The Effect of Firm Performance on Management
Turnover
Thesis
Master of Science in Business Administration Specialisation: Finance
University of Groningen
Faculty of Economics and Business
Submission: 05-05-2011
Abstract
This paper investigates the relationship between firm performance and top management turnover. It expands current research to the unique environment of the Netherlands, which, until recently, was described as a system in which shareholders had very little say in the corporate governance of a firm. This paper finds that CEO turnover is significantly related to market adjusted stock returns while other top management turnover is significantly related to accounting returns. These results imply that top management turnover is affected by firm performance and that the CEO is evaluated and disciplined differently to the rest of the board of management.
JEL classification
G3, G34,G38
Keywords
1. Introduction
Corporate governance has been a topic of interest for several years. Many scandals have taken place in companies due to management seeking to fulfil their own interests instead of those of the company and its shareholders (for example, the accounting scandals of Ahold1, Enron2, Worldcom3, etc.). Currently, organizations are paying a great deal of attention to corporate control mechanisms and how to improve them. Management’s interests are aligned with those of shareholders through a mix of explicit and implicit incentives. Explicit incentives such as high pay, bonuses and stock options have a large effect on efforts. However, implicit incentives such as the threat of being fired, the possibility of being removed through a takeover or proxy fight, the possibility of being put on a tight leash during financial distress, the prospect of being appointed to new boards of directors or of receiving offers for directorship in more prestigious firms (Tirole [2006]) also influence the actions and decisions taken by managers. This paper focuses on one implicit but strong incentive of management to behave: the threat of dismissal. I investigate whether members of top management in Dutch firms are, in fact, dismissed after bad performance.
The Dutch corporate governance climate varies greatly with that of the U.S. and U.K., where most research on this topic has been carried out. In Anglo-Saxon countries, a complementary intervention of both internal and external control mechanisms is used to discipline management (Shleifer and Vishney [1997]). In the Dutch corporate environment, external control mechanisms such as the threat of takeover used to be very limited due to takeover defences which were particularly strong in the Netherlands (Kabir et al. [1997]). Although much has changed in recent years in this respect, external control mechanisms in continental Europe, including the Netherlands, are still limited and less powerful than in the U.S. and U.K.
The main area of research of this paper is the relationship between firm performance and top management turnover in the Netherlands in recent years. Does bad performance lead to top management turnover? Turnover is a visible internal control mechanism and the end product of corrective action. Through this we can determine whether the supervisory board satisfies the disciplinary task appointed to it and whether the internal control mechanisms work sufficiently to bridge the gap in shareholder/management interests. In testing this it is also necessary to take into
1
See de Jong et al. [2005a] for full insights into and discussion of the scandal.
account the effects of certain board characteristics on turnover, such as the size of the supervisory board and the independence of its members with respect to the company and board of management members. Another aspect to be tested is whether firm size affects the turnover of management.
The Netherlands provides an interesting environment to investigate corporate governance because of its unique two-tier board structure. It is suggested that supervisory board members are more capable of monitoring and advising management than investors (Maury [2006]). However, top management board members could also become entrenched, making it less likely that the supervisory board will punish them through dismissal. Most research on corporate governance in the Netherlands dates to before the introduction of the Dutch corporate governance code. This code has improved the Dutch corporate environment, implying greater disciplining of management and sensitivities of turnover to performance. I expect an inverse relationship between prior firm performance and the probability of top management turnover. Poor firm performance should lead to changes in top management.
A regression model is used to investigate the relation between management turnover and firm performance. Firm performance is measured by both stock returns and accounting returns. An event study is another possible method used to analyse the relationship.However, by applying a regression model, cases in which there was bad firm performance but no turnover are also taken into account, giving a more complete analysis of the relationship.
The period in which turnover is analysed is between the years 2006 and 2009. These years include years of crises and recession. I expect that this may amplify the results. In good economic climates, most firms perform well and shareholders are content. However, in times of crisis, shareholder scepticism increases and investors are more conscious of poor performance. Badly governed firms feel the blow the hardest and shareholder pressure may be greater. Chang and Wong [2009] find a negative relationship between profitability and CEO turnover in loss-making firms, but no such relationship in profit-loss-making firms. This reflects the greater incentives of shareholders to discipline management in bad times.
little say in the firm.It also expands on current literature involving entire management boards or executive directors and not solely the CEO. This paper provides insights into and evaluates the disciplinary power and monitoring ability of the supervisory board. The results could be useful in evaluating the workings of the current corporate governance system and in changing the regulatory environment of corporate governance.
The rest of this paper proceeds as follows. Section 2 reviews previous research on the topic of management turnover and firm performance and provides a brief overview of the Dutch corporate governance system. Section 3 gives a description of the method of testing used in this paper. Section 4 provides a description of the data used and descriptive statistics. The main results and robustness tests are presented in section 5. Section 6 concludes.
2. Literature review
2.1. Corporate governance and management turnover
The Berle and Means [1932] paradigm sparked a wave of research in the field of corporate governance. This paradigm highlights the agency problem that arises as a result of conflicts of interest between the managers and owners of a firm (Jensen & Meckling [1976]; Shleifer and Vishny [1997]). These studies focus on the United States where investor protection is high, shareholders are dispersed and the free-rider problem is significant. Managerial performance is then maintained by a combination of external and internal control mechanisms. As stated by Tirole [2006], there are many explicit and implicit incentives for managers to perform well and in the interests of the firm’s shareholders. One such implicit incentive, also considered an internal control mechanism, is the threat of being fired. An essential role of good corporate governance is to identify and terminate badly performing managers. The efficiency of the corporate governance system in place can be studied by analysing whether inefficient managers are removed from their positions after bad decisions and poor performance.
results imply that firm’s boards create sufficient incentives for managers to act in the interest of its shareholders. Kaplan and Minton [2006] build on this research by investigating internal (board driven) and external(through takeover and bankruptcy) CEO turnover. The results conclude that internal turnover is related to three components of stock performance. A limitation of the previously mentioned papers is that they do not include accounting measures. Warner et al. [1988] acknowledge that accounting measures are generally the actual measure used when evaluating managerial performance. Goyal and Park [2002] include earnings performance as well as stock performance in their analyses and find that both measures are related to CEO turnover. Cosh and Hughes [1997] find similar results for the U.K. The above-mentioned papers all find significant results for the relation of performance on CEO turnover in the U.S. and U.K.
The corporate governance systems in Anglo-Saxon countries, however, vary greatly to those in other countries. Renneboog [2000], Maury [2006] and Kaplan [1994] research executive turnover in Belgium, Finland and Germany respectively. Despite the large structural differences in corporate governance systems, these studies reveal similar results on turnover to those based on U.S. firms. Due to the governance differences, these articles lend more attention to accounting performance measures as opposed to stock performance measures. Each article reveals an increase in management turnover in response to poor firm performance. Suchard et al. [2001] explain that one of the main differences between the results in Anglo-Saxon and other countries is the time lag between bad performance and management dismissal. In the U.S. and U.K., turnover is significantly related to current performance and the time lag is very small. This is due to the market-driven environment of Anglo-Saxon countries. In Japan, Germany and Australia, however, it is the lagged performance that is significantly related to turnover. The market setup in non Anglo-Saxon countries, due to lack of market pressure for short term performance, may not initiate board action against the CEO. These results are supported by Kaplan [1994] and Maury [2006]. This paragraph shows that there should be a relation between performance and turnover in European counties, and therefore, also the Netherlands. It is likely that a lag will be present between bad performance and actual management dismissal.
in operating performance before the resignation of a CEO and large increases following the resignation. Finally, the paper by Cools and van Praag [2007] focuses on the Dutch market and addresses some of the shortcomings of prior event studies. They conclude that dismissals of badly performing executives are wealth maximizing.
Most authors focus their research on CEO turnover (Goyal and Park [2002]; Brookman and Thistle [2009]). Kaplan [1994], Maury [2006] and Renneboog [2000] also include models with other members of top management in their research and find similar turnover results to those shown by CEOs. Eriksson et al. [2001] find that in Denmark, CEO dismissal and large changes in the board are strongly intertwined. However, in their years of research, the CEO is turned over twice as much as the board majority. Fee and Hadlock [2004] find that the rate of forced turnover for non-CEOs is at least as great as that for CEOs. The sensitivity, however, of turnover to firm performance is smaller for non-CEOs. Denis and Denis [1995] find that top executive changes are more important events than those involving other members of the top management team. They fail to find significant abnormal returns after management changes not involving the top executives. Given these results, I would expect that in my sample of Dutch firms, performance will have a significant effect on board turnover excluding the CEO.
In light of the above discussion, we can see that there is strong evidence supporting the hypothesis that both CEO and other top management turnover are related to firm performance in both Anglo-Saxon as well as non Anglo-Saxon countries. Adams et al. [2010] state that boards have become tougher and that the probability of CEO dismissal is trending upward.
Another important aspect in CEO turnover is CEO tenure. Brookman and Thistle [2009] use two opposing theories in their discussion of the effect of tenure on turnover. The matching theory implies that the risk of termination first increases with tenure and then decreases because the well-matched CEOs will stay with the firm while the poorly matched CEOs leave. The CEO power theory implies that the risk of termination decreases as CEOs gain tenure. This is because CEOs have built up a power base through investment choices or board selection leading to entrenchment. The authors find evidence for the matching theory. Goyal and Park [2002], however, find evidence of entrenchment, which gives support to the CEO power theory as they find a negative relationship between CEO turnover and CEO tenure.
Table 1
Comparison of turnover-performance studies
This table gives a broad summary of the articles studied on turnover. Performance measures, when possible, are classified into stock returns, accounting returns and dummy variables representing exceptionally bad performance.
Study Country Period Turnover type
Event or regression Performance measures Coefficient of logit model
US 1963 - 1978 CEO Regression Stock returns -1.37**
Warner et al. (1988)
Market return 1.84**
Event Stock prices -
Weisbach (1988) US 1974 - 1983 CEO Regression Stock returns -0.64**
Kaplan (1994) Germany 1981 - 1989 CEO Regression Stock returns -1.49
Accounting return 11.1
Dummy bad perf 1.29**
Stock returns -0.08**
Management
boarda Accounting return 0.49
Dummy bad perf 0.09***
UK 1989 - 1994 Executive Regression Stock return -0.95*
Cosh and Hughes
(1997) Accounting return -0.77**
Belgium 1989 - 1994 Stock return -0.78***
Renneboog (2000)
Top
managementb Accounting return -0.004
Dummy bad perf -0.32*
Suchard et al. (2001) Australia 1989 - 1995 CEO Regression Stock return -0.44*
US 1992 - 1996 CEO Regression Stock return -1.69***
Goyal and Park
(2002) Accounting return -2.27
US 1985 - 1988 CEO Event Stock returns -
Denis and Denis
(2005) Accounting return -
Maury (2006) Finland 1993 - 2000 CEO Regression Stock returns -0.3***
Accounting return -1.34
Dummy bad perf -0.78**
Boardb Stock return -0.05***
Accounting return -0.52***
Dummy bad perf -0.22***
US 1992 - 2006 CEOc Regression Stock return -0.07***
Kaplan and Minton
(2006) Accounting return -0.15*
1991 - 2000 -
Cools and van Praag (2007)
Nether-lands
Top executives Event
a OLS regression
b Tobit model
b
Probit model
2.2. Dutch corporate governance environment
In contrast to the United States, firms in the Netherlands, as well as those in many other non Anglo-Saxon countries, generally have concentrated shareholders (La Porta et al. [1999]). The corporate governance in the Netherlands is based on an ‘insider’ system and relies mainly on internal monitoring. Characteristics of such a system are a small number of listed firms, concentrated share ownership and relatively low levels of takeover activities (Franks and Mayer [2001]). Dutch firms with more than 100 employees, a legally installed work council and book value of shareholders’ equity in excess of 11.4 million euros are legally required to organise as a structured regime (Structuurregime) (de Jong et al., [2005b]). This structured regime involves a two-tier board structure consisting of a management board (Raad van Bestuur) and a supervisory board (Raad van Commissarissen). The supervisory board evaluates important decisions, appoints and dismisses the management board and draws up the yearly accounts. The supervisory board has an important role in the corporate governance of Dutch firms. Members of the supervisory board are chosen by co-optation and ordinary shareholders have little say in the appointment or removal of supervisory board members or members of the board of management. The supervisory board has two main tasks, each relating to a different theory. Agency theory suggests that managers only work to satisfy their own benefits and do not work in the interest of the shareholder (Tirole [2006]). Therefore, a supervisory board’s role is to monitor the performance and actions of the manager to ensure that he/she acts in the best interest of the shareholders. Stewardship theory or resource dependence theory implies that supervisory boards are needed for their counsel, advice and connections. These two roles are also incorporated into the duties of the supervisory board as mentioned in the Dutch Corporate Governance code. Millstein and MacAvoy [1998] find a shift from passive (counselling) to more active (monitoring) boards since mid-1990s. It is this monitoring role related to agency theory that is the issue of interest in this paper.
conclude that actions taken by the supervisory board with respect to dismissals of executives are shareholder wealth maximizing.
A potential downside to a two-tier board is the possibility of entrenchment. In two-tier boards, shareholder power is vested in the supervisory board. Shareholder votes, and therefore, control, are limited in these firms (Cziraki et al. [2010]; Renneboog and Szilagyi [2007]). The internal monitoring present in the Netherlands reflects a relationship-based system in which managers are less subject to short-term pressures from the market. However, critics claim that this system is highly vulnerable to entrenchment (Denis and McConnell, [2003]). The fact that new members are elected to the supervisory board through co-optation does not improve the situation. Faleye [2007] investigates the influence of classified boards, that is, boards in which the members serve different term lengths and are not elected in one period, on firm value. He finds that staggered elections to boards reduce the probability of an involuntary turnover and the sensitivity of turnover to firm performance. This is confirmed by van Ees [2003]. Van der Goot and van het Kaar [1997] state that the management board has a large influence on appointments to the supervisory board. This can increase the dependency of the supervisory board on the management board. Eriksson [2001] includes a two-tier board dummy in the regressions on turnover for Danish firms and finds that a two-tier board is significantly negatively related to CEO turnover. He finds no relationship between the two-tier board dummy and management turnover. Therefore, the probability of CEO turnover decreases when moving from a one-tier to two-tier board system. While a two-tier structure may improve the performance to turnover sensitivity, it is also found to decrease the probability of turnover.
Previously, firms in the Netherlands were renowned for the abundance of anti-takeover defences and lack of shareholder protection (van der Goot and van het Kaar [1997]).In 2004, the Dutch corporate governance environment was reformed as a result of two important modifications: i) the introduction of the Dutch corporate governance code; ii) and the enforcement of the Structured Regime Reform Act, which primarily cut back the authority of the supervisory board and gave shareholders more power (Cziraki et al. [2010]).
The Dutch corporate governance code was drawn up and published in 2003 and revised in 2008. The code contains principles and best practices for listed companies. Compliance is voluntary; however, since the end of 2004, listed companies in the Netherlands are obliged to refer to the code in their annual report. Here they have to indicate the extent to which they have complied with the principles and best practices of the code.
the policy of management and the general course of affairs of the corporation and to assist management with advice. Some important best-practice provisions concerning the supervisory board are: i) that all but one member of the supervisory board be independent4; and ii) any member of the management board may not be a chairman of a supervisory board. These provisions deal with the negative effects of board dependency and duality found by many authors ([Deutsch [2005]; Goyal and Park [2002]; Adams et al. [2005]). The code also requires that the supervisory board meet at least once a year, without the presence of the management board, to discuss the functioning of the management board as an organ of the company and the performance of its individual members. The Netherlands Board Index (www.spencerstuart.com, 04-04-2011) shows that 87% of the companies with a supervisory board complied with this provision in 2007. In 2005, 83.5% of companies complied. This shows increasing supervision of the supervisory board. According to Akkermans et al. [2007], there is high compliance to the Dutch corporate governance code. This would imply that we may expect a high sensitivity of turnover to performance due to these provisions for listed firms.
It is found that codes have a good influence on the internal workings of firms. Dayha et al. [2002] investigate the effects of the Cadbury code in the United Kingdom. They find that CEO turnover increased following issuance of the code and that the negative relationship between CEO turnover and performance became stronger. There is also an increased sensitivity of turnover to performance in firms that adopted the code. Authors such as Kabir et al. [1997] and Chirinko et al. [2004] explain the Dutch environment as one in which managers have great power, have adopted numerous takeover defences and where investors have little to no say in the organisation. The effect of the Dutch corporate governance code and the Structured Regime Reform Act could have had a positive impact on this environment making the turnover relationship stronger than before. An improvement on the turnover sensitivity may contrast with the findings of van Ees et al. [2003]. They conclude that non-natural management turnover rates are generally found to be smaller in the Netherlands.
Other board characteristics are also shown to influence forced dismissals. The classical paper by Weisbach [1988] examines the different effects of inside and outside directors on the performance to turnover sensitivity. He finds that outsider dominated boards, those that are likely to be independent of management, are much more effective than insider dominated boards in removing poorly performing management. Outsider dominated boards respond to corporate performance when making CEO turnover decisions. Deutsch [2005] confirms these findings. The more independent supervisory boards are, the more effective they are in firing badly performing
managers (Hermalin and Weisbach [1998]; van Ees et al. [2003]). Several other authors confirm this (Suchard et al. [2001]; Hwang and Kim [2009]).
Board size can also have an effect on monitoring ability. Yermack [1996] measures the impact of board size on the relationship between CEO turnover and firm performance. He finds that firms with smaller boards have a stronger relationship between poor performance and CEO turnover than firms with larger boards. While large boards have more problem-solving capabilities (Coles et al. [2008]) they are often ineffective in monitoring. Van Ees et al. [2003] and Knop and Mertens [2010] confirm the ineffectiveness of large, overcrowded supervisory boards in the Dutch context.
2.3. The financial crisis
To my knowledge, no research has as yet been done to investigate the effect of the recent financial crisis on corporate governance. Clarke [2004] does explain the cyclical nature of crises and regulations. The article explains how reform and increased regulation occur during periods of recession and corporate collapse, and how interest in the workings of governance mechanisms diminishes in good times and times of expansion. This implies that investors and other monitors are less occupied in monitoring management’s actions in good economic times. Chang and Wong [2009] confirm this by finding a negative relationship between pre-turnover profitability and CEO turnover in times of financial losses and no such relationship when they are making profits. They also find that post-turnover profitability improved in loss making firms, but not in profit-making firms. This shows the differences in the incentives of shareholders and other stakeholders to monitor in good times as opposed to bad times. It is therefore plausible that turnover has increased since or during the 2008 financial crisis.
3. Methodology
has a two-tier board; secondly, that turnover is positively related to firm size; and thirdly, that CEO turnover is related to CEO tenure, either positively or negatively.
3.1. Research model
The regressions used to test these hypotheses are as follows:
Forced CEO Turnoveri,t = a + b(performance)i,t
+ c(CEO tenure)i,t
+ d(dummy two-tier board)i,t
+ e(supervisory board size)i,t
+ f(independence dummy)i,t
+ g(firm size)i,t + εi,t
Forced Management Turnoveri,t = a + b(performance)i,t
+ c(dummy two-tier board)i,t
+ d(supervisory board size)i,t
+ e(independence dummy)i,t
+ f(firm size)i,t + εi,t
To test the relationship between management turnover and performance I use regressions with panelled data, with turnover as the dependent variable and performance the independent variable. Panelled data represents a combination of cross-sectional (company) and time series data (years), resulting in a two-dimensional dataset. This combination has the advantage of increasing the number of degrees of freedom and thereby the power of the test by employing information on the dynamic behaviour of a large number of entities at the same time (Brooks [2008]). My data set consists of 41 cross-sections over a time series of four years, giving a total number of 164 observations for turnover.
Forced CEO turnover is a binary variable, therefore a limited dependent variable model must be used. Pooled data is used in all analyses of CEO turnover as the statistical software, Eviews, does not allow for panel effects with limited dependent variable models. This is not expected to have large implications on the results5. When using limited dependent variable
5
models, there are two options available as to which model to use, logit or probit model. In theory, the two models are very similar and will produce very similar results. Hence, the choice of method is usually arbitrary (Brooks [2008]). However, the different models may produce non-negligibly different results when the split of the dependent variable is unbalanced, ie. Turnover=1 only, or less than, 10% of the time. This is the case in the current model as turnover is not expected to occur as often as non-turnover. In line with the majority of literature on this topic (Goyal and Park [2002]; Dayha [2002]; Maury [2006], etc.), I use a logit model to analyse CEO turnover. In the results section, I compare these results with the results obtained when using a probit model.A left-centred tobit model is also used to control for the fact that forced turnover is censored. A tobit model assumes that part of the dependent variable is unobservable. In this case, CEOs are either dismissed or not. While CEOs that remain in the company are classified the same, there could be large differences in their performance. Dismissal of a CEO is a last resort, as is explained in section 1. A left-centred tobit model controls for this.
Management turnover is a fraction of the board turned over in a given year, therefore pooled OLS regressions are used for the turnover of the board of management.As I am interested in the changing effects per firm, I also apply panel techniques to the regression. It is not possible to carry out fixed effects regressions on the entire model as some variables such as firm board structure and the independence of the supervisory board6 are for the most part constant over the four years of the sample. This would result in a near singular matrix. Because I am still interested in the impact of these variables on the dependent variable and do not want to exclude them from equations, I use a random effects model to control for cross-sectional variations over time. A random effects model uses a maximum-likelihood estimator. The use of random effects adjusts the outcomes of the statistical tests for firm heterogeneity. As with CEO turnover, left-centred tobit models are also used for the board of management turnover equations. To control for possible interaction effects between the performance proxies, regressions are also done on separate performance measures for both CEO turnover and management turnover.
3.2. Proxies used
3.2.1. Dependent variable – forced turnover
The dependent variable represents forced turnover. It is often very difficult to distinguish between voluntary and involuntary resignations as they are rarely published as such. Various
6 A regression with fixed effects is carried out excluding these variables. No differences in the p-values of
authors take different approaches in dealing with this issue. Some authors ignore it (Kaplan [1994]). Many authors investigate the reason given for the departure of a CEO or member of the management board and classify these as either voluntary or involuntary (Warner et al. [1988]; Neumann and Voetmann [2005]). Others combine the reasons given for the change with age to control for retirement cases (Dahya et al. [2002]; Maury [2006]). Most authors have similar definitions for forced turnover. In my opinion, the definition used by Brookman and Thistle [2009] is the most comprehensive and is in line with other literature (Denis and Kruse [2000]; Parrino et al. [2004]). Therefore, consistent with Brookman and Thistle [2009], I classify departure as forced if the departing member of management is younger than 65 and the reason given is conflict, resigned or no reason is given. Departure is classified as not forced when the director is over the age of 60 and the reason given is retirement. Brookman and Thistle [2009] use the cut-off age of 60 as many retirement plans allow retirement without penalty at the age of 60. A dummy variable is constructed that is equal to one when a CEO is ‘forcefully’ turned over and zero otherwise.
Forced top management turnover is calculated as a fraction of board members forcefully turned over in a given year. It is expected that the worse the performance, the larger the fraction of management turned over in a given year. The CEO is excluded from this fraction because the effects of CEO turnover are measured separately.
For comparison, regressions on all CEO turnover (that is, including both forced and non-forced turnover) and all management turnover are also done. The results are discussed in section 5.1. To check the robustness of the results to the classification, I use a different definition in the robustness section, 5.2.3.
3.2.2 Independent variable – Performance
As a result of the previously mentioned arguments concerning stock price measures and accounting measure and following the example of Renneboog [2000], Maury [2006] and Kaplan [1994], I use a combination of stock returns and accounting data. Three measures of firm performance are used which are consistent with those used by Maury [2006]: i) the market adjusted stock return, ii) change in operating profits to total assets, and iii) a dummy variable equal to one if the operating income is negative. These three measures each capture a different aspect of performance and thereby create a complete image of the relation.
The market adjusted stock return is a better predictor of performance than raw return, as management is then not held responsible for trends in the overall market (Weisbach [1988]; Goyal and Park [2002]). Some may argue that industry adjusted returns would be more appropriate as some industries are affected more and differently by certain shocks. Kaplan and Minton [2006] and Dayha et al. [2002], for example, use industry adjusted returns to measure stock price performance. Dahya et al. [2002] also check the results against the market model excess returns and CAPM excess returns and find no significant differences in the p-value of the coefficients. Furthermore, due to the size of the dataset used in this research, the calculation of industry adjusted returns is complicated. Using one digit ICB classification would have the result that only one firm is present in this industry. This could present a high measurement bias.
Operating profit/loss is used as a measure of accounting performance as it prevents changes in capital structures and tax treatment from distorting the performance measure (Defond and Hung [2003]). Renneboog [2002] also states that the use of operating income rather than net earnings after tax reduces the impact of possible earnings manipulations.
Finally the dummy variable for operating profit is used to measure the probability of turnover when a firm’s performance is exceptionally bad. This measure is also used and found to be significant by Kaplan [1994].
As all three measures capture a certain aspect of managerial performance, I expect both the market adjusted stock return measure and the operating profit measure to be negatively related to turnover. The dummy variable for operating profit should be positively related to turnover as this is given the value of 1 when there is a loss.
and include a performance measure over a three-year period. This is also done by Dahya et al. [2002]. To check the robustness of this measurement, I also carry out regressions on performance measured over one year. The regressions also include one-year lags of market adjusted stock returns and operating profit to total assets. The results are presented in section 5.2.
3.2.3. Control variables
As shown in the literature review in section 2.1, CEO tenure is shown to influence turnover.A long tenure could suggest that the CEO has become entrenched, which would imply a negative relationship with turnover. However, a long tenure could also mean that a CEO is close to retirement, implying a positive relationship between tenure and performance (Goyal and Park [2002]). I control for CEO tenure as the number of years a CEO has held that position. This is also done by Goyal and Park [2002]. If long tenure implies that the CEO is close to retirement, the probability of turnover will increase. If, on the other hand, long tenure means that the CEO has become entrenched, a negative relation is expected between tenure and turnover, implying that the probability of turnover will decrease. Goyal and Park [2002] report a negative relationship between tenure and turnover, suggesting that turnover is less likely as tenure increases. This gives support for the entrenchment view. Brookman and Thistle [2009], on the other hand, find that CEOs typically do not become entrenched. Tenure is not included as a control variable in the regressions for the board of management as it is not believed to have such an influence on management turnover as it does on CEO turnover. It is not believed that members of the board of management can become highly entrenched.To my knowledge, no article on this topic includes research into the effect of management tenure on turnover. Furthermore, including the tenure of each member of the board of management is complicated and outside the scope of this research.
The board structure of a firm is shown to be important in controlling and monitoring top management (Maury [2006]; Cools and van Praag [2007]). Although most Dutch firms have a two-tier board, there are some exceptions7. To control for this, I include a dummy variable equal to one if the company has a two-tier board and zero otherwise. This is also done by Eriksson [2001]. The author finds a negative relation between a two-tier board structure and turnover. I expect firms with two-tier boards to be stricter in dismissals and to have a higher probability of turnover.
It is well established that the number of outsiders or independent non-executive directors has a significant effect on turnover (Weisbach [1988]; Deutsch [2005]). A large degree of
independence of the supervisory board is a requirement of the Dutch corporate governance code. Not all firms fully comply with this best practice provision. Therefore I control for this by including a dummy variable, equal to one for independent supervisory boards, in the regression.
I include board size as a control variable because of the significant relationship found by Yermack [1996] between board size and turnover. Following van Ees et al. [2003], I measure board size by the log of the number of supervisory board members. I expect that a larger supervisory board will be less efficient than a smaller one and therefore, that it will have a negative relation to turnover.
The larger a firm is, the more likely it is that the managers will be turned over (Eriksson et al. [2001]). Therefore, the effect of firm size on turnover is also controlled for by including the natural logarithm of sales (Maury [2006]).
As discussed in section 2.3, the financial crisis may have had effects on the turnover of top management and CEOs. I conduct separate tests including a crisis dummy to test the impact of the financial crisis on turnover.
4. Data and descriptive statistics
4.1. Data
The sample consists of all non-financial Dutch companies listed on the Amsterdam Stock Exchange (AEX) and the Amsterdam Midcap Index (AMX) at the start of 2010 for which full financial information is available for the sample period. Financial firms are excluded because of the differences in accounting methods and the regulatory environment of these firms.
Altogether, 50 firms were traded on the AEX and AMX in 2010. 5 of these (2 AEX and 3 AMX) are financial institutions and are therefore excluded from the sample. One company is excluded from the sample as a result of a merger in 2007 and three companies are excluded due to lack of financial information. This leaves a total sample of 41 companies over the period 2006 to 2009, yielding a total of 164 observations.
4.2. Measurement and description of variables
Data on the composition of the board of management (Raad van Bestuur) and years of birth are obtained from the companies’ annual reports. The data therefore reflect the composition at the end of each given year and turnover is recorded from one annual report to the next. As in Dahya et al. [2002], I note each change in a name of the CEO or president as CEO turnover and a disappearance of a name from the management board, as management turnover. If a CEO is replaced by another member of management, it is counted as CEO turnover and not management turnover. All turnover is internal as no firms in the final sample merged or were acquired in the sample period.
In order to measure the disciplining ability of the supervisory board through management turnover, it is important to be able to properly detect forced turnover. I try to identify forced turnover by searching the annual reports and press releases of the company for an explanation. As explained in section 3.1, I classify departure as forced if the departing member of management is younger than 65 and the reason given is conflict, resigned or no reason is given. Departure is classified as not forced when the director is over the age of 60 and the reason given is retirement. If it is stated that the director will continue in the company or has left the company for a specific position elsewhere, turnover is also classified as voluntary. CEOs that continue in the company often become members of the supervisory board. As this would not be done if the supervisory board was not content with CEO performance, it is not deemed as forced. The reasons given for departures can be grouped into several categories.Table 2 shows the reasons given for departure of members of management. In 32.8% of all turnover cases, no reason, or an uninformative reason, was given for turnover. In nearly 30% of the cases, retirement was the reason given. As previously mentioned, this is not classified as forced unless the manager is under the age of 60. Other reasons given include conflict, to join another company and bad performance.
A binary variable for forced CEO turnover is constructed. This variable is equal to one in the period that the CEO is dismissed and this dismissal is classified as forced, and zero otherwise. Management turnover is calculated as the fraction of the management board that has been ‘forcefully’ dismissed, during a given year. The dependent variable for the management equations is always between zero and one. As mentioned in the methodology section, section 3.1, the CEO is not included in the management turnover fraction. This measure of board turnover is also used by Maury [2006].
majority of the board of management (excluding CEO) is ‘forcefully’ changed in one year. This gives an unconditional probability of 7.9% of the majority of the board being turned over involuntarily.
Table 2
Reasons given for departures of CEO and top management. Based on observations from 41 firms over a 4 year period. Reasons are gathered from press releases and grouped into several general categories. This table splits CEO and management
turnover. It shows the number of cases of turnover in each category, that number as a percentage of total turnover in that group, and how many of these cases are classified as forced (using the definition given in Table A1 of the appendix), for both CEO and management turnover. The final column expresses the reasons given as a percentage of both CEO and management
combined. CEO management Reason Number % no. of forced Number % no. of forced % of total turnover No reason 3 16.67% 3 15 40.54% 13 32.73% Retirement 4 22.22% 1 12 32.43% 1 29.09% Uninformative reason 1 5.56% 1 6 16.22% 6 12.73% Conflict 2 11.11% 2 2 5.41% 2 7.27%
Continued in the company 4 22.22% 0 0 0.00% 0 7.27%
To join another company 2 11.11% 0 2 5.41% 0 7.27%
Bad performance/ousted 2 11.11% 2 0 0.00% 0 3.64%
Total 18 100.00% 9 37 100.00% 22 100.00%
4.2.2. Firm Performance
I use three measures of firm performance consistent with those used by Maury [2006]: i) the market adjusted stock return; ii) change in operating profits to total assets; and iii) a dummy variable equal to one if the operating income is negative. In order to calculate the market adjusted stock return, year-end firm stock prices are taken from the DataStream database. Dividends, retrieved from annual reports, are added to the prices of each firm. If the dividends are expressed as a different currency, the annual reports are searched for a conversion date and the euro amount is calculated. From these prices, the two-year holding stock return is then calculated for t-2 to t: rit
= (pit – pit-2)/pit-2.The total return index of the AEX, the AEXR index, is obtained from the NYSE
Table 3
Descriptive statistics on governance and performance characteristics
Summary statistics for financial and board characteristics for Dutch non-financial firms. Performance measurements from t-2 to t represent data over the period 2004-2009. Other data is over the period 2006 -2009. The sample includes 41
listed firms.
Mean Median Minimum Maximum Std. Dev. # obs.
Panel A: Performance characteristics
Change in OP/TA (t-2 to t) 0.00 0.01 -0.49 0.27 0.08 164
Operating loss dummy (t-2 to t) 0.12 0.00 0.00 1.00 0.33 164
Stock returns (t-2 to t) 0.34 0.27 -0.92 3.62 0.76 164
Market adjusted stock returns (t-2 to t) 0.31 0.16 -0.74 3.09 0.64 164
Sales (MEUR) 13.60 2.77 0.09 329.35 40.87 160
Total assets (MEUR) 12.61 2.42 0.18 202.91 31.93 164
Panel B: CEO characteristics
Age 55.07 56 42.00 76.00 6.79 164
Age (of departing CEOs) 59.33 58.5 51.00 76.00 5.93 18
Age (of forcefully dismissed CEOs) 56.44 57.00 51.00 63.00 4.50 9
Tenure 5.29 4.00 0.00 39.00 5.92 163
Tenure (of departing CEOs) 6.29 3.00 0.00 39.00 9.03 17
Tenure (of forcefully dismissed CEOs) 4.88 4.00 0.00 14.00 4.22 8
Panel C: Board characteristics
Two-tier board 0.88 1.00 0.00 1.00 0.33 164
Independent supervisory board 0.84 1.00 0.00 1.00 0.37 164
Size supervisory board 6.68 6.00 13.00 18.00 2.65 164
Size management board 3.56 3.00 1.00 8.00 1.32 164
Data concerning operating profits and total assets are obtained from the Orbis database. Operating profits is operating income minus operating expenses. The change in operating profits is calculated from t-2 to t. The difference is then scaled by the average of the book value of total assets over t, t-1 and t-2. As with the stock returns, the change in operating profits to total assets is calculated using information from 2004 onwards. Book value is used as it is not as prone to high volatility as the market value is. The operating profit dummy is equal to 1 if the firm has experienced a loss between t and t-2.
and a median of 16%. Sales, used to measure firm size, have a mean of 13.6 million euro’s per year. The median is 2.77 million euro’s.
4.2.3. Governance and control variables
As described in section 3.3, I control for CEO tenure as the number of years a CEO has held that position. Tenure is calculated using the year of appointment to that particular position. The average CEO tenure of the data sample is 5.29 years. The Netherlands Board Index (www.spencerstuart.com, 04-04-’11) reports average CEO tenure of 4.7 years as of October 1st 2008. This index is a consensus of 100 of the largest companies listed on the Dutch segment of the NYSE-Euronext. This average is slightly lower for AEX and AMX companies. The difference in results may be due to an outlier present in the current sample. The maximum tenure reported in the sample is 39 years and the median of the sample is 4 years. The median is more in line with the results reported by the NBI. The average (median) tenure of all departing CEOs is 6.29 (3) years. Furthermore the average (median) tenure of forcefully dismissed CEOs is 4.88 (4 years). CEO age is also investigated. The average (median) age of all CEOs is 55.07 (56) years. The mean (median) age of departing CEOs is logically higher at 59.33 (58.5) years. However, this sample includes many CEOs leaving the firm because of retirement. When these are filtered out and only forced CEO turnover is investigated, an average (median) age of 56.44 (57) years is found.
The information for the two-tier boards dummy is gathered from the annual reports. The descriptive statistics show that 88% of the companies included in the sample have a two-tier board structure. This is consistent with the findings reported in the Netherlands Board Index (NBI).
The Dutch corporate governance code requires that a statement of compliance (or lack thereof) is included in the company’s annual report with respect to the independence provision. This provision states that all but one member of the supervisory board must be independent. This information is therefore gathered from the annual reports. A dummy variable is constructed that is equal to one if a firm complies with the best practice provision III.2.2 of the Dutch Corporate Governance code and zero otherwise. Table 3 shows that 84% of the supervisory boards are defined as independent over the period 2006 to 2009. The NBI reports 87.8% of the supervisory boards to be independent in 2008.
Table 4
Correlation matrix independent variables
Correlations are based on Dutch listed, non-financial firms over the period 2006-2009. The sample includes 41 firms and 164 observationsa. Operating profit/Total assets Operating loss dummy CEO tenure Two-tier Board Supervisory board size Independent supervisory
board Firm size
Market adjusted stock return 0.278 -0.137 -0.1132 0.112 -0.077 0.094 -0.074
Operating profit/Total assets -0.499 0.104 0.433 -0.154 -0.002 -0.068
Operating profit dummy -0.130 0.024 0.100 0.066 -0.058
CEO tenure 0.198 -0.298 0.199 -0.234
Two-tier board -0.601 0.466 -0.521
Supervisory board size -0.372 0.743
Independent supervisory board -0.307
a See table A1 of the appendix for variable definitions
Table 4 presents the correlations between the independent and control variables. There is a large correlation between firm size and supervisory board size (0.74) and between board structure and supervisory board size (-0.6). This could pose a problem in the multivariate regressions and will be taken into account when analysing the results. Most other correlations are smaller in magnitude (under 0.5) and are not expected to pose serious problems.
If multicollinearity exists, the model will present a high goodness-of-fit, but the individual coefficients will have high standard errors. This would imply that few or none of the individual coefficients would be statistically significant. The regression would also become very sensitive to small changes in the specification. Brooks [2008] proposes several options to deal with multicollinearity. The first option is to ignore it, if the model is otherwise adequate. The second option is to drop one of the collinear variables, given that this can also be argued from a theoretical perspective. Thirdly, one can transform the highly correlated variable into a fraction, again, given that theory permits this. Lastly, one could gather more data and expand the sample size. I investigate dropping the size of supervisory board variable, as this is, in my opinion, the least theoretically important variable in the regression.
5. Empirical results
Table 5 shows the results of the regressions on CEO turnover. Panel A of table 5 depicts regressions on all CEO turnover. The first model in each panel is a logit regression and the second model shows the tobit analysis. The tobit analysis is used because of censoring of the dependent variable as discussed in section 3. As can be seen from the table, no variables are significant in the model representing all CEO turnover. This could be because turnover that is not forced is included in the model and voluntary turnover should have no relation to performance or other governance and firm characteristics.
Table 5
CEO turnover and firm performance
Regressions of CEO turnover on performance during the period 2006-2009. The sample consists of non-financial firms listed on the AEX and AMX. The regressions are based on 159 observationsa. Observations differ to initial sample size because of missing sales and tenure data. All logit and tobit results are robust to heteroscedasticity (Huber/White). The
p-values of the coefficients are shown in parentheses.
Panel A: All CEO
turnover Panel B: Forced turnover
Logit Tobit Logit Tobit
Intercept -0.227 -0.071 -0.503 0.407
(0.95) (0.98) (0.92) (0.93)
Market adjusted returns -0.181 -0.162 -1.37 -1.272
(0.62) (0.62) (0.07)* (0.06)*
Change in operating income/total
assets -3.33 -3.098 -7.187 -6.66 (0.53) (0.53) (0.28) (0.29) Operating loss 0.15 0.119 -0.582 -0.542 (0.89) (0.90) (0.68) (0.68) Tenure -0.024 -0.024 -0.285 -0.265 (0.87) (0.86) (0.25) (0.24) Two-tier board -1.155 -1.054 -0.191 -0.159 (0.20) (0.19) (0.87) (0.88)
Independent supervisory board -0.391 -0.35 0.108 0.09
(0.63) (0.64) (0.91) (0.93)
Supervisory board size 0.202 0.17 1.276 1.156
(0.89) (0.90) (0.58) (0.61)
Firm size (log of sales) -0.065 -0.06 -0.243 -0.221
(0.77) (0.77) (0.41) (0.42)
Pseudo R² 0.06 0.16
Akaike information criterion 0.93 0.56
* p < 10%
** p < 5%
*** p < 1%
Panel B, representing only forced turnover, appears to have a better goodness of fit. The pseudo R2 for the logit regression is higher at 0.16. In this model, the market adjusted stock return is negatively related to turnover and significant at the 10% level. This means that, according to the logit model, a one percent decrease in the market adjusted returns would lead to a 0.4 percent increase in the probability of CEO turnover. This is consistent with many previous studies, such as Suchard et al. [2001], Renneboog [2000], Maury [2006], etc. CEO turnover is not significantly related to the change in operating profits to total assets. Kaplan [1994] and Maury [2006] also fail to find a significant relationship between accounting performance and CEO turnover. The dummy variable equal to one when an operating loss occurs, representing exceptionally bad firm performance has no relationship with CEO turnover. This is a puzzling result. One possible explanation is that CEOs are turned over only when outside parties (shareholders) exert pressure on the firm for changes. This explanation would give support to the entrenchment theory posed in section 2.It also implies that shareholders have more influence in the governance of Dutch firms than implied by Cools and van Praag [2007] and that shareholder control is not “virtually absent”.
Separate tests were also conducted for each performance measure8. The results are highly similar and also show a significant, negative relationship between market adjusted stock returns and CEO turnover. The pseudo R2 for these regressions are slightly lower implying that the full model including all performance measures is a better predictor for turnover.
As mentioned in section 3, a probit model can produce slightly different results when the split of the dependent variable is not balanced. Therefore, a probit model is also used for comparison9. The results are highly similar, and as with the logit model, market adjusted stock returns is the only significant variable on forced CEO turnover. The coefficient is significant at the 5 percent level.
Table 6 shows the analysis of management board turnover. Again, panel A shows all management turnover, while panel B presents the fraction of the board of management turnover corrected for natural turnover (according to the definition of forced turnover stated in table A1 of the appendix). There are qualitatively few differences between panel A and panel B. This is probably because more than half of the observations of management turnover are classified as forced. This therefore also influences the results in the full sample.
Table 6
Top management turnover and firm performance
Regressions of top management turnover on performance during the period 2006-2009. The sample consists of non-financial firms listed on the AEX and AMX. Management turnover is the number of members of the board
of management turned over as a fraction of total board members, excluding the CEOa. The regressions are based on160 observations. Observations differ to initial sample size because of missing sales data. The
p-values of the coefficients are shown in parentheses. Tobit results are robust to heteroscedasticity (Huber/White).
Panel A: All turnover Panel B: Forced turnover
OLS
Random
effects Tobit OLS
Random
effects Tobit
Intercept 0.467 0.464 0.474 0.347 0.343 0.328
(0.04)** (0.05)* (0.59) (0.1)* (0.12) (0.79)
Market adjusted returns 0 0.002 0.022 0.005 0.006 0.096
(0.98) (0.94) (0.85) (0.86) (0.82) (0.60) Change in operating income/total assets -0.695 -0.681 -3.076 -0.708 -0.703 -5.168 (0.02)** (0.02)** (0.01)** (0.01)** (0.01)** (0.00)*** Operating loss -0.011 -0.01 -0.079 -0.016 -0.018 -0.249 (0.86) (0.88) (0.74) (0.78) (0.76) (0.53) Two-tier board -0.173 -0.171 -0.571 -0.128 -0.126 -0.612 (0.02)** (0.03)** (0.08)* (0.06)* (0.08)* (0.23) Independent supervisory board 0.042 0.039 0.071 0.049 0.049 0.13 (0.43) (0.49) (0.79) (0.33) (0.36) (0.77)
Supervisory board size -0.113 -0.111 -0.6735 -0.08 -0.077 -0.797
(0.17) (0.19) (0.02)** (0.29) (0.34) (0.06)*
Firm size (log of sales) -0.002 -0.003 0.045 -0.004 -0.004 0.038
(0.87) (0.87) (0.43) (0.81) (0.81) (0.65)
R² 0.08 0.07 0.07 0.07
Akaike information criterion 1.15 0.93
* p < 10%
** p < 5%
*** p < 1%
a See table A1 of the appendix for variable definitions
to be higher by 0.7 percent. These results are in line with those of Maury [2006], who also find this relationship for management turnover and fail to detect it for CEO turnover. As in the CEO analyses, tobit models are used in the management turnover analyses to correct for censoring of the dependent variable. In this model the significance of certain variables has increased. The performance measure change in operating profits to total assets is now significant at the 1% level and the coefficient has greatly increased to -5.17. This means that turnover is predicted to increase by 5 percent with every 1 percent decrease in operating profits to total assets. In the tobit model, supervisory board size is important, while this is insignificant in both the OLS and random effects models. In the tobit model, larger supervisory boards are less likely to dismiss members of the board of management. This could be as a result of board inefficiencies as discussed in section 3.3.
In all but one models in table 6, the two-tier board structure is significant for management turnover. A negative relationship was found meaning that the presence of a two-tier board reduces the fraction of the board of management turned over. The coefficient, however, is quite small. According to the OLS model, the presence of a two-tier board reduces board of management turnover by 0.13. This result corresponds with that found by Eriksson [2001]. These effects are greater for the full turnover sample, when not correcting for natural turnover. This could give support to the relationship-based environment to which a two-tier board is associated.
As with the CEO regressions, separate tests were done on each performance measure. These results are very similar to the combined model and are therefore not shown. Logit regressions of board majority turnover on performance are also tested10. The dependent variable is then equal to one when 50 percent or more of the board were forcefully removed in one year. No variables were found to be significantly related to board majority turnover.
Because of the correlation problem presented in section 4.2.3, I carry out regressions excluding supervisory board size on both CEO turnover and management turnover. The goodness of fit for these models are lower, while the results remain qualitatively unchanged. Because of this similarity, I conclude that the variable for supervisory board size does not pose a major problem and it is therefore included in all reported regressions. The same was done for the two-tier board as this also had high correlations. As with supervisory board size, the results remain qualitatively unchanged.
5.2. Robustness tests
5.2.1. Performance variables
In order to get more insight into the effect of yearly returns and the reaction time after bad performance, I construct performance variables measured over a shorter period (t-1 to t). I also construct one-year lags of these variables. The results of the regressions on forced CEO and board management can be found in the appendix, table B1. The logit regressions for CEO turnover reveal that there is a lag between bad performance and turnover. The result of the one-yearmarket adjusted return is slightly above the conventional significance levels while the one-year lag of this stock return is highly significant. The coefficient of -6.36 implies a large effect on turnover. The strong effect of the lagged performance measure is in line with the findings of Suchard et al. [2001], who state that non Anglo-Saxon countries tend to delay dismissal of badly performing CEOs. Management turnover, on the other hand, is significantly related to the current measure of performance. The lagged variable gives a highly insignificant result. It appears that management disciplining is more immediate than CEO disciplining. This differs from the results of other authors who find that the lags of the performance measures are more strongly related to management turnover than the current measure (Kaplan [1994], Maury [2006]). My results imply that supervisory boards are slower in dismissing CEOs then they are in dismissing other members of the management board.
5.2.2. Time effects
To control for market-wide shocks over time, I repeat the analysis of both CEO turnover and management turnover to include year dummies. The inclusion of year dummies leaves the results qualitatively unchanged11.
Section 2.4 discusses the possible impact of crises on turnover. The previous financial crisis of 2008 may also have an effect on turnover. Therefore, I include dummy variables to test the impact of this crisis on turnover in the Netherlands. I construct a dummy variable equal to one to represent the years before the crisis, 2006 and 2007 (before crisis dummy), and a dummy equal to one to represent the years after the crisis has struck, 2008 and 2009 (after crisis dummy). The results are presented in appendix, table B2. The performance measures discussed in the main results remain qualitatively unchanged for both CEO and management turnover. It appears from
table B2 that the crisis did not have any impact on the CEO turnover. The coefficients for both dummies are highly insignificant. There appear to be significant differences, however, in management turnover before and after the crisis. According to the OLS results, the likelihood of forced management turnover was 7% lower before the crisis. The effect of the after crisis dummy is inconsistent. The OLS regression reveals a significant effect on turnover, while the tobit regression predicts an insignificant relation.
5.2.3. Definition of forced turnover
The definition of forced turnover used is, in my opinion, a very comprehensive one. Some may argue that the definition is too strict and that it may exclude some cases of forced turnover. In section 5.1 I have already compared forced turnover to all turnover. The results in table 5 for all management turnover (panel A) have a better goodness of fit than the results for only forced management turnover (panel B). This could imply that the definition of forced turnover could exclude some observations of when the board of management is involuntarily dismissed. I will now turn to a different definition of forced turnover. Weisbach [1988] states that companies rarely announce the true reason behind resignations and that the reasons given are non-informative. Therefore, he ignores the reasons stated in press releases and Forbes, and uses only death and changes preceding takeovers as suitable reasons for not classifying the turnover as forced. He further only controls for the retirement age of 65. Because information on CEO age is only accurate to one year, he uses 63 as the cut-off age for retirement. I follow this method by Weisbach [1988] and reclassify both forced CEO turnover and management turnover in this way. The results of the analysis can be found in the appendix, table B3.
The results of the CEO logit regression reveal only one significant variable. With this new definition of forced turnover, market adjusted stock returns are no longer related to forced turnover. Now tenure is significantly, negatively related to turnover at the 10% level. The higher the tenure of a CEO, the less likely it is that he/she will be fired. This gives support to the entrenchment theory discussed in section 3.3. As for management turnover, an OLS regression with the new definition of forced turnover still gives a significant coefficient for the change in operating income to assets. Board size is also significantly, negatively related to turnover with this definition.
5.2.4. The inclusion of the CEO in management turnover
by calculating the forced management turnover fractions including the CEO. The results are similar in that the change in operating income is still statistically significant at the 1 percent level.
5.2.5. Endogeneity
Endogeneity is a serious problem when measuring variables on corporate governance. It can result in difficulties when interpreting the effect of the governance structure on turnover. Hermalien and Weisbach [2003] and Adams et al. [2010] pay much attention to this issue. Maury [2006] controls for this problem by analysing within-firm variation in variables over time, using firm-level fixed-effects regressions. For my current data sample it is not possible to include these specifications as it would lead to a near singular matrix. This is because, as stated in section 3.1, firm effects such as board structure and independent supervisory board are largely fixed over the sample period. A fixed effects model is carried out excluding these variables and no qualitative differences were found in the results12. This implies that endogeneity does not pose a large problem in this model.
6. Conclusions
This paper empirically investigates the relationship between top management turnover and firm performance in Dutch firms between the years 2006 and 2009. The relation to other board and firm characteristics, such as board structure, supervisory board size, independence of the supervisory board members and firm size on turnover are also tested. This research is the first of its kind to investigate the Dutch corporate governance environment after the reforms in 2004. Firm performance is measured through both accounting measures and market measures. The results show that CEO turnover is significantly related to stock price performance, while board of management turnover (excluding CEO) is related to accounting performance. Disciplinary action, through dismissal, is taken against CEOs when market adjusted stock returns are negative. The rest of the board of management is disciplined when operating profits to total assets decrease. The literature including top management turnover as well as CEO turnover find largely similar results for management as for CEOs (Coughlan and Schmidt [1985], Warner et al. [1988]). Some authors (eg. Fee and Hadlock [2004]) find that the sensitivity of turnover to firm performance is less for management than for CEOs. My results present a different outcome. Management
turnover is highly related to firm performance. However, members of the board of management are evaluated differently to CEOs. These differences between management and CEO are unique.
There is some evidence of relation-based internal monitoring in the Netherlands when analysing CEO turnover. The significant relationship between CEO turnover and lagged market performance reflects this. A surprising result, which is not in line with the current literature, is that the variable representing extremely bad performance, the operating loss dummy, is highly insignificant in all regressions.
The results on the control variables used in the analysis are not consistent. CEO turnover does not seem to be affected by these variables. CEO tenure is the only variable that gives significant results and is negatively related to CEO turnover. This could provide support for the entrenchment problem posed in section 2.2. Management turnover does seem to be affected by certain control variables (give examples), but the significance of their coefficients depends on the model specification and are therefore inconsistent between models.
The results highlighted above provide an insight into the workings of the Dutch corporate governance system. They show that the internal control mechanisms are effective. Management is disciplined by the supervisory board and the Dutch corporate environment portrayed by van der Goot and van het Kaar [1997] has greatly changed. The significant relationship between stock returns and CEO turnover implies that shareholders do have an influence in Dutch firms.
