Will CEO duality influence board turnover?

(1)

Master Thesis

Will CEO Duality Influence

Board Turnover?

Author: Ailin Huang (1137 4446)

MSc. in Finance, Corporate Finance Track

Amsterdam Business School, University of Amsterdam

Faculty of Economics and Business

Supervisor: Torsten Jochem, Ph.D. Assistant Professor of Finance

(2)

Acknowledgment

I would like to express my sincere gratitude to my thesis supervisor, Torsten Jochem, for great support during the writing process. I am thankful for his feedback as well as patience to help me for a better thesis.

And special thanks for my family and friends, who support me along the way, as they always do. Whenever I need help, or someone to listen, they are always there.

Statement of Originality

This document is written by Student Ailin Huang who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

I examine the effect of the combination of CEO and board chairman (CEO duality) on board turnover. The results show if the firm has CEO duality, the board members are more likely to stay stable, when instrumental variable (IV) methodology is applied and share returns is used as firm performance proxy. And the results also suggest that the sensitivity of board turnover to firm performance is lower when the the duties of CEO and board chairman are bestowed on only one individual. My findings basically support the view of agency theory.

(4)

List of Tables and Figures

Tables Table 1: Year distribution of observations and average board turnover rate per year...35

Table 2: Descriptive statistics partitioned by board turnover rate... 36

Table 3: Variable statistical description...37

Table 4: OLS regressions... 39

Table 5: Fixed Effects Analysis...41

Table 6: Instrument Variable 2SLS Analysis... 43

Table 7: Probit Analysis... 45

Table 8: Tobit/IV Tobit Analysis... 45

Appendix 1: Strength of instrument variable...49

Figures Figure 1: Time series graph of average board turnover rate...35

(6)

1. Introduction

Board turnover and other topics related to board refreshment, like mandatory retirement ages, term limits for board room services, are highly debated these years across the boardrooms and in the public. In fact, I find some firms do not change any board member for years, which is a very interesting phenomenon. Institutional Shareholder Services (ISS) conducted a survey annually among investors and non-investors to identify the board refreshment problem. The report published recently (ISS, 2017) suggested that 51% investors in the survey were worried about the lengthy tenure of board members and questioned the independence of long-serving directors. A growing number of investors and activists have increasingly voiced out and requested more extensive use of refreshment mechanisms. While the public are concern about the relative low board turnover rate, the management and other insiders have different opinions. Only 6% of the management executives replied that their board members were considering diverse measures to ensure director turnover (Spencer Stuart, 2016).

And when a firm has a specific leadership arrangement, like CEO duality, will this arrangement affect the board refreshment in the firm? Some people think the answer would be affirmative. In fact, CEO duality is rather common in companies. The phenomenon of CEO serving as chairman of the board at the same time prevailed in 67% of S&P 500 boards in 2006, but the number has decreased to 52% in 2016 (Spencer Stuart, 2016). Despite the fact that the percentage has declined, still more than half of the S&P 500 companies have not separated the chairman and CEO roles. Some activists request those companies to separate the two roles for more independence on board and better corporate governance. However, the relationship between CEO duality, firm performance and corporate governance is complex, and the empirical evidence is somehow conflicting. Most of the previous research relied on two contrasting theories: agency theory and stewardship theory. The former theory, which was first suggested by Fama and Jenson (1983), argues that the firm can

(7)

implement a series of checks and balances in corporate governance mechanisms when the roles of CEO and chairman are separated. The managers will then have more difficulty in entrenching themselves and thus the practice can lead to a better firm performance. Some proponents of the latter theory, such as Donaldson and Davis (1991), argue that the duality role can make managers act in their own self-interests and reduce agency loss such as abundant director compensation and excessive communication.

While some empirical evidence shows that CEO duality decreases CEO turnover rate, the direct empirical evidence is scant when it comes to the relationship between CEO duality and board turnover rate. This paper attempts to fill this void by examining the board turnover reaction to the CEO duality dummy when we split the companies into two groups according to their board turnover rate.

I apply Ordinary Least Square (OLS) and Instrument Variable (IV) methodology to evaluate the effect of CEO duality on board turnover dummy, and also try to figure out the sensitivity of CEO duality to firm performance in the period of Year 2008 -2015. After applying IV, which is CEO ratio, the result suggests that CEO duality lower the probability of board turnover. And the sensitivity of board turnover to firm performance is significantly lower when the CEO is also the chairman of the board. However, these results are based on the validity of IV.

My paper investigates the effect of CEO duality on board turnover, which provides more empirical evidence on the influence of CEO duality in a new perspective. To the best of my knowledge, this paper is the first to exploit the problem of board turnover in this new angle.

Another contribution is on board refreshment. I think board turnover rate is also an indicator of board refreshment. If the turnover rate is low, the board members are not changed a lot, which means the board is not refreshed. While it is believed that board refreshment is significant to improve the efficiency of corporate governance

(8)

and firm performances, empirical results are very limited. And my paper can fill a gap in this area.

This paper is organized as follows. Section 2 elaborates a review of the theories, previous empirical studies and my hypothesis. Section 3 describes the sample selection process, variable definition, and also methodology in my analysis. Section 4 presents descriptive statistics on independent variables. In Section 5, I discuss empirical results as well as robustness checks. A summary of my findings, limitation of this paper and further discussion are presented in Section 6.

2. Literature Review and Hypothesis

2.1 Board Turnover

The board of a corporation performs the critical function of monitoring and advising the top management (Fama and Jenson, 1983). This corporate governance mechanism is designed to decrease principal-agent problem and make management to be more responsible for shareholders. In details, the board directors consult and advise CEO and the management team on strategies for development of the company. Besides, the board members set up various committees on the board in order to oversee legal requirements or regulatory compliance. And the board directors also evaluate the management’s performances at regular intervals.

In theory, it is believed that the mechanism of corporate governance will be more effective if the directors of board have good balance of both independence and knowledge.

When it comes to independence, several key regulations promulgated by New York Stock Exchange (NYSE) and the Securities and Exchange Commission (SEC), for example, the Sarbanes-Oxley Act (SOX) of 2002, require companies to have more

(9)

independent directors on the board and committees. NYSE defines independent directors in the following way: “no director qualifies as 'independent' unless the board of directors affirmatively determines that the director has 'no material relationship' with the listed company, either directly or as a partner, shareholder or officer of an organization that has a relationship with the company.” In other words, an independent director cannot be employed by the company or have affiliation with the company. Therefore, independent directors are widely considered to be helpful to monitor the management and provide outsider perspectives as well. Lots of studies have investigated the effectiveness of independent directors and suggest positive results indeed (Hickman and Byrd, 1992; Nguyen and Nielsen, 2010; Miletkov et al., 2015), although some other papers (Bhagat and Black, 1998) find different results.

Directors become more experienced on board after several years. As directors stay longer in the company, they will get more familiar with the business and gain more knowledge; hence the cost of acquiring information is relatively lower, which leads to a better firm performance. Previous empirical evidence has found supports to this supposition (Coles et al., 2008). However, the intimacy of board-management relationship grows year by year and can weaken the intensity of board monitoring, which leads to declining performance value. (Fracassi and Tate, 2011; Kim and Hwang, 2009).

So effective board can only enjoy the benefits of director independence while not neglect the power of knowledge or skill sets at the same time. It reflects the balance of the two elements. Board turnover, which reflects the balance to a certain extent, may work as an indicator of corporate governance.

In fact, researchers have done a lot in the field of director turnover and CEO turnover. Warner et al. (1988) and Denis et al. (1995) find a negative relation between firm performance and CEO turnover. Jenter and Kanaan (2014) find out that CEOs are significantly more likely to leave their positions after bad industry performance instead of bad market performance. Yermack (2004) suggests that outside directors

(10)

will experience higher turnover rate following poor firm performance. When it comes to director turnover, Coles and Hoi (2003) point out that directors of firms who opted out the Pennsylvania Senate Bill 1310 were more likely to retain their seats on the board. Harford (2003) finds out that the outside directors are more likely to hold their board seats after a completed offer when he investigates the director turnover in M&A activities.

2.2 Dual leadership structure

CEO duality refers to the situation when a firm's CEO also serves as the chairman of the board at the same time (Boyd 1995). The role of CEO is to develop strategies and ensure the firm performance is consistent with business principles while the role of chairman is to advise and monitor the management team, including CEO, so to ensure the effective operation of the board (Brickley et al., 1997). There are two widespread theories to explain the effect of CEO duality on corporate governance, one is agency theory, and the other is stewardship theory. The proponents of agency theory think that when CEO is also the chairman, he or she can gain a stronger power. And the function of board monitoring could be diminished and challenged (Mallette and Fowler, 1992). The board independence may be constrained, which will influence the function of proper oversight and governance role of the board (Lorsch and MacIver, 1989; Fizel and Louie, 1990; Dobrzynski, 1991; Millstein, 1992). On the other hand, supporters of duality claim that the combination of two roles in one person can improve effectiveness because of reduction of monitoring and controlling costs. As CEOs do not need to deal with the potential rivalry between the chairmen and themselves, they can reduce confusion between the board and management, and then focus more on improving corporate performance. McGrath (2009) suggests that CEO duality in a firm creates unity across managers and directors, and allows the CEO to better align with the interest of shareholders. In fact, a debate on CEO duality has

(11)

been going on for years as to whether CEO duality is good for firms. And there is also up and down on the trend of separating the two roles of CEO and chairman in real practices. Larcker et. al. (2015) examine the details of leadership structures of the publicly traded companies, which includes CEO duality status, over the period of 1996 to 2015. They find that the CEO duality status is not always stable. Only about a third companies did not change the structure in those years. While some companies separated the two roles under the pressure of shareholders, some companies refused to do that, for example, Exxon Mobil. Others would recombine the two roles afterwards, like the Walt Disney Company.

Most of the discussion on CEO duality focus on its effect on firm performance, and the previous empirical results are mixed. Pi and Timme (1993) study the effect in bank industry and suggest that cost efficiency and return on assets (ROA) are lower for the banks whose CEO is also the chairman. Brickley et al. (1997) find a different result that those firms with CEO duality perform better. Baliga et al. (1996), however, suggest that there is little evidence of operating performance differences around changes in duality status. Finkelstein and D'Aveni (1994) investigate in 3 specific industries and find that board vigilance is positively associated with CEO duality. CEO duality may work just like a double-edged sword and can have two-way effects on the firm performance and internal control. Instead of preferring a certain mode of theory, Ryan et al. (2015) claim that companies should consider learning mechanisms and retention objectives when it comes to board structures. There is no one fixed rule for all companies.

2.3 Development of Hypothesis

Previous research has shown CEO power can influence the nomination process of board directors. Shivdasani and Yermack (1999) study whether CEO involvement in the selection of new directors influences the board nomination process, and find the

(12)

answer is affirmative. Even after issuing SOX in 2002, a recent research (Clune et al., 2014) also shows that there is continuing recognition of CEO influence in the director nomination process. This is one of important reasons for NYSE to require all directors of nomination committee of its listed companies should be independent directors. However, some previous research indicates that independent directors may not be ‘independent’ actually, as CEO and board directors may have close private intimacy (Fracassi and Tate, 2011; Kim and Hwang, 2009). In fact, Goyal and Park (2002) do a similar research on the relationship between CEO duality and CEO turnover rate and find out that the sensitivity of CEO turnover is lower when a director is both CEO and chairman. Poorly performing CEOs may stay longer on the board and management position when the company has CEO duality and CEO has a stronger influence on the succession process. Therefore I propose a hypothesis that the turnover rate of board directors can also be influenced when the company has dual leadership structure. H1. If a company has CEO duality, the board of a firm is more likely to stay stable and hence less board turnover.

Ocasio (1994) find that CEO duality decreases the likelihood of CEO turnover but it does not interact with firm performance. Goyal and Park (2002) revisit the problem but find different results. They find that the duality role can weaken the effect of poor firm performance on CEO turnover by half, which means that when a company has CEO duality, the CEO is 50% less likely to leave the firm. Likewise, it is reasonable to propose a hypothesis that the sensitivity of board turnover to firm performance can be influenced and also lower when the firm has CEO duality.

H2: The sensitivity of board turnover to firm performance is significantly lower when the CEO is also the chairman of the firm.

(13)

3. Data and Methodology

3.1 Sample selection

I access the director turnover and other director information in S&P 1500 firms, including CEO duality, director turnover and director age from the Institutional Shareholder Services (ISS) RiskMetrics database. As the ISS database has changed the collection procedures from Year 2007, I only make use of the data over the period of Year 2007- Year 2015 . To measure the compensation of board members, especially the ownership of the firm of each directors, I get access to Compustat (Capital IQ) Execucomp database. To investigate the firm performance and other firm characteristics, I use the company financials from database of the Center for Research in Security Prices (CRSP)/Compustat Merged database over the same period. Likewise, I use CRSP’s database to obtain stock price and generate yearly share returns as well as market-adjusted share returns.

Since I compile my sample from different databases, I cannot match all the companies among each database as there are some missing values in the database. I delete those observations which cannot be matched and then generate a less misleading sample.

To measure the change of board directors, the observations in Year 2007 have to be omitted as I do not obtain complete records of Year 2006 and thus cannot generate the board turnover change in Year 2007. As I will make use of the value of previous year (lag value), for example, share returns value of last year, some missing value are expected to be generated in the first year in my sample. For instances, for those companies which have been on S&P 500 from Year 2008 to Year 2015, the values of previous year for Year 2008 are missing as I could not gather that information in my sample. In order to do a reliable analysis, I will delete those observations when they appear in my sample in the first year.

(14)

Besides, I also delete those observations with missing values. It can happen when the records of companies are not gathered if there are gaps in the time series. For example, if a company appear in my sample in Year 2008 and Year 2010 but does not have records in Year 2009, then the value of share returns (or other variables) of previous year will be missing as well. So far I find 43 observations in this case in my sample, and thus I delete them. Then it leaves us total 8,489 records in my sample.

3.2 Key Variables

3.2.1 Board turnover

Consistent with previous papers (Harford, 2003; Bates et al., 2016) , the variable board turnover is defined as a change in the identity of the directors. Board turnover rate is calculated as the number of directors who leave the company in that year over total number of the board members of the same year. For each director, if a director does not have his or her name in the director list of any one year compared to the previous year, then the director is considered as leaving the company in that year. Many reasons can trigger director turnover. Some directors may leave a firm because of diseases or death, some leave for retirement, and others may leave due to bad performance of the firm or other personal issues.

When the problem of board turnover is discussed here, the normal succession of board directors is not considered. It means that the situation of director turnover because of “normal retirement” is not considered in my calculation. The retirement age in 95% of S&P 500 companies is set as 72 or higher (Spencer, 2016). It suggests that when a director reaches the age of 72, he or she will retire and not become a member of the board again. Based on this criteria, I regard the retirement age as age 72 in my sample and thus exclude the director turnover with the director age over 71, in order to rule out the observations which may involve with “normal retirement”

(15)

(Although probably there are still plenty of retirements of directors below the age of 72). For the situation of diseases or death, although it is also a natural process or an accident to a person and can hardly be influenced by corporate governance, I do not exclude any observation as I cannot get access to all true information of director health from public media.

To avoid the influence of extreme special situations of some firms, I will also exclude some other observations that cannot be applied to the research. As a board with 3 or less directors is considered as too small to validate the analysis, those firms will not be included in my sample, and so 3 observations are deleted. As I discussed before, the directors who are in normal succession like retirement will be excluded. Since I only have records from Year 2007 to Year 2015, I do not know the board turnover rate of Year 2007, so those observations fell in Year 2007 will also be deleted.

Finally I obtain 8,489 observations (8,489 firm-years records), with 4,433 observations (4,434 firm-years records) which do not experience board turnover and 4,055 observations (4,055 firm-years records) whose board turnover rate is not zero during Year 2008 to Year 2015.

3.2.2 Dual role of CEO-chairman (dummy)

Following the definition of the study of Goyal and Park (2002), Dual role of CEO-Chairman is coded as 1 if the director is both a CEO and chairman at the same time and 0 otherwise. It is an important indicator of corporate governance and board leadership as per my discussion before.

3.2.3 Firm Performance of the previous year

Firm performance can be measured in different ways. Some previous studies, for example, Warner et al. (1988), Kaplan (1994), use share return to measure firm performance in their papers discussing the relationship between firm performance and

(16)

changes of management members. The amount of share returns is measured as the cumulative share returns of that period in their papers. Other researchers, for example, Shen and Cannella (2002b) use ROA to indicate the firm performance when they discussed the problem about CEO succession, which is also related to management turnover. ROA is calculated as the firm’s net income divided by total assets in their paper, just like many other studies do. And other researchers use adjusted-ROA or adjusted share returns to control the industry or market influences. Goyal and Park (2002) use market-adjusted share returns and industry-relative earnings to estimate the sample firm’s return when they discuss whether CEO duality will affect CEO turnover and the sensitivity of CEO turnover to firm performance.

As there is no clear evidence to tell the best way to measure firm performance, I will use 3 ways in this paper: ROA, share returns and market-adjusted share returns. ROA, share returns and market-adjusted share returns will be lagged to reflect firm performance of previous year.

The first measure is ROA, following the measure provided by Shen and Cannella (2002b). The basic element of ROA can be easily obtained from the database of company financials.

The second measure is yearly share return. After identifying the share price of first trading day and last trading day of the company in a calendar year, I can easily calculate the yearly share return.

The third measure I apply is similar with the second one, while I also control the market influence. It is calculated as yearly share returns minus average market share returns in the same calendar year.

(17)

3.3 Control variables

3.3.1 Firm size

Prior literature related to board turnover, like the one of Dalton (2006) and Cannella and Shen (2001), suggest that firm size is an important determinant of board turnover. Cannella & Shen and Dalton define firm size as the natural log of the firm’s total assets. But Firm size can be defined in many ways.

In this paper firm size is represented by log value of market value of the firm to reflect the changes of firm size in a more direct way.

3.3.2 Independent director ratio

Bereskin and Smith (2014) investigate the effectiveness of existing corporate governance mechanisms. They assume when the rate of board turnover increases, which is related to the situation of directors resigning or not being renominated after the exposure of backdating practices, the governance mechanisms are working. They also find that companies in which independent directors receive backdated stock options have higher levels of turnover.

Based on the agency theory, more independent directors on board, board directors will be less likely to stay on board. So I include independent director ratio as a control variable in my analysis. The ratio is calculated by independent director number over total number of the board members in a specific year.

3.3.3 Company director ownership

An individual’s stock ownership serves as a proxy for power in the firm and may also influence turnover (Shen & Cannella, 2002b). While some well-known firms like Facebook or Google set some specific classes that divorce ownership and control, more companies do not have such separation mechanism. Under this circumstance,

(18)

when directors also own more shares of the company, they may want to stay in the board and have a greater influence on the company management and operation.

So the stock ownership of directors may influence the board turnover rate, and I think it is necessary to control this variable in my model. Following Dalton (2006), the company director ownership is measured as the percentage of the firm equity which are held by directors.

3.3.4 Maximum age of board directors

CEO age is normally considered to be a control variable in the previous studies related to CEO turnover.

Likewise, I will include the variable related to age of board members in my model. The age of director may influence turnover rate, as older directors may be more likely to retire or resign due to personal health considerations. So when there are more old directors on the board, the board may have a higher turnover rate.

However, most of the previous researches just made use of the individual age, like the age of CEO or the age of a director. In order to represent the age of the board members, I will use the maximum age of any director among the board members instead of average age of all directors in the company. Following Ocasio (1994), Zhang and RajagopalanAge (2003), this variable is measured as the natural log of the individual’s age in years.

Before adopting the variables in the regression, I also do the Person correlation Table to check the multicollinearity problem of the variables (do not show results here). Most of the correlations are small in magnitude, with absolute correlation coefficients less than 0.3, while two variables to measure firm performance, yearly share returns and market-adjusted share returns are highly correlated with each other (Pearson correlation coefficient = 0.80). It is not difficult to understand as the

(19)

market-adjusted share returns is calculated based on share returns value. As I will not put the two variables in the same regression, the problem can be neglected. According to the deduction in the paper of Goyal and Park (2002), multicollinearity will not be a big problem in the multivariate analysis in Section 5, and thus we could adopt the variables in my regression analysis.

3.4 Methodology

In this section, I use two empirical tests to investigate the relationship of board turnover and CEO duality status. First, I will examine the parameters in the regression model by applying ordinary least squares (OLS) method. Secondly, I will use instrumental variable (IV) in an attempt to mitigate endogeneity concerns about prior results.

3.4.1 OLS regression

Goyal and Park (2002) investigate the relationship of CEO turnover and board leadership, which mainly focus on CEO duality status. The 3 key variables in their OLS regression are firm performance, CEO duality dummy and CEO turnover rate (binary variable).

In light of their findings, the main relationship analyzed in this paper is the effect that CEO duality may have on board turnover. I identify the dependent variable as a binary variable related to board turnover status. As I mentioned before, CEO-Chairman duality dummy, firm performance proxies and other control variables will be included in the regression. The specification for investigating the effect is as follows: t i t i t i t i t i p d p control iables d bi,t =01 , 2 ,13 ,  ,1 var , (1)

(20)

where i indexes firms, t indexes year, bi,t is a binary dependent variable. It is

coded as 1 when board turnover rate is zero and 0 otherwise, di,t _{is referred as the}

dummy variable of dual role of CEO-Chairman. It is coded as 1 when the CEO of the company is also the chairman of the same year, and 0 otherwise. pi, t 1 _{is estimated}

as firm performance of previous year. The indicator can be any of the 3 performance proxies: ROA, share returns and market-adjusted share returns. di_,tpi_,t₁ is an

interaction term between the CEO-Chairman duality dummy and the firm performance.  is an unobserved error term that influences board turnover.i,t

3.4.2 IV regression

Goyal and Park (2002) apply some methods to improve the reliability of their test, including adding different control variables in the regression, and checking robustness of the models by each year. I think these methods are valuable to mitigate concerns with potential endogeneity problems, so I follow the basic idea of their methods to check the robustness of my model as well.

But I also adopt another typical method IV to try mitigate the endogeneity bias. I will estimate the sensitivity of board turnover to CEO duality status, using a two-stage IV regression approach to deal with the potential bias. Bhagat and Bolton (2013) identify CEO ratio as an instrumental variable for CEO duality dummy to explore the impact of SOX Act on the relationship between corporate governance and company performance in their paper. The IV, CEO ratio, is defined as the ratio of the CEO among the total board directors. For example, if there is total 10 directors on the board, then the CEO ratio will be 10% (1/10=10%). Some previous research (Hallock (1997), Westphal & Khanna (2003)) emphasize the role of networks among CEOs that serve on boards and the adverse impact on the corporate governance of such firms. Based on this indication, Bhagat and Bolton (2013) use this variable as IV for CEO duality, which they think is an indicator for corporate governance. The more directors on the

(21)

board, the less likely that CEO will be the chairman of the board at the same time. I follow this idea of IV in my paper. I will use the CEO ratio as an IV in the first-stage fitted regression. The first stage decomposes CEO duality into a component caused by CEO ratio.

The two-equation model for investigating the effect are as follows:

t i t i t i t i t i p d p control iables d bi,t =0 1 , 2 ,13 ,  ,1 var , (2) t i t i t i t i t i t i z p p control iables d, 01 , 2 ,13d,  ,1 var , (3)

where zi,t represents the IV, CEO ratio, an exogenous variable that determines

CEO duality status. The impact of CEO duality on the director turnover rate is measured by 1 and  . These two coefficients are related to CEO duality status,3

and also my research focus. i,t is an unobserved error term. The definition of other

variables are the same as equation (1).

While applying IV method in the linear regression model, two criteria should be considered: (1) the instrument variable must have a “strong” correlation with endogenous explanatory variable, conditional on other covariates. (2) the instrument affects the dependent variable only through the target independent variable. In other words, the relevance condition and exclusion restriction should be considered properly.

For strong relevance criteria, Bound et al. (1993; 1995) point out the IV may be misleading if it is only weakly correlated with the endogenous variables. Stock, Wright and Yogo (2002) suggest a useful thumb rule regarding to weakness of IV. The rule of thumb indicates that when F-statistic against the null that the excluded instruments are irrelevant in the first-stage regression is larger than 10, the relevance of IV and independent explanatory variable is strong and more likely not to fail.

(22)

Consistent with Bhagat and Bolton (2013), I follow this common rule of thumb to test the models and check the relevance of the IV. In detail, if the Stock Yogo critical F value is above 10 in the first-stage regression, then the IV in the regression is considered as “strong” enough. The details of all the IV regression validation is shown in Appendix 1. Most of the F-statistic that I test in our regression models are higher than 10, which suggest the IV I use is strong enough.

For exclusion restriction, one can hardly prove it in practice. Ideally the IV should be perfectly random to everything that affects the dependent variable other than the one investigated independent variable. However, it is rather difficult to obtain such IV in the real life. Bhagat and Bolton (2013) introduce the IV CEO ratio in their paper, which is motivated by extant literature, but they admit that it is difficult to confirm that the IV is uncorrelated with error terms absolutely. And they do not discuss the exclusion restriction in details. Normally the number of CEO is always one, and the board number is relatively stable in a certain company as the bylaw of corporate were normally set up years ago. So the CEO ratio will not fluctuate a lot in one company. In this paper, when there are more directors on the board, CEO ratio then becomes smaller, which suggest the CEO network influence is smaller. In this case, a smaller probability to have CEO-Chairman duality in the company will be considered as better corporate governance, based on the agency theory. CEO ratio, as I point out above, is an indicator of corporate governance. I include some other variables, like independent director ratio and company director ownership, which are used as the other two indicators of corporate governance in the paper of Bhagat and Bolton (2013), as control variables. By including control variables that are well known and proved to be related to corporate governance, CEO ratio cannot be correlated to the dependent variable (board turnover rate dummy) through other unobservables other than through the CEO duality dummy. By doing that, however, I still can hardly eradicate the possibility of the correlation and prove the exclusion restriction, although there is also no evidence to show CEO ratio affects board

(23)

turnover rate through other channels except CEO duality. This is also one of the limits of my paper. So my results are only valid when the exclusion restriction condition is met.

4. Descriptive statistics

In Table 1, I summarize the average yearly turnover rate of the sample company from Year 2008 to Year 2015. As I mentioned above, I do not have the turnover rate of Year 2007 in my sample, so I will not present it. Then leaves total 8,489 year-observations from Year 2008 to Year 2015 in my sample.

As we can see from the Graph 1, the average turnover rate was around 7% per year and quite stable from Year 2009 to Year 2013. It is kind of counter intuitive because shareholders asked directors to step down and took responsibility of the crisis, which was one of the themes of shareholder activism after the crisis in Year 2008. Although in Year 2008 the board turnover rate is higher, around 8.2%. Interestingly, the turnover rate increased dramatically above 8% and reached 9.6% in Year 2014. Since the board turnover rate can be influenced by a lot of different factors, the trend of its change is not very clear and predictive. But the rate is expected to not be very high in a row, as shareholders would prefer good directors to stay on board and help the company run smoothly.

Besides, the total number of company of Year 2008 in Table 1 is less than the number of other years. One reason for this difference may be that some data are filtered out because of non-matching problem in different databases. So the data in Year 2008 may be influenced because of the changing rules of the database in Year 2007. Or another possible explanation is massive delisting of companies before Year 2008.

(24)

In Table 2, I further divide the sample into several groups to check whether the value of board turnover rate of the observations varies a lot among different groups. While the observations with zero board turnover rate is categorized as a single group, other observations are divided into 5 groups, based on the amount of their board turnover rate.

Except for group 1, which has lowest turnover rate among the 5 groups with non-zero board turnover rate, the sample number of company is distributed quite even. The number of all these groups are all between 690 to 770. The mean value and median value are similar in Group 1 to Group 4. All observations in these 4 groups have an average board turnover rate lower than 0.2, and we can only experience excessive volatility of board turnover rate in group 5. Most of the observations fall in the group with no board turnover rate, which explains the average value of 0.08 of all year-sample.

Then I further divide the observations by year and the result is shown in Table 3 and Graph 2. Compared with other groups, the number of companies in the group with zero board turnover rate was decreasing for most of the years since Year 2010, especially a dive from Year 2013 to Year 2014. Meanwhile we can see a surge in group 1. The trend was going up for other groups during that period too, although there were also some back and forth in different years.

Before I go into detailed analysis, it is important to check the quality of data input and see if we can understand the characteristics of different variables. So I prepare the summary table with all the samples and also divide them into two groups, those with zero board turnover rate in any one year and those companies with board members changes in any one year. And Table 3 presents us the results.

We can find out that for all the year-company samples during Year 2008 to Year 2015, 52.23% (4,434/8,489) of the companies did not change their board director

(25)

members during that period, which means their board turnover rate was equal to zero. The median value and average value of the variable representing CEO duality are relatively very close to 0, which means the CEOs of most companies did not have duality role during this period. But the firm with zero turnover rate has a higher average value, which means more firms have CEO duality in this group. I also compare whether the difference between the two groups is significant or not. And the result shows the difference is significant, with a t-statistics value of 27.35. This is not surprising as I think the CEO duality may have influenced on the status of board turnover rate, which I will explore further in the analysis.

When it comes to the 3 indicators for firm performance, the mean value of market-adjusted share returns of previous year is the same between the two groups. The difference of Return on Asset (ROA) of previous year is significant for the two groups. The same situation works for share returns of previous year. In all, the group with zero board turnover rate has a higher and significant average value than its counterparts when the market influence is not controlled.

When it comes to firm size, the observation number drops from 8,489 to 7,897 because the values of some observations are missing. And the result shows that the firms with zero board turnover rate have a smaller market value. The same situation happens to the variables indicating the maximum age on the board, which also generates an missing value.

And when it comes to the other control variables, including company director ownership and independent director ratio, the difference between the two groups of companies is also significant. For the maximum age on the board, the observations have younger board directors when they belong to the group with zero board turnover rate.

(26)

5. Empirical results and robustness checks

5.1 OLS model analysis

Table 4 reports the results of normal OLS regression model to show the effect of CEO duality and other variables on board turnover status of the company.

The dependent variable is a dummy variable that takes a value of 1 if board turnover rate is zero and a value of 0 otherwise . Independent variables include the CEO duality dummy (equal to 1 if the CEO of the company is also the chairman of the board and 0 otherwise), 3 firm performance indicators (ROA of previous year, share returns of previous year, market-adjusted share returns of previous year), interaction term between the CEO duality dummy and any one of the performance indicators, log value of market value to show firm size, independent director ratio of the board, ownership of the board members, and log value of maximum age of board directors to represent the age status of the board members.

Some previous study shows that there is a negative relation between management turnover and firm performance (Weisbach, 1998; Parrino, 1997). Following Goyal and Park (2002), I add the interaction terms between the CEO duality dummy and the firm performance of prior year to the regression, in order to find out whether the company duality status will influence the sensitivity of board turnover to performance. When the coefficient is positive, it suggests that the sensitivity of board turnover to firm performance is lower for firms when their CEO is also the chairman of the board than those observations when CEO and chairman of the board are different.

Columns 1-3 of Table 4 show the OLS results for different performance measures separately, while columns 4-6 include control variables in the regression. We can know that the coefficient of CEO duality dummy is not significant although it is positive in column 1, 4-6, while the coefficient is positive and significant at 10%

(27)

level in column 2 and 3. Consistent with the previous result, the coefficient of firm performance of previous year for the first two measures is positive and significant, which means that the firms with a better performance have a lower board turnover rate. The result shows that when a company have a good firm performance of previous year, its board member will be more likely to stay stable. The probability in this regression will increase by 22.2% for one standard increase in ROA of prior year. In column 2, when the firm performance indicator is changed into share returns of previous year, the probability decreases to 2.2%. And when the stock market is adjusted to the market variation, then the result is not significant although positive, which is shown in column 3.

Column 4-6 report the estimated coefficients when the control variables are added. We can see that the coefficient of independent director ratio is negative and significant, which suggests that when there are less independent directors as board members, the board directors will stay more stable instead of changing more often. I think this is related to the controversial role of independent directors. While some studies show negative influence of independent directors on corporate governance, other research finds different results as well. For company director ownership, the coefficient is all significant at 1% level, which suggests that the company have lower probability to change the board members when directors on board own a higher portion of the company stock share. When it comes to the age influence, after filtering out those directors who have more potential to retire in a “normal” process instead of “abnormal” succession, I can obtain those observations which I consider have non-retirement-related turnover. The negative coefficient of column 4-6 suggests that when the board members are older, it is more likely to have board turnover. The result is also intuitive. Another variable firm size is all positive and significant.

In all, the results in Table 4 do not support my hypothesis of the role of duality on board turnover. However, these results may be influenced potential endogeneity problems of the models, so I will include fixed effects and use instrument variable

(28)

method to test my hypothesis again.

5.2 Industry and Year Fixed Effects

I apply two fixed effects, year fixed effects and industry fixed effects in my regression. By including fixed effects (group dummies), I can control for the average differences across different years and industries in any observable or unobservable predictors, such as differences in quality in various industries and in different time. The fixed effect coefficients are expected to absorb all the across-group actions. Based on the theory of fixed effect, the threat of omitted variable bias is expected to be reduced.

Table 5 presents the results of regression analysis for the models in Table 4, but control fixed effects at the same time. Column 1-3 reports the OLS results, which are based on 3 different firm performance measures and do not include all the control variables while column 4-6 include all control variables. Compared with the result of Table 4, the R-squared greatly increases after controlling the fixed effects in all of 6 columns. In conclusion, the results are similar to what I have found in the OLS baseline regression analysis, with a higher statistic accuracy.

In summary, the sign and significance of the variables are generally consistent with those reported in Table 4.

5.3 IV 2SLS model analysis

Table 6 shows the 2SLS results after I apply the instrument variable (IV). The dependent variables, key independent variables and control variables are the same as those in Table 4. After applying the IV, which is CEO ratio in my analysis, the magnitude of coefficient has changed a lot. In order to better interpret the results, I

(29)

also obtain the standard deviation of the coefficient of CEO duality dummy, which is shown in a bracket under the t-statistics value. Columns 1-3 show the OLS results separately with no fixed effects, while columns 4-6 control year and industry fixed effects in the regression.

We can tell from the results that the coefficient of CEO duality is all positive and significant at 1% level of every column, which is different from the OLS baseline model in Table 4. It may suggest that when a CEO is also the chairman of the board, the board turnover rate will be more likely to stay as zero, or in other words, the board will stay more stable compared to their counterparts which do not have CEO duality.

In column 1, the coefficient of CEO duality dummy suggests that one standard deviation increase in duality dummy will also increase the probability for the board to stay stable by 2.209. For the firm performance, the positive and significant coefficient suggests that the board turnover rate will decrease when the firm performance, represented by ROA of previous year, is better. Goyal and Park (2002) check the relationship between board leadership and CEO turnover and find that the turnover rate sensitivity to firm performance can be influenced by board leadership. The coefficient of interaction term here represents the relationship between CEO duality and firm performance. It is statistically significant at 5% level but negative, which is against our predicted sign. And the coefficients represented turnover rate sensitivity to ROA turn negative when CEO is also the chairman of the board, which also contradicts to our hypothesis.

Column 2 of Table 6 presents the results of 2SLS regression that use yearly share return as firm performance measures. When there is one standard deviation increase in on CEO duality dummy, the probability of the board members staying stable will increase by 1.76, a slightly smaller than the result of column 1. When there is an 100% increase in share returns, the probability of board staying stable will increase 29.3%. For observations with CEO duality, however, a 100% increase in share returns decreases the likelihood of board turnover by 482.4% (6.732-2.201+0.293=4.824),

(30)

which is positive and different from the result of column 1.

The coefficient of duality, firm performance and interaction term of column 3 is similar to the one of column 2, with a smaller magnitude. Although the sign of interaction term is different in the 3 columns, the positive and statistically significant coefficients on any one of the performance indicators suggest that when the firm performance is good, the likelihood of board director turnover will decrease.

In summary, these results support my hypothesis that CEO duality will make the board become more stable with a lower board turnover rate. When it comes to the board turnover sensitivity to firm performance, my hypothesis is true when I use share returns or market-adjusted returns as firm performance indicator.

As I mentioned before, I will include fixed effects in this IV model. This is what column 4-6 shows. Similar with column 1-3, the coefficients of CEO-Chairman duality dummy are all positive and statistically significant at 1% level, although the they have smaller magnitude than previous columns. It is mainly due to improvement of the precision of regression.

5.4 Alternative specifications and Robustness checks

While applying IV measure can mitigate the influence from potential endogeneity bias, I further reexamine the effect of CEO duality with a probit analysis and replace the age variable by calculating it in different ways. Besides, I also change the dependent variable as board turnover rate to see whether the outcomes support my hypothesis.

5.4.1 Alternative probit analysis

When the dependent variable is a binary variable, probit analysis or logit analysis are widely used. In Goyal and Park’s paper (2002), they have a logit analysis of CEO

(31)

turnover regressed on the CEO-Chairman dummy variables and other independent variables. As probit or logit analysis is limited when I apply IV at the same time in STATA, here I only compare the results with those generated from OLS baseline model to see whether there is a huge variation between the two. Theoretically the coefficients are similar between the two models, both the magnitude and sign.

Table 7 reports the results of the average marginal effects of the probit model. Column 1-3 show the results of OLS model with control variables but no fixed effect, while column 4-6 shows regression results with fixed effects. Their counterparts are column while the column 4-6 of Table 4 and column 4-6 of Table 5. As we can see, the sign and significance are very alike between the two, but probit model generates a smaller magnitude.

In summary, the results of probit model are similar with my previous OLS results.

5.4.2 Alternative analysis with board turnover rate

I also examine whether the results are also robust when the binary dependent variable in my previous analysis is changed as board turnover rate. Table 8 represents the results of OLS regression and IV regression which all include fixed effects. Technically I just change the dependent variable of the column 4-6 of Table 4 and column 4-6 of Table 6. As the value of dependent variable ranges from 0 to 1, I use tobit model in my regressions.

The coefficients of duality dummy turn negative in Table 8. However, the results do not contradict to my previous findings. When the coefficient is negative, it means that the board turnover rate is smaller and the board directors are more stable when the titles of CEO and chairman of board are bestowed on the same person. While the coefficients are all insignificant in the first 3 columns in Table 8, they turn significant and negative in column 4-6 when I apply IV in my regressions. And all the coefficients of firm performance proxies are significant and negative when the

(32)

indicator applies ROA and share return. It suggests that when the firm performance is good last year, the board turnover rate will be smaller. This is consistent to the results of most previous papers in this area. For the interaction term, the coefficients are not significant at all in column 1-3, while they all turn positive and significant when I apply IV later on in column 4-6.

In conclusion, the results are consistent to my previous findings in Table 4, 5 and 6.

5.4.3 Dealing with average age of the board members

I also examine the robustness of my findings when “age of the board members” is defined in different ways. I use the maximum age of board members to represent the board member age as one of the control variables in my previous analysis. I also check whether my results are consistent with other 3 different definitions: average age of board members after deleting those directors whose age is or over 72 in the sample; maximum age of board members before deleting those directors whose age is or over 72 in the sample; average age of board members before deleting those directors whose age is or over 72 in the sample.

As age will mechanically increase the probability of turnover rate, I filter out those directors who look more likely to retire and obtain a “non-retirement” sample. When the variable is changed to average age from average age, the results are qualitatively identical to those reported in Table 4,5 and 6. When I include all the directors in my sample, the results do not change a lot (not reported in the paper).

6. Conclusion, Limitations and further discussion

This paper attempts to investigate if the board turnover rate is influenced when the company has CEO duality. Firm performance is the most commonly discussed topic

(33)

in the papers related to CEO duality. However some researchers also worked on other problems such as the succession effects and CEO entrenchment. To my best knowledge, Goyal and Park’s Paper (2002) is the most recent paper that try to predict the relationship between CEO duality and CEO turnover. In light of their paper, I investigate the possible effect of CEO duality in board succession process.

If the results are consistent to agency theory, it is predicted that the sensitivity of board turnover to firm performance is lower in the firms which bestows the titles of CEO and chairman on the same person. According to my findings, when IV (CEO ratio in my paper) is applied in our model and market-adjusted share returns is used to indicate firm performance, the sensitivity of board turnover to firm performance is lower for the firms with CEO duality leadership. Besides, CEO duality has negative effect on board turnover, which means that the probability of board turnover is higher when firms do not separate the CEO and chairman positions.

However, the results will be effective if the IV methodology is valid, although some research actually argue the validity of instrument variables is only a matter of degree. As per my above discussion, I can hardly prove that the IV is only correlated to the dependent variable (board turnover rate dummy) through the CEO duality dummy other than error terms, and show the the exclusion restriction is already met.

Another limitation of this paper is the potential sample selection bias. I include the year-firm observation from Year 2007 to Year 2015 to check the results. When others make use of observations in different period, the result may be different. I think the result will be more effective and trustworthy when a longer period is applied. Besides, I only consider the S&P 1,500 firms in my paper. What I find may not work for all other companies, although it is a good start to have the first exploration in this area. There are a few questions and topics worth further discussion. I find out some companies go back and forth in the CEO duality problem. One famous example I discuss above is Walt Disney, which dropped CEO duality in the first place but then

(34)

later reused it. So more empirical work is needed to figure out possible explanations for it. Another interesting problem is to examine the effect of board refreshment policies, such as term limit policy, retirement age setting etc.. As far as I know, there is very limited work on this policies although they are highly discussed for years. One possible reason is limited data accessible to the public. But still we can use some proxies for board refreshment, for example, board turnover rate in this paper, to examine the influence of these policies.

(35)

Table 1 Year distribution of observations and average board turnover rate per year

This Table reports the composition of our sample of board turnover rate from 1 Jan., 2008 to 31 Dec., 2015, amounting to a population of 8,849 observations

Fiscal Year No. of Observations Board turnover rate

Y2008 783 8.20% Y2009 1,049 7.50% Y2010 1,077 6.20% Y2011 1,059 6.60% Y2012 1,148 6.90% Y2013 1,153 7.20% Y2014 1,167 9.60% Y2015 1,053 8.30% Total 8,489 7.56%

Figure 1 Time series graph of average board turnover rate

This figure shows the composition of our sample of board turnover rate from 1 Jan., 2008 to 31 Dec., 2015.

(36)

Table 2 Descriptive statistics partitioned by board turnover rate

This Table reports the descriptive statistics of board turnover rate from 1 Jan., 2008 to 31 Dec., 2015 based on 6 different groups. The observations with zero board turnover rate is categorized in group 0, other observations are divided into 5 groups based on the amount of their board turnover rate. Group 1 is the group with the lowest average board turnover rate among group 1 to group 5, while group 5 is the group with the highest average board turnover rate.

Board Turnover

Rate Group N Mean Median Minimum Maximum

0 4,434 0.00 0.00 0.00 0.00 1 1,074 0.09 0.09 0.10 0.13 2 765 0.11 0.11 0.13 0.20 3 773 0.14 0.14 0.17 0.29 4 697 0.19 0.18 0.22 0.38 5 746 0.30 0.27 0.89 1.00 Total 8,489 0.08 0.00 0.89 1.00

Figure 2 Time series graph of the number of observations in 6 groups

This figure shows the observation number change of 6 different groups by year from 1 Jan., 2008 to 31 Dec., 2015. The observations with zero board turnover rate is categorized in group 0, other observations are divided into 5 groups based on the amount of their board turnover rate. Group 1 is the group with the lowest average board turnover rate among group 1 to group 5, while group 5 is the group with the highest average board turnover rate.

(37)

Table 3 Variable statistical description

This Table reports descriptive statistics for all firm-years sample from 1 Jan., 2008 to 31 Dec., 2015. Board turnover rate is the annual board turnover. The variable dual role of CEO-Chairman (duality dummy) is coded as 1 if the company has CEO-Chairman duality and 0 if otherwise. Share returns of previous year is measured as yearly cumulative stock price change in a calendar year. Market-adjusted share returns of previous year is measured as yearly share returns minus average market share returns in the same year. Firm size is calculated as log value of market value of the firm in a specific year. Independent director ratio is measured by independent director number over total number of the board members in a specific year. Company director ownership is calculated as the percentage of the firm equity which are held by directors.

Variable name

Full sample Board turnover rate NOT

equal to 0 sample Board turnover rate equal to 0 sample T test N Median Mean Std. N Median Mean Std. N Median Mean Std. t-Statistics Dual role of CEO-Chairman (duality dummy) 8,489 0.00 0.07 0.26 4,055 0.00 0.07 0.25 4,434 0.00 0.08 0.27 27.35 ROA of previous year 8,489 0.05 0.05 0.09 4,055 0.04 0.04 0.10 4,434 0.05 0.05 0.09 10.12 Share returns of previous year 8,489 0.07 0.10 0.46 4,055 0.07 0.09 0.46 4,434 0.08 0.11 0.46 14.04

(38)

Table 3 Variable statistical description (continued) Variable name

Full sample Board turnover rate NOT_{equal to 0 sample} Board turnover rate equal to 0 sample T test N Median Mean Std. N Median Mean Std. N Median Mean Std. t-Statistics Market-adjusted stock returns of previous year 8,489 0.01 0.04 0.42 4,055 0.02 0.04 0.42 4,434 0.01 0.04 0.42 7.58 Firm Size 7,897 7.77 7.92 1.54 3,769 7.94 8.11 1.62 4,128 7.61 7.75 1.44 64.17 Independent director ratio 8,489 0.82 0.79 0.11 4,055 0.83 0.80 0.11 4,434 0.80 0.78 0.11 10.79 Company director ownership 8,489 0.01 0.03 0.06 4,055 0.01 0.02 0.05 4,434 0.01 0.03 0.07 10.79 Maximum age of

the board directors 8,488 70.00 68.85 2.64 4,054 70.00 69.03 2.74 4,434 69.00 68.69 2.53 78.63 Log value of

maximum age of the board directors

8,488 4.25 4.23 0.04 4,054 4.25 4.23 0.04 4,434 4.23 4.23 0.04 86.66 Board size 8,489 9.00 9.39 2.39 4,055 10.00 9.94 2.46 4,434 9.00 8.88 2.20

(39)

Table 4 OLS regressions

This Table presents estimates from regressing board turnover dummy during 2008-2015 on the dummy variable of CEO duality, 3 performance measures (ROA, share returns and market-adjusted share returns), and other control variables. Each column reports estimates from a single regression, with t-statistics in parentheses. ***, **, * denote significance at the 1, 5 and 10 percent levels respectively. The dependent variable is a dummy variable. It is coded as 1 if the observation has no board turnover rate and 0 if otherwise. The variable dual role of CEO-Chairman (duality dummy) is coded as 1 if the company has CEO-Chairman duality and 0 if otherwise. Share returns of previous year is measured as yearly cumulative stock price change in a calendar year. Market-adjusted share returns of previous year is measured as yearly share returns minus average market share returns in the same year. Duality dummy X ROA of previous year is an interaction term between duality dummy and one of the performance measures (the interaction term for other two measures is similar). Firm size is calculated as log value of market value of the firm in a specific year. Independent director ratio is measured by independent director number over total number of the board members in a specific year. Company director ownership is calculated as the percentage of the firm equity which are held by directors.

Predicted Sign

(1) (2) (3) (4) (5) (6)

Dependent Variable: dummy variable, board turnover rate equals to zero Dual role of CEO-Chairman (duality dummy) + 0.034 (0.023) 0.039* (0.021) 0.036* (0.021) 0.023 (0.024) 0.030 (0.022) 0.027 (0.022) Duality dummy X ROA of previous year + 0.039 (0.259) 0.072 (0.266) ROA of previous year + 0.222*** (0.056) 0.315*** (0.058) Duality dummy X share returns + -0.039 (0.041) -0.045 (0.041) Share returns of previous year + 0.022* (0.013) 0.041*** (0.014) Duality dummy X market-adjusted share returns of previous year + -0.033 (0.044) -0.038 (0.045) Market-adjusted share returns of previous year + 0.008 (0.014) 0.022 (0.015)

(40)

Table 4 OLS regressions (continued) Predicted

Sign

(1) (2) (3) (4) (5) (6)

Dependent Variable: dummy variable, board turnover rate equals to zero

Firm Size ? -0.033*** (0.004) -0.030*** (0.004) -0.029*** (0.004) Independent director ratio - -0.281*** (0.059) -0.290*** (0.060) -0.290*** (0.060) Company director ownership + 0.327*** (0.126) 0.357*** (0.129) 0.358*** (0.128) Log value of maximum age of the board directors

- -0.371** (0.153) -0.368** (0.153) -0.367** (0.153) Constant N/A 0.509*** (0.007) 0.517*** (0.007) 0.519*** (0.007) 2.553*** (0.644) 2.526*** (0.644) 2.520*** (0.644) Observations 8,488 8,488 8,488 7,895 7,895 7,895 R-squared 0.002 0.001 0.000 0.026 0.024 0.023

(41)

Table 5 Fixed Effects Analysis

This Table presents OLS regression results by regressing board turnover dummy during 2008-2015 on the dummy variable of CEO duality, 3 performance measures (ROA, share returns and market-adjusted share returns), other control variables, and also year and industry fixed effects. Each column reports estimates from a single regression, with t-statistics in parentheses. ***, **, * denote significance at the 1, 5 and 10 percent levels respectively. The dependent variable is a dummy variable. It is coded as 1 if the observation has no board turnover rate and 0 if otherwise. The variable dual role of CEO-Chairman (duality dummy) is coded as 1 if the company has CEO-Chairman duality and 0 if otherwise. Share returns of previous year is measured as yearly cumulative stock price change in a calendar year. Market-adjusted share returns of previous year is measured as yearly share returns minus average market share returns in the same year. Duality dummy X ROA of previous year is an interaction term between duality dummy and one of the performance measures (the interaction term for other two measures is similar). Firm size is calculated as log value of market value of the firm in a specific year. Independent director ratio is measured by independent director number over total number of the board members in a specific year. Company director ownership is calculated as the percentage of the firm equity which are held by directors.

Predicted Sign

(1) (2) (3) (4) (5) (6)

Dependent Variable: dummy variable, board turnover rate equals to zero Dual role of CEO-Chairman (duality dummy) + 0.025 (0.024) 0.034 (0.022) 0.032 (0.022) 0.015 (0.025) 0.027 (0.023) 0.025 (0.022) Duality dummy X ROA of previous year + 0.127 (0.261) 0.196 (0.262) ROA of previous year + 0.257*** (0.062) 0.342*** (0.065) Duality dummy X share returns + -0.027 (0.041) -0.029 (0.042) Share returns of previous year + 0.019 (0.015) 0.031* (0.016) Duality dummy X market-adjusted share returns of previous year + -0.017 (0.044) -0.019 (0.044) Market-adjusted share returns of previous year + 0.018 (0.015) 0.030* (0.016)

(42)

Table 5 Fixed Effects Analysis (continued) Predicted

Sign

(1) (2) (3) (4) (5) (6)

Dependent Variable: dummy variable, board turnover rate equals to zero

Firm Size ? -0.028*** (0.005) -0.022*** (0.005) -0.022*** (0.005) Independent director ratio - -0.285*** (0.066) -0.296*** (0.066) -0.296*** (0.066) Company director ownership + 0.490*** (0.128) 0.498*** (0.128) 0.498*** (0.128) Log value of maximum age of the board directors

- -0.353** (0.157) -0.342** (0.157) -0.343** (0.157) Constant N/A 0.507*** (0.019) 0.522*** (0.018) 0.521*** (0.018) 2.403*** (0.661) 2.343*** (0.659) 2.346*** (0.659)

Year Fixed Effects N/A YES YES YES YES YES YES

Industry Fixed Effects

N/A YES YES YES YES YES YES

Observations 8,488 8,488 8,488 7,895 7,895 7,895

(43)

Table 6 Instrument Variable 2SLS Analysis

This Table presents 2SLS regression results by regressing board turnover dummy during 2008-2015 on the dummy variable of CEO-Chairman duality, 3 performance measures (ROA, share returns and market-adjusted share returns), and other control variables, when applying the CEO ratio as the instrumental variable in the first stage. CEO ratio is defined as the ratio of directors as CEOs among the total board directors. Each column reports estimates from a single regression, with t-statistics in parentheses and standard deviation of the coefficient in bracket. ***, **, * denote significance at the 1, 5 and 10 percent levels respectively. The dependent variable is a dummy variable. It is coded as 1 if the observation has no board turnover rate and 0 if otherwise. The variable dual role of CEO-Chairman (duality dummy) is coded as 1 if the company has CEO-Chairman duality and 0 if otherwise. Share returns of previous year is measured as yearly cumulative stock price change in a calendar year. Market-adjusted share returns of previous year is measured as yearly share returns minus average market share returns in the same year. Duality dummy X ROA of previous year is an interaction term between duality dummy and one of the performance measures (the interaction term for other two measures is similar). Firm size is calculated as log value of market value of the firm in a specific year. Independent director ratio is measured by independent director number over total number of the board members in a specific year. Company director ownership is calculated as the percentage of the firm equity which are held by directors.

Predicted Sign

(1) (2) (3) (4) (5) (6)

Dependent Variable: dummy variable, board turnover rate equals to zero Dual role of CEO-Chairman (duality dummy) + 8.463*** (2.860) [2.209***] 6.732*** (1.862) [1.760***] 6.622*** (1.820) [1.731***] 7.400*** (2.093) [1.931***] 6.120*** (1.429) [1.600***] 5.952*** (1.380) [1.556***] Duality dummy X ROA of previous year + -45.620** (18.210) -39.200*** (13.370) ROA of previous year + 2.331*** (0.704) 1.865*** (0.519) Duality dummy X share returns + -2.201*** (0.709) -1.973*** (0.569) Share returns of previous year + 0.293*** (0.072) 0.258*** (0.056) Duality dummy X market-adjusted share returns of previous year + -1.024** (0.432) -0.883** (0.357)

(44)

Table 6 Instrument Variable 2SLS Analysis (continued) Predicted

Sign

(1) (2) (3) (4) (5) (6)

Dependent Variable: dummy variable, board turnover rate equals to zero Market-adjusted share returns of previous year + 0.146*** (0.039) 0.121*** (0.032) Firm Size ? -0.031* (0.017) -0.043** (0.017) -0.042** (0.017) -0.044*** (0.017) -0.052*** (0.017) -0.048*** (0.016) Independent director ratio - -0.958*** (0.335) -0.899*** (0.278) -0.918*** (0.280) -0.810*** (0.280) -0.735*** (0.226) -0.744*** (0.226) Company director ownership + -0.406 (0.525) -0.993* (0.586) -1.041* (0.590) -0.160 (0.444) -0.705 (0.478) -0.733 (0.474) Log value of maximum age of the board directors

- 0.146 (0.645) 0.112 (0.577) 0.063 (0.570) 0.186 (0.566) 0.168 (0.533) 0.094 (0.516) Constant N/A 0.335 (2.699) 0.621 (2.410) 0.850 (2.381) -0.071 (2.371) 0.240 (2.221) 0.422 (2.153)

Year Fixed Effects N/A YES YES YES

Industry Fixed Effects

N/A YES YES YES

Will CEO duality influence board turnover?

Master Thesis