and departure expectedness. The broad classification captures whether CEO departures are forced/voluntary. The detailed classification categorises into three categories for each variable, specifically forced/semi-involuntary/voluntary. The distribution between the categories of departure reasons is shown in parentheses, with the number of observations in the final sample.
A potential limitation of relying on media reports is that misreporting may lead to inaccurate classifications. This limitation is acknowledged, but we believe the implied third-party objectivity over the firm’s disclosure framing outweighs any concern.
3. Hypothesis development
The effect of CEO departure announcements on shareholder wealth is a function of the real and information effects, as provided by signalling theory (Warner et al., 1988). Similarly, the effect of CEO departure announcements on volatility is a function of two main effects. Clayton et al. (2005) suggest ability and strategy effects, where the ability effect relates to uncertainty about the ability of the incoming CEO and the strategy effect relates to uncertainty around possible changes to corporate strategy.
The strategy effect does not capture all signals resulting from the CEO departure announcement. In instances where the CEO is not immediately replaced, uncertainty arises about the level of productivity decreases and missed opportunities caused by the absence of a permanent leader (Intintoli, 2010). CEO turnovers therefore may also signal news relating to current and future firm performance, and trigger uncertainty in the market due to reasons unrelated to the former CEO’s departure.
Following from the above argument, volatility increases are a function of: (i) an ability effect, which relates to uncertainty relating to the human capital of the incoming CEO; and (ii) a signalling effect, which relates to uncertainty stemming from signals associated with the departure announcement, such as future strategy, the firm’s current condition and its future opportunities. Consistent with this proposition, the literature has found significant stock return volatility increases following announcements of CEO departures (Clayton et al., 2005; Intintoli, 2010).
Consistent with this prior literature, information effect theory posits that new information released to the market, assuming it is price sensitive and the market is efficient, will shift about the consensus view of the price of the asset to its new fundamental value. During the time it takes to disseminate the information, investor uncertainty is expected to increase, as the new information is digested and expectations about future performance are assessed (Francis et al., 2002; Krinsky and Lee, 1996; Venkatesh, 1989). In addition, during the price discovery process, the variance of expectations fluctuates. This fluctuation can be further exacerbated by increases in adverse selection costs due to the presence of information processors with superior ability (Kim and Verrecchia, 1994), or by the activities of noise traders (Rogers et al., 2009).
Thus, a period of heightened volatility should follow CEO departure announcements as market participants exchange information with each other to predict the impact of the departure on firm value and revise expectations on firm value according to additional signals in the departure announcement. Accordingly, the following hypothesis is proposed:
H1: Stock return volatility increases upon the announcement of a CEO departure.
The expected impact of the announcement of a CEO departure is not as clear cut when examining the share price implications of the disclosure. While information effect theory suggests that mean expectations, and therefore price, will change, the predicted direction of the change will depend on the manner in which the announcement is viewed by the market. If it is assumed that the interpretation of the announcement as being ‘good’ or ‘bad’ is entirely random, then the expected value signed abnormal returns would likely be zero, and indistinguishable from the return on a matched firm for which there is no news at the time.
However, given the discussion above, a directional prediction can be made. Given that the signal to the market of the announcement of a CEO departure is more often than not negative, it is expected that, on average, the cumulative abnormal returns will be negative compared to a matched-control firm. Therefore, the following hypothesis is proposed:
H2: Cumulative abnormal return is on average lower for firms when they announce a CEO departure relative to a matched firm.
When considering the difference in volatility for forced and voluntary CEO turnovers, the aggregate ability and signalling effects must be considered. Firstly, it can be argued that the CEO ability effect is higher for voluntary CEO departures compared to forced departures. As there are costs associated with replacing a CEO, it is expected that the board would dismiss the CEO only when the ability of the current CEO is significantly lower than the ability of a replacement CEO. On the other hand, for a voluntary CEO departure, whether the departing CEO is replaced by a CEO with a similar or higher ability depends on candidate availability. Following this argument, it is likely that the CEO ability effect on market uncertainty is higher for voluntary CEO departures compared to forced departures.
In contrast, the signalling effect is higher for forced than for voluntary CEO departures. Voluntary resignations are less likely to signal unknown information about the firm to the market. Conversely, forced dismissals may signal information previously unknown to the market such as poor firm prospects, scandals, board conflicts or changes in corporate strategy. In particular, a change in strategy is a common motivation for the board’s firing of an existing CEO (Clayton et al., 2005). Therefore, it can be argued that the signalling effect on market uncertainty is higher for forced CEO departures.
Considering the CEO ability and signalling effects together, these opposing effects suggest that their relative strengths determine whether forced or voluntary departure announcements are associated with a higher level of volatility increase following CEO departure announcements. The literature finds higher stock return volatility following forced departure announcements (Clayton et al., 2005; Intintoli, 2010), indicating that the magnitude of the signalling effect is larger than the CEO ability effect. Consistent with this literature, the following hypothesis is proposed:
H3: Stock return volatility is higher when the CEO’s departure is forced than when it is voluntary.
To the extent that forced departures are likely to signal severe internal problems, changes in company direction, poor firm prospects, scandals and board conflicts (Clayton et al., 2005), they are likely to be interpreted by the market as more negative compared to voluntary departures. While voluntary departures are also likely to signal negative information to the market, it is expected to be more pronounced for forced CEO departures. As a result, the following hypothesis is proposed:
H4: Abnormal returns around the time of the announcement of the CEO’s departure are lower when the departure is forced than when it is voluntary.
An alternate hypothesis could be proposed to that in H4. If it were assumed that the announcement of a forced CEO departure signalled to the market that the firm is at ‘rock bottom’, it would be expected that the abnormal return would be greater than if the departure were voluntary. This reaction would be due to investors attempting to buy stocks of forced departure firms while they are at their low, and before firm performance is expected to improve.
4. Data and research design
4.1. Sample selection
The sample is drawn from all companies listed on the Australian Stock Exchange (ASX) at the time of the departure announcement for the financial years ended 1999–2009. These dates were selected as they represent the period for which searchable ASX announcements are available on Datanalysis. Using the keyword search for ‘managing director’, ‘chief executive’ or ‘CEO’, all ASX announcements in the category ‘company administration – director appointments/resignations’ were obtained. These filtered announcements were then screened for news of CEO departures to ensure accuracy and completeness in the sample identification process, as the keyword search does not distinguish between CEO departures and departures of other directors where the CEO/MD was also mentioned in the announcement. Consequently, the initial sample consists of CEO departure announcements identified by this process.
Where multiple announcements of a CEO departure were identified, only the first official announcement is included in the sample. For example, several CEO departures were officially announced months before the departure and again on the departure date itself. Where the CEO signalled a possible intention through unofficial means, only the subsequent official announcement is included in the sample.
The following information was hand-collected for each announcement: (i) announcement date (date of release by ASX to the public); (ii) whether the successor was immediately appointed and, if so, whether they were from within the firm (insider) or sourced from another company (outsider); (iii) whether the departing CEO was a company founder; (iv) whether simultaneous departures of other directors were mentioned in the same announcement; and (v) whether the announcement was classified as market sensitive by the ASX. In addition, the time of the announcement was manually collected using Factiva and SIRCA Australian Equities ASX announcement search functions in order to identify announcements released outside trading hours.
The initial sample comprised 647 CEO departures over the period 1999–2009. Of these, 146 had insufficient information in the announcement, 98 lacked stock price data and 3 could not be matched with a control firm. This results in 400 usable observations for univariate analysis. A further 141 observations were missing additional data required for multivariate analysis, which is based on 259 observations. The sample attrition is summarised in Table 2.
Sample selection.
This table provides the reasons for observation attrition in the sample of CEO resignations over the period 1999–2009.
Table 3 provides the number of CEO resignation announcements by year for both the univariate and final samples. The majority of resignations are voluntary, with 61.3% (62.5%) of the univariate (final) sample being in this category. A larger number of resignations occurred in 2008 and 2009 than other years (73 and 54 announcements, respectively, for the univariate sample).
Number of CEO departures by year.
This table provides the number of CEO resignation announcements by year over the period 1999–2009. The observations are divided into the univariate and final samples, and categorised as either forced or voluntary departures.
To show the effect of the classification system adopted in this paper, forced departures are reclassified consistent with prior literature. When departures are categorised solely according to the reasons stated in the company announcement, 28 departures are categorised as forced, compared to 155 under the classification system summarised in Table 1. In other words, the method adopted in prior research would have correctly identified only 18% of the CEO departures which, all things considered, were likely to have been forced.
4.2. Market uncertainty and returns
Stock return volatility can be measured in a number of ways. Two existing studies investigating the effect of CEO departures on market uncertainty (Clayton et al., 2005; Intintoli, 2010) both measure return volatility as the daily standard deviation of market-adjusted returns over the event period. While this measure has been used widely by volatility studies (for example, Bushee and Noe, 2000; Leuz and Verrecchia, 2000), it does not capture intra-day stock return variation. In other words, this measurement leads to a loss of readily available information (e.g. daily high and low prices) contained in intra-day stock movements.
Further, the focus of Clayton et al. (2005) and Intintoli (2010) is on long-run volatility. Long-run volatility primarily reflects the underlying economics of the firm, as opposed to the short-term changes in volatility that reflect the uncertainty surrounding future expectations of the firm at the time of the announcement. As such, a parsimonious intra-day measure of volatility is preferred in this study.
In order to more efficiently capture intra-day volatility effects, we use the Garman and Klass (1980) ‘best analytic scale-invariant estimator’ as the measure of stock return volatility. This measure accounts for all readily available information, specifically the daily high (H) and low (L) prices as well as the opening (O) and closing (C) prices. This estimator is therefore a more efficient estimator for stock return volatility, being unbiased with minimum variance (Garman and Klass, 1980; Jackson, 2011), and is defined by Equation (1):
where a = ln(H i,t /O i,t ); b = ln(L i,t /O i,t ); x = ln(C i,t /O i,t ) for firm i on day t, and the required inputs for price data are obtained from the Core Research Database.
To provide a cleaner comparison with prior literature, stock return volatility is also measured using the standard deviation of market-adjusted returns over the event window [–7, +7]. Market-adjusted returns are used in calculating cumulative abnormal returns (CAR) over the announcement period. The ASX All-Ordinaries return index, sourced from Datastream, is used to proxy for the market return.
4.3. Hypothesis testing
Both univariate and multivariate tests are conducted. Event time is measured in days and the event window is [–7, +7], where day 0 is the date of the ASX announcement. This event window captures the immediate effects following the announcement, and allows examination of the time it takes for volatility to return to more normal levels.
H1 predicts that stock return volatility increases upon the announcement of a CEO departure. Permutation analysis has been employed to test the significance of the difference in means between the sample of firms announcing a CEO departure and a control firm sample matched on firm size (market capitalisation), industry and performance (return on equity), using financial data provided by Aspect.5,6
H3 predicts how the context of the CEO’s departure affects the magnitude of the volatility increase following the announcement. Specifically, H3 predicts that the volatility increase is higher for forced compared to voluntary departures.
To investigate H3, both univariate and multivariate tests were conducted. Univariate testing was performed using permutation analysis as described for H1, testing the significance in means between voluntary and forced departures. Multivariate testing was also conducted to control for announcement-specific and firm-specific factors. The following ordinary least squares (OLS) regression was estimated to explain stock return volatility on a day-to-day basis over the event window:
The dependent variable, VOL t , represents market uncertainty, as proxied by stock return volatility for day t. The main variable of interest, FORCED, is a dummy variable representing the departure type (1 = forced, 0 = voluntary). H3 predicts β1 is positive.
As noted by Zhang (2006), uncertainty with respect to implications of new information for a firm’s value stems from two sources: the volatility of a firm’s underlying fundamentals and inadequate provision of information. Therefore, both announcement-specific controls and firm-specific controls are included in the regression model. Consistent with information effect theory, the level of detail regarding the CEO’s departure provided in the departure announcement impacts upon market uncertainty, as the provision of more information allows market participants to more accurately predict firm value.
SIMULT refers to whether there are simultaneous departures of other directors, and is assigned a value of 1 for instances of simultaneous departures and 0 otherwise. CEO departures that are accompanied by simultaneous departures of other directors signal higher instability and a larger loss of human capital by the firm, and may be associated with a higher level of market uncertainty relating to current and future firm performance. FOUNDER represents CEOs who have been CEO since incorporation. If the departing CEO is also a company founder, market participants may be more uncertain about the ability of the incoming CEO to lead the firm. A control variable has been included where company founders are coded as 1 and 0 otherwise. TIMING indicates whether or not a successor is immediately appointed and announced in the same ASX announcement, with immediate appointments coded as 1 and 0 otherwise. Announcements of CEO departures where the successor is not immediately appointed could potentially lead to a higher level of uncertainty regarding future firm performance due to the inability of market participants to predict the successor’s ability and preferences.
SUCTYPE is a dummy variable where 1 represents succession by an insider, as prior research has found that the origin of the successor affects market uncertainty (Clayton et al., 2005). Finally, as the literature has found that announcements issued outside market trading hours are associated with a higher level of volatility compared to announcements issued during trading hours (French and Roll, 1986), CLOSED has been included as a control variable, taking a value of 1 if the announcement is made outside trading hours.
Firm-specific variables have been included to control for fundamental measures of uncertainty in the firm’s environment. As firms with a higher information environment are associated with more publicly available firm information and hence lower information uncertainty, ANALYSTS (the number of analysts following the firm) has been included to proxy for the firm’s information environment. Prior research has suggested that firms with higher risk are associated with higher estimation uncertainty (Chordia et al., 2007). Firm risk is controlled by the inclusion of BETA or systematic risk, which is estimated by OLS from daily share market data. Firm performance is controlled for with ROE (return on equity), and the volatility of past performance is proxied by EARNVOL, which is measured by the natural logarithm of the standard deviation of ROE for the previous three years. This has been found to be an important driver of market uncertainty, as it makes it more difficult for the market to predict the impact on firm value (Pastor and Veronesi, 2003). Company size has been controlled for with MKTCAP, which is the natural logarithm of market capitalisation.
Significant association between volatility and trading volume in Australia has been documented in prior research. Specifically, trading generates volatility by itself as traders react to each other (Ruppert, 2004). Trading volume has been controlled for with TRADEVOL t , which is the natural logarithm of trading volume on day t.
In addition, prior-day volatility has been controlled for with VOLt–1. Controlling for the previous day’s volatility isolates the effect of stock return volatility on the day being tested. As well, there is evidence that the best explanation for volatility on day t is the volatility for day t–1, as ‘large changes (in stock prices) tend to be followed by large changes, of either sign, and small changes tend to be followed by small changes’ (Jackson, 2011; Mandelbrot, 1963: 418).
In summary, the regression model noted in Equation (2) includes control variables both for announcement-specific and firm-specific variables. Although prior studies have not controlled for such announcement-specific effects, the arguments above show that these factors potentially impact upon market uncertainty.
Cumulative abnormal returns surrounding the announcement of a CEO departure are examined. Univariate tests, using permutation analysis, are used to test the significance in the difference in means in the cumulative abnormal returns between CEO departure firms and the matched-control firm, as well as between forced and voluntary departures. To empirically test H4, a similar multivariate model to Equation (2) is estimated:
All variables in Equation (3) have the same definitions as in Equation (2), excepting the dependent variable. Here, the focus is on CAR, which is measured over four different horizons: [–1, +1], [–7, +7], [–1, +3] and [–1, +7].
5. Results
Table 4 provides the descriptive statistics for the sample of 259 CEO departure announcements employed for the multivariate analysis. Announcement-specific variables indicate that 44.8% of CEO departure announcements are accompanied by a simultaneous announcement of the successor (TIMING). Of these 44.8%, over two-thirds are inside successions (SUCTYPE). Simultaneous departures of other directors occurred for 6.2% of the sample (SIMULT), and 8.9% of departing CEOs were company founders. Announcements are as likely to be disclosed outside trading hours as during the trading day, with 52.1% being publicly released when the market is closed (CLOSED).
Descriptive statistics.
This table reports descriptive statistics for the variables used for multivariate analyses conducted for the final sample of 259 observations drawn from CEO departure announcements from 1999 to 2009. FORCED is whether the CEO departure was forced by firm circumstances; SIMULT whether there were simultaneous departures of other directors; FOUNDER is whether the departing CEO is also the company founder; TIMING whether the successor is simultaneously announced; SUCTYPE is whether the successor is from within (=1) or outside (=0) the firm; CLOSED is whether the announcement was released outside of trading hours; ANALYSTS is the number of analysts following as a proxy for information environment; BETA is firm risk as measured by ordinary least squares (OLS) beta; ROE is the return on equity for the financial year prior to the announcement; EARNVOL is the natural logarithm of earnings volatility, measured as the standard deviation of ROE for the past three years; and MKTCAP is the natural logarithm of market capitalisation.
Firm performance, as measured by ROE, is skewed negatively, with a mean (median) of 0.004 (0.093). This can be attributed to the sample of forced departures, which generally occur in poorly performing firms. After taking the natural logarithm, both earnings volatility (EARNVOL) and market capitalisation (MKTCAP) are approximately normally distributed.
Table 5 presents Pearson and Spearman correlations for variables included in the regression model. FORCED is positively correlated with SIMULT and EARNVOL, suggesting that forced departures tend to occur in firms with unstable performance and where the board changes are substantial. The positive correlation with ANALYSTS can be explained by a need for firms with higher analyst following to undertake corrective CEO action in response to poor firm performance.
Correlation matrix.
This table presents the correlation matrix for the final sample of 259 observations of CEO departure announcements from the period 1999–2009. Pearson (Spearman) correlations are presented and are shown above (below) the diagonal. FORCED is whether the CEO departure was forced by firm circumstances; SIMULT whether there were simultaneous departures of other directors; FOUNDER is whether the departing CEO is also the company founder; TIMING is whether the successor is simultaneously announced; SUCTYPE is whether the successor is from within (=1) or outside (=0) the firm; CLOSED is whether the announcement was released outside of trading hours; ANALYSTS is the number of analysts following as a proxy for the information environment; BETA is firm risk as measured by ordinary least squares (OLS) beta; ROE is the return on equity for the financial year prior to the announcement; EARNVOL is the natural logarithm of earnings volatility, measured as the standard deviation of ROE for the past three years; and MKTCAP is the natural logarithm of market capitalisation. Correlations significant at the 1, 5 and 10% levels are represented by ***, ** and *, respectively.
As expected, EARNVOL and BETA are positively correlated, illustrating that the more unstable the firm’s performance, the higher its risk.
5.1. Prior departure classification
To confirm the validity of the sample, the analysis was first run using a categorisation consistent with prior literature. This categorisation resulted in the identification of 28 forced departure announcements. After data limitations described in Table 2, there are 15 (244) forced (voluntary) departures used to run the analysis.
Firstly, Table 6 presents the univariate results of the differences in daily stock return volatility, trading volume and returns for the split of forced and voluntary departures based on classification rules from prior literature. The results are qualitatively consistent with those presented by Clayton et al. (2005) and Intintoli (2010), despite the use of short-term volatility. Specifically, surrounding the CEO departure announcement, stock return volatility is statistically significantly greater for forced departures compared to voluntary. A similar pattern is exhibited for daily trading volume as well.
Daily stock return volatility, trading volume and returns surrounding CEO departure announcements: classification rules from prior literature.
This table presents the means for daily stock return volatility, daily trading volume and returns surrounding ASX announcements of CEO departures. This sample consists of 259 CEO departure firms. Forced represents firms that release an announcement of a forced CEO departure on day 0, while Voluntary represents firms that release an announcement of a voluntary CEO departure on day 0. This sample consists of 15 forced departures and 244 voluntary departures based on classification rules used in prior literature. Volatility is estimated using the Garman and Klass (1980) ‘best analytic scale invariant estimator of volatility’, trading volume refers to the number of shares traded on each day, Std dev returns is a measure of volatility using the standard deviation of market-adjusted returns over the period [–7, +7], and CAR is the cumulative abnormal return. Significance level at the 1, 5 and 10% levels are represented by ***, ** and *, respectively, measured using a resampling without replacement test (permutation test) using 1000 iterations.
For the cumulative abnormal returns surrounding the departure announcement, forced departures are more negative than voluntary departures across all estimation windows. At the extreme, the cumulative abnormal returns over the period [–7, +7] is –13.58% for forced departures, compared with –4.24% for voluntary departures.
Multivariate results are next presented in Table 7. 7 Firstly, Panel A estimates the regressions on daily stock return volatility. It is only the volatility on day 0 that increases. The coefficient on Forced is 0.0018; however, due to a lack of power in the regressions, this result is not statistically significantly.
Daily stock return volatility following CEO departure announcements: classification rules from prior literature.
This table presents the results of running the model estimated in Equation (2), replicating Tables 7 (Panel A), 8 (Panel B) and 11 (Panel C) using the decision rules to categorise forced and voluntary departures. This sample consists of 15 forced departures and 244 voluntary departures based on classification rules used in prior literature. For brevity, only the coefficients on Forced and the Adjusted R2 are presented. All other control variables are qualitatively the same as in Tables 7, 8 and 11. Two-tailed p-values are reported in parentheses.
Using alternate specifications on stock return volatility (Panel C), this pattern is repeated. Cumulative volatility over days [–1, +1] is statistically significantly greater for forced rather than voluntary announcements (0.0043, p-value 0.0362), as is the standard deviation of returns over the period [–7, +7] (0.0127, p-value 0.0628).
Finally, the analysis of cumulative abnormal returns is presented in Panel B of Table 7. Again, the results reported from using the classification scheme from prior literature show consistent results. At the extreme, the three-day abnormal returns surrounding the departure announcement [–1, +1] on Forced is –0.0502 (p-value 0.0968).
5.2. Univariate analysis
Hypothesis 1 posits that stock return volatility increases following announcements of CEO departures. Table 8 reports the daily average stock return volatility for days [–7, +7] for the sample of CEO departure announcement firms and a sample of control firms, matched on industry, size and prior performance. 8 Figure 1 presents these results graphically. Examination of the results shows that CEO departure firms experience a sharp spike in volatility on day 0, which gradually returns to normal levels by day +3. Panel A of Table 8 reveals that the volatility level on day 0 is more than double that of pre-announcement levels. 9 In contrast, the results do not indicate any significant volatility increase for the sample of matched-control firms, in line with information effect theory.
Daily stock return volatility and trading volume surrounding CEO departure announcements: sample and matched-control firms. (a) Daily volatility. (b) Trading volume.
Daily stock return volatility, trading volume and returns surrounding CEO departure announcements.
This table presents the means for daily stock return volatility (Panel A), daily trading volume (Panel B), and returns (Panel C) surrounding ASX announcements of CEO departures. CEO departure firms represent firms that release an announcement of a CEO departure on day 0, and the matched-control firm sample is a matched-pair control, matched on industry, performance (return on equity (ROE)) and size (market capitalisation). This sample consists of 259 CEO departure firms and 259 matched-pair control firms. Forced represents firms that release an announcement of a forced CEO departure on day 0, while voluntary represents firms that release an announcement of a voluntary CEO departure on day 0. This sample consists of 97 forced departures and 162 voluntary departures. Volatility is estimated using the Garman and Klass (1980) ‘best analytic scale invariant estimator of volatility’, trading volume refers to the number of shares traded on each day, Std dev returns is a measure of volatility using the standard deviation of market-adjusted returns over the period [–7, +7] and CAR is the cumulative abnormal return. Significance level at the 1, 5 and 10% levels are represented by ***, ** and *, respectively, measured using a resampling without replacement test (permutation test) using 1000 iterations.
Comparisons between CEO departure firms and matched-control firms show that the difference in average volatility is not significant until day –1. Moreover, the difference on day 0 reaches a statistically significant level of 0.00119, around seven times the pre-announcement difference. Examination of the actual volatility level on day 0 shows the CEO departure firms (0.00192) experience a volatility level of more than twice that of matched-control firms (0.00073). This provides support consistent with H1 that CEO departure announcements are followed by statistically significant increases in stock return volatility.
Panel C of Table 8 provides the results for the returns-based measures. Consistent with the finding using daily stock return volatility, the standard deviation of market-adjusted returns over the period [–7, +7] is not statistically significantly greater for the CEO departure firms. This highlights the importance of using a daily volatility metric when examining short-term shifts in volatility around a corporate disclosure, as there is a loss of information to calculate volatility using only closing prices.
The cumulative abnormal returns across all windows are statistically significantly different for the CEO departure firms compared to their matched-control firms. The information effect theory predicts that in the absence of any information, mean expectations should not change and no abnormal return is expected. Over the period [–7, +7], the control sample support this prediction, with a cumulative abnormal return of –1.82%. On the other hand, the cumulative abnormal return over the period [–7, +7] for the CEO departure sample is –4.81%. Results for the other cumulative abnormal return estimation periods are consistent with this result, although their magnitude is smaller. Overall, results are consistent with H2 in that abnormal returns surrounding the announcement of a CEO departure are lower compared to a control sample.
A similar trend is found for trading volume (Panel B of Table 8 and Figure 1(b)). CEO departure firms experience a significant increase in trading volume around the announcement date, while the matched-control firms maintain a steady trend. The difference in trading volume between CEO departure firms and control firms on day 0 is almost six times the difference in the pre-announcement period. Specifically, trading volume on day 0 for CEO departure firms (2,233,882) is nearly twice that of the control firms (1,299,255).
Following the finding that CEO departure announcements lead to increases in stock return volatility, H3 predicts that stock return volatility increases are higher for forced compared to voluntary departures. Results comparing forced and voluntary CEO departure announcements are presented in Table 8 and Figure 2.
Daily stock return volatility and trading volume surrounding CEO departure announcements: forced and voluntary departures. (a) Daily volatility. (b) Daily trading volume.
Panel A of Table 8 shows that voluntary departure announcements are associated with a smaller increase in volatility compared to forced departure announcements, for which volatility increases to nearly three times its pre-announcement level. Furthermore, although differences between forced and voluntary are significant throughout the event period, the difference on day –1 of 0.00138 is double that in the pre-announcement period, almost doubling again to a peak of 0.00201 on day 0, and then returning to normal levels by day +2. Taken together, these preliminary results are consistent with H3 and with prior research (Clayton et al., 2005; Intintoli, 2010). Using the standard deviation of market-adjusted stock returns (Panel C of Table 8) as the proxy for stock return volatility is consistent with this result.
Similarly, trading volume increases by a larger amount for forced compared to voluntary departure announcements: the increase in the difference between forced and voluntary departures rises from pre-announcement levels of 1,272,219 to a peak of 2,399,265 on day 0. We conclude that announcements of forced departures are associated with both a greater market reaction and greater market uncertainty following the event, compared to voluntary departure announcements.
The difference between the mean cumulative abnormal return for forced and voluntary CEO departures is consistent with H4. Forced departures have consistently more negative effects across all returns windows examined. At the extreme, over the period [–7, +7] the cumulative abnormal return for forced departures is –9.22%, compared to –2.24% when it is voluntary. This reinforces the interpretation of Clayton et al. (2005) that the signalling effect is greater for forced departures.
5.3. Multivariate analysis
To examine the differential impact of departure reason on stock return volatility, Equation (2) is estimated on a daily basis for days [–2, +2], with results presented in Table 9. H2 predicts that forced departure announcements are associated with a higher level of volatility compared to voluntary departure announcements. In line with univariate results, and as predicted, FORCED has a positive coefficient on day 0, significant at the 5% level (0.0011, one-sided p-value 0.0262). Its economic significance can be demonstrated by examining the means of daily stock return volatility on day 0 for all CEO departure announcement firms (0.00199) and the sample of voluntary resignations (0.00117). However, FORCED is only weakly significant on day +1 (0.0005, one-sided p-value 0.0804) and day +2 (0.0005, one-sided p-value 0.0704), indicating that market uncertainty is quickly resolved. Overall, this result is consistent with the predictions made in H3.
Daily stock return volatility following CEO departure announcements.
This table presents the results of running the model estimated in Equation (2), where day 0 is the announcement date. The total sample in the regression models includes 259 observations. Daily volatility is estimated as the Garman and Klass (1980) ‘best analytic scale-invariant estimator of volatility’. FORCED is a dummy variable for whether the departure announcement was forced or voluntary; SIMULT is a dummy variable indicating whether there were simultaneous departures of other directors; FOUNDER is a dummy variable for CEOs who are also the company founder; TIMING is a dummy variable referring to whether or not a successor was simultaneously appointed; SUCTYPE is a dummy variable indicating whether the successor is an insider or an outsider; CLOSED is a dummy variable representing whether the announcement was released during trading hours; ANALYSTS refers to the number of analysts following; BETA indicates firm risk as measured by ordinary least squares (OLS) beta; ROE is a measure of firm performance provided by return on equity; EARNVOL is the natural logarithm of earnings volatility, measured as the standard deviation of ROE in the past three years; MKTCAP is the natural logarithm of market capitalisation, used to proxy for firm size; VOLt–1 controls for prior-day volatility; and TRADEVOL is the natural logarithm of trading volume for the same day. EARNVOL, MKTCAP, ROE and TRADEVOL have been winsorised at the 1st and 99th percentiles. Two-tailed p-values are reported in parentheses.
Surprisingly, most of the announcement-specific variables were not found to significantly influence market uncertainty. The exception is SIMULT, where the announcement of simultaneous director departures shows a statistically significant, positive coefficient for the days surrounding the announcement. More radical board changes are usually associated with major events in the company’s history, such as a complete change in the company’s strategic direction. Overall, this suggests that additional disclosure of departure-specific information does not significantly affect market uncertainty.
Of the firm-specific controls, firm size (MKTCAP) is marginally statistically significant and negative over the period, indicating that daily stock return volatility is less pronounced for larger firms. Both the prior day’s volatility (VOLt–1) and trading volume (TRADEVOL) are positive and consistently significant at less than the 1% level. Consistent with Jackson (2011), volatility from the previous day is the most significant factor in explaining current volatility.
Table 10 presents the results of estimating Equation (3), examining the differential impact of departure reason on cumulative abnormal returns. After controlling for other factors expected to impact on the abnormal returns, there is only a statistically significant difference in returns over a longer window. Abnormal returns over the period [–7, +7] are lower by –4.34% for forced departures (one-tailed p-value 0.0177) and –3.54% lower over the period [–1, +7] (one-tailed p-value 0.0257). This suggests that, even though the volatility effect is resolved with a few days, it takes a longer period for the price reaction to occur, possibly due to additional information becoming available following the ASX announcement itself.
Cumulative abnormal returns following CEO departure announcements.
This table presents the results of running the model estimated in Equation (3) where day 0 is the announcement date. The total sample in the regression models includes 259 observations. Cumulative abnormal returns (CAR) are estimated using market-adjusted returns. FORCED is a dummy variable for whether the departure announcement was forced or voluntary; SIMULT is a dummy variable indicating whether there were simultaneous departures of other directors; FOUNDER is a dummy variable for CEOs who are also the company founder; TIMING is a dummy variable referring to whether or not a successor was simultaneously appointed; SUCTYPE is a dummy variable indicating whether the successor is an insider or an outsider; CLOSED is a dummy variable representing whether the announcement was released during trading hours; ANALYSTS refers to the number of analysts following; BETA indicates firm risk as measured by ordinary least squares (OLS) beta; ROE is a measure of firm performance provided by return on equity; EARNVOL is the natural logarithm of earnings volatility, measured as the standard deviation of ROE in the past three years; MKTCAP is the natural logarithm of market capitalisation, used to proxy for firm size. EARNVOL, MKTCAP and ROE have been winsorised at the 1st and 99th percentiles. Two-tailed p-values are reported in parentheses.
Consistent across all windows, SIMULT is negative. This implies that the signalling effect is stronger when there are multiple departures from the board. In a similar vein, the positive result on ROE and negative coefficients on EARNVOL imply that when past performance has been weaker, or there has been a higher volatility in earnings, the signal to the market is that the firm is in a poor financial position.
On the other hand though, TIMING is consistently positive, and statistically significant, with the exception of the [–7, +7] window. When a successor is announced simultaneously with the departure, the market views events in a more positive light. The implication is that the market prefers firms to quickly resolve the succession question and establish the firm’s future strategic and operational approach.
5.4. Comparison of CEO departure classification
To demonstrate the improvement in the classification system of forced departures used in this study, we compare results using a categorisation consistent with prior literature and classification system corroborated with media reports. The univariate differences in daily volatility between the two classification systems are qualitatively similar, both with a statistically significant difference on days –1 to +2. However, it is the magnitude of the volatility measures that are important. For the forced departures, daily volatility peaks on day 0 at a value of 0.00410 for the classification, consistent with prior literature, around 1.3 times the estimate of volatility with the improved classification system. Likewise, for the voluntary departures, volatility peaks on day 0 at 0.00231, over 1.5 times the estimate from the improved classification. The inference drawn from this is that while the difference in volatility between forced and voluntary departures yields qualitatively the same results, the magnitude of volatility for both groups is over-estimated when reliance is placed solely on company announcements to determine departure reason. A similar pattern emerges for the trading volume.
For the cumulative abnormal returns surrounding the departure announcement, the estimates are over-stated relying solely on company announcements to classify departure reasons. At the extreme, the returns over the period [–7, +7] for forced departures are –13.58%, over 1.5 times higher than returns estimated with the improved classification system. In addition, the difference in returns between forced and voluntary departures is greater in magnitude, applying a classification system employed in prior literature. In the case of returns over the period [–1, +1], the difference between the groups using the classification rules in prior literature (–0.0559, Table 6) is almost twice the difference compared with the improved classification (–0.0285, Table 8).
Multivariate results are consistent with this interpretation. Daily stock return volatility on day 0 is 1.6 times greater when using prior classification schemes compared to a scheme corroborated with external verification. Cumulative volatility under a classification scheme used in prior literature over days [–1, +1] and the standard deviation of returns over the period [–7, +7] are 1.2 and 1.7 times the coefficient from the improved classification scheme, respectively. Likewise, cumulative abnormal returns follow the same pattern across each estimation window. At the extreme, the three-day abnormal returns surrounding the departure announcement [–1, +1] on Forced is 4.3 times greater than that reported using a scheme without sole reliance on company disclosures.
In summary, using a classification with potential bias that takes company announcements at their face value can lead to erroneous conclusions about the stock return volatility and cumulative abnormal returns. In such a classification system, ‘forced’ departures will contain the extreme cases, and ‘voluntary’ will contain some cases where the departure was, in fact, forced. The net result of this is that the volatility and (negative) cumulative abnormal returns will be higher for both ‘forced’ and ‘voluntary’ classifications, with the difference between the groups appearing similar to a classification system that uses third-party accounts to verify the ‘true’ departure reason. This indicates that using a biased classification of determining the nature of a CEO departure may lead to inflated estimates of the difference in volatility and abnormal returns. Rather than biasing towards a no result, it appears conclusions are inflated. Reliance by market participants on results that do not consider the true underlying reason for a CEO departure may lead to erroneous conclusions.
5.5. Robustness checks
A number of tests have been conducted to confirm the results are robust to alternative methods. Table 11 respecifies Equation (2) with four alternate measures of the dependent variables: (i) the cumulative volatility over days [–1, +1]; (ii) log difference of volatility on day +1 to day –1; (iii) the log difference of volatility on days [+1, +2] to days [–2, –1]; and (iv) the log difference of volatility on day 0 for the sample of CEO departure announcement firms to the matched-control sample. For all four specifications, FORCED has a significantly positive impact on volatility. Table 11 also uses the standard deviation of market-adjusted returns over the period [–7, +7] to proxy for stock return volatility, more consistent with the approach taken by Clayton et al. (2005) and Intintoli (2010). Again, the tenor of the results does not change with this alternate proxy.
Modelling stock return volatility on day 0 with alternative dependent variables.
This table presents the results of running the model estimated in Equation (2), using alternate specifications of the dependent variable, where day 0 is the announcement date. The specifications of the dependent variable are the cumulative volatility between days –1 to day +1; the natural logarithm of the ratio of volatility on day +1 to day –1; the natural logarithm of the ratio of volatility on days +1 and +2 to the volatility on days –1 and –2; and the natural logarithm of the ratio of volatility for the sample of CEO departure announcement firms to the volatility of a sample of matched-control firms. Volatility is estimated as the Garman and Klass (1980) ‘best analytic scale-invariant estimator of volatility’, except for Std dev returns, where the standard deviation of market-adjusted returns over the period [–7, +7] are used. FORCED is a dummy variable for whether the departure announcement was forced or voluntary; SIMULT is a dummy variable indicating whether there were simultaneous departures of other directors; FOUNDER is a dummy variable for CEOs who is also the company founder; TIMING is a dummy variable referring to whether or not a successor was simultaneously appointed; SUCTYPE is a dummy variable indicating whether the successor is an insider or an outsider; CLOSED is a dummy variable representing whether the announcement was released during trading hours; ANALYSTS refers to the number of analysts following; BETA indicates firm risk as measured by ordinary least squares (OLS) beta; ROE is a measure of firm performance provided by return on equity; EARNVOL is the natural logarithm of earnings volatility, measured as the standard deviation of ROE in the past three years; MKTCAP is the natural logarithm of market capitalisation, used to proxy for firm size. EARNVOL, MKTCAP and ROE have been winsorised at the 1st and 99th percentiles. Two-tailed p-values are reported in parentheses.
As the method of classification of CEO departure reason is subjective, the announcements were categorised into three levels – forced, semi-involuntary and voluntary – as outlined in Table 1. Semi-involuntary departures are characterised by departures that are forced by the circumstances but the announcement does not refer specifically to dismissal by the board. The results, presented in Table 12, show a statistically insignificant coefficient on VOLUNTARY, and a positive and weakly statistical significance on FORCED (one-sided p-values of 0.0848 and 0.0684 on days 0 and +1, respectively). However, FORCED is statistically significantly greater than VOLUNTARY, as evidenced by the F-tests. Therefore, it can be concluded that the results are not overly sensitive to the categorisation method we use.
For further robustness checks, the above analysis was replicated removing the semi-involuntary CEO departures, with the results presented in Table 12. The results remain unaltered, with the coefficient on FORCED being positively and statistically significant. This indicates, consistent with the main analysis, that stock return volatility is indeed greater for CEO departures deemed to be forced by the company.
Sensitivity of model to departure reason classification.
This table presents the results of running the model estimated in Equation (2) for day -1 to day +1, and the cumulative volatility between day -1 and day +1, where day 0 is the announcement date. Volatility is estimated as the Garman and Klass (1980) ‘best analytic scale-invariant estimator of volatility’. FORCED refers to instances of CEO dismissals by the board, while VOLUNTARY refers to instances of departures initiated by the CEO for reasons unrelated to the firm. Departures initiated by the CEO but for the benefit of the company are classified as ‘semi-involuntary’; all other variables are as previously defined. The right-hand columns exclude ‘semi-involuntary’ departures. The intercept in the left-hand (right-hand) columns capture semi-involuntary (voluntary) departures. A total of 259 (209) observations are included in the regression in the left-hand (right-hand) columns. Two-tailed p-values are reported in parentheses.
To confirm that the effect found is driven by CEO uncertainty alone, the analysis was re-run omitting those observations with simultaneous departures of other directors. In doing so, the tenor of the results did not alter. The results are robust to a variety of other specifications as well. Specifically, they do not depend on the way in which we measure firm size (market capitalisation or total assets), firm performance (ROE or return on assets (ROA)) or risk (firm beta or leverage). Likewise, the results are unaffected by inclusion of several other control variables, such as the classification of the announcement by the ASX as market sensitive, other ASX announcements made by the firm around the time of the CEO departure announcement, the simultaneous announcement of an interim successor, CEO tenure and age.
Finally, a control is included for the period of the Global Financial Crisis (years 2008 and 2009). While the coefficient on this is positive and significant, indicating that volatility was greater in that period, it did not change our conclusion on the incremental impact of a forced departure.
6. Concluding remarks
As CEOs typically have primary responsibility for the allocation of company resources and hence significantly affect firm value, a CEO departure announcement should create uncertainty relating to how firm value is affected. Our study is motivated by the limited research surrounding how CEO departure announcements impact on market uncertainty, and the biased manner in which departures have been classified as forced or voluntary. We examine how departure type affects stock return volatility following a CEO departure announcement in Australia, using an enhanced method to classify departures. We also examine the cumulative abnormal returns on the share market surrounding such an announcement.
Four-hundred CEO departure announcements issued on the ASX between the financial years 1999 and 2009 were examined. Using media reports to determine departure context, each CEO departure was categorised into comprehensive departure reasons. Overall, results provide strong evidence that stock return volatility increases following CEO departure announcements, consistent with the information effect hypothesis. Results examining volatility differences between forced and voluntary departure announcements find that, as predicted, forced departure announcements are associated with volatility nearly three times that of voluntary departure announcements on the announcement date. This is in line with prior literature and the reasoning that forced departure announcements have an additional signalling effect, which creates uncertainty about the firm unrelated to the CEO, such as fresh evidence of internal board conflicts and previously unknown poor firm performance.
Our results provide a number of practical implications. Firstly, we highlight the importance to firms to quickly resolve the issue of succeeding CEO in the event of a forced departure, and to establish the strategic and operational approaches of that succeeding CEO. Secondly, we highlight the importance of the business press in understanding the context of company disclosures. Company disclosures are naturally framed in a positive light by the firm; however, placing too much reliance on company announcements without having knowledge of the unbiased facts may lead to erroneous conclusions. We demonstrate the importance of the business press by comparing a forced CEO classification system based on company announcements and media reports with a classification based solely on company announcements.
Our study has several limitations. Firstly, the classification method remains subjective. While all care has been taken to provide consistency and clarity, the understanding of the context gained from media reports may not provide an accurate indication of how market participants interpret the departure announcement, because their accumulated understanding of the firm is more context specific. Secondly, confounding announcements during the event period are likely to be present for forced departure announcements, as major events leading up to a CEO’s dismissal are likely to be have been disclosed and to have affected market uncertainty and share prices.
We make several contributions. Firstly, we extend what has been a limited literature on the effects of CEO departure announcements on stock return volatility. Secondly, we use a regression model that includes announcement-specific explanatory variables. Thirdly, we measure stock return volatility more efficiently by extracting information from intra-day price movements. Finally, we use what we believe is a better way to distinguish between forced and voluntary CEO departures.
Footnotes
Acknowledgements
We acknowledge the comments of Eli Amir, Philip Brown, Liz Carson, Peter Clarkson (editor), Jeff Coulton, Wen He, Baljit Sidhu, Mark Wilson, Ava Wu, an anonymous reviewer and participants at the 2010 Australian National University (ANU) Honours Conference. All errors remain our own responsibility. Cheung acknowledges financial assistance provided by the Ros Kelly Honours Scholarship at the University of New South Wales (UNSW). The views expressed in this paper are not necessarily the views of the Department of the Treasury, Australia.
Funding
This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
Date of acceptance of final transcript: 2 May 2012.
Accepted by Associate Editor, Peter Clarkson (Accounting).