University of Amsterdam
MSc Finance, specialization: Asset Management
Master Thesis
Title:
The mitigating effect of the board of directors on the relationship between CEO
overconfidence and acquisitiveness
Name: Rick Zwarthoed
Date: 14-08-2017
2 Statement of Originality
This document is written by Student Rick Zwarthoed 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.
Abstract:
In this thesis the mitigating effect of the board of directors on the relationship between CEO overconfidence and acquisitiveness is researched. Overconfident CEOs overestimate their ability to generate value from potential takeover targets. Thus, on average overconfident CEOs make more value-destroying acquisitions. A strong board of directors can act as a counterbalance to an overconfident CEO. These predictions are analysed by studying the different effects in separate regressions. Firstly the effect of CEO overconfidence on acquisitiveness is researched. Secondly the mitigating effect of board of directors on this relationship is studied. This is done by investigating a sample of 395 of the largest firms based on revenue in the U.S. in the period 2007 – 2016. CEOs are indicated as overconfident when in spite of their under diversification, hold stock options of their own firm when they are deep in-the-money. The mitigating effect of the board is researched by creating a Board-Index based on board characteristics that are expected to increase firm performance. This study finds weak evidence for the positive relationship between CEO
overconfidence and acquisitiveness. Moreover, this study also finds weak evidence for a mitigating effect of a strong board on this positive relationship between CEO overconfidence and
acquisitiveness. Overconfident CEOs under weak boards are found to positively impact
acquisitiveness and overconfident CEOs under strong board are found to not significantly affect acquisitiveness.
3
Table of content
1 Introduction ... 4
2 Literature review ... 5
3 Methodology ... 10
3.1 Measuring CEO overconfidence ... 11
3.2 Measuring Board – Index ... 12
3.3 Empirical Specification ... 14
3.3.1 Regression 1 ... 14
3.3.2 Regression 2 ... 16
4 Data and descriptive statistics ... 18
5 Results ... 23
5.1 Results of regression 1 ... 23
5.1.1 Longholder overconfidence measures ... 23
5.1.2 Overconfident-Year measures ... 24
5.2 Results of regression 2 ... 25
5.2.1 Longholder overconfidence measures ... 25
5.2.2 Overconfident-Year measures ... 26
6 Robustness checks ... 28
6.1 Overconfidence measures ... 28
6.2 Alternative explanations ... 29
6.3 Endogeneity ... 29
6.4 Firm fixed effects ... 30
7 Conclusion and discussion ... 30
Reference list ... 35
4
1
Introduction
Overconfidence of individuals and among executives is a character trait that has been getting an increasingly amount of attention in the recent years. The first people to create a solid measure for overconfidence among CEOs were Malmendier & Tate (2005a). This measure has since been used in many recent studies on CEO attitude and how it negatively affects certain corporate policies. One of those policies being acquisitiveness, which is also studied by Malmendier & Tate (2008) in a later research. In their revised version of their 2005 paper (Malmendier & Tate, 2005b) they stress the importance of the board of directors in dealing with this CEO overconfidence. The board of directors plays an important role in the possible mitigation of the negative effect of CEO overconfidence on several corporate policies. Although the literature on CEO overconfidence is steadily growing, the literature on the possible counterbalancing effect of the board of directors remains hardly studied. This relatively new area of research provides a lot of ground for interesting studies. In this study the possible mitigating effect of a strong board of directors is researched. The affected corporate policy that is studied is acquisitiveness. The main central question in this thesis is as follows: Does supervision of a strong board mitigate the effect of CEO
overconfidence on acquisitiveness?
To answer this question, this study provides an unique index-measure for board strength, derived from prior research on board composition and board characteristics and their relationship to managerial power and firm performance. This board-index measure divides a sample of 395 of the largest U.S. firms (based on yearly revenues) into ones having weak boards and ones having strong boards. Furthermore, this study questions the
assumption made by many studies in this field. This assumption is that overconfidence is persistent through the tenure of the CEO. In this thesis an additional overconfident measure is created that allows it to deviate from year to year. This gives some additional insights in the existing field of CEO overconfidence. Finding this mitigating effect of a strong board makes way for other new studies in the field of CEO overconfidence and its effects on different policies. It would also stress the importance of the board of directors of firms and encourage companies to alter their board compositions. It could also change regulations when it comes to the composition and structure of the board.
This thesis is organized as follows. In section 2 a literature review of previous studies is given. These studies research related topics such as individual overconfidence, different
5
effects of CEO overconfidence on corporate policies, board composition and managerial power. Section 3 the methodology of this study is presented. This section discusses what measures are used and how they are composited. Furthermore it presents the empirical model that is used to do the research. In section 4 the data and descriptive statistics are presented. This section provides summary statistics of different variables and measures and gives insight into how the data for this study is collected. In section 5 the main results of the regressions are presented. This section discusses the implications of the findings and gives economic meaning to them. Section 6 provides robustness checks and critically assesses the findings and potential problems. Finally, section 7 concludes and discusses the study and provides directions for further research.
2
Literature review
A CEO is probably the most important person within a firm. But every person has its own characteristics, vision, style and other traits, which affect the decisions they make. A study by Graham, Li and Qui (2012) finds evidence that managerial characteristics affects policies, actions and outcomes of a company. A characteristic trait that has been receiving much attention in recent studies is overconfidence.
According to the psychology literature, individuals display overconfidence for three reasons. Firstly, Individuals show overconfident behaviour when it’s about the outcome of a project they are highly committed to (Weinstein, 1980). Secondly, when the reference point is abstract, overconfidence is likely to be the strongest. (Alicke et al., 1995). Thirdly, when individuals believe that they are in control, they are more likely to be overconfident about the outcome (Weinstein, 1980). They basically overestimate their ability to influence outcomes. When it comes to overestimating future outcomes, some studies refer to this as “overoptimistic” instead of “overconfident”, but when following literature on self-serving attribution the label “overconfidence” can be chosen as they are rather believed to overestimate their own ability to influence outcomes, than they are generally overestimating exogenous outcomes.
In the study of Ben-David, Graham & Harvey (2013) overconfidence among executives is researched. They find that executives are extremely miscalibrated; the distributions of the forecasts they make are too narrow. They state that miscalibration is a
6
form of overconfidence and that this is persistent among executives. They also find that when executives are miscalibrated when it comes to forecasting the stock market, they also show miscalibration when it comes to forecasting the value of their own firm’s projects. Lastly, they find that miscalibrated executives follow more aggressive policies, invest more and use more debt financing. This paper finds clear evidence of overconfidence among executives and also displays some of the negative downsides of this executive
overconfidence.
Jonsson & Allwood (2003) argue that there is individual stability over time when it comes to overconfidence. This means that if an individual is found to be overconfident, this overconfidence is persistent through time. Ben-David, Graham & Harvey (2007) also find that overconfident is a persistent characteristic among executives and state that this falls in line with the theory of individual stability over time. Gervais & Odean (2001) however state that overconfidence is dynamic on both an individual level and in aggregate and that overconfidence can change with success and failure. They find that investors are most overconfident early in their careers. With experience and self-assessment they become more realistic and less overconfident. If this theory holds up for executives as well, it would counter the theory on individual stability of overconfidence.
Other papers dig deeper into the negative aspects of overconfidence and focus solely on the prime decision maker of the firm, the CEO. Malmendier & Tate (2005a) found a way to measure this overconfidence by observing the timing of stock options exercises. If CEOs fail to exercise their stock options, even though they are deep-in-the-money and their portfolio is under diversified, they are said to be overconfident. These CEOs believe their stocks are undervalued and overestimate their ability to increase the price of the stock. In their study Malmendier & Tate (2005a) also argue that overconfident CEOs systematically overestimate the return of their investment projects. They state that overconfident CEOs are reluctant to issue new equity, because they believe their own company’s stock to be
undervalued. This leads to overconfident CEOs underinvesting when there is a shortage of internal cash flows and overinvesting when there is an abundance of internal cash flows available. This means that overconfident CEOs are always investing under or above the optimal investment-level.
A paper by Galasso & Simcoe (2011) researches the effect of CEO overconfidence on innovation. They also use the stock-option exercise measures proposed by Malmendier &
7
Tate to measure overconfidence. Innovation is measured as the amount of citation-weighted patents. They find a robust positive relationship between overconfidence and innovation, where the effect is largest in competitive industries. Their results suggest that overconfident CEOs are more likely to innovate and take their firms into new technological directions. Malmendier & Tate (2008) also study the effect of CEO overconfidence on
acquisitiveness. They argue that CEOs not only overestimate their ability to generate returns, but also overestimate their ability to generate value from potential takeover targets. So on average, overconfident CEOs undertake mergers that are value-destroying. They test this idea using their previously mentioned measures and find that CEOs that are indicated as overconfident are more likely to conduct mergers. The effect is found to be largest in companies with large amount of cash and untapped debt capacity. Lastly, they find that the market reacts in a negative way to takeover bids. This effect Is also found to be stronger for overconfident CEOs.
These outcomes of studies researching the effect of CEO overconfidence on different policies, suggest that overconfidence is a bad characteristic for a CEO. The idea that firms should avoid hiring an overconfident CEO is underlined by the study of Campbell et al. (2011). They argue that it is hard to determine whether a CEO is overconfident before the CEO is hired. In their study they find that once a CEO is hired and is found to be
overconfident, he is more likely to be terminated than a CEO that doesn’t show this
behaviour. This implies that companies and shareholders, don’t like it when a CEO exhibits overconfident behaviour. This result however only holds up under good corporate
governance.
This importance of corporate governance and then in particular the board of directors to counterbalance an overconfident CEO is also stated by Malmendier & Tate (2005b). They state that if overconfident CEOs required outside approval from a strong board, the overconfidence could be counterbalanced. They argue that this board should be active and be aware of the firm’s opportunities. It should also encourage the CEO to
undertake projects that create value for the firm. The board members should be directors who are not afraid to speak up in the boardroom and have a strong opinion. Malmendier & Tate (2005b) however conclude that there is not yet a reliable measure of high-quality boards and that this would make an interesting topic for further research.
8
board is determined by easy-observable board characteristics that are proven to have a positive effect on firm performance. It is assumed that if a certain characteristics is found to have a positive effect on firm performance, it will also contribute to a stronger board of directors.
One of the studies that finds a significant relationship between board characteristics and corporate performance is done by Hermalin, Benjamin & Weisbach (2003). They find that board size has a negative effect on corporate performance. They put forth the idea that when boards become too large, agency problems within the board, such as director free-riding, increase. This makes the board more symbolic and less part of the management process. Yermack (1996) empirically tests this idea and finds evidence for it. In his paper the relationship between board size and Tobin’s Q is examined. His sample consists of U.S. firms and he controls for variables that are likely to affect the Tobin’s Q. The results of that
research suggest that there exists a significant negative relationship between board size and Tobin’s Q.
Another characteristic that is found to have a significant relationship with firm performance is board diversity. Carter, Simkins & Simpson (2003) examine the relationship between board diversity and firm value for Fortune 1000 firms. They define board diversity as the percentage of African Americans, Asians, Hispanics and women on the board of directors. They state the idea that board diversity stimulates innovation and creativity. Secondly they put forth the idea that diversity leads to more effective problem-solving. Carter, Simkens & Simpson argue that diversity may initially produce more conflicts, but because of the variety of perspectives that emerge, decision makers have to evaluate other alternatives and be more careful in exploring the consequences of the alternatives. Finally, diversity on the board could better the global relationships of the company, through better understanding of other cultures among the executives. After controlling for other corporate governance measures, size and industries, Carter, Simkens & Simpson find a significant positive relationship between the fraction of women or minorities on the board and firm value.
A study by Pearce II & Zahra (1991) states that power of a CEO over the board is a potential threat to a strong board. There are couple of board characteristics that have found to impact the managerial power of a CEO. Board characteristics that are proven to decrease the managerial power of a CEO, are assumed to also contribute to a stronger board of
9
directors. Bebchuk, Fried & Walker (2002) argue that managerial power increases when a CEO is also the chairman of the board. Carter, Simkins & Simpson (2003) examine this CEO/Chairman duality and research its effect on firm value. They find a significant negative relationship between the duality and firm value. A CEO also filling the role as chairman of the board, increases managerial power and is expected to decrease firm value.
Whether the CEO is the only insider on the board, can also affect managerial power. Adams, Almeida & Ferreira (2005) state that if an inside director besides the CEO is on the board, he is more likely to participate in the important decision-making together with the CEO. CEOs that are not the only insiders on the board thus have less influence and power over the board. This decreases the managerial power and increases the power of the board. Managerial power can also be increased by the fact that some individual directors on the board are either too old or too busy to be an active an alert member of the board. Core, Holthausen & Larcker (1999) state the idea that directors become less effective when they serve on too many boards and as they grow older. Many companies already require a mandatory retirement for directors at the age of 70. They also suggest that many directors don’t attend to their duties adequately, because they serve on too many boards. According to NACD (1996) guidelines a director should retire when he reaches the age of 70 and a director is too busy when he is on 3 or more boards. So when directors are too old or attending too many boards, they lose their focus, which increases the power of the CEO. The characteristics discussed above, all have a significant effect on firm performance and/or managerial power. In this study these characteristics are pooled together into an index. This index will represent a measure of board strength. Characteristics that have a positive effect on firm performance and/or a negative on managerial power are assumed to have a positive effect on board strength. Characteristics that have a negative effect on firm performance and/or a positive effect on managerial power are assumed to have a negative effect on board strength. The idea put forth by Malmendier and Tate (2005b) is that the board of directors can act as a counterbalance to overconfident behaviour of CEOs. Where overconfident CEOs are more likely to overinvest, innovate and make acquisitions, a strong board could have the ability to prevent this behaviour from happening.
In this study one of those outcomes of CEO overconfidence is researched. The effect of CEO overconfidence on acquisitiveness is investigated and the index for board strength is used to see whether a strong board has the ability to mitigate this effect of CEO
10
overconfidence on acquisitiveness. This gives a research with two different regressions. Where one regression is done to research the effect of CEO overconfidence on
acquisitiveness, and the other to research whether strong board supervision mitigates this effect of overconfidence on acquisitiveness. This would imply dividing the sample of overconfident CEOs into two categories. One being overconfident CEOs under weak board supervision and the other being overconfident CEOs under strong board supervision. From this idea two hypotheses are derived:
Hypothesis 1: CEO overconfidence has a significantly positive effect on
acquisitiveness.
Hypothesis 2: Supervision of a strong board mitigates the effect of CEO
overconfidence on acquisitiveness.
In addition to these two hypotheses, two predictions based on the existing literature are made.
Prediction 1: The outcome of the first regression is expected to follow the result of
the study of Malmendier & Tate (2008). They find a significantly positive effect of CEO overconfidence on acquisitiveness. This effect is also expected to be found in the first regression.
Prediction 2: The outcome of the second regression is expected to follow the idea
that a strong board can counterbalance an overconfident CEO. This means that strong board supervision mitigates the effect of CEO overconfidence on
acquisitiveness. The effect of overconfident CEOs under strong board supervision on acquisitiveness is thus expected to be found not to be significantly different from zero. The effect of overconfident CEOs under weak board supervision on
acquisitiveness is then logically expected to be strongly positive.
3
Methodology
From the data and descriptive statistics section arises a panel dataset of 2771 firm-year observation, 629 distinct CEOs and 352 acquisitions. Accounting for more than 1 acquisitions in a particular firm-year, leaves 264 firm-years where an acquisition has been made. In this
11
section it is discusses how CEO overconfidence is measured and what board characteristics the Board-Index is made of. Furthermore, the empirical specification for both regression are discussed, hypotheses are drawn and expectation based on previous literature are given.
3.1 Measuring CEO overconfidence
Previous literature of Hall and Murphy (2002) finds that under diversified, risk-averse executives should not hold their option until the expiration date. Contracts that CEOs have, usually contain large quantities of options and stocks instead of just a compensation in cash. CEOs are prohibited to perfectly hedge the risk of those shares by selling short stocks or the company they lead. This is to maximize the incentive effect of the contract. The contract of the CEO also limit their ability to sell their stocks and exercise their option (Hall & Murphy, 2002). This leads to CEOs having too much of their own companies stock in their portfolio and thus being under diversified.
In the paper of Malmendier & Tate (2008) a CEO is indicated as overconfident if he ever holds an option until the expiration date during his tenure as a CEO. But because the detailed information on the timing of option exercises they used to construct the confidence measure is not available, the paper of Campbell et al. (2011) is roughly followed instead. In this thesis CEOs are indicated as overconfident if they hold stock options that are more than 100% in the money. Option moneyness is computed as follows: The realizable value per option is computed as the total realizable value of the exercisable options (ExecuComp variable OPT_UNEX_EXER_EST_VAL) divided by the number of exercisable options
(OPT_UNEX_EXER_NUM). Then the per-option realizable value is subtracted from the stock price at the fiscal year end (PRCC_F) to obtain an estimate of the average exercise price of the options. The average percent moneyness of the options equals the per-option realizable value divided by the estimated average exercise price. Note that because it is needed to identify CEOs who chose to hold options that could have been exercised, the variables from ExecuComp (OPT_UNEX_EXER_EST_VAL and OPT_UNEX_EXER_NUM) that only include exercisable options are used. Once a CEO is indicated as overconfident based on the above measure, he is indicated as overconfident for all his CEO-years in the dataset. The variable “Longholder100” is then 1 for all the firm-years of that CEO. An extra measure is the
“Longholder200” overconfidence measure. In this case, a CEO is indicated as overconfident if the average percent moneyness of his options is higher than 200%. Only the “extreme” overconfident CEOs are now indicated as overconfident.
12
For both the 100% and 200% cut-off another additional measure for overconfidence is created. This measure Is not derived from other papers, but is drawn from the idea that overconfidence is dynamic (Gervais & Odean, 2001). The variable “Overconfident-Year-100” and “Overconfident-Year-200” are created. They are both measured the same way as Longholder100 and Longholder200. The only difference is that they only equal 1 in the years that the average percent moneyness of a CEOs options is higher than 100 and 200. This creates a situation that allows the overconfidence of a CEO to fluctuate from year to year.
3.2 Measuring Board – Index
To capture board strength a Board-Index is created. This index consists of board
characteristics that were shown to have a positive impact on firm performance. In this thesis it is assumed that when a board characteristic has a positive effect on firm performance, it will constitute a stronger board. In total 7 different variables are put into the Board-Index, resulting in an index numbering 0 to 6. Firms scoring an index from 0 – 3 are indicated as having a weak board and firms scoring an index from 4 – 6 are indicated as having a strong board. A variable called “Strong_Board” is created, which equals 1 if a firm has a Board-Index from 4 - 6 in a particular year and 0 if the score is from 0 – 3. The Board-Index consists of the following 6 points:
1. Board size (Number of directors on the board). Board size has a negative effect on corporate performance (Hermalin, Benjamin & Weisbach, 2003) and can thus be seen an indicator of board strength. More directors on the board is expected to decrease board strength. This negative relationship is put into a binary variable following different papers on the topic of overconfidence such as Malmendier & Tate (2008) and Malmendier & Tate (2005a) that use it as a measure of corporate governance. This dummy variable is 1 if number of directors on the board is between 4 and 12 and 0 otherwise. Since there are no boards that have less than 4 directors on the board in the sample, this is a good measure to capture the negative relationship between board size and firm performance.
2. Board diversity (Percentage of ethnicities on the board of directors besides Caucasian). Board diversity has a positive impact on firm value Carter, Simkins & Simpson (2003) and can thus be seen as an indicator of a board strength. More
13
diversity within the board is expected to increase board strength. To capture this board diversity, a binary variable “Director_Ethnicity” is created. This variable is 0 if the ethnicity of a director is Caucasian and 1 if it is otherwise. The amount of
directors with a different ethnicity is divided by the total amount of directors on the board, so that there is not too much correlation with the board size variable. The Board-Index variable “Board Diversity” is 1 if the percentage is higher than 10% and 0 if it is equal to or lower than 10%. This cut-off percentage of 10% is based on 10% being the median of the variable.
3. Number of females (Percentage of women on the board of directors). The paper of Carter, Simkins & Simpson (2003) also finds that more women on the board of directors has a positive impact on firm value through gender diversity benefits. More women on the board also increases board diversity and is therefore expected to increase board strength. The binary variable “Female” is created and is either 1 if a director is female or 0 if a director is male. The amount of females on the board is divided by the total amount of directors on the board, so that there is not too much correlation with the board size variable. Board-Index variable Female is 1 if the percentage is higher than 15% and 0 if it is equal to or lower than 15%. This cut-off percentage of 15% is based on 15% being the median of the variable.
4. CEO is chairman (CEO is chairman of the board of directors). Bebchuk, Fried & Walker (2002) state that the CEO-Chairman duality have a negative effect on the strength of the board. A binary variable called “CEO_Chairman” is created. If the annual title of the CEO says that he doesn’t hold the title of chairman of the board, the variable is indicated as 1. If the CEO also is the chairman, the variable is 0. If a CEO is not the chairman of the board, this is expected to have a positive impact on board strength. 5. CEO is only Insider (CEO only insider on the board). Adams, Almeida & Ferreira (2005)
state that if an inside director besides the CEO is on the board, he is more likely to participate in the important decision-making together with the CEO. CEOs that are not the only insiders on the board thus have less influence and power over the board. This decreases the managerial power and increases the power of the board and thus the board strength. If the CEO served on the board as a director, the
variable “EXECDIR” is indicated as 1. The variable ONE_INSIDER is created and equals 1 If on the board only one director has the “CLASSIFICATION” of an insider (E) and 0
14
otherwise. After this the variable CEO_INSIDER is created. This variable equals 0 if both EXECDIR and ONE_INSIDER are 1. If this is not the case than the variable equals 1.
6. Old or Busy Directors (Percentage of directors that are too old or busy). Core,
Holthausen & Larcker (1999) state the idea that directors become less effective when they serve on too many boards and as they grow older. Many companies already require a mandatory retirement for directors at the age of 70. They also suggest that many directors don’t attend to their duties adequately, because they serve on too many boards. According to NACD (1996) guidelines a director should retire when he reaches the age of 70 and a director is too busy when he serves on 3 or more boards. The variable “TOO_OLD” is created which equals 1 if a director is 70 years or older and 0 otherwise. Afterwards the variable “TOO_BUSY” is created which equals 1 if a director serves on 3 or more boards. Then the variable “OLD_OR_BUSY” is created which equals 1 if either one or both of the “TOO_OLD” and “TOO_BUSY” variables equal 1 and 0 otherwise. The total number of too busy and/or old directors are divided by the total amount of directors, so that there is not too much correlation with the board size variable. Lastly the variable “NOT_OLD_BUSY” is created, which equals one if the percentage of directors that are too old or busy is below the median of 30% and 0 otherwise.
3.3 Empirical Specification
In this thesis two regressions are performed. The first regression focusses on the relationship of CEO overconfidence on acquisitiveness. In the second regression the Board-Index is
added.
3.3.1 Regression 1
To test the effect of CEO overconfidence on acquisitiveness, the paper of Malmendier & Tate (2008) is followed. The following general regression specification is used:
𝑃𝑟{𝑌𝑖𝑡 = 1|𝑂𝐶𝑖𝑡, 𝑋𝑖𝑡, } = 𝐺(𝛽1+ 𝛽2𝑂𝐶𝑖𝑡+ 𝑋𝑖𝑡′𝛽𝑥) (1) Where OC is the measure of overconfidence and X is the set of control variables. The measures of overconfidence that are used are Longholder100, Longholder200,
15
Overconfident-Year-100 and Overconfident-Year-200. The subscript i stands for a particular company and the subscript t stands for a particular year. Again the paper of Malmendier & Tate (2008) is mostly followed and the set of controls includes Tobin’s Q, firm size, cash flow, percentage ownership and year fixed effects. Y is a binary variable that takes the value of 1 when the CEO of the firm made at least one successful acquisition in a particular firm-year. In this paper it is assumed that G is the logistic distribution. The null-hypothesis (H0) of this first regression is that the coefficient of on overconfidence, 𝛽2, equals zero. The alternative hypothesis (H1) is that 𝛽2 is significantly different from zero. It is expected in this regression that results are comparable to that of Malmendier & Tate (2008) since the methods of this regression are based on their paper. The coefficient is thus expected to be significantly positive.
The goal of this regression is to measure the effect of a managerial trait,
overconfidence on the dependent variable, acquisitiveness. Two kinds of variation can be used to identify this effect. Cross-sectional variation and variation between CEOs within the same company. The first variation is researched by comparing overconfident and non-overconfident CEOs across different firms with each other. This could lead to finding an effect of overconfidence on acquisitiveness. The second variation is researched by
comparing overconfident an non-overconfident within a firm with each other. This could also lead to finding an effect of overconfidence on acquisitiveness.
Equation (1) is estimated using three estimation methods: 1. Logit regression.
2. Logit regression with random effects. 3. Logit regression with fixed effects.
Method 1 and 2 both make use of both the variations previously mentioned. A logit
regression however, fails to account for the possibility firm-specific effects on the estimates. The problem with the second method, logit regression with random effects, is that it models the effect of the firm on acquisitiveness instead of the CEO on acquisitiveness. The third method to estimate Equation (1) is a logit regression with fixed effects. This procedure only uses variation between CEOs within the same firm. Only firms that change CEO in the sample period are observed and those CEOs have to differ in overconfidence. So either the first CEO is indicated as overconfident, and the second is non-overconfident, or the other way around.
16
Although this does decrease the amount of observations, it also eliminates the possible firm effect on acquisitiveness. The remaining differences in acquisitiveness are attributed to the CEOs themselves. For the Overconfident-Year measures this estimation method is not used, since no CEO is indicated as overconfident for every year in the dataset.
If the null-hypothesis is rejected in this regression, it would mean that CEO overconfidence either has a significant positive or negative effect on CEOs likelihood of making an acquisition. If the coefficient 𝛽2 is found to be significantly positive, this result would follow the paper of Malmendier & Tate (2008) who find a significant positive relationship between managerial overconfidence and acquisitiveness. If the coefficient is found to be either significantly negative or not significantly different from zero, this result would oppose the findings of the previously mentioned paper. In this thesis it is expected that coefficient 𝛽2 is found to be significantly positive. For the control variables, the
expectations are based on the paper of Malmendier & Tate (2008). The effect of Tobin’s Q is expected to be negative, because acquisitions can be seen as substitute for profitable investment opportunities. If a firm doesn’t have many profitable opportunities, it is more likely to make acquisitions. Size expected to not be significantly different from zero, since all the firms in the sample are already selected on highness of their revenues. All the companies in the samples are of large size by selection. Cash flow is expected to have a positive effect on acquisitiveness, because CEOs of firms with more internal cash flows are more likely to make investments (Malmendier & Tate, 2005). Percentage of ownership is expected to not have a significant effect on acquisitiveness.
The Longholder overconfident measures are widely used by researchers in the field of overconfidence. The additional Overconfident-Year measure may question the assumed features of CEO overconfidence. A stronger positive significance of this measure over the existing Longholder measure could mean that there is a possibility that the overconfidence of CEOs fluctuate from year to year and that this affects their likelihood of making an acquisition.
3.3.2 Regression 2
To test if board strength changes the effect of CEO overconfidence on acquisitiveness, the following regression specification is used:
17
Where again OC is the measure of overconfidence and X is a set of control variables. The measures of overconfidence that are used are Longholder100, Longholder200,
Overconfident-Year-100 and Overconfident-Year-200. The set of controls includes Tobin’s Q, firm size, cash flow, percentage ownership and year fixed effects. The variable that is added to the regression is the variable “SB”, which stands for Strong Board. This variable is added as an extra control variable, but also as an interaction variable with overconfidence. Now there will be a distinction between CEO overconfidence on acquisitiveness under weak and strong boards. The subscript i stands for a particular company and the subscript t stands for a particular year. Y is once more a binary variable that indicated whether an acquisition has been made, and G is assumed to be a logistic distribution as in regression 1. From this regression two different hypotheses are derived. The first null-hypothesis (H0) is that the coefficient of CEO overconfidence under weak boards, 𝛽2, equals zero. The alternative hypothesis (H1) is that 𝛽2 is significantly different from zero. The second null-hypothesis (H0) is that the coefficient of CEO overconfidence under strong boards, (𝛽2 + 𝛽4), equals zero. The alternative hypothesis (H1) is that (𝛽2+ 𝛽4) is significantly different from zero. The goal of this regression is to measure whether or not the effect of CEO
overconfidence on acquisitiveness differs under different board strength. As in regression 1, there are two different kinds of variation. Cross-sectional variation and variation between CEOs within the same firm. Equation (2) is estimated using the same three estimation methods discussed in the regression 1.
If the first null-hypothesis is rejected in this regression, it would mean that under weak boards CEO overconfidence either has a significant positive or negative effect on CEOs likelihood of making an acquisition. If the second null-hypothesis is rejected in this
regression, it would mean that under strong boards CEO overconfidence either has a significant positive or negative effect on CEOs likelihood of making an acquisition. In this thesis it is expected that a strong board mitigates the effect of overconfidence on
acquisitiveness. So if a significantly positive relationship between CEO overconfidence and acquisitiveness is found, it is expected to exist among CEOs under weak boards. 𝛽2 is thus expected to be significantly positive and (𝛽2+ 𝛽4) is expected to not be significantly different from zero. For the control variables, expectations are the same as in regression 1. The only control variable that is added is Strong Board. Since it is expected that a strong board mitigates the effect of overconfidence on acquisitiveness, the control variable on itself
18
is not expected to be significantly different from zero. This is due to the idea that a strong board can counter an overconfident CEO, but among firms where the CEO isn’t
overconfident, the effect can easily be different.
The expectation for the Overconfident-Year measures can’t be drawn from any other paper in the field of CEO overconfidence, therefore they are the same as for the Longholder measure. They however are used as an additional measure to control for the idea that overconfidence could be dynamic. An assumption that most papers researching CEO overconfidence disregard. The results of these measures can provide extra insight into the existing field of CEO overconfidence.
4
Data and descriptive statistics
In this thesis a sample of 395 large publicly-traded firms from the United States in the period 2007 – 2016 are analysed. This period is chosen because data on directors from the
Institutional Shareholder Services database (ISS) is only available from 2007. This gives a sample period of 10 years. For a company to be included in the sample, it must appear at least 6 times in the top 500 US companies with the highest revenue in the period 2007 – 2016. This method of sample selection is derived from the paper of Malmendier & Tate (2008). They select firms based on listings of US companies in Forbes magazine. If a firm appears at least four times on one of the lists it is selected for the sample. One of the variables that the Forbes top 500 largest US companies is based on is revenues. Since data on Forbes magazine lists are not available, an own top 500 list is created based on revenues. Firstly, all available companies are added to the dataset. Data on firm specific
variables are collected from the Compustat database. Only companies whose ISO country code is USA are kept in the dataset. All firms for which data on revenues is missing are also dropped out of the dataset. Data on CEO specific variables are collected from the Exucomp database. These datasets are merged with each other. Only the firms that have both Compustat data and ExecuComp data available are kept in the dataset.
The variables that are collected from the Compustat database are Current ISO Country Code (FIC), Total Assets (AT), Common Shares Outstanding (CSHO), Debt in Current Liabilities – Total (DLC), Long-Term Debt – Total (DLTT), Revenue (RVT), Price Close – Annual – Fiscal (PRCC_F), Depreciation and Amortization (DP), Income Before Extraordinary Items
19
(IB) and Property, Plant and Equipment – Total (Net) (PPENT), Debt – Convertible (DCVT), Liabilities – Total (LT), Preferred Stock – Liquidating Value (PSTKL), Deferred Taxes and Investment Tax Credit (TXDITC). From these variables, firm size and Tobin’s Q are derived. Firm size is measured as the logarithm of total assets (log(AT)). Tobin’s Q is measured as the ratio of market value to book value of assets. The market value of assets is measured as total assets (AT) plus market equity minus book equity. Market equity is defined as the fiscal closing price (PRCC_F) times the common shares outstanding (CSHO). Book equity is measured as total assets (AT) minus preferred stock (PSTKL) minus total liabilities (LT) plus convertible debt (DCTV) plus deferred taxes (TXDTIC). Capital is measured as property, plants and equipment – total net (PPENT). And cash flows are measured as income before
extraordinary items (IB) plus depreciation (DP).
The variables that are collected from the ExecuComp database are Annual CEO Flag (CEOANN), Annual Title (TITLEANN), CEO served as director during the fiscal year (EXECDIR), Executive’s Age (AGE), Unexercised Exercisable Options (OPT_UNEX_EXER_NUM), Estimated Value of In-the-Money Unexercised Exercisable Options (OPT_UNEX_EXER_EST_VAL), Gender (GENDER) and Shares Owned (SHROWN_TOT).
Within the dataset, the firms are sorted on revenue per year from highest to lowest. The top 500 companies per year are kept in the database. All other observations are
dropped out. This gives 500 distinct firms over a 10 year period, which makes the total sample consist of 5000 firm-year observations. Hereafter the dataset is sorted on the GVKEY identifier and year. The number of times a company appears in the dataset is now counted, which comes down to the number of times a firm appears in a top 500 highest revenue companies per year. If a firm appears more than 5 times in the dataset, it is kept in the sample. This gives a sample of 474 distinct companies in a 10-year period. This results in 4740 firm-year observations
The data in the sample is joined by variables on director information. Data on directors is collected from the Institutional Shareholder Services database (ISS). The variables that are drawn from ISS are Ethnicity (ETHNICITY), Board Affiliation
(CLASSIFICATION), Female (FEMALE?), Number of Other Major Company Boards
(OUTSIDE_PUBLIC_BOARDS) and Director Age (AGE). Firm-years that don’t have director information available are dropped from the sample. This decreases the sample to 431 firms. From these variables the Board-Index data is generated.
20
Dealing with CEO that have zero exercisable options, the paper of Campbell et al. (2011) is followed. In their analysis using the option-based measures, they omit CEOs that unclassified in the dataset. Removing all the observations with unclassified CEOs, leaves a sample of 395 firms and 2771 firm-year observations.
Lastly the data on mergers & acquisitions for the 395 firms is collected from the Thompson One database and merged with the dataset. The variables that are drawn from the Thompson One database are Percentage of Shares Acquired (OFSHARESACQ) and Value of Transaction in Millions (VALUEOFTRANSMIL). Since there are many acquisitions available from the database, only the acquisitions that meet the following criteria are added to the dataset:
1. The nation of the target is the U.S. 2. The acquisition is completed.
3. The acquirer acquires more than 50% of the shares in the transaction. 4. The deal value of the transaction is at least 5% of the value of the acquirer.
Since this thesis focusses on firms that are located in the US, acquisitions abroad are left out of the dataset. Criterion 3 is made to make sure that the acquirer is the biggest shareholder of the target company after the transaction. Criterion 4 is taken from the paper of
Malmendier & Tate (2008). They omit mergers in which the value of the target is less than five percent of the acquirers value. They state that this selection criterion is especially important since acquisitions of small units of another firm differ substantially from the acquisition of large NYSE firms and may not require the acquiring company’s CEO direct involvement.
Applying the criteria to the acquisitions gives a set of 352 acquisitions across 395 companies and 2771 firm-years. In table 1a a summary statistics of acquisition data is provided. The average of CEOs that has made an acquisition during their tenure is reported in the first row and is 0.39, which means that on average 39% of the CEOs made one or more acquisitions during their reportable tenure as a CEO1. The standard deviation is 0.46, which makes sense, since the variable can be either 0 or 1.. The average of CEOs that made an acquisition during a certain firm-year is reported in the second row and is 0.10. This means
1 Note that for some CEOs only a part of their tenure is researched the dataset, since some years of their
21
that in a certain firm-year around 10% of the CEOs made an acquisition. The standard deviation is 0.29. Total acquisitions made by CEO is reported in row 3. The average amount of acquisitions a CEO makes in his reportable tenure is 0.57. The standard deviation is 0.90, which means that most CEOs made either 0 or 1 acquisitions during their tenure. This is indeed the case, since around 80% of the CEOs made either 0 or 1 acquisitions in the dataset. Acquisitions made by CEO per year is reported in row 4. On average CEOs made 0.14 acquisitions per year, with a standard deviation of 0.26. In row 5 the value of
transaction in reported. The average value of an acquisition is 3818 million. Because all the small acquisitions were dropped from the dataset, this number is logically high. The standard deviation is 6210 million, which means that the value of the transaction still deviates heavily per acquisition.
In table 1b a summary statistics of firm data is given. In the first row data on firm assets is provided. On average firms have total assets of approximately 60 billion with a standard deviation of roughly 191 billion. This says that the firms differ a lot in assets and since the median is 15 billion, which is significantly smaller than the mean, there are a couple large upward outliers in total assets. The same situation applies for Capital in the second row. The average is about 8 billion, with a standard deviation and standard deviation of around 17 and 2 billion. In row 3 to 5 information on cash flows is provided. On average firms have cash flows of roughly 3 billion with a standard deviation of 5 billion. Normalizing cash flows by capital results in a mean of 0.85 with a standard deviation of 2.92. Normalizing by assets results in a mean of 0.09 and standard deviation of 0.07. In the last two rows, information on Tobin’s Q and firm size are provided. The average Tobin’s Q is 1.77 with a standard deviation of 0.95. The average size of a company is 9.83 with a standard deviation of 1.33.
In table 1c a summary statistics of CEO data is provided. In the first row the CEO age is presented. On average a CEO is 56.98 years old with a standard deviation of 5.88. In the second row the number of shares owned is provided. On average a CEO owns around 5 million shares. The standard deviation is approximately 58 million and the median is 910 thousand. So there are some large outliers that drive up the average of the variable. In the third row the percentage of ownership is provided. This leads to the same situation as with the number of shares owned. The mean is 0.98% with a standard deviation and median of 2.8% and 0.28%. In the next two rows information on the Longholder variables are given. In
22
the 2771 firm-years 50% of the firm-years have a CEO that is indicated as overconfident based on the Longholder 100 variable and for the Longholder 200 variable this is 18%. In the last two rows information on the Overconfident-Year variables are given. In the 2771 firm years, 21% of the firm-years have a CEO that is indicated as overconfident based on the Overconfident-Year-100 measure and for the Overconfident-Year-200 measure this is 5% A summary statistics of the Board-Index data is provided in table 1d. In the first row data on the firms board size for every firm-year is provided. The mean board size is 10.90, with a standard deviation of 2.02. In the second row the number of females on the board is provided. On average there are 1.94 females on a company’s board. The standard deviation is 1.09, which means that most firms have 1-3 females on the board. The number of
ethnicities besides Caucasian on the board is provided in the third row. The average number of other ethnicities on the board is 1.36, with a standard deviation of 1.12. This means that most firms have 0-2 people with ethnicities other than Caucasian on the board. In row 4 the number of insiders on the board is provided. On average there are 1.40 insiders on the board, with a standard deviation of 0.76. This means that most board have either 1 or 2 insiders on the board. In the fifth row information on the chairman of the board is provided. The mean of CEO is chairman of board is 0.65, which means that in 65% of the 2771 firm-years, the CEO is also the chairman of the board. The standard deviation is 0.48. In row six and seven the director age and number of outside boards the director serves on are
provided. The average age of a director on the board is 62.38 year, with a standard deviation of 7.61. The average number of outside boards a director serves on is 1.11, with a standard deviation of 1.09, which means that most director serve on 0-2 outside boards. In table 2a a correlation table of the Board-Index variables is given. None of the binary variables are highly correlated with each other, so they won’t form a problem when it comes to multicollinearity.
Lastly, in table 2b a correlation table of various key variables is given. The only variables that are highly correlated with each other are Longholder 100 with Overconfident-Year-100 and Cash flows with size. The first one is expected, since both measures are based on the same variables and won ‘t form a problem since they are not simultaneously
regressed. For the latter it won ‘t lead to problems of multicollinearity, since both are just used as control variables.
23
5
Results
In this section the results from the two regression are discussed. The main results are discussed in this section. The remainder of the statistical analysis is given in the robustness checks section.
5.1 Results of regression 1
5.1.1 Longholder overconfidence measures
Table 3a and 3b contain the results of the Longholder overconfidence measures. Since regression 1 is based on the paper of Malmendier & Tate (2008), the results of this
regression are compared to their results. For both the Longholder measures the outcome of the control variables is roughly the same, so only one is discussed extensively. The first control is Tobin’s Q, which is used to control for investment opportunities. This control is found to be strongly significantly negative for all the estimation methods except for the fixed effect logit. This is roughly the same result Malmendier & Tate (2008) get from their
regression. A significantly negative Tobin’s Q can implicate that CEOs see an acquisition as a substitute for an investment opportunity. When investment opportunities are low, a CEO is more likely to make acquisitions. Translating this into odds ratios, when looking at the year fixed effect logit (2) regression, it can be seen that a CEO is 0.5074 less likely to make an acquisition for every unit increase of the Tobin’s Q. The control variable Firm Size is found to be significantly negative for logistic regression and the logistic regression with year fixed effects, not significantly different from zero in the random effects regression and strongly positive for the logit regression with fixed effects. The negative values of the firm size variable follow the paper of Malmendier & Tate (2008). This negative value was not
expected, but can be explained by the mechanical fact that there is a with-in firm time series variation. Since assets are measured in the beginning of the year, a year in which an
acquisition has been made, has fewer assets, and thus a smaller firm size, than the next year, in which most likely no acquisitions are made. Cash flows is found to not be significantly different from zero, which counters both the results of Malmendier & Tate (2008) as the theory that more cash flow leads to more acquisition activity through financing
considerations or because cash flow proxies recent success for the company. Percentage of ownership is found not to be significantly different from zero for most cases expect for the logit (1) of the Longholder200 measure, where it is found to be significant at a 10%
24
find a nonsignificant result. Stock ownership was added to control for incentive effect. The insignificant result of this measure implies that incentivising a CEO with more options is not a good measure to prevent a CEO from making acquisitions.
For the Longholder measures, the results vary between the 100-cutoff and the 200-cutoff. For the Longholder100 measure, the impact of overconfidence on acquisitiveness is found to be positive, but it is not significant. The Longholder200 measure however, is also found to be positive, but significant at a 10% significance-level for all the specifications except the fixed effect logit. These findings do deviate from what was expected and from the findings of Malmendier & Tate (2008). They find the Longholder measure to already be strongly significant at a 67-cutoff. The results of this study imply that the Longholder
measure is only significant at a 10%-level with a cut-off of 200%. This suggest that only CEOs that hold options that are more that 200% in-the-money are expected to be more likely to make an acquisition. So only for CEOs that are indicated as ‘extremely’ overconfident, the relationship between overconfidence and acquisitiveness is found to be positive and significant. Translating this into odds ratios and looking at the fixed year effects logit (2) regression, it can be seen that holding everything else constant, an (extremely)
overconfident CEO is 1.3985 times more likely to make an acquisition than a non-overconfident CEO.
The problem with the fixed effect logit is that it removes so much observations, that it becomes hard to get reliable results. This problem is discussed more extensively in the robustness checks section. In this section the results of the fixed effect logit regression are no longer mentioned, since it only finds some significance for the Tobin’s Q.
5.1.2 Overconfident-Year measures
Table 3c and 3d contain the results of the Overconfident-Year overconfidence measures. This measure allows overconfidence of CEOs to deviate from year to year. This measure makes a distinction (in comparison to the Longholder measures) between CEOs that consistently hold options that are deep in the money and CEOs who for instance once or twice did not exercise options that were deep in the money and were indicated as overconfident for their whole observed tenure.
Since the outcome of the control variables is roughly the same as those of the Longholder measure regression, these results are not discussed again. Although the control variable don’t provide any new insights, the Overconfident-Year measures do. Where the
25
Longholder100 measure was insignificant for all regression specifications in the first
regression, the Overconfident-Year-100 measure is found to be significantly positive for the logit (1) and the random effects logit specification. Both at a 5%-significant level. When year fixed effects are added in the logit (2) regression, this significance disappears. This suggest that CEOs that fail to exercise options that are more than 100% in the money are more likely to make an acquisition in that particular year. For the logit (1) specification the odds of a CEO making an acquisition, holding everything else constant, increase 1.4443 times when he or she is indicated as overconfident in that year. For the random effect logit the odd ratio is 1.4881. When a control for year fixed effects is added, the significance disappears. This could imply that in certain years more acquisitions are made in general and that in those same years CEOs in general are less likely to exercise their options that are deep-in-the-money. A failure to account for this year fixed effect in logit (1) and random effect logit could be the reason for the significance of those variables.
For the Overconfident-Year-200 the situation somewhat changes. When CEOs are indicated as overconfident at a cut-off of 200%, which basically qualifies them as extremely overconfident, the outcome of all the logit specification become more positive and
significant. The logit (1) and random effect logit are now significant at a 5%-level and even controlling for year fixed effects keeps the variable significant at a 10%-level. This suggest that CEOs that fail to exercise options that are more than 200% in the money are more likely to make an acquisition in that particular year. So if a CEO is indicated as (extremely)
overconfident during a certain year he is 1.9703 times more likely to make an acquisition in that particular year than a non-overconfident CEO. Controlling for year fixed effects this odds ratio becomes 1.6878 and controlling for random effects it becomes 2.0692.
5.2 Results of regression 2
5.2.1 Longholder overconfidence measures
Table 4a and 4b contain the results of the Longholder measures for regression 2. Since this regression is not done by any other study, the results of the overconfidence and interaction measures are harder to compare with previous research. Most control variables however can be compared in the same way as done in the previous section. In both the
Longholder100 as the Longholder200 regression can be seen that the outcome of the control variables don’t provide any new insight. The same explanation as for regression 1 holds up
26
here too. The new control variable Strong Board is not found to be significant, which was expected beforehand. This implies that a strong board on itself doesn’t necessarily increase or decrease the likelihood of an acquisition. Both the Longholder100 as the Longholder100 * Strong Board variables are not found to be significant. Comparing this with the results of the Longholder100 in regression, it makes sense, since there the Longholder100 measure also wasn’t found to be significant. The separation of overconfident CEOs under strong and weak boards, does increase the odds ratio of CEOs under weak boards and decrease the odds ratio of those under strong boards. This effect was expected to be found. The results are however, not significant.
For the Longholder200 measure the situation is different. The Longholder200 measure is overall more positive and more strongly significant than the Longholder100 measure in regression 1. This means that when there is a separation between overconfident CEOs under strong boards and weak boards, the overconfident CEOs under weak boards are found to have a significantly positive impact on acquisitiveness, even when a control for year fixed effects and random effects are added to the regression. The coefficient for
overconfident CEOs under strong boards, which is the coefficient of the interaction variable of Longholder200 and Strong Board added to that of the Longholder200 measure, is found to not be significantly different from zero. This implies that CEOs that are indicated as overconfident and are under supervision of weak boards are more likely to make
acquisitions than non-overconfident CEOs. Overconfident CEOs that are under supervision of strong boards are not more likely to make acquisitions than non-overconfident CEOs. This means that for the Longholder200 measure, a strong board mitigates the effect of
overconfidence on acquisitiveness. This implies that a strong board is only important when CEOs are extremely overconfident based on their option exercising. In these situations a strong board can counter the overconfidence of the CEO and decrease the likelihood of them making potentially value-destroying acquisitions. Taking the logit (2) regression with fixed year effects as an example. A CEO that is indicated as overconfident on a 200% cut-off, ceteris paribus, is 1.4561 times more likely to make an acquisition than a CEO that is
indicated as non-overconfident on that measure. 5.2.2 Overconfident-Year measures
Table 4c and 4d contain the results of the Overconfident-Year measures. When the
27
and new insights is created. The outcome of the control variables is again more or less similar to the other outcomes of the control variables. For which the explanations are already discussed. The Overconfident-Year measure at the 100% cut-off gives some unexpected, though insignificant results. The positive outcome of the Overconfident-Year variable was expected, but the fact that the interaction variable between Overconfident-Year and Strong Board is also positive, is unexpected. This would imply overconfident CEOs based on the 100% cut-off under supervision of strong boards are more likely to make acquisitions that overconfident CEOs under weak boards. This result however is not found to be significant.
The outcome of the Overconfident-Year at a 200% cut-off measure and it’s interaction with the Strong Board variable do provide interesting results. First of all the Overconfident-Year variable is found to be strongly positive and significant at either a 5%- or 1%- significance-level. The interaction variable between Overconfident-Year and Strong Board is negative, which it is expected to be. Although this interaction variable is not
significant, it does however cancel out the significance when the coefficients are added up to derive the coefficient for overconfident CEOs under strong board supervision. So the
coefficient for overconfident CEOs under weak board supervision is found to be strongly positive and significant and the coefficient for overconfident CEOs under strong board supervision is found to not be significantly different from zero. Translated into odds ratios, CEOs that are indicated as overconfident in a particular year based on a 200% cut-off under weak board supervision are found to have a significantly higher odds ratio than 1. CEOs that are indicated as overconfident under strong board supervision are found to have an odds ratio that is not significantly different from 1. So an overconfident CEO that is supervised by a weak board, controlling for year fixed effects and holding everything else constant, is 2.3955 times more likely to make an acquisition in that particular year than a
non-overconfident CEO. An non-overconfident CEO that is supervised by a strong board, is not more likely to make an acquisition in a particular year than a non-overconfident CEO. These results suggest that a strong board can mitigate the positive effect of (extreme) overconfidence on acquisitiveness, assuming confidence levels of CEOs can deviate from year to year.
28
6
Robustness checks
In this section the nature of different variables and specifications are discussed.
Furthermore, certain endogeneity problems, alternative explanations and other problems encountered are discussed.
6.1 Overconfidence measures
In most studies researching overconfidence a cut-off of 67% or 100% is used. In this paper a cut-off of 100% is used to be sure to find a positive effect of overconfidence on
acquisitiveness without controlling for a strong board. The results for the 2 overconfidence at a 100% cut-off researched in two regressions are not what was expected. Of the 4 times the measures are investigated, only the Overconfident-Year-100 measure in the second regression is found to be significantly positive. This significance however disappears when controlling for year fixed effects. Since this paper followed a lot of guidelines from the paper of Malmendier & Tate (2008) it is worrisome that even at a cut-off of 100% the
overconfidence measure is not found to be significantly positive on acquisitiveness. Whether this happened, because another sample period was used, a slightly different sample
selection method was used, there were small differences in defining the overconfidence measure or because the acquisition were accumulated another way is hard to say. The fact that the outcome of the overconfidence measure at a 100% cut-off don’t resemble the outcome of Malmendier & Tate (2008) could be troublesome to the robustness of this study. Another possible threat to the robustness of the results is the amount of missing observations. CEOs are indicated as overconfident if they’ve ever held stock options that were 100%- or 200% in the money. If, for instance, a CEO was active for a company for 10 years, but of those 10 years, there is missing data on 6 of them, this could alter results. A CEO could for instance be indicated as non-overconfident, while he should’ve been indicated as overconfident based on the 6 missing CEO-years. Or a CEO is indicated as overconfident based on the 4 years, but made an acquisition in one of the missing CEO-years. This would also bias results and form an threat to the robustness of the results. A solution to this problem is allowing the overconfidence of a CEO to deviate from year to year. This is done by the Overconfident-Year variables. Now missing data forms less of a problem, because overconfidence is not persistent through time. A certain CEOs tenure is not investigated as a whole, but a CEO tenure is separately investigated from year to year.
29
6.2 Alternative explanations
There could also be different explanations for why there is a positive relationship between holding stocks and acquisitiveness. The first possible explanation is derived from a study of Carpenter & Remmers (2000) who state the idea that CEO stock options revolve around asymmetric information. This means that CEOs could have inside information on for instance the foresight of an acquisition. If a CEO expects their stocks to go up due to that upcoming acquisition, they will be more likely to hold their stocks even though they are already deep in the money. In that same research Carpenter & Remmers (2000) conclude that inside
information is not an issue when it comes to the timing of executive stock exercises. Another explanations is derived from the paper of Harford (2005). He states that most merger waves are driven by industry shocks. So if a shock across multiply industries drives a great amount of mergers in a particular year and another macro-economic shock make CEOs hesitant to exercise their deep-in-the-money stock options, then this could also explain the positive relationship between overconfidence and acquisitiveness. This effect is however already taken into account when controlling for year fixed effects. The effect does drive down the significance of the Overconfident-Year measures, so the effect exist, but doesn’t explain all the merger activity.
An idea that could also explain the is put forth by Malmendier & Tate (2008). They state that the education, background and personal traits of the CEO could also affect both their option exercise and merger behaviour. Data limitation prevented the usage of those control in this study. This could be a danger to the robustness of this study. Malmendier & Tate (2008) find that some of those characteristics are found to be significant, but that they don ‘t alter the significance of the overconfidence measures. Moreover, the correlation between for instance finance education and the overconfidence measure was found to be very low.
6.3 Endogeneity
A common problem faced when researching topics such as corporate governance and board characteristics, is the threat of endogeneity. The Board-Index generated in this study,
consists of board characteristics that have an empirically proven effect on firm performance. This problem is stated by Hermalin & Weisbach (2003), who argue that when studying the relationship between board characteristics and firm performance, endogeneity is a general threat. There could for instance be a reversed causality where firm performance impacts
30
certain board characteristics instead of the other way around.
This is however not the relationship that this study investigates. The relationship between board characteristics and firm performance is only used to choose certain characteristics for the Board-Index. The relationship that is researched is that the on between board characteristics and the ability of the board to mitigate CEO overconfidence. Hermalin & Weisbach (2003) also state that when it comes to board of directors and particular tasks, it is less likely that endogeneity of board structure will affect the results. Furthermore, board characteristics are only used as a interaction with the
overconfidence measures. The direct effect of board composition on acquisitiveness is used in the regression, but merely as a control variable. This also makes the likelihood of
endogeneity being a problem smaller.
6.4 Firm fixed effects
The logit (4) fixed effect estimation method, controls for firm fixed effects. This method rules out the firms that have the same CEO during all the observable firm-years. Only firms that change CEO during the sample period and change from an overconfident CEO to a non-overconfident CEO are kept in the sample. This ensures that the result cannot be caused by firm specific effects that drive up the likelihood of acquisitions. Only keeping the above mentioned firms, however leaves out so many firm-years from the sample, that finding a reliable results and significance becomes difficult. In addition to this problem comes the missing observations issue mentioned in section 6.1. This missing variables is an larger threat to robustness when the sample becomes this small. Not being able to reliably control for firm fixed effects is a significant threat to the robustness of the results. A problem that could have been prevented by substantially increasing the sample of the study.
7
Conclusion and discussion
In this thesis the effect of CEO overconfidence on acquisitiveness is studied. The main focus of the thesis is whether the strength of the board of directors has a mitigating effect on this relationship between CEO overconfidence and acquisitiveness. To research this question, the empirical tests are divided in two separate regression specifications. The first one researches the assumed relationship between CEO overconfidence and acquisitiveness. The second one divides the sample of overconfident CEOs in two categories, the first category being
