November 3, 2008 by Pavlo M. Levchuk University of Groningen

(1)

1

CEO CLONING AS A MEASURE OF BOARD COMPOSITION

ENDOGENEUSLY DETERMINED BY CEO POWER AND

STAKEHOLDER PRESSURE

November 3, 2008

by Pavlo M. Levchuk University of Groningen

(2)

2 CONTENTS CONTENTS ... 2 INTRODUCTION ... 3 CONCEPTUAL MODEL... 6 Board composition ... 6

CEO-Board power struggle ... 7

Relative tenure ... 8

Duality ... 8

External environment and stakeholders’ pressure ... 9

Active institutional monitoring ... 9

Presence of a large stockholder ... 10

Corporate social responsibility ... 11

METHODOLOGY ... 13

Data ... 13

Time dimension ... 13

Operationalization of variables and measurement issues ... 14

ESTIMATION ISSUES ... 19 Multicollinearity ... 19 Heteroskedastisity ... 20 Serial correlation ... 21 Endogeneity... 22 RESULTS ... 25

Managerial power approach ... 25

External pressure approach ... 25

Other findings... 27

DISCUSSION ... 28

BIBLIOGRAPHY ... 31

APPENDIX 1 – MULTICOLLINEARITY ... 35

APPENDIX 2 – SERIAL CORRELATION ... 40

(3)

3

INTRODUCTION

This paper builds upon theoretical frameworks of principal-agent theory (Ross, 1973; Eisenhardt, 1989), bargaining framework (Hermalin and Weisbach, 1998), managerial power approach (Bebchuk, Fried and Walker, 2002) and stakeholder theories of the firm (Freeman, 1984; Clarkson, 1995; Post, Preston and Sachs, 2002) to develop a conceptual and econometric models for determining corporate board composition where board composition is an outcome of the CEO-board bargaining and the external pressures imposed by the stakeholders.

Organizational behavior and social psychology studies (Baskett, 1973; Wexely and Nemeroff, 1974) provide ample empirical evidence of a positive relationship between demographic applicant-employer similarity and the employer’s perceived quality of the applicant. Thus, applicants who are demographically similar to the employer will have a higher probability of getting the job, then other applicants (Lawrence, 1997). In such a way, employers can ‘clone’ themselves throughout the organization – a process, which was named by Kanter (1977) as ‘homo-social reproduction’. If we apply this framework the corporate governance setting, then the demographic characteristics of the board members will be most similar to those individuals, who can influence the nomination decision process. Officially, shareholders hold the legal right to elect board members. However, in practice, and as supported by numerous studies (Shivdasani and Yermack, 1999; Main, O'Reilly and Wade, 2002; Lorsch and MacIver 1989), it is often CEOs who can exert the most influence upon the nominating process by proposing and endorsing the candidates.

(4)

4

and the CEO, the greater the probability that the CEO will be awarded a more generous compensation scheme.

On the other hand, a number of recent studies suggest that external stakeholders can successfully limit the CEO’s ability to use managerial power in his/her own interest. Particularly studies of CEO compensation by Almazan, Hartzell and Starks, (2005) and Cyert, Kang and Kumar, (2002) have found that the presence of institutional investors and large shareholders, negatively affect the amount of CEO compensation awarded. The study of Bebchuk, Fried and Walker (2002) is particularly interesting in this context. The authors argue that excessive managerial power and consequently, abnormal compensation packages, come at a certain cost for the CEO, which they have termed as ‘the outrage costs’. One can think of the outrage costs as the CEO’s fear of causing a financial scandal, which would result in reputation damage, fraud litigation, etc. Therefore, outrage costs are high in companies where shareholders strongly demand accountability and transparency from the management. Thus institutional investors and large shareholders can mitigate CEO power by raising the outrage costs, thus making it less attractive for the CEO to abuse his power, whether for raising his/her pay, or for setting the tone and composition of the corporate board. Another interesting contribution in this area was made by Roosenboom P. (2005) who found that large shareholders can successfully oppose CEOs in bargaining on board composition to increase the number of independent directors. He argues that large shareholders can restrict CEOs ability to determine board composition. Indeed, there are a number of ways, how external stakeholders can influence board characteristics. They can achieve this by demanding better corporate social responsibility from the company, introducing codes of governance and placing their representatives on the board, thus altering the socio-demographic characteristics of the board.

(5)

5

conceptually appealing then previous models because it accounts for both internal and external environments of the company. Testing this model and propositions, CEO cloning hypothesis and its implications will be the main subject of further discussion.

(6)

6

CONCEPTUAL MODEL Board composition

It seems natural to begin by introducing the central construct, which the model is attempting to explain. Internal mechanisms of board composition have received a lot of attention in resent empirical research studies. However, board composition is an abstract concept, which is not measurable. Different studies of organizational demography used various techniques to express group composition as for instance, in terms of demographic distance expressed as an average dissimilarity between individual’s and group’s characteristics (Boone, Olffen, Witteloostuijn, and Brabander 2004) or in terms of demographic similarity between individual members of the group (Westpal and Zajac 1995).

In this model, my point of reference is the CEO of the firm. Consequently, I have defined board composition in terms of socio-demographic similarity between CEO and the board members’ average profile.1 This proxy for board composition, will be further referred to as the ‘CEO cloning’. The term ‘CEO cloning’ is used because hypothetically speaking, the highest degree of demographic similarity between the CEO and the board members implies that the board will consist exclusively from CEO ‘clones’. Demographic distance between CEO and the board members is measured along a number of dimensions. These dimensions are age, gender, nationality, education and recruitment path. Detailed definition of the CEO cloning index is discussed in the methodology chapter.

In the model, board composition measured by CEO cloning index, is endogenously determined as the organization’s response to the preferences of the CEO, and the requirements imposed by the stakeholders in the external environment where the firm operates. In general terms the relationship can be expressed as follows:

c = f (p, s)

where CEO cloning c, is some function of (1) managerial power of the CEO vis-à-vis the board of directors p, and (2) influence of the stakeholders s. This model of board composition is more theoretically satisfying and appealing then previously developed models, because it assumes that board composition is jointly determined by both the internal forces within the organization itself, and the forces in the external

1

(7)

7

environment of the firm. From this perspective, board dynamics resembles the evolution of a living organism whose characteristics (board composition) are shaped by the inherited gene code (CEO managerial power) and the irritants from the natural habitat (demands from the company’s stakeholders) to which the organism is constantly adapting. I believe that such evolutionary view of organizational structure is a closer representation of the underlying ‘reality’ that ultimately determines composition of the corporate boards. I will proceed by examining the relationships (1) and (2) in detail.

CEO-Board power struggle

(8)

8

bargaining framework by Hermalin and Weisbach (1998) can explain reasonably well the changes in board composition.

Relative tenure

A number of scholars have documented the effect of tenure on the managerial power (Alderfer 1986, Finkelstein, 1992) arguing that long tenure results in accumulation of the firm specific know-how, reputation and expertise, which subsequently enhance managerial influence and authority throughout the organization. In the corporate governance setting, findings confirm that CEO tenure relative to that of the board (directors’ average tenure) has the same effect on the CEO power (Wade, O’Reilly and Chandrat, 1990; Westpal and Zajac, 1995). The longer a CEO holds his position, the harder it is for the board members (particularly for those who were appointed before the CEO) to question or challenge his/her authority.

Hypothesis 1: CEO tenure relative to the board’s average is positively associated with CEO cloning.

Duality

Another essential aspect of corporate governance is the structural arrangement between CEO and the board of directors. Agency theory predicts that CEO duality, which occurs when the positions of the CEO and the chairman of the board are held by the same person, promotes CEO entrenchment by reducing board independence and monitoring effectiveness (Jensen and Meckling, 1976). And although findings of Finkelstein and D'aveni (1994) demonstrated positive association between board vigilance and CEO duality, which contradicts the logic of agency theory, CEOs holding both positions will have both greater informal and delegated authority in director selection and recruitment decisions, thus minimizing board’s involvement in the director recruitment process.

(9)

9

External environment and stakeholders’ pressure

Up until now we have discussed the factors both structural and psychological that may affect board composition, from the inside of the firm. However there is a missing link in this bargaining framework – external stakeholders. It does not mean that the bargaining framework is flawed, rather that there are other agents in the bargaining game whose preferences have not been accounted for in the model yet. As illustrated in Figure 2, the factors in the external environment of the firm, which may help us better explain variations in the board composition, are: institutional concentration, presence of a large stockholders and social corporate responsibility. While the first two, represent the preferences of various investors (shareholders), the latter represents overall preference in terms of demand for social corporate responsibility which is demanded not only by the owner but also by the other involved parties such as trade unions, government, stock exchange (codes of good governance), etc. This stakeholder view of the firm goes back to the studies of strategic management by Freeman (1984) who described operations of a firm as a concentration of interdependent relationships between involved parties – ‘stakeholders’, which was later developed into a ‘stakeholder theory’ of the firm, employed by academicians in numerous studies (e.g. Clarkson, 1998; Mills and Weinstein, 2000; Post, Preston and Sachs, 2002). Recently, in a study of Dutch board roles and member recruitment procedures, using comparative institutional analysis, Ees and Postma pointed out: “…board roles and procedures are socially constructed; that is, they are embedded in a wider institutional context that determines the commitment and bargaining positions of relevant stakeholder groups” (2005:93). All of these studies emphasize the importance of both internal and external environments when analyzing the firm and organizational and governance structure, where board composition is not an exception.

Active institutional monitoring

(10)

10

P. Morgan (bank holding company) have successfully lobbied for the removal of the CEO at a number of poor-performing corporations in U.S. among which were Kodak, IBM, American Express and General Motors. An even more direct evidence of how institutional investors can successfully ‘bargain’ on board composition is presented in an empirical study by Baker and Gompers (2003) who studied the role of venture capital, which is usually supplied by investment banks and other financial institutions, in a sample of 1,116 IPO firms. Interestingly, the authors express board composition as an outcome of a bargaining game between CEO and outside shareholders, and find that the number of outside independent directors is positively associated with the presence of venture capital and the reputation of the venture firm (shareholder power); and negatively associated with CEO tenure and voting control (managerial power). In a similar study on executive compensation by Almazan, Hartzell and Starks (2005), whose findings are consistent with the view that active institutional investors provide more intense monitoring of corporate management. Particularly, pay-for-performance sensitivity is positively associated with the concentration of active institutional investors. All of these findings suggest that active institutional investors are influential actors in the bargaining game and can successfully lobby for their own preferred board composition. Applying this line of argumentation together with similarity-attraction principle, we expect the demographic composition between CEO and the board to be greater in the presence of an active institutional investor, and the greater the ownership of the investor the greater the distance.

Hypothesis 3: Active institutional investors’ ownership is negatively associated with CEO cloning.

Presence of a large stockholder

(11)

11

CEOs in management controlled firms can easier negotiate contracts with reduced employment risk by manipulating incentive alignment structure, the monitoring arrangements, and to a lesser extent governance structure. Implications of these findings are clear, in the absence of ownership concentration, CEOs have greater control of their own contractual arrangements at the expense of firm performance and shareholder returns, on the other a large stockholder can limit such behavior. Evidence from the executive compensation study by Tosi and Gomez-Mejia (1989) is consistent with this view, who concluded that in owner-controlled firms major stock holders and board of directors had greater control of the CEO compensation than in the management controlled firms, where CEO pay arrangement was beyond their influence. Another study of considerable importance directly examines the effect of equity ownership of the largest outside shareholder on the CEO-board negotiations of the CEO compensation. The authors find that the CEO equity compensation is negatively related to the (1) equity ownership of the largest external shareholder, (2) equity ownership of the board, and (3) the firm’s risk of default. Similarly, in the context of this study, we can expect that the CEO’s ability to influence board composition will be limited, the greater the equity ownership of the largest external shareholder.

Hypothesis 4: equity ownership of the largest external shareholder is negatively associated with CEO cloning.

Corporate social responsibility

(12)

12

composition. For this reason, stakeholders are especially interested in the board composition and often try to affect the recruitment process, in such a way, that could best protect their interests. This increased concern in how responsible firms are in protecting interests of the stakeholders, has sparked an explosive growth in the demand for corporate social responsibility ratings (Márquez and Fombrun, 2005). In this context the best way to think of corporate social responsibility is by using the analogy with the ‘outrage costs’ introduced by Bebchuk, Fried and Walker (2002). Outrage among the stakeholders can be produced by the actions of the CEO, as for instance trying to get more control of the nomination process, manipulate his/her own compensation arrangement or proposing a candidate for a position in the board who comes from the CEO’s social network and thus can easily be influenced. Outrage produced by such actions can significantly damage CEO’s reputation and in the extreme can lead to CEO’s dismissal. The same actions of the CEO will generate greater outrage if the level of corporate social responsibility is high. Therefore corporate social responsibility can be seen as external firm environment created by the stakeholders, which poses constraints for the CEO to realize his preferred board composition.

(13)

13

METHODOLOGY Data

This study is based on a sample of 190 observations, containing information on 38 US firms in the time period 2002 – 2006. The sample has been randomly selected from the list of Fortune’s 500 largest companies. Most of the information has been collected from the SEC filings published on the website of US Securities and Exchange Commission. Information on the board of directors, company ownership and personal director information has been extracted from DEF 14A filings. Firm-level information related to size and performance has been extracted from 10-K filings. Personal information about education of directors has been primarily collected from online directories and social mapping websites2. Additionally, annual reports and company websites were used as complimentary sources of information. As a proxy for company’s corporate social responsibility I have used use Dow Jones Sustainability World Index (DJSI World).

This research relies heavily on the availability of personal information on directors and executives. Therefore only the largest companies, which receive significant public attention, were suitable for analysis, while smaller companies tend to provide less information about their directors. Even though the sample has been randomly selected, such convenience sampling technique has direct implications on the generalizability of the research. It will be difficult to draw conclusions and indicate implications for the corporate governance of the smaller firms. Furthermore due to the fact that the corporate governance environment formed by the US institutional framework, differs significantly from the rest of the world, it would be difficult to extrapolate conclusions of this study on other countries. Nevertheless this study will provide a framework which can be used to study determinants of the board composition in other samples.

Time dimension

As argued by Pfeffer (1997), in order to fully understand the determinants of demographic group composition, it is essential to examine demographic composition at a specific point in time and the subsequent demographic flows into and out of the group. Boone, Olffen, Witteloostuijn, & Brabander (2004), also believe that group

2

(14)

14

dynamics can be best understood by studying its metabolism i.e. entry and exit of individuals over a period of time. On the other hand, a snapshot analysis can also provide interesting insights for understanding the determinants of demographic group composition and as explained by Hermalin and Wesbach: “Perhaps the most natural way to examine the factors affecting board composition is to look cross-sectionally at the firm level factors…” (2003:17).

Considering these arguments I have decided to use panel data for this research, which allowed me to benefit from both time series and cross-sectional approaches. Furthermore, using panel data helps in dealing with the endogeneity, which is one of most common and pervasive problems in social and psychological science research.

Operationalization of variables and measurement issues

Table 1 summarizes information on the measurement of the variables used in the model and the predicted sings of their beta coefficients. Further in this section I discuss in greater detail measurement of the dependent variable and a number of problems related to the measurement of some independent variables.

(15)

15

TABLE 1 - OPERATIONALIZATION OF VARIABLES

Name Description Effect Hypothesis

Dependent variable

ci

CEO cloning is a continuous variable, which measures the distance between CEO and the board of directors along age, gender, education and recruitment dimensions.

Variables measuring managerial power

pi1 Relative tenure, calculated by subtracting CEO tenure from the average _{director tenure.} + H1

pi2 Duality is a binary variable which assumes value of 1 if the CEO is also the _{chairman of the board and 0 otherwise.} + H2

Variables measuring external pressure

si1

DJSI is a proxy for corporate social responsibility expressed as a binary variable assuming 1 if the company is listed in the Dow Jones

Sustainability Index and 0 otherwise.

– H3

si2

Institutional monitoring is calculated as a percentage of total stock ownership held by active institutional investors among the firm’s five largest institutional owners.

– H4

si3

Presence of a large stockholder is a binary variable coded as 1 when at least 5 per cent of the outstanding equity is held by an external owner and 0 if no single individual, institution or group holds at least 5 per cent, or it is held by an internal manager.

– H5

Control Variables

xi1 Board size, number of board members. xi2 Firm size, number of employees

(16)

16

into a single general proxy of demographic distance between an individual and a group.

This research paper will utilize a combination of methods used in the studies mentioned above. A single general proxy of demographic similarity between CEO and the board of directors called ‘CEO cloning’ will be compiled. This proxy will provide an indication of how successful the CEO was in cloning himself into the board, i.e. enforcing his/her own characteristics on the board by advocating board members with the same demographic characteristics. To compile the CEO cloning index I will use one continuous and five categorical measures – age, gender,

nationality, area of expertise, internal hire and CEO-network, respectively. Age similarity is calculated using the analog of Euclidean measure of distance, commonly used in relational demography research (e.g. Tsui, Egan, and O'Reilly, 1992), here defined as:

∑

= − n i j j i n S S )2 (

where Si is the CEO age, Sj is the age of the board director j and n is the number of

directors on the board. Following Westpal and Zajac (1995), I will transform the outcome of this equation into a measure of similarity by subtracting each firm’s coefficient form the highest value in the sample.

(17)

17

Using a single cumulative measure of demographic similarity, rather than a number of separate measures, has one peculiar advantage, besides eliminating the need to run separate regressions for each measure. As argued by Boone, Olffen, Witteloostuijn, and Brabander (2004) demographic distance between an individual and a group can be best measured by aggregating uncorrelated distances along a number of dimensions. The reason for that is because changes in the group composition do not affect all demographic indicators of the group in a uniform way (Priem, Lyon and Dess, 1999). Therefore demographic distance between an individual and the group can be best and more conveniently captured using a single cumulative proxy.

As outlined by Shieifer and Vishny (1986), measuring multiple institutional owners is often plagued by the free-rider problems. To avoid this limitation I will employ the measure of concentration of the institutional investors' ownership, which was developed by Almazan, Hartzell and Starks (2005) by separating institutional investors into two categories of active and passive investors. The classification distinguishes between passive and active institutional investors, where bank trust departments, insurance companies, endowment funds, self managed corporate pension funds and public pension funds are passive, while investment companies and advisor

TABLE 2 – DESCRIPTIVE STATISTICS

Mean Median Max. Min. Std. Dev. Skewness Kurtosis n

(18)

18

are active institutional monitors.

(19)

19

ESTIMATION ISSUES Multicollinearity

Most of the data used in this research, was not obtained as a result of a controlled experiment but was rather collected for administrative and reporting purposes of the firm. Thus there is no guarantee that the data contains enough information to identify and estimate individual effects of the explanatory variables with sufficient precision, because some of the variables may move together in a systematic way – a condition termed multicollinearity. Related problem is when the explanatory variables exhibit little variation, leading to a less precise estimation of its coefficients, particularly when the number of variables increases relative to the sample size. High sensitivity for including variables in the regression and changing direction of the effects of duality and institutional monitoring, suggest for the existence of multicollinearity problem in the data. Collinearity can be detected in a number of ways. One way is to use sample correlation coefficients between pairs of explanatory variables. Generally

TABLE 3 – CORRELATION MATRIX

This table provides pairwise correlations. CEO cloning, is the distance between CEO and the board of directors along age, gender, education and previous employment dimensions. Relative tenure is measured as difference in years by subtracting CEO’s tenure from the board’s average. Duality is a binary variable which assumes value of 1 if the CEO is also the chairman of the board and 0 otherwise. DJSI is a binary variable coded as 1 if the company is listed in the Dow Jones Sustainability Index. Institutional monitoring is calculated as a percentage of total stock ownership held by active institutional investors among the firm’s five largest institutional owners. Presence of a large stockholder is a binary variable coded as 1 when at least 5 per cent of the outstanding equity is held by an external owner and 0 if no single individual, institution or group holds at least 5 per cent, or it is held by an internal manager. P-values are given in parentheses.

clon r_tenure duality djsi_w inst_mon l_stock b_size

(20)

20

a correlation coefficient greater than 0.8, indicates a strong linear association and a potentially harmful collinear relationship. However collinear relationships may exist among more than two explanatory variables. In such cases the presence of multicollinearity can be detected by estimating auxiliary regressions where one of the explanatory variables becomes a dependent variable. After examining the correlation matrix in Table 3 we can see that all the correlation coefficients of duality have a relatively small effect in terms of magnitude and are mostly insignificant. However, the auxiliary regressions, reported in Appendix 1 indeed have identified harmful collinear relationships exhibited by duality and institutional monitoring. These were highly correlated with each other and some other explanatory variables.

The best way of dealing with multicollinearity is to obtain more and better data to provide richer information on the studied relationships. Unfortunately, within the time horizons of this study it was not feasible to obtain better data or use a larger sample. Adding non-sample information was not an option either, due to (1) the lack of understanding in modelling board psychology and (2) the associated bias problem if the introduced restrictions are not true.

Heteroskedastisity

Discussion of heteroskedasticity is relevant since the data contain both cross-sectional and time series information. Heteroskedasticity exists when the residuals from the least square procedure increase (or decrease) systematically, in their absolute value, together with an increase (or decrease) in some other explanatory variable. If that is the case then the variation of the dependent variable around its mean follows some systematic pattern, meaning that the variance and the uncertainty about the dependent variable are not constant. As a result the least square estimator is not BLUE (Best Linear Unbiased Estimator), although it is still linear and unbiased. Since I cannot assert theoretically that a particular form of heteroskedasiticty exists, I have used the Goldfeld-Quandt test to check for heteroskedasticity problems.

Relative tenure and institutional monitoring were used as a point of reference for sorting and dividing the sample. The results of the tests are shown below. The sample has been sorted separately by relative tenure and by institutional monitoring. Both

samples were split in two groups of sub-samples with high (

σ

ˆ₁2) and low ( 2

2

ˆ

σ

)

(21)

21

0,17. Computing GQ =

σ

ˆ₁2/ 2

2

ˆ

σ

gives us 0,875. The critical values from the

F-distribution at 95% confidence level, with T1 = T2 = 88 degrees of freedom is 1,423.

The null hypothesis of homoskedasticity is rejected when GQ > Fc . Since 0,875 is

less than 1,423 we reject the alternative of heteroskedasticity. Similarly, estimated error variances when arranging the sample according to institutional monitoring are

2 1 ˆ

σ

_{= 0,161 and} 2 2 ˆ

σ

_{= 0,167. GQ = 0,923 and F}_c₌_{1,422. Since 0,923 is less than}

1,422 we reject the alternative hypothesis of heteroskedasticity.

Serial correlation

Serial correlation occurs in time-series studies when the errors associated with a given time period carry over into future time periods. The consequences of autocorrelation are: (1) OLS is no longer efficient among linear estimators (2) standard errors computed using the textbook OLS formula are incorrect and generally understated, and (3) if there are lagged dependent variables on the right-hand side, OLS estimates are biased and inconsistent. Furthermore, since prior residuals help to predict current residuals, we can take advantage of this information to form a better prediction of the dependent variable.

In the context of this study, discussion of serial correlation is particularly relevant because of the board’s slow ‘metabolism’. Board composition appears to change slowly over time, in fact, data indicates that boards replace on average 1,7 directors each year. Given that the average board size is 11,6 members, the majority of directors will serve on the board for at least next three years. This means that board composition in any subsequent year can be largely explained by board composition in previous years, which is a good theoretical argument to include lags of the dependent variable in the right hand side of the regression equation. This will also help to mitigate the consequences of autocorrelation.

(22)

22

are: dL = 1,687 and dU = 1,839. Since dw > 4 – dL or 2,553 > 2,313 I accept the

alternative hypothesis of negative serial correlation.

To mitigate the effects of negative AR(1) serial correlation and to improve the explanatory power of the model I have included t – 1 and t – 2 lags of the dependent variable on the right-hand side of the equation. Output of this regression is reported in Table 2 of Appendix 2. However, when the lagged dependent variables appear among the regressors, the DW test is no longer valid and cannot be used. Unfortunately, version 5.1 of EViews does not provide a full set of specification tests for panel equations, therefore I have constructed a test using residuals obtained from the panel estimation, reported in Table 2. Residuals from this specification were saved and regressed on the lagged residuals. Under the null hypothesis that the original idiosyncratic errors are uncorrelated, the residuals from this equation should have an autocorrelation coefficient of – 0,5. Output in Table 3 of Appendix 2 reports an estimate of – 0,43 which appears to be close to the null hypothesis value. A formal Wald hypothesis test, reported in Table 4 of Appendix 2, rejects the alternative that the original idiosyncratic errors are serially correlated.

Endogeneity

One of the most pervasive problems in social and psychological science research is endogeneity. In their survey of literature on boards of directors, Hermalin and Weisbach (2003) point out that one of the most common hurdles in empirical work on corporate governance is that almost all of the variables of interest are endogenously determined, causes can almost always be effects.

(23)

23

positively affect CEO power, which will translate into greater CEO – board similarity. On the other hand, one can argue that stronger CEOs will be capable of negotiating a position of the board’s chairman for themselves.

The problem of endogeneity is closely related to the problem of establishing causality. One can observe and detect correlation whereas establishing cause and effect is a completely different matter altogether. There is always a chance that the variation in both variables was caused by the third, unobservable, variable. In his treatise on human nature (1740) and later in his enquiry concerning human understanding (1977), David Hume raises the question of how we arrive at the concept of causation. He argues that causation is not a recording of something observed in the external world but rather a result of purely metal processes of the mind, it is what mind assigns to ‘constantly conjoined’ experiences when A reliably precedes and B reliably follows. Therefore causal relationships are fundamentally grounded in the theory and rational reasoning. Based purely on theoretical arguments, the CEO cloning model, built upon managerial power approach and similarity attraction principle, suggests that the relationship between CEO power and board-CEO similarity will exhibit reciprocal causality. Theoretically this is not a problem, reversed causality relationship is a very common phenomenon in nature. Problems arise when one wants to measure the effects of endogenous variables empirically.

(24)

24

sustainability rating and thus decrease the company’s chances of being listed in the Dow Jones Sustainability Index.

(25)

25

RESULTS Managerial power approach

As summarized in Table 1, I used relative tenure and CEO-chairman duality as measures of the managerial power. Managerial power approach predicts that CEO cloning will be positively related to both relative tenure and CEO-Chairman duality, as stated in the hypothesis 1 and 2. Results reported in Table 3, are partly consistent with the predictions of the managerial power approach. Relative tenure is positively correlated with cloning at the 5% significance level. The correlation coefficient for relative tenure is 0,16. For duality, correlation coefficient is negative, and statistically insignificant.

Let us now look at the regression results, reported in Table 4. Results for relative tenure are consistent with predictions of the framework and significant at the 1% level, in all the models where it was used. When examined separately, duality is positively correlated with cloning at the 1% significance level. Surprisingly, when tested together with relative tenure in Model 6, duality appears negative and insignificant. When including all the independent variables in Model 8, duality coefficient appears with the correct sign but statistically insignificant. These results provide strong support in favor of hypothesis 1 that CEO cloning is more pronounced in the boards where the CEO tenure is relatively higher than the board’s average. While hypothesis 2 cannot be accepted because results provide insufficient evidence to accept that cloning is higher in those companies where CEOs occupy position of the chairperson. Inconsistent results of duality on cloning in different models can be explained by collinear relationships which will be discussed later in this chapter.

External pressure approach

(26)

26

that DJSI negatively affects CEO cloning at the 1% significance level both, when tested separately in Model 3, and together with other variables in Models 7 and 8.

Presence of the large shareholder also has a negative effect on CEO cloning. The effects are significant at the 5% level when tested separately in Model 5 and at the 1% level in Models 7 and 8. These results partly support propositions of the external

TABLE 4 – REGRESSION ANALYSIS

Dependent variable: Log (CEO Cloning)

Estimated coefficients from multiple regressions employ panel data from 38 largest US firms for in the time period 2002 – 2006. Board size and number of employees are control variables. Lags of the dependent variables are included to control for serial correlation. Standard errors are computed using robust methods in which observations are clustered by cross-sections. P-values are given in parentheses. Models 1 - 5 test the separate effects of each of the independent variables on the dependent. Models 6 and 7 groups explanatory variables based on managerial power and external pressure approaches respectively. Model 8 includes all the explanatory variables in the regression. Model 9 excludes potentially harmful collinear relationships from the regression.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9

Relative tenure 0,010 0,010 0,009 0,009 (0,000) (0,000) (0,000) (0,000) Duality 0,069 -0,017 0,002 (0,000) (0,250) (0,940) DJSI -0,057 -0,057 -0,055 -0,051 (0,000) (0,001) (0,003) (0,004) Institutional Monitoring 0,002 0,003 0,002 (0,009) (0,001) (0,056) Presence of a large shareholder -0,042 -0,059 -0,062 -0,060 (0,027) (0,003) (0,000) (0,002) Control Variables Board size -0,031 -0,028 -0,028 -0,028 -0,028 -0,031 -0,028 -0,030 -0,030 (0,000) (0,000) (0,000) (0,000) (0,000) (0,000) (0,000) (0,000) (0,000) Number of employees

(27)

27

pressure approach. We accept hypothesis 3 and 5. CEO cloning is less pronounced if the company is listed in the DJS Index and if at least 5% of the stock ownership is concentrated in the hands of a single shareholder. We expected to see negative relationship between active institutional monitoring and CEO cloning. To our surprise, regression coefficients showed up positive and statistically significant at the 1% level in Models 3 and 7; and at 10% level in Model 8. Consequently we cannot accept hypothesis 4.

Other findings

Even though the interpretation of control variables is not the subject of this study, I felt it necessary to discuss it briefly. Board size and number of employees were used to control for the size differences. Board size has negative and significant effect on cloning in all models which implies that larger boards have larger diversity thus more distant from the CEO. On the contrary, size of the company is positively associated with CEO cloning. A possible explanation of the positive sign could be that larger companies have the most successful and demanded CEOs, thus making them more powerful vis-à-vis the board. However the relationship is statistically insignificant in all the regression models. Table 4 also reports the adjusted R2 for each of the examined models. The values appear relatively high, around 50 - 60%. This is normal for studies that include time-series data. However, when using econometric models with correlated errors, as explained earlier, one should avoid interpreting the results of R2.

(28)

28

DISCUSSION

In this paper, I have set out to investigate the proposition that strong CEOs can clone themselves in the boards of directors, thereby undermining the board’s monitoring function, unless there are pressures in the external environment that can limit the CEOs cloning ability. Decreased monitoring ability of the board could subsequently lead to abnormal compensation packages and sub-optimal performance of management thus destroying shareholder value.

To test the existence of CEO cloning I used personal information on directors and firm-specific data from 38 largest US companies over a 5-year period using. A theoretical model consisting of five hypotheses has been developed to analyze the relationships between cloning and characteristics of the firm.

As mentioned in the previous section, the empirical results provide only partial support for the managerial power approach and the external pressure approach as proposed by the theoretical framework developed in this paper. We have failed to confirm two out of five hypothesis, which leads us to believe that the theoretical framework needs adjustment. Possible improvements will be suggested at the end of the chapter.

(29)

29

Wade Chandratat 1990). Gaining support among the board members may eventually lead to decreased monitoring ability of the board, letting the CEO take decisions that are not always in the best interest of the shareholders. Examples could be generous compensation contracts as found by (Zajac Westphal 1999) or decreased profitability of the firm, i.e. management, chasing for greater bonuses will strive to maximize sales instead of profits, consequently decreasing the shareholder value.

However the following research suggests that these consequences of cloning could be avoided. We have confirmed that the effects of cloning are less observed in companies with high social corporate responsibility standards. Meaning that CEOs have less control relative to the board in appointment process and it becomes more difficult for them to clone themselves in companies with high social corporate responsibility and sustainability standards. This is due to the fact that companies which following best practices in corporate governance will ensure proper balance of forces influencing director appointment. Such companies will have stronger boards that will impose limits on the CEO power. Presence of a large shareholder also seems to disrupt the CEOs ability to clone. Apparently, large shareholders are successful in promoting their own representatives, which will mostly have other socio-demographic qualities than the CEO. Thus by decreasing the sociod-demographic similarity, large stockholders can limit CEO power and increase board independence. These findings are consistent with (Eisenhardt and Bourgeois 1988) who show that demographic similarity is the foundation for coalition formation. Powerful and large shareholders will try to enlarge their support in the board by advocating appointment of directors with similar demographic characteristics.

(30)

30

knowledge. Then the similarity between board and CEO would increase, because new directors are likely to have similar characteristics with the CEO, assuming that CEO himself possesses the required firm- and industry-specific knowledge. Under these conditions, we would expect active institutional monitoring to have positive effect on CEO cloning.

This study has documented a number of interesting relationships, which provide valuable information for investors, policy makers and businesses; and has important implications for the corporate governance in public companies. We have seen how CEO cloning can result in potentially harmful managerial practices such as abnormal salaries and bonuses, which destroy shareholder value. We have also shown which company characteristics can limit CEO power and limit such opportunistic behavior. Namely, CEO rotation, high corporate social responsibility and sustainability standards and presence of a large shareholder will negatively affect the cloning, thus limiting its harmful consequences.

(31)

31

BIBLIOGRAPHY

Alderfer C. P. 1986. The invisible director on corporate boards. Harward Business Review, 64: 38-52.

Almazan, A. Hartzell, J. C. and Starks, L. T. 2005. Active institutional shareholders and costs of monitoring: Evidence from executive compensation. Financial Management, 34(4): 5-34.

Baker, M. and Gompers, P. A. 2003. The Determinants of Board Structure at the Initial Public Offering. Journal of Law and Economics, 46(2): 569-98.

Banning, K. 2004. Corporate governance and the new chief executive: How institutionalized power affects the agency contract. Corporate Ownership and Control, 2(1): 73-85.

Baskett, G. D. 1973. Interview decisions as determined by competency and attitude similarity. Journal of Applied Psychology, 57: 343-345.

Beatty, R. P. and Zajac, E. J. 1994. Managerial incentives, monitoring, and risk bearing: A study of executive compensation, ownership, and board structure in initial public offerings. Administrative Science Quarterly, 39(2): 313-335.

Bebchuk L. A., Fried, J. M. and Walker, D. I. 2002. Managerial power and rent

extraction in the design of executive compensation. The University of Chicago Law Review, 69: 751-846.

Berle, A. and Means, G. 1932. The modern corporation and private property. New York: Mac-Millan.

Boone, C., Olffen, W. van, Witteloostuijn, A. van, Brabander, B. De. 2004. The genesis of top management team diversity: Selective turnover among top management teams in Dutch newspaper publishing 1970-94. Academy of Management Journal, 47(5): 633-656.

Boot, A. W. and Macey, A. J. R. 1999. Objectivity, proximity and adaptability in corporate governance. Working Papers (William Davidson Institute), University of Michigan Business School. Available at SSRN: http://ssrn.com/abstract=216390

Byrne, D., Clore, G. L. and Worchel, P. 1966. The effect of economic similarity-dissimilarity as determinants of attraction. Journal of Personality and Social Psychology, 4: 220-224.

Clarkson, M. E. 1995. A stakeholder framework for analyzing and evaluating corporate social performance. Academy of Management Review, 20(1): 92-117.

(32)

32

Cyert, R. M., Kang, S. and Kumar, P. 2002. Corporate governance, takeovers, and top-management compensation: Theory and evidence. Management Science, 48(4): 453-469.

Ees, H., van and Postma, T. J. B. M. 2004. Dutch boards and governance. International Studies of Management and Organization, 34(2): 90-112.

Eisenhardt, K. M. 1989. Agency theory: An assessment and review. Academy of Management Review, 14(1): 57-74.

Finkelstein, S. and D'aveni, R. A. 1994. CEO duality as a double-edged sword: How boards of directors balance entrenchment avoidance and unity of command. Academy of Management Journal, 37(5): 1079-1108.

Freeman, R. E. 1984. Strategic management: A stakeholder approach. Boston: Pitman Publishing.

Hallock, K. F. 1997. Reciprocally interlocking boards of directors and executive compensation. Journal of Financial and Quantitative Analysis, 32(3): 331-344.

Hermalin, B. E. and Weisbach, M. S. 1988A. The determinants of board composition. RAND Journal of Economics, 19(4): 589-606.

Hermalin, B. E. and Weisbach, M. S. 1998B. Endogenously chosen boards of directors and their monitoring of the CEO. American Economic Review, 88(1): 96-118.

Hermalin, B. E. and Weisbach, M. S. 2003. Boards of directors as an endogenously determined institution: A surevey of economic literature. Economic Policy Review (Federal Reserve Bank of New York), 9(1): 7-26

Hume, D. (1740). A Treatise of Human Nature (1967, edition). Oxford University Press, Oxford.

Hume, D. (EHU) (1777). An Enquiry concerning Human Understanding. Nidditch, P. N. (ed.), 3rd. ed.

Jensen, M. and Meckling, W. H. 1976. Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics, 3(4): 305-360.

Kahn, C. and Winton, A. 1998. Ownership structure, speculation, and shareholder intervention. Journal of Finance, 53(1): 99-129.

Kanter, R. M. 1977. Men and women of the corporation. New York: Basic Books.

Kaufman, A. and Englander, E. 2005. A team production model of corporate governance. Academy of Management Executive, 19(3): 9-22.

(33)

33

Lawrence, B. S. 1997. Opening the black box of organizational demography. Organizational Science, 8:1-22.

Lorsch, J. and MacIver, E. 1989. Pawns or potentates: The reality of America’s corporate boards. Harvard Business School Press, Boston.

Main, B. G., O'Reilly III, C. A. and Wade, J. 2002. The CEO, the board of directors and executive compensation: economic and psychological perspectives. Industrial and Corporate Change, 4(2): 293-332.

Mills, R. W. and Weinstein, B. 2000. Beyond shareholder value – reconciling the shareholder and stakeholder perspectives. Journal of General Management, 25(3): 79-93.

Montgomery, C. A. and Kaufman, R. 2003. The board's missing link. Harvard Business Review, 81(3): 86-93.

Pagan, A. R., 1986. Two stage and related estimators and their applications. Review of Economic Studies, 57: 517-538.

Pfeffer, J. 1997. New directions for organizational theory: Problems and prospects. New York: Oxford University Press.

Post, J. E., Preston, L. E. and Sachs, S. 2002. Managing the extended enterprise: The new stakeholder view. California Management Review, 45(1): 6-28.

Priem, R. L., Lyon, D. W., Dess, G. G. 1999. Inherent limitations of demographic proxies in top management team heterogeneity research. Journal of Management, 25(6): 935-953.

Roosenboom P. 2005. Bargaining on board structure at the initial public offering. Journal of Management Governance, 9(2): 171–198.

Ross, S. A. 1973. The economic theory of agency: The principal's problem. American Economic Review, 63(2): 134-139.

Schleifer, A. and Vishny, R. W. 1994. Large shareholders and corporate control. Journal of Political Economy, (94)3: 461-488.

Schneider, B. 1987. The people make the place. Personnel Psychology, 40: 437-453.

Shivdasani, A. and Yermack, D. 1999. CEO involvement in the selection of new board members: An empirical analysis.” Journal of Finance, 54: 1829-1854.

StatSoft, Inc. 2006. Electronic statistics textbook.

http://www.statsoft.com/textbook/stathome.html. Multiple regression:

(34)

34

Tosi Jr., Henry L. and Gomez-Mejia, L. R. 1989. The decoupling of CEO pay and performance: An agency theory perspective. Administrative Science Quarterly, 34(2): 169-189.

Tsui, A. S. and O’Reilly III, C. A. 1992. Being different: Relational demography and organizational attachment. Administrative Science Quarterly, 37: 549-579.

Tsui, A. S., Egan, T. D. and O'Reilly III, C. A. 1992. Being different: Relational

demography and organizational attachment. Administrative Science Quarterly, 37(4): 549-579.

Wade, J., O'Reilly III, C. A. and Chandratat, I. 1990. Golden parachutes. CEOs and the exercise of social influence. Administrative Science Quarterly, 35(4): 587-603

Westphal, J. D. and Zajac, E. J. 1995. Who shall govern? CEO/board power, demographic similarity, and new director selection. Administrative Science Quarterly, 40(1): 60-83.

(35)

35

APPENDIX 1 – MULTICOLLINEARITY

Table 1

Dependent Variable: R_TENURE Method: Panel Least Squares Date: 10/30/08 Time: 01:05 Sample: 2002 2006

Cross-sections included: 38

Total panel (balanced) observations: 190

Variable Coefficient Std. Error t-Statistic Prob.

DUALITY 8.347109 1.616561 5.163497 0.0000 DJSI_W 2.736370 1.706569 1.603434 0.1111 INST_MON -0.195513 0.083936 -2.329302 0.0213 L_STOCK 0.195805 1.609440 0.121660 0.9033 STAFF 5.31E-06 1.30E-05 0.408340 0.6836 B_SIZE 0.119405 0.200802 0.594640 0.5530 C -6.986480 3.068120 -2.277121 0.0243

Effects Specification

Cross-section fixed (dummy variables)

(36)

36 Table 2

Dependent Variable: DUALITY

Method: ML - Binary Probit (Quadratic hill climbing) Date: 10/29/08 Time: 16:30

Sample: 2002 2006

Included observations: 190

Convergence achieved after 7 iterations

Covariance matrix computed using second derivatives

Variable Coefficient Std. Error z-Statistic Prob.

R_TENURE 0.109701 0.021357 5.136501 0.0000 DJSI_W 0.707400 0.324430 2.180443 0.0292 INST_MON 0.005469 0.010518 0.519978 0.6031 L_STOCK 0.179041 0.325342 0.550316 0.5821 STAFF -3.65E-06 1.41E-06 -2.596782 0.0094 B_SIZE 0.076932 0.049401 1.557301 0.1194 C -0.256746 0.690435 -0.371862 0.7100

Mean dependent var 0.673684 S.D. dependent var 0.470103 S.E. of regression 0.411216 Akaike info criterion 1.066815 Sum squared resid 30.94501 Schwarz criterion 1.186443 Log likelihood -94.34747 Hannan-Quinn criter. 1.115275 Restr. log likelihood -119.9924 Avg. log likelihood -0.496566 LR statistic (6 df) 51.28980 McFadden R-squared 0.213721 Probability(LR stat) 2.59E-09

(37)

37 Table 3

Dependent Variable: DJSI_W

Sample: 2002 2006

R_TENURE -0.046972 0.019511 -2.407428 0.0778 DUALITY 0.679211 0.299862 2.265078 0.0235 INST_MON -0.015491 0.013630 -1.136544 0.2557 L_STOCK -0.610344 0.297496 -2.051606 0.0598 STAFF 2.71E-07 7.97E-07 0.339329 0.7344 B_SIZE 0.056531 0.036749 1.538306 0.1240 C -1.537584 0.557710 -2.756958 0.0058

Mean dependent var 0.178947 S.D. dependent var 0.384321 S.E. of regression 0.367475 Akaike info criterion 0.909320 Sum squared resid 24.71196 Schwarz criterion 1.028947 Log likelihood -79.38538 Hannan-Quinn criter. 0.957779 Restr. log likelihood -89.26078 Avg. log likelihood -0.417818 LR statistic (6 df) 19.75079 McFadden R-squared 0.110635 Probability(LR stat) 0.003067

(38)

38 Table 4

Dependent Variable: INST_MON Method: Panel Least Squares Date: 10/30/08 Time: 01:06 Sample: 2002 2006

R_TENURE -0.188236 0.080812 -2.329302 0.0213 DUALITY -0.633293 1.727886 -0.366513 0.7145 DJSI_W 0.905163 1.687894 0.536268 0.5926 L_STOCK 3.919398 1.544660 2.537386 0.0122 STAFF -2.61E-08 1.28E-05 -0.002043 0.9984 B_SIZE 0.144125 0.196904 0.731959 0.4654 C 4.679194 3.039700 1.539360 0.1259

Effects Specification

Cross-section fixed (dummy variables) Period fixed (dummy variables)

(39)

39 Table 5

Dependent Variable: L_STOCK

Sample: 2002 2006

R_TENURE -0.039883 0.017309 -2.304182 0.7017 DUALITY 0.269485 0.287682 0.936745 0.3489 DJSI_W -0.628778 0.289427 -2.172492 0.5144 INST_MON 0.051901 0.016017 3.240450 0.0012 STAFF 1.47E-06 8.15E-07 1.799614 0.9344 B_SIZE -0.083235 0.037834 -2.200013 0.1484 C 1.438698 0.494388 2.910058 0.0036

Mean dependent var 0.831579 S.D. dependent var 0.375229 S.E. of regression 0.336929 Akaike info criterion 0.799575 Sum squared resid 20.77442 Schwarz criterion 0.919202 Log likelihood -68.95964 Hannan-Quinn criter. 0.848034 Restr. log likelihood -86.14101 Avg. log likelihood -0.362945 LR statistic (6 df) 34.36274 McFadden R-squared 0.199456 Probability(LR stat) 5.72E-06

(40)

40

APPENDIX 2 – SERIAL CORRELATION

Table 1

Dependent Variable: LOG(CLON) Method: Panel Least Squares Date: 10/27/08 Time: 20:03 Sample: 2002 2006

Variable Coefficient Std. Error t-Statistic Prob. C -0.484789 0.042274 -1.146781 0.0000 R_TENURE 0.006850 0.001339 5.117396 0.0000 DUALITY -0.047714 0.035865 -1.330362 0.1855 DJSI_W -0.038309 0.014171 -2.703278 0.0077 INST_MON -0.000547 0.000810 -0.675217 0.5006 L_STOCK -0.036132 0.041309 -0.874667 0.3832 STAFF -9.69E-08 1.39E-07 -0.698728 0.4859 B_SIZE -0.035193 0.001103 -3.191757 0.0000 Effects Specification

(41)

41 Table 2

Dependent Variable: LOG(CLON) Method: Panel Least Squares Date: 10/27/08 Time: 20:00 Sample (adjusted): 2004 2006 Cross-sections included: 38

Variable Coefficient Std. Error t-Statistic Prob. C -0.921880 0.118121 -7.804550 0.0000 R_TENURE 0.009372 0.002176 4.306947 0.0001 DUALITY 0.001708 0.022746 0.075075 0.9404 DJSI_W -0.055383 0.017720 -3.125494 0.0027 INST_MON 0.002476 0.001273 1.944843 0.0561 L_STOCK -0.062251 0.013085 -4.757508 0.0000 STAFF 3.08E-08 4.81E-08 0.641734 0.5233 B_SIZE -0.030385 0.001506 -20.18050 0.0000 LOG(CLON(-1)) -0.253341 0.132016 -1.919023 0.0594 LOG(CLON(-2)) -0.100273 0.054217 -1.849458 0.0689 Effects Specification

R-squared 0.760356 Mean dependent var -0.975835 Adjusted R-squared 0.583389 S.D. dependent var 0.163003 S.E. of regression 0.105211 Akaike info criterion -1.367857 Sum squared resid 0.719510 Schwarz criterion -0.191772 Log likelihood 126.9679 F-statistic 4.296585 Durbin-Watson stat 2.771270 Prob(F-statistic) 0.000000

Table 3

Dependent Variable: RESID02 Method: Panel Least Squares Date: 10/27/08 Time: 20:17 Sample (adjusted): 2005 2006 Cross-sections included: 38

RESID02(-1) -0.425047 0.097091 -4.377828 0.0000

(42)

42 Table 4

Wald Test:

Equation: SERIAL_CORR2

Test Statistic Value df Probability

F-statistic 0.595957 (1, 75) 0.4426 Chi-square 0.595957 1 0.4401

Null Hypothesis Summary:

Normalized Restriction (= 0) Value Std. Err.

0.5 + C(1) 0.074953 0.097091

(43)

43

APPENDIX 3 – ENDOGENEITY

Table 1

Dependent Variable: R_TENURE Method: Panel Least Squares Date: 10/30/08 Time: 01:00 Sample (adjusted): 2004 2006 Cross-sections included: 38

Variable Coefficient Std. Error t-Statistic Prob. DUALITY 6.842626 1.660045 4.121952 0.0001 DJSI_W -0.067092 1.545010 -0.043425 0.9655 INST_MON 0.106416 0.090250 1.179118 0.2427 L_STOCK -0.946480 1.688335 -0.560600 0.5770 STAFF 8.77E-06 1.43E-05 0.614251 0.5412 B_SIZE 0.258961 0.244977 1.057084 0.2944 LOG(CLON(-1)) -0.836337 2.659047 -0.314525 0.7541 LOG(CLON(-2)) 1.354107 2.745268 0.493251 0.6235 ROA 5.135262 8.781512 0.584781 0.5608 CEO_A 0.549231 0.100568 5.461270 0.0000 C -39.48657 7.524706 -5.247590 0.0000 Effects Specification

(44)

44 Table 2

Dependent Variable: DUALITY

Sample (adjusted): 2004 2006

Included observations: 114 after adjustments Convergence achieved after 9 iterations

Variable Coefficient Std. Error z-Statistic Prob. R_TENURE 0.270341 0.065115 4.151741 0.0000 DJSI_W 0.481545 0.532252 0.904732 0.3656 INST_MON -0.006598 0.015545 -0.424405 0.6713 L_STOCK -0.076488 0.553658 -0.138150 0.8901 STAFF -6.13E-06 2.31E-06 -2.653979 0.0080 B_SIZE 0.004872 0.077920 0.062524 0.9501 LOG(CLON(-1)) -0.794200 0.955019 -0.831607 0.4056 LOG(CLON(-2)) -0.527652 0.971641 -0.543052 0.5871 ROA 12.61234 3.931747 3.207821 0.0013 CEO_A 0.050067 0.035276 1.419294 0.1558 C -2.939300 2.511845 -1.170176 0.2419 Mean dependent var 0.631579 S.D. dependent var 0.484506 S.E. of regression 0.363759 Akaike info criterion 0.921534 Sum squared resid 13.62900 Schwarz criterion 1.185554 Log likelihood -41.52746 Hannan-Quinn criter. 1.028685 Restr. log likelihood -75.02454 Avg. log likelihood -0.364276 LR statistic (10 df) 66.99416 McFadden R-squared 0.446482 Probability(LR stat) 1.68E-10

(45)

45 Table 3

Dependent Variable: DJSI_W

Variable Coefficient Std. Error z-Statistic Prob. R_TENURE -0.047388 0.032477 -1.459145 0.1445 DUALITY 0.309608 0.397625 0.778644 0.4362 INST_MON -0.027177 0.019493 -1.394225 0.1632 L_STOCK -0.184447 0.451329 -0.408675 0.6828 STAFF -8.26E-08 1.17E-06 -0.070742 0.9436 B_SIZE 0.051230 0.062695 0.817139 0.4138 LOG(CLON(-1)) -1.242338 0.935380 -1.328163 0.1841 LOG(CLON(-2)) -1.362554 0.904873 -1.505796 0.1321 ROA 0.301802 2.607445 0.115746 0.9079 C -3.979700 1.342281 -2.964877 0.0030 Mean dependent var 0.192982 S.D. dependent var 0.396382 S.E. of regression 0.377625 Akaike info criterion 1.011017 Sum squared resid 14.83044 Schwarz criterion 1.251034 Log likelihood -47.62797 Hannan-Quinn criter. 1.108427 Restr. log likelihood -55.91914 Avg. log likelihood -0.417789 LR statistic (9 df) 16.58234 McFadden R-squared 0.148271 Probability(LR stat) 0.055673

(46)

46 Table 4

Dependent Variable: INST_MON Method: Panel Least Squares Date: 10/30/08 Time: 01:50 Sample (adjusted): 2004 2006 Cross-sections included: 38

Variable Coefficient Std. Error t-Statistic Prob. R_TENURE 0.156022 0.189660 0.822640 0.4126 DUALITY -0.533557 2.409984 -0.221394 0.8252 DJSI_W -3.741326 2.493104 -1.500670 0.1365 L_STOCK 9.263544 2.844289 3.256893 0.0015 STAFF -1.04E-05 4.76E-06 -2.183697 0.0313 B_SIZE -0.177704 0.441875 -0.402159 0.6884 LOG(CLON(-1)) -4.785467 5.948657 -0.804462 0.4230 LOG(CLON(-2)) -11.26882 5.781791 -1.949020 0.0540 ROA -3.059212 15.95637 -0.191724 0.8483 C -9.466989 8.794222 -1.076501 0.2842 Effects Specification

Period fixed (dummy variables)

(47)

47 Table 5

Dependent Variable: L_STOCK

Variable Coefficient Std. Error z-Statistic Prob. R_TENURE -0.078202 0.041989 -1.862434 0.0625 DUALITY 0.200286 0.562402 0.356127 0.7217 DJSI_W -0.728842 0.586930 -1.241787 0.2143 INST_MON 0.255633 0.090889 2.812592 0.0049 STAFF 8.32E-06 7.05E-06 1.179252 0.2383 B_SIZE -0.158269 0.079026 -2.002731 0.0452 LOG(CLON(-1)) 3.249503 1.917775 1.694413 0.0902 LOG(CLON(-2)) -1.265203 1.832011 -0.690609 0.4898 ROA -5.040232 3.282411 -1.535528 0.1247 C 3.572840 1.848326 1.933013 0.0532 Mean dependent var 0.850877 S.D. dependent var 0.357782 S.E. of regression 0.289254 Akaike info criterion 0.638874 Sum squared resid 8.701477 Schwarz criterion 0.878892 Log likelihood -26.41585 Hannan-Quinn criter. 0.736284 Restr. log likelihood -48.01503 Avg. log likelihood -0.231718 LR statistic (9 df) 43.19837 McFadden R-squared 0.449842 Probability(LR stat) 1.98E-06

(48)

48 Table 6

Variable Coefficient Std. Error t-Statistic Prob. R_TENURE 0.008297 0.009063 0.915489 0.3634 DUALITY 0.007602 0.078971 0.096260 0.9236 DJSI_W -0.060146 0.065050 -0.924605 0.3586 INST_MON 0.002531 0.003196 0.791766 0.4314 L_STOCK -0.064848 0.062157 -1.043292 0.3007 STAFF 4.92E-08 5.14E-07 0.095746 0.9240 B_SIZE -0.030570 0.008497 -3.597737 0.0006 LOG(CLON(-1)) -0.252632 0.090928 -2.778374 0.0072 LOG(CLON(-2)) -0.102419 0.096608 -1.060158 0.2931 RESID_TENURE 0.001354 0.010480 0.129170 0.8976 C -0.925823 0.170554 -5.428325 0.0000 Effects Specification

(49)

49 Table 7

Variable Coefficient Std. Error t-Statistic Prob. R_TENURE 0.013675 0.005122 2.669641 0.0096 DUALITY -0.124401 0.125571 -0.990683 0.3256 DJSI_W -0.051139 0.053161 -0.961966 0.3397 INST_MON 0.002381 0.003137 0.759161 0.4505 L_STOCK -0.064212 0.058232 -1.102689 0.2743 STAFF -4.87E-08 4.93E-07 -0.098784 0.9216 B_SIZE -0.033678 0.008757 -3.845911 0.0003 LOG(CLON(-1)) -0.281780 0.093078 -3.027346 0.0036 LOG(CLON(-2)) -0.135515 0.098917 -1.369995 0.1755 RESID_DUAL 0.129171 0.110789 1.165919 0.2480 C -0.850581 0.176969 -4.806372 0.0000 Effects Specification

(50)

50 Table 8

Variable Coefficient Std. Error t-Statistic Prob. R_TENURE -0.008386 0.007326 -1.144773 0.2566 DUALITY 0.153356 0.082417 1.860724 0.0674 DJSI_W -1.462147 0.516681 -2.829883 0.0062 INST_MON -0.006806 0.004528 -1.503166 0.1377 L_STOCK -0.173606 0.068952 -2.517771 0.0143 STAFF -1.09E-07 4.70E-07 -0.232778 0.8167 B_SIZE -0.007274 0.011583 -0.627939 0.5323 LOG(CLON(-1)) -0.756048 0.202828 -3.727534 0.0004 LOG(CLON(-2)) -0.675940 0.228875 -2.953311 0.0044 RESID_DJSI 1.402627 0.512674 2.735904 0.0080 C -1.890325 0.387962 -4.872450 0.0000 Effects Specification

(51)

51 Table 9

Variable Coefficient Std. Error t-Statistic Prob. R_TENURE 0.002503 0.015848 0.157942 0.8750 DUALITY 0.028298 0.087830 0.322197 0.7484 DJSI_W 0.111997 0.379971 0.294752 0.7691 INST_MON 0.046821 0.099714 0.469552 0.6403 L_STOCK -0.472405 0.923683 -0.511436 0.6108 STAFF 4.69E-07 1.10E-06 0.425883 0.6716 B_SIZE -0.023310 0.017964 -1.297588 0.1991 LOG(CLON(-1)) -0.037435 0.493636 -0.075836 0.9398 LOG(CLON(-2)) 0.394446 1.115925 0.353470 0.7249 RESID_INSMON -0.044279 0.099516 -0.444943 0.6579 C -0.486097 0.993645 -0.489205 0.6264 Effects Specification