Curb your Confidence
Curbing acquisitive behavior of overconfident CEOs through
improved governance
_____________________________________________
An empirical research into the effect of the
Sarbanes-Oxley Act
_____________________________________________
Master’s Thesis 15 ECTS
Author:
Jim Pel
MSc Business Economics,
10442103
Finance
Statement of Originality
This document is written by Student Jim Pel 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
Abstract
In this thesis I try to find the effect of improved corporate governance on curbing acquisitive activity of overconfident CEOs. I use the installation of the Sarbanes-Oxley Act as a natural shock to
improved governance. The main question that this paper tries to answer is: Does SOX have a stronger effect in restricting acquisition activity for firms with overconfident CEOs? I use data on CEO compensation to construct an overconfidence measure, based on the average money-ness of stock options held by the CEO. The model used to empirically research the main question is a probit model with a difference-in-difference estimator that measures the different effect for overconfident CEOs relative to rational CEOs. The results indicate that SOX does have a negative effect on the probability of an overconfident CEO performing an acquisition. However, there is no significant effect found for the improvement of acquisition performance by SOX. The effect on overconfidence is found to be nonmonotonic. Thus for higher levels of overconfidence, the significance of the effect of SOX decreases. This is indicative that SOX is particularly successful for moderately overconfident CEOs. The effect of SOX is also decreasing for high levels of G-Index, which indicates weak corporate governance. So for poorly governed firms, the effect of SOX is not sufficient to curb acquisitiveness of overconfident CEOs.
1
Table of Contents
Abstract………0
1. Introduction………2
2. Literature Review………4
2.1 Mergers and Acquisitions ………4
2.2 Overconfidence………5
2.3 Overconfidence and Acquisitions………6
2.4 Corporate Governance and Acquisitions………8
2.5 Sarbanes-Oxley Act………..9
2.6 Implications and Hypotheses………10
3. Methodology………11 3.1 Difference-in-Difference Model……….11 3.2 Governance Index 3.3 Measuring Overconfidence………13 3.4 Acquisitions………..14 3.5 Performance……….15
4. Data and Descriptive Statistics……….17
5. Results………21
5.1 Regression Results Model 1………21
5.2 Regression Results Model 2………25
6. Robustness checks……….27
7. Conclusion and Discussion………32
2
1. Introduction
In early 2000, the American stock market crashed when the bubble burst. Following this crash, some high profile scandals came to light. Companies such as Enron and Worldcom were committing fraud on a big scale (Clark, 2005). Literature suggests that such malpractices were caused by poor
corporate governance and managerial hubris. (O’connor, 2003). In response to these scandals, the Bush Administration enacted the Sarbanes-Oxley Act (SOX) in 2002. The act was aimed to fix the audit process and increase board independence (Clark, 2005). This in turn could help to restrain managers from taking on excessive risk, such as with the Enron case (Banarjee et al, 2014).
The negative effects associated with managerial hubris, often called overconfidence, are widely confirmed by economic literature. CEOs that are classified as overconfident overestimate project returns, while underestimating the risk involved in these projects. This leads to investment distortions (Malmendier & Tate, 2005), a higher CEO turnover rate (Campbell et al, 2011) and more value destroying Mergers and Acquisitions (Malmendier & Tate, 2008). All of these effects are detrimental to firm value. However, some literature suggests that overconfidence also has positive effects. Galasso and Simcoe state that overconfident CEOs are more likely to pursue innovation. This is confirmed by Hirshleifer et al, who add that overconfident managers achieve greater innovative success for a given amount of R&D investment. In psychology literature, confidence is also linked with better decision making (Insabato et al, 2010). Moreover, Goel and Thakor say that some degree of overconfidence can mitigate the problem of underinvestment by risk-averse managers, so that moderate overconfident managers can improve firm value compared to non-confident
managers. However, too much confidence leads to overinvestment. There is thus a non-monotonic relationship between managerial overconfidence and firm value. According to them, corporate governance plays a big role in steering overconfident managers to exert the optimum level of investment (2008).
The ways in which corporate governance affects the behavior of overconfident CEOs is a subject that is not yet well understood. It is important that the negative effects associated with overconfidence are attenuated, while the positive effects are exploited. One of the main negative effects from overconfidence is that they perform significantly more acquisitions and that these are often value destroying (Malmendier & Tate, 2008). On the other hand, the mergers and acquisitions literature suggests that improved corporate governance can significantly decrease acquisition activity (Harford et al, 2012). However, managerial biases are often ignored in this type of research. One of the existing acquisition theories that does include managerial biases, the hubris theory, suggests that optimistic managers are part of the reason why corporate takeovers often have
3 negative effects on firm value (Roll, 1989). Therefore, improved governance should mainly have the effect that it retains acquisitions made with this motive. This could improve the performance of acquisitions in general.
In my thesis, I will investigate whether improved governance is effective in restraining acquisitions made by overly optimistic, or overconfident CEOs. I will use the Sarbanes Oxley Act as a natural shock for improved government for all firms. The main question is: Does SOX have a stronger effect in restricting acquisition activity for firms with overconfident CEOs. Answering this question has multiple implications. Firstly, it can provide some more clarity regarding the effect of corporate governance on attenuating the negative effects associated with overconfident CEOs. Moreover, it can add to the evaluative research on the effectiveness of SOX. The Act is aimed to increase transparency and control, in order to improve investment decisions. If it is effective in doing so, there must be a decrease of acquisitions done by overconfident CEOs, which are often found to have a negative effect on firm value. Furthermore, it can re-evaluate some of the findings in the article by Malmendier and Tate, which uses data from 1984 until 1994. This was before the introduction of SOX. It is interesting to see if such a policy that is aimed to improve corporate governance has had a significant decreasing effect on acquisitions done by overconfident CEOs. Finally, it adds to the literature on the effect of corporate governance and acquisition performance.
I will combine multiple data-sets, accessible through Wharton WRDS. First, I will collect data on CEO compensation, in order to construct a measure for overconfidence. The construction of this measure is based on the method of Banarjee et al (2014). I will also collect the necessary financial data for the standard control variables used in the acquisition literature. Moreover, data on acquisitions will be collected along with information about the firm’s corporate governance. The statistical analysis will be done using a Probit model, combined with a Difference-in-Difference estimator. The control, or non-treated- group for this estimator will be constructed using the G-Index, which is an indication of effective corporate governance.
The outline of the rest of this thesis is as follows: the next chapter reviews the existing literature on mergers and acquisitions, overconfidence and corporate governance. From here, hypotheses are derived. Chapter three will provide the methodology used for answering the main question. Subsequently, I will list the data sources used and provide some descriptive statistics for the variables of interest in the fourth chapter. Chapter 5 shows the results from the empirical analysis, along with robustness checks in chapter 6. Finally, chapter 7 provides the conclusion and discussion of this thesis.
4
2. Literature Review
This chapter will provide a literary background to the research question and investigate the main theories that are important for this thesis. Next, the hypotheses will be derived to answer the main question.
2.1 Mergers and Acquisitions
According to Varaiya, there are four major conclusions that can be drawn from the extensive research done regarding mergers and acquisitions. Firstly, shareholders of target firms earn large positive abnormal returns due to the market premium paid by acquiring firms. Secondly, the abnormal returns for acquiring firms are non-positive. Thirdly, the total positive gains from
corporate takeovers may be positive but are relatively small. Finally, there is no general consensus among researchers about the sources of these gains (1988). Mueller says that economic theory only has partial success in explaining what causes acquisitions to be made. Especially the fact that the acquiring firm generally receives non-positive return seems to go against shareholder wealth maximization. That is, if the CEO aims to increase maximum wealth for the shareholders, he would not partake in value destroying acquisitions (1969). Jensen on the other hand, finds that corporate takeovers generate overall positive gains. They state that the target firm shareholders benefit, while the acquiring firm shareholders do not lose (1983). However, Roll claims that such statements about the overall positive gains from mergers and acquisitions are overestimated. He even doubts if they exist at all (1989).
A possible explanation for the fact that there is no overall consensus among researches whether corporate takeovers generate positive returns or not is given by Seth et al. They say that this is because little is known about the sources of potential gains or losses. They distinguish three types of motives driving corporate takeovers. The first motive is to create potential synergies. This occurs when the value of the combined firms exceed the sum of the value of the two individual firms. Secondly, managerialism can induce companies to acquire other firms. This happens when managers embark in acquisitions to maximize their own utility instead of the shareholders and thus knowingly overpay for the target firm. Thirdly, they find that managerial hubris can be the reason behind acquisitions. While the first motive, to create synergies, adds value for both firms, the last two motives are value destroying. Especially for the acquiring firm (2002).
This thesis will mainly focus on the last motive, managerial hubris, that drives corporate takeovers. Managerial hubris as a motive for corporate takeovers is first described by Roll (1989). His
5 so called “Hubris Hypothesis” theorizes that managers, that are overly confident in their own ability to add value to a potential target firm, might end up overpaying for acquiring that firm. That is, managers often overestimate their ability to generate synergies and thus overpay for the target firm. Varaiya also finds that acquiring firm often overpay for the target firm and provides an alternative explanation, which is called the Winner’s Curse. The idea behind the Winner’s Curse is that whoever wins a sealed-bid auction is the most likely to have overvalued the object and has the highest probability of overpaying. Varaiya states that the Winner’s Curse also exists in the context of mergers and acquisitions, because an acquisition can be seen as a sealed-bid auction. The potential acquiring firms are the bidders, while the potential target firm is the object to be sold. The firm that ends up acquiring the target firm is the one that assigns the highest value to that firm and is likely to pay too much. Indeed, Varaiya concludes that the Winner’s Curse is one of the reasons why
acquiring firm often pay more than the expected takeover gains (1988).
The Hubris Hypothesis is actually linked with the Winner’s Curse. Compte argues that competition among bidders induces a selection bias in favor of the optimistic bidders. The optimistic bidders are thus more vulnerable to the Winner’s Curse. This is in line with the Hubris Hypothesis. Bidders should therefore incorporate this bias in their bid by a downward adjustment conditional on winning. However, bidders that are overconfident in the precision of the signal, will underestimate the possible estimation errors they make. This leads them to adjust the bid downwards
insufficiently and so they still end up paying too much (2004). Managers that are either overly confident about their own ability to add value, or underestimate the possible valuation errors they make are thus more likely to overpay for the target company. To get a better understanding of how overconfidence can affect corporate investment and acquisitive behavior, the next section will define overconfidence from a psychologic perspective and will describe how it can affect the behavior of CEOs.
2.2 Overconfidence
Overconfidence has historically been the subject of a large psychology literature. According to Moore and Haley, there are three ways in which psychologists define overconfidence. First, it is defined as an overestimation of one’s performance. Secondly, one’s performance relative to others and thirdly as an overestimation of the precision of one’s own believes (2007).
The first definition means that people in general think better about themselves than they actually are. An example of this is given by Clayson. He finds that students systematically
overestimate the grades they got for their exams (2005). The second definition means that people have the tendency to look at themselves more favorably than others. An individual tends to think he
6 or she is better than average. This so called “better-than-average-effect” is stated in many
psychology articles. One prime example of this is illustrated by Svenson (1980). He held a survey among US and Swedish drivers. A surprising 93 percent of US drivers thought themselves to be better drivers than average. Whether people are asked about desirable traits (Alicke, M. D., 1985) or their life expectancy (Weinstein, 1980), the better-than-average-effect is found. The third definition of overconfidence is present when one overestimates the precision of his or her beliefs. Moore and Haley call it over precision (2007). This type of overconfidence is often shown when people are asked to estimate a numerical answer to a question and build an X percent confidence interval around that estimate. In their paper, Soll et al (2004) ask their participants to estimate the length of the Nile river or the year of the first hot air balloon. Then the participants must provide an interval such that they are X percent sure the correct answer is in that interval. In reality, the correct answer laid between the interval in much less than X percent of the time. The participants thus
overestimated the precision of their estimations.
The reason that overconfidence is relevant for financial literature is that overconfidence is found among many CEOs. Goel and Thakor even state that overconfidence is more likely to be present in managers. A main cause of this finding is that overconfident people are more likely to be promoted to CEO, due to the selection process of promotion. People that take more risk are more likely to be noticed and selected for promotion. CEOs are therefore more likely to be overconfident. They add that this can have many implications for corporate decision making (2008).
2.3 Overdconfidence and Acquisitions
Although psychology has long recognized overconfidence, it has gained more attention from economists in the past decades. Roll (1986) is one of the first to investigate how personal managerial traits can affect corporate finance. He argues that managerial hubris can partially explain economical inefficiencies in the context of mergers and acquisitions. With this research, Roll uses psychological theories about human nature to partially explain economical inefficiencies.
Malmendier and Tate also investigate how managerial hubris, or overconfidence, can have an effect on corporate governance. They too link this psychological phenomenon to investment inefficiencies. They argue that overconfident CEOs overestimate project returns, while
underestimating the risk involved in these projects. This has certain implications for their investment decisions, because it means that the CEO’s decision whether or not to invest also depends on the type of financing available. When internal financing, such as cash, is available, the overconfident CEO overestimates the project return and thus overinvests. However, when the firm needs external financing for their project, the overconfident CEO underestimates the project risk and thus perceives
7 the cost of equity to be too high. This is because the market expects lower returns than the CEO and thus requires a larger return on equity as theorized by Modigliani and Miller. In this case, the overconfident CEO underinvests. Finally, the overconfident CEO will accept debt to externally finance its project even if the CEO perceives the cost of debt to be too high. It will favor debt to equity, because the cost of debt does not include the perceived upside of the project (2005). These results show that in order to analyze the effect of overconfidence on corporate investment, one should take into account the type of financing that is available and whether the firm is equity-dependent.
Another paper by Malmendier and Tate (2008) further analyses investment distortions caused by overconfidence in the light of mergers and acquisitions. They empirically show that overconfident CEOs with access to abundant internal resources are 65 percent more likely to
conduct acquisitions than non-confident CEOs and that they overpay for the target firms. This means that they are more likely to undergo value-destroying mergers and acquisitions. According to them, this is caused by two main reasons. Firstly, the overconfident managers overestimates the value of the potential target firm. This is a manifestation of the “better-than-average-effect”, because he thinks he can lead the company better than the existing management. Secondly, he overestimates the value of his own company, because he overestimates his only abilities to lead it. Combined, these two forms of overconfidence leads a manager to overpay for the target company. The market reaction to these acquisitions are also significantly more negative than with non-overconfident CEOs.
Brown and Sarma (2007) also research the effects of overconfidence on acquisitive behavior. They argue that not only the extent of overconfidence is important, but also the CEO’s ability to impose his ideas on the firm’s decisions, called CEO dominance. The article finds that CEO dominance has a positive effect on acquisitions. So the more dominant the CEO is, the higher the chance of undertaking acquisitions. This is particularly the case for diversifying acquisitions. They find that the likelihood of an overconfident CEO performing a diversifying acquisition almost doubles for a 10 percent increase in CEO dominance. However, a possible shortcoming of this article is that they measure CEO dominance as the ratio of CEO compensation to Assets. Later research by Otto shows that CEO compensation and CEO overconfidence are correlated. He finds that overconfident CEOs receive less total compensation than their non-confident counterparts, because they overvalue stock option based salary (2014). This means that there is possibly some multicollinearity between overconfidence and CEO dominance, which can bias the results found by Brown and Sarma. However, CEO dominance can be viewed as a measure for how much power the CEO has and how the interaction between the CEO and the board of directors is.
8 How this interaction affects investment is further explored by Goel and Thakor (2008). They investigate the dynamics between corporate governance, CEO overconfidence and corporate
investment. They find that overconfidence has a non-monotonic effect on firm value. The idea is that rational CEOs are risk-averse, which causes them to sometimes not take on positive NPV projects. Rational CEOs thus underinvest. A certain degree of overconfidence might increase investment to the optimal investment level, and thus increase firm value. The effectiveness of board governance plays an important role. They claim that an increase in the quality of corporate governance reduces overinvestment made by overconfident CEOs, which has a positive effect on firm value. This is in line with the theory of Brown and Sarma (2007).
2.4 Corporate Governance and Acquisitions
To see through which mechanisms improved corporate governance can lead to a more optimal level of investment, it is important to identify what good corporate governance is. Gompers et al construct an Index to proxy the effectiveness of corporate governance, the so called G-Index. The provisions they use all measure the balance of power between shareholders and management. The more rights shareholders have, the lower their index is and the better the corporate
governance. They find that a low Corporate Governance Index is associated with higher firm value. Moreover, they argue that one of the main causes for this result lies in the acquisitive behavior of management. A low index implies lower management power, which might lead to less value destroying acquisitions. This in turn improves firm value (2003).
Baghat and Bolton provide another empirically successful measure of good corporate governance: board independency. They state that this measure is in many cases an equally good corporate governance proxy as the G-Index. Independency of boards are important, because they have the task to monitor management and evaluate certain corporate decisions. A more
independent board is more able to objectively judge these decisions. This can have a positive effect on the value of investment (2007). Byrd and Hickman also support the importance of board
independency for good corporate governance. They conclude that bidding firms with independent, outside directors that hold at least 50 percent of the seats in the board of directors, experience higher abnormal returns after the announcement date of an acquisition (1992).
This link between corporate governance and corporate takeovers is thoroughly investigated in the corporate finance literature and, among others, found by Harford et al. They research the effect of corporate governance on cash holdings and find that lower governance leads to lower cash holdings, due to higher investment. This is in line with the spending hypothesis. That is, for a given amount of cash, ceteris paribus, firms with weaker corporate governance will spend their cash more
9 quickly than firms with better governance. They add that most of the cash is spent on acquisitions (2012). One of the reasons that this cash is quickly spent on acquisitions is given by Bliss and Rosen, who find that a CEO receives higher compensation after mergers and acquisitions, even though they often destroy value. There is thus an incentive for the CEO to spend his cash reserves on acquisitions (2000). This is backed by Harford and Li, who find that acquiring CEOs are better off due to increased personal wealth in 75 percent of the cases, despite the negative effects for shareholders. However, they claim that firms with strong corporate governance do not reward their CEOs for acquiring other firms so that there is no personal incentive for them to undergo these acquisitions (2007). It can thus be argued that improved corporate governance can ensure that CEOs spend less of their cash
reserves on acquisitions.
2.5 Sarbanes-Oxley Act
This thesis will investigate the effect of improved corporate governance on the acquisitive behavior of overconfident CEOs. One must therefore identify an event that improved corporate governance. Goel and Thakor provide such an event. According to them, the Sarbanes-Oxley Act is an exogenous shock that improves corporate governance, which can be used to empirically test the interaction between board governance and CEO overconfidence.
The Sarbanes-Oxley Act (SOX) was passed by Congress in 2002 in order to enhance corporate governance and restore public trust. It introduced changes in management’s reporting responsibilities as well as the responsibilities of the auditor (Zhang, 2007). Banarjee et al (2014) add that SOX reduces the amount of power the CEO has over the board, due to increased transparency and independency. In other words, it reduces CEO dominance. This can attenuate the effect of managerial overconfidence regarding corporate investment. Linck et al (2009) add that after SOX, the board consisted of more financial experts and lawyers. This is also an indication that the quality of control will improve.
Clark distinguishes three types of reforms. Firstly, SOX aims to improve the audit process. These audit-related changes are mainly to prevent conflicting interests between firms and the audit company due to lasting relationships. Furthermore, it induces managers to improve the internal audit. A fair amount of managers said that getting ready for these audit changes already improved their management information systems as a whole. The second type of SOX reforms is Board-related changes. These measures encourage changes to the composition of the board of directors. More specifically, the goal is to increase board independence. This in turn should improve their
performance as judgmental monitor of managers. Clark states that the fact that managers know they are being monitored more, induces them to improve their information acquisitioning regarding
10 investing projects. For example, if they know they have to present a plan regarding the acquisition of a certain company to a strong, independent board of directors, it might push them towards plans more favorable for shareholders. The last category of SOX reforms involve financial disclosures to investors and the public in general. These reforms are aimed to improve the firms transparency. The idea behind it is that it will empower shareholders. Giving the shareholders more rights to vote, sue and buy- and sell shares might result in better investment decisions (2005).
2.6 Implications and Hypotheses
This section will derive hypotheses from the literature review. The main question to be answered is whether improved corporate governance has an effect on the acquisitive behavior of overconfident CEOs. SOX will be used as a natural experiment for improved governance, because it is aimed to improve the audit process, create more independent boards and improve transparency. Firstly, It is shown that overconfidence induces managers to engage in more acquisitions. However, the amount of dominance or power the CEO has over the board of directors also plays an important role in the acquisition behavior. Improved audit, Independent board and increased transparency all reduce the power of the CEO. Furthermore, theory suggests that overconfident managers overinvest when internal funds, such as cash is available. Improved governance decreases investment from cash reserves, and thus is expected to decrease investment by overconfident CEOs. Therefore the first hypothesis is:
1) The introduction of SOX leads to a decrease in acquisitions done by overconfident CEOs.
Secondly, it is argued that overconfident managers are more likely to fall for the winner’s curse. This is due to an underestimation of their estimation error of the target firm value. Therefore, they often pay a premium that is too high, which in turn leads to negative abnormal returns. It is also shown that SOX improves the control of the board due to higher independency. This creates an incentive for the CEO to come up with a more favorable plan for the shareholders. Moreover, the board consist of more financial experts after SOX, which might also improve the acquisition valuation. Therefore the second hypothesis:
11
3. Methodology
This chapter describes the methodology used to answer the main question. First, the regression model is explained along with the empirical specification. Next, the type of data needed for the variables used will be described.
3.1 Difference-in-Difference Model
The main question will be empirically researched using a Difference-in-Difference model (DiD) The DiD approach is widely used by researchers and particularly useful to estimate the effect of a certain policy change or intervention that does not affect everybody in the same way or at the same time, such as the introduction of the Sarbanes-Oxley Act. DiD essentially identifies four different groups. First it divides the treated and the non-treated. Treated are especially affected by the policy, while non-treated are affected less. Secondly, the pre-treatment and the post-treatment group. This divides the observations before- and after the policy is introduced. This creates four groups: the post-treatment treated, post-treatment non treated, pre-treatment treated and pre-treatment non treated. The model estimates the effect of the post-treatment, treated group compared to the other three groups (Lechner, M. 2011). The general expression for a DiD, linear estimator is as follows, where X denotes a set of explanatory variables, :
Y= X𝛽 +𝛽1Post + 𝛽2Treat + 𝛽12Post*Treat + +𝜀
E(y| X, Treat = 1, Post = 1) = X𝛽 +𝛽1 + 𝛽2 + 𝛽12
E(y | X, Treat = 1, Post = 0) = X𝛽 + 𝛽2
E(y | X, Treat = 0, Post = 1) = X𝛽 +𝛽1
E(y | X, Treat = 0, Post = 0) = X𝛽
Probit DiD Model
In this thesis, I investigate the effect of SOX on acquisitive behavior. In other words, if the probability of an overconfident CEO performing an acquisition decreased after this policy. The dependent variable is acquisition, which is a binary variable. Therfore, a probit regression is used. Y is the binary variable for Acquisition {𝑦 = 0 𝑖𝑓 𝑛𝑜 𝑎𝑐𝑞𝑢𝑖𝑠𝑖𝑡𝑖𝑜𝑛 𝑖𝑛 𝑦𝑒𝑎𝑟 𝑡𝑦 = 1 𝑖𝑓 𝑎𝑐𝑞𝑢𝑖𝑠𝑖𝑡𝑖𝑜𝑛 𝑖𝑛 𝑦𝑒𝑎𝑟 𝑡
12 The standard Probit Model looks the following way:
Pr (Y=1 | X) = F
(Xi 𝛽)
Where F is the cumulative normal distribution and Xi is a set of variables that have an effect on the outcome variable Y. In this case, the explanatory variables include the DiD estimator model as described above. Therefore the combined probit, DiD model is:
Pr (Y=1 | X) = F
(X𝛽 +𝛽1Post + 𝛽2Treat + 𝛽12Post*Treat + +𝜀)
Finally, this research does not only look at the effect of the treatment, SOX, on a treated group. It investigates the differences within the treated group. While the treated group consists of rational, as well as confident CEOs, the effect on the latter is particularly of interest for this research. Therefore, another interaction variable, Overconfidence*Post*Treated, is created to specifically estimate the effect for overconfident CEOs within the treated group. This variable is called OverconfSOXTreat. In addition to this variable, the singular interaction variables Overconf*SOX and Overconf*Treat are included as well. The variable that tests hypothesis 1 is OverconfSOXTreat.The following regression is used to estimate the effect of SOX on the acquisitive behavior of overconfident CEOs:
Model 1)
Pr (Acq𝑖=1| ConfTreated𝑖)=
F(𝛽0+𝛽1Conf + 𝛽2SOX + 𝛽3Treat +
𝛽4SOX*Treat + 𝛽5Overconf*Treat + 𝛽6Overconf*SOX +
𝛽6Overconf*SOX*Treat + Xi, +𝜀)
The time-frame during which this model is tested is between 1992 and 2012, ten years before and ten years after the introduction of SOX. The dependent variable Y is the binary variable Acquisition, which equals 1 if a successful acquisition is performed. Conf is a dummy variable for overconfidence. It is argued that overconfident CEOs perform more acquisitions than their rational counterparts, so the expected coefficient is positive. SOX is a dummy variable for the treatment period. SOX was introduced in 2002, so the variable SOX equals 1 if the observation is in 2002 or later. A negative sign is expected to be found here, as this means that overall, acquisitions decreased after the
introduction of SOX. This is expected due to increased corporate governance. Treated is a dummy variable that seperates the treated- and non-treated group, where the value 1 is assigned to the treated group. A positive coefficient means that the treated group performed more acquisitions during the entire observation period (1992-2012). SOXTreated is the DiD estimator for all types of CEOs. A negative sign is expected here, which would mean that after SOX, the treated group performed less acquisitions than the non-treated group. Overconf*SOX measures the effect of SOX on overconfident managers for the non-treated group, while Overconf*Treat measures the effect for overconfidence in the treated group. ConfTreated is an interaction variable Conf*SOXTreated. This
13 measures the specific effect for overconfident CEOs that were affected by the treatment. The coefficient is expected to be negative, which would imply a decrease of acquisitions done by
overconfident CEOs relative to rational CEOs in the treated group. This coefficient is essential for the first hypothesis. Xi is a set of control variables.
3.2 Governance Index
First, it is important to identify the treated group. The treated and non-treated groups are affected differently by the treatment. In the case of SOX, the main goal was to improve corporate
governance. As described earlier, it is due to improved governance that the acquisitive behavior of overconfident CEOs is expected to change. Therefore, the non-treated group consists of companies that, prior to SOX, already adhered to SOX standards. In other words, companies that already had strong corporate governance before SOX was introduced. These companies will not be affected in the same way as companies with weak corporate governance. It is therefore also vital to obtain data on the strength of a company’s corporate governance. One such corporate governance measure is the so called G-Index, by Gompers Ishii and Metrick (2001). This index is based on 24 governance rules and is used as a proxy for shareholder’s rights. A low index indicates strong corporate governance and good shareholder rights. Companies with low G-Index will be used as a control group, because they already had strong corporate governance and lower CEO dominance before SOX was introduced. I will use companies with a G-Index of lower than 7 as my control- or non-treated group, because this is the bottom 25th percentile. The companies with a higher G-Index will
be indicated as the treated group.
3.3 Measuring Overconfidence
Moreover, a measure of overconfidence is needed. Malmendier and Tate provide such a measure by looking at the timing of option exercise behavior by managers. CEOs often receive a large share of their compensation in the form of stock and option grants. A rational risk-averse investor should diversify his portfolio of stock holdings. However, the CEO is often limited in undertaking such divestures due to contractual agreements. He is therefore often overexposed to his own firm’s idiosyncratic risk and should sell his stock options early to avoid this risk. Malmendier and Tate find that certain CEOs in their observations do not behave in such way. Instead they hold on to their stock options for too long. These CEOs get classified as overconfident because they seemingly have too high expectations of the performance of their own company (2005, 2008).
The dataset used by Malmendier and Tate is proprietary, so it can’t be used for this thesis. However, Banarjee et al use a slightly different method that is easier to replicate. They look at the
14 extent to which CEOs hold vested options that are in the money and use this as a measure of
overconfidence. That is, the more these options are in the money, the more confident the manager is (2015). The money-ness of options is often not directly provided, so this must be calculated by dividing the stock closing price by the average strike price of stock options that the CEO holds. The average strike price can in turn be estimated using the method of Core and Guay (2002). They calculate the strike price by first dividing the value of exercisable, but unexercised, options by the number of exercisable, unexercised options held by the CEO. This yields estimates of how far in the money the average stock option is. Next, this estimate is subtracted from the stock closing price to obtain the average strike price. Finally, the strike price is divided by the closing price to get an estimate of the average ness of the stock options held by the CEO. The higher the money-ness of these stock options, the more overconfident the CEO is expected to be. In order to construct this measure, data on CEO compensation is necessary along with stock price information.
One downside of using this method is that all options are assumed to be in-the-money. Therefore the true strike price will be underestimated if the CEO’s portfolio includes out-of-the-money stock options. However, this thesis will look at stock option behavior of CEOs relative to others in the sample. That is, the top 25th percentile of in-the-money stock option holders will be
classified as overconfident. So even though the strike price will be underestimated in some instances, this will be the case for all types of CEO.
3.4 Acquisitions
Next, data on acquisitions is obtained. This includes information about the target- and acquiring firm, the means of financing used for the acquisition, the execution date and the announcement date of the acquisition. Then, this information is merged with data on CEO compensation and financial data for each company in order to investigate the effect of overconfidence on acquisitions. The set of control variables that will be used are standard in the acquisition literature. According to Malmendier and Tate, this includes Size, Tobin’s Q, Cashflow, total vested options and corporate governance (2005).
The post-treatment treated group thus contains CEOs at companies with a high G-Index, measured after the introduction of SOX. The acquisitive behavior of these CEOs will be measured relative to the other CEOs to estimate the effect of SOX. Once the date on acquisitions, CEO
compensation, firm-level variables, stock price returns and G-Index is obtained, it will be merged to construct a dataset including all the necessary variables. The CEO observations that miss crucial data on CEO compensation will be dropped. Next, all companies that do not have any indication of G-Index or firm-level variables will be dropped as well, as these companies do not have the necessary
15 control variables and can’t be assigned to the treated- or non-treated group. The end result is a database that contains a CEO observation for each year, in which he either performed an acquisition or not. The probabilty for this acquisition will be analyzed using the necessary explanatory variables. For example, Dorrit Bern is the CEO of Charming Shoppes Inc, from 1996 until 2007. During this time, he once performed an acquisition in 1999. Dorrit is labeled as overconfident, based on the overconfidence measure being in the top 25th percentile. The G-Index of the company is 14, which
means relatively weak corporate governance. After 2002, Dorrit did not do anymore acquisitions. This could be an example of an overconfident CEO that was restrained by increased corporate governance after the introduction of SOX when controlled for Size, Q and vested options.
3.5 Performance
Measuring the long-run performance of acquisitions is often done using event studies, usually from 100 days before until 100 days after the event. However, many things can influence the performance of a company during that period of time, such as other investment decisions or structural changes . It is thus difficult to measure the direct effect of acquisitions on long-time firm performance. Therefore, a proxy can be used for acquisitions that are more likely to be value destroying, such as diversifying mergers (Malmendier, Tate. 2005). Diversifying mergers have been found to have a negative effect on firm value and are therefore more likely to be value-destroying (Lang & Stulz, 1993). The effect of overconfidence on acquisition performance will be measured, using Diversifying mergers as a proxy for bad performance. I classify an acquisition as diversifying if the target
company has a different Fama and French Industry Code. The empirical specification is the same as for Acquisitions, except that the independent variable now is diversifying acquisitions:
Denote y as diversifying acquisition, {𝑦 = 1 𝑖𝑓 𝑎𝑐𝑞𝑢𝑖𝑠𝑖𝑡𝑖𝑜𝑛 𝑖𝑛 𝑦𝑒𝑎𝑟 𝑡 𝑖𝑠 𝑑𝑖𝑣𝑒𝑟𝑠𝑖𝑓𝑦𝑖𝑛𝑔 𝑦 = 0 𝑖𝑓 𝑎𝑐𝑞𝑢𝑖𝑠𝑖𝑡𝑖𝑜𝑛 𝑖𝑛 𝑦𝑒𝑎𝑟 𝑡 𝑖𝑠 𝑛𝑜𝑡 𝑑𝑖𝑣𝑒𝑟𝑠𝑖𝑓𝑦𝑖𝑛𝑔
Model 2)
Pr (Div Acq𝑖=1| ConfTreated𝑖)=
F(𝛽0+𝛽1Conf + 𝛽2SOX + 𝛽3Treat +
𝛽4SOX*Treat + 𝛽5Overconf*Treat + 𝛽6Overconf*SOX +
𝛽6Overconf*SOX*Treat + Xi, +𝜀)
The type of data needed in order to estimate this regression model is identical as before. The only thing that is different is the dependent variable, diversifying acquisitions. In this case, only the observations are used when an acquisition was actually done (so acq=1). Then the effect of SOX on diversifying mergers done by overconfident CEOs is analyzed. Conf is expected to have a positive sign, because it is argued that overconfident CEOs perform more value-destroying acquisitions. SOX is expected to have a positive effect on acquisition performance. This corresponds with a negative
16
coefficient. The coefficient of Treat is expected to be positive, because weak corporate governance is linked to more value-destroying acquisitions. However, after the introduction of SOX, diversifying mergers are expected to drop. So the sign for SOXTreated might be negative. Hypothesis 2 argues that after the introduction of SOX, the performance of acquisitions done by overconfident CEOs improves. If this is the case, the sign of the coefficient of ConfTreated should be negative. That is, after SOX, overconfident CEOs performed less value destroying acquisitions.
17
4. Data and Descriptive Statistics
The data needed for this research is retrieved from several standard data-bases. Firstly, the data on CEO compensation is obtained from the Execucomp database. CEO annual compensation from the date range 1992-2012 is collected. The overconfidence measure is constructed by dividing the value of exercisable, but unexercised, options (item name opt unex exer ext val) by the number of
exercisable, unexercised options (item name: opt unex exer num) held by the CEO. Next, this number is substracted from the closing stock price. This variable is provided by the Compustat database (item 199). This is the estimation of the average strike price of options held by the CEO. The the stock closing price is then divided by the average strike price to obtain an estimate of the average money-ness of options held by the CEO. This is used as a measure for CEO overconfidence. Execucomp provides roughly 200 thousand observations for CEO compensation between 1992 and 2012.
The Compsustat database is also used to collect the necessary standard control variables used in the acquisition literature: size, tobin’s Q, cashflow and options vested. These variables are collected and transformed copying the example of Malmendier and Tate (2005). Size is expressed as the natural logarithm of assets (item 6). Tobin’s Q measures the ratio of the market- and book value of assets. Market value of assets is calculated by multiplying common shares outstanding (item 25) by the fiscal year closing price (item 199). The book value of assets is assets (item 6) minus total liabilities (item 181) minus preferred stock (item 10) plus deferred taxes (item 35) plus convertible debt (item 79). Next, cashflow is defined earnings before extraordinary items (item 18). This is normalized by capital, which is found as property, plants and equipment (item 8). Finally, the number of option awards granted to the CEO is provided by Execucomp (item name:
option_awards_num). After the control variable data is merged with the CEO compensation data, the CEO observations for which the necessary information is missing, is dropped. This leaves 30 thousand observations.
The acquisition data is provided by Thomson One. A selection is made of US based acquiring firms. Information about the target firm, the acquiring firm, the acquisition date and the payment method is acquired. This data is merged by company CUSIP code with the previously collected data from Execucomp and Compustat.
Lastly, the G-Index is gathered from the ISS database, formerly known as RiskMetric. The index is only given in the years: 1995, 1998, 2000 and 2002. Therefore, the average G-Index is calculated for these years. Based on the average G-Index, the company is assigned in the treated- or
18 treated group. The companies for which the G-Index is missing for these years are dropped. This results in a merged database of more than 17 thousand observations.
Variable Description
Overconfidence Measure Closing price fiscal year divided by average strike price options Acq Dummy variable = 1 if acquisition was done
Size Natural logarithm of Assets in dollars
Cashflow Cashflow normalized by plants, property and equipment Q Tobin's Q = market value assets/book value assets OptionsVested Total number of CEO's vested options
Overconf Dummy variable = 1 if overconfidence measure >= 1.82 SOX Dummy variable = 1 if the fiscal year is 2002 or later
Treat Dummy variable = 1 if the company has a G-Index of 7 or higher SOXTreat Interaction variable SOX*Treat
OverconfTreat Interaction variable Overconf*Treat OverconfSOX Interaction variable Overconf*SOX OverconfSOXTreat Interaction variable Overconf*SOX*Treat
Diversifying Dummy variable = 1 if the acquisition was diversifying according to Fama French industry code
Table 1. Variable descriptions
Variable Obs Mean Std.Dev Min Max
Overconfidence Measure 17.471 1,62 0,84 1,00 4,50 Acq 17.471 0,07 0,03 0,00 1,00 Size 17.471 3,39 0,76 0,02 6,50 Cashflow 17.471 0,55 0,50 0,03 1,59 Q 17.471 1,66 0,83 0,84 3,49 OptionsVested 17.471 150,00 100,00 0,00 700,00 Overconf 17.471 0,25 0,33 0,00 1,00 SOX 17.471 0,51 0,40 0,00 1,00 Treat 17.471 0,64 0,45 0,00 1,00 SOX Treat 17.471 0,30 0,47 0,00 1,00 OverconfTreat 17.471 0,16 0,34 0,00 1,00 OverconfSOX 17.471 0,13 0,21 0,00 1,00 OverconSOXfTreat 17.471 0,06 0,49 0,00 1,00 Diversifying 1250 0,22 0,39 0,00 1,00
Table 2. Summary Statistics
Table 2 shows the summary statistics of the dependent and independent variables of interest for this study. The Overconfidence Measure is used as a proxy for overconfidence. The table shows that the average money-ness of stock options held by CEOs in this sample is 1,62 or 162 percent. Note that this is likely to be an overestimation of the true money-ness of stock options due to the earlier assumption that all the options are in the money. The majority of variables are dummy variables.
19 The minimum of these variables is thus 0 and the maximum 1. The mean can be interpreted as a percentage of the total observations. For example, the mean of Acq is 0,072. This means that roughly 7,2 percent of all observations of the dummy variable equal 1. The exact amount of
acquisitions done in this sample is 1250. The mean of SOX is 0,51, so 51 percent of the observations is after 2002 and 49 percent is before. 64 percent of the companies are in the treated group with a G-Index of 7 or lower. Furthermore, OverconfTreated shows how many observations there are overconfident CEOs in the sample. This amounts to 6 percent of the total observations, so 1048 observations.
Acquisitions: (OC) CEO rational CEO Total
Pre SOX 349 510 859
Post SOX 99 292 391
Total 448 802 1250
Table 3: Acquisition frequency
Table 3 shows the total amount of acquisitions done in the sample from 1992 until 2012 by overconfident (OC) CEOs and rational CEOs. In the sample, a total of 1250 acquisitions were
performed. 859 of those were done before SOX was introduced, compared to 391 acquisitions after SOX. This is a decrease of 55 percent. Before SOX, 41 percent of the acquisitions were done by Overconfident CEOs. This is remarkable, because only the top 25th percentile is defined as
overconfident. It could be a sign that overconfident CEOs performed more acquisitions than rational CEOs, controlling for other variables. This result is also found by Malmendier and Tate (2005) After SOX was introduced, both types of managers did less acquisitions. The decrease in acquisitions done by overconfident CEOs is an impressive 72 percent after the introduction of SOX. Acquisitions done by rational CEOs decreased by 43 percent from 510 pre SOX to 292. SOX thus appears to have a stronger effect for overconfident CEOs. From the 391 acquisitions in the post-SOX era, 25 percent was done by overconfident CEOs. This is thus a clear decrease from pre SOX and it appears that overconfident managers did not perform more acquisitions after SOX than their rational counterparts.
20
Treated Non-Treated
Pre SOX Post SOX Total Pre SOX Post SOX Total
OC CEO 221 47 268 128 52 180
Rational
CEO 378 279 557 132 113 245
Total 599 226 825 260 165 425
Table 4: acquisitive behavior for treated vs non-treated firms
Table 4 provides the different effect from SOX on the acquisitive behavior of both types of CEOs for treated- and non-treated companies. Firstly, treated companies performed 825 acquisitions in total. This amounts to 66 percent of all acquisitions done in the sample. The summary statistics table shows that 64 percent of observations are from the treated companies. The treated group thus seems to perform relatively more acquisitions than the non-treated group.
Secondly, the table illustrates acquisition frequency before and after SOX for both groups. For the treated companies, 599 of the 825 acquisitions were done pre SOX. This decreased with 62 percent to 226 after SOX was introduced. In the non-treated group, 260 acquisitions were made before SOX. This number decreased with 37 percent to 165 post SOX. The treated companies thus experienced a relatively larger drop in acquisitions after SOX than the non-treated group. Another remarkable observation is that post SOX, relatively more acquisitions were done by overconfident managers in the non-treated- than in the treated group; 32 percent compared to 21 percent respectively.
Finally, the table compares the decrease in acquisitions from overconfident CEOs in the treated- and non-treated group after the introduction of SOX. Overconfident managers from the treated group did 78 percent less acquisitions post SOX, whereas rational managers decreased the number of acquisitions with 60 percent. So while SOX led to a substantial reduction in acquisitive behavior for overconfident CEOs for both groups, the treated group experienced a larger decrease than the non-treated group
21
5. Results
5.1 Regression Results Model 1
Probit Probit Probit Probit Probit Probit Probit
(1) (2) (3) (4) (5) (6) (7) Size 0,2964 0,3644 0,3647 0,3614 0,3804 0,3790 (8,55)*** (10,15)**** (10,12)*** (10,00)** (10,30)*** (10,22)*** Cashflow 0,0941 0,1898 0,1868 0,0602 0,2302 0,0873 (2,08)** (4,06)*** (4,03)*** (1,20) (4,91)*** (1,69)** Tobin's Q 0,1166 0,1140 0,1075 0,1032 0,078 0,0788 (4,06)*** (3,96)*** (3,65)*** (3,46)*** (2,54)*** (2,59)*** Options Vested 0,0002 0,0002 0,0002 0,1617 0,0003 0,0002 (2,12)** (1,64)* (1,73)* (-1,31) (2,67)*** (2,01)** Overconf 0,3315 0,2953 0,1808 0,1562 0,1487 0,2614 0,1372 (8,60)*** (6,91)*** (4,10)*** (3,24)*** (3,09)*** (2,95)*** (2,79)*** SOX -0,4304 -0,5151 -0,5248 -1,1230 -1,1403 (-6,65)*** (-7,11)*** (-7,25)*** (-6,25)*** (-6,15)*** Treat 0,1155 0,1033 0,1309 0,1378 0,1011 (1,80)* (1,51) (1,93)** -1,07 (1,47) SOXTreat -0,1748 -0,0638 -0,0675 -0,1249 -0,0612 (-2,19)** (-0,71) (-0,75) (-0,63) (-0,67) ConfSOX 0,3014 0,3319 0,0359 0,2535 (2,88)*** (3,18)*** (2,04)** (2,36)** ConfTreat 0,0441 0,0214 0,2023 0,0435 (0,62) (0,30) (0,24) (0,60) ConfSOXTreat -0,4271 -0,4045 -0,3736 -0,3485 (-2,75)*** (-2,62)*** (-2,36)*** (-2,21)**
Industry Fixed Effects no no no no yes no yes
Year Fixed Effects no no no no no yes yes
Observations 17.471 14.728 14.728 14.728 14.728 14.728 14.728
Table 5: Regression output, dependent variable = Acquisition
Table 5 contains the estimation results of Equation 1. The coefficients shown are marginal effects. This means that a unit increase in the dependent variable X leads to a change equal to the coefficient of the dependent variable in Y. However, for interaction variables in a non-linear regression such as Probit, this is not the case. Therefore, only In column 1, the only independent variable used is the dummy variable for Overconfidence. In regression 1 until 7, the dependent variable is binary, where 1 indicates a successful acquisition performed by a firm in the fiscal year. Column one shows the estimation on the overconfidence measure dummy. The coefficient is positive and strongly significant. The coefficient is 0,33, which means that overconfident managers
22 are roughly 33 percent more likely to perform an acquisition than a rational CEO. However, no control variables are used, so that this is probably an overestimation of the effect.
Column 2 adds several control variables that are standard in the acquisition literature
according to Malmendier and Tate (2005). Firstly, size is added and has a positive, significant effect on acquisitive behavior. That means that firms with more assets at the beginning of the fiscal year are more likely to perform an acquisition. Cashflow also has a positive, significant marginal effect on acquisitive behavior. So in the sample, firms with larger cashflows at the previous fiscal year, are more likely to do an acquisition. A higher cashflow means more funds are available to the firm. This can lead them to perform more acquisitions (Jansen, 1988). The coefficient for Tobin’s Q is also significantly positive. Tobin’s Q measures the ratio of the market value of assets to the book value. Firms with higher Tobin’s Q have a positive effect on acquisitions in this sample. This result is in line with earlier research done by Doukas (1). He finds that firms with high tobin’s Q are more likely to have positive gains from acquisitions. So it is reasonable to assume that these companies would engage in more acquisitions. Finally, the number of vested options held by the CEO is significant and positively related to acquisitions, although the effect is small The same result is found by
Malmendier and Tate (2005).
In column 3, the binary variables for the treated companies and the introduction of SOX are included. Treat equals one if the observation is from a company in the treated group. The binary variable SOX equals one if the observation is done in 2002 or later. Furthermore, the interaction variable SOXTreat is the DiD-estimator. So this measures the effect of SOX on acquisitive behavior in the treated companies, for both overconfident- and rational CEOs. The coefficient for SOX is
significantly negative. This implies that after SOX is introduced, acquisitions decreased. Because the interaction variable SOXTreat is included, SOX only measures the effect on companies that are not treated. So for firms with good corporate governance, the introduction of SOX still has a restraining effect on acquisitive behavior. The coefficient of Treat estimates the effect of a company being in the treated group prior to SOX. So the negative and significant coefficient implies that before SOX, companies with worse corporate governance performed more acquisitions than companies with better corporate governance. This result is supported by the literature review in chapter 2. There it is suggested that companies with worse corporate governance spend their internal funds more quickly and that a large portion of these funds are spent on acquisitions (Harford et al, 2012). The DiD-estimator SOXTreat specifically estimates the effect of SOX on the treated companies, post SOX. The negative, significant coefficient implies that SOX has a negative effect on the treated companies, compared to the non-treated companies. That is, the effect of curbing acquisitions is stronger for companies with lower corporate governance. This result was expected because it was theorized that
23 SOX is aimed to improve corporate governance. Because the non-treated companies are those who had better corporate governance to begin with, the introduction of SOX should have a lower effect for them. This result is also supportive to earlier research that suggests that corporate governance and acquisitive behavior are related. Note that the DiD-estimator does not include the different effect for overconfident- relative to rational CEOs.
Column 4 includes the variables needed to estimate the different effect of SOX on
overconfident- and rational CEOs. Two singular interaction variables ConfSOX and ConfTreat are included. In addition, a double interaction variable, ConfSOXTreat, is used as a difference-in-difference-in-difference estimator (DDD). The latter variable is the variable of interest for the estimation of hypothesis 1. The interaction variable ConfSOX measures the acquisitive behavior of overconfident managers in the non-treated group, post SOX. The coefficient is significant and positive. Thus, post SOX, overconfident managers are more likely to do acquisitions. So even though SOX might have a restraining effect on acquisitions by overconfident CEOs, they still perform more than rational CEOs for companies with a lower G-Index. Another interpretation of this coefficient could be that overconfident managers increased their acquisition activity post SOX in the non-treated companies. However, table 3 shows that acquisitions largely decreased post SOX.
Furthermore, this is not supported by literature. On the contrary, literature suggests that improved corporate governance leads to less acquisitions .
ConfTreat has a positive, but non-significant sign. Therefore, no clear positive effect is found on acquisitive behavior of overconfident managers in the treated group, prior to SOX. This can either mean that overconfident CEOs did not perform significantly more acquisitions in the treated group compared to the non-treated group, or that within the treated group, overconfident
managers are not more likely to perform acquisitions than rational managers. The first explanation is more likely, because the single variable Overconf shows that there is a positive effect for
overconfidence for non-treated groups, prior to SOX. For companies with higher corporate governance, overconfident CEOs are thus more likely to participate in an acquisition than rational CEOs. Therefore it is safe to assume that there must also be a positive effect in the treated, weaker governance, groups. Literature suggests that bad corporate governance is associated with more, not less, acquisitions. This suggests that non-treated companies, even though they had better corporate governance standards before SOX, did not do a significantly better job at restraining the
overconfident CEOs from undertaking more acquisitions.
The variable ConfSOXTreat is a triple interaction variable that can be interpreted as the DDD-estimator. This variable estimates hypothesis 1 and is crucial in answering the main question of this
24 thesis. It estimates the effect of SOX on acquisitive behavior of overconfident CEO relative to rational CEOs. The sign of the coefficient is negative and significant. This means that SOX has a stronger effect of curbing acquisition activity for overconfident CEOs than for rational CEOs, as is predicted by hypothesis 1. The reason why the drop in acquisition activity of overconfident CEOs is likely to be caused by SOX, is because the effect is stronger for treated-, than non-treated companies. These groups were separated based only on their different G-Index, which is a proxy for corporate governance. Other unobserved changes that occurred after 2002 and possibly affected acquisition activity have probably impacted both groups in the same way. SOX affected both groups in different ways, and so the treated group isolates the effect from SOX. Table 4 shows another result that supports this claim. In column 3, it is shown that the DiD-estimator is negative and significant, so for all types of CEOs, treated firms experienced a decrease in acquisitions due to the introduction of SOX. In column 4, the coefficient of this DiD-estimator becomes non-significant. That is because the interpretation of the coefficient changes due to the addition of the interaction variables in column 4. The variable SOXTreat must now be interpreted as the effect of SOX on treated companies, post SOX, for rational CEOs. The fact that the coefficient went from significant to non-significant implies that there is no negative effect found from SOX on the acquisitive behavior of rational CEOs, when controlled for overconfident CEOs. This result is further evidence that SOX has a significantly larger effect on the acquisition activity of overconfident CEOs than of rational CEOs.
Column 5, 6 and 7 add industry fixed effects, year fixed effects and both year- and industry
fixed effects, respectively. Malmendier and Tate state that there is variation of acquisition activity over time and between industries. Although they capture some of the significance of ConfSOXTreat when both added, the coefficient stays significant at the 5 percent level. The change in effect of other coefficients is negligible, so the results found in column 1 until 4 still apply when controlling for year- and industry variation of acquisitive behavior. Column 7 lists the final estimation results of the dependent variables, controlled for industry- and fiscal year variations. The control variables Size, Cash flow, Tobin’s Q and Options Vested are all significant and have a positive effect.
Furthermore, overconfident CEOs have an increased probability of performing an acquisitions in the sample for non-treated firms prior to SOX. The DiD-estimator SOXTreat is non-significant when controlled for the effect of SOX on overconfident CEOs, which implies that there is no effect found for SOX on rational CEOs. The DDD-estimator is significant and negative, which means that SOX has a significant, negative effect on the acquisitive behavior of overconfident CEOs.
25
5.2 Regression Results Model 2
Probit Probit Probit Probit
(1) (2) (3) (4) Size 0,2815 0,3063 0,3060 (2,89)** (3,10)** (3,10)** Cashflow -0,6226 -0,0068 -0,0043 (-2,15)** (-0,04) (-0,03) Tobin's Q 0,0017 -0,0101 0,0066 (0,02) (-0,13) -0,01 Options Vested 0,0002 0,0002 0,0002 (0,53) (0,68) (0,70) Overconf 0,3037 0,3404 0,3045 0,3248 (1,88)* (1,86)* (1,68)* (1,75)* SOX -0,1426 -0,2626 (-0,70) (-1,06) Treat 0,2336 0,2605 (1,46) (1,47)* SOXTreat -0,1766 0,0665 (-0,70) (0,22) ConfSOX 0,3239 (0,97) ConfTreat -0,0497 (-0,29) ConfSOXTreat -0,8405 (-1,61) Industry Fixed
Effects yes yes yes yes
Observations 1.250 1.250 1.250 1.250
Table 6: Regression output, dependent variable = Diversifying
Table 6 shows the result of the regression output, when the dependent variable is the binary variable Diversifying, which equals 1 if the acquiring company is in a different Industry group than the target company of an acquisition. In column 1, Only the overconfidence measure is used as an independent variable. The effect is positive and significant, which implies that overconfident managers performed more diversifying acquisitions. Malmendier and Tate argue that diversifying acquisitions are a proxy for value destroying acquisitions. The positive coefficient means that in the sample, overconfident managers performed more value destroying acquisitions, which is in in line with the results found in earlier research (Malmendier & Tate, 2003). In column 2, the control variables are added. Only size is significant and positive, which means that bigger firms performed more diversifying acquisitions in the sample. Column 3 and 4 add the variables for the effect of SOX in general and for different types of CEOs respectively. The coefficient for SOXTreat is
non-26 significant, so no effect is found for SOX on diversifying mergers. The coefficient of ConfSOXTreat is negative, but lacks significance, with a p-value of 0.12. There is thus no significant effect found for SOX on the performance of acquisitions done by overconfident CEOs. However, this does not mean that no such effect exists. Diversifying acquisitions is used as a proxy for performance. This means that in this regression, many proxies are used which might cause noisy estimations of the real effect. Besides, it might be that after SOX, acquisitions are better evaluated by the board of directors, so that diversifying acquisitions are less associated with negative effects for firm value. If this is the case, the performance of acquisitions still increased, even though no significant effect is found for diversification.
27
6. Robustness Checks
This thesis studies the effect of SOX on overconfident CEOs. In the empirical research, some assumptions are made that can affect the results found in table 5. First of all, the overconfidence measure is crucial. The dummy variable for overconfidence is equal to 1 if the proxy for
overconfidence is greater than 1.80, because this was the top 25th percentile of the overconfidence
measure in this sample. The table below checks the robustness for varying levels of overconfidence.
Moneyness Coefficient P>|z| 130% -0,0993 0,469 140% 0,3149 0,234 150% -0,1770 0,205 160% -0,2645 0,081 170% -0,2301 0,115 180% -0,4300 0,007 190% -0,3476 0,042 200% -0,3326 0,052 210% -0,2589 0,100 220% -0,2250 0,232 230% -0,2110 0,271
Table 7: coefficients of ConfTreatSOX for different measures of money-ness
This table shows the probit regression estimation coefficients of the DDD-estimator ConfSOXTreat for different values of the overconfidence measure. That is, the different values for which the dummy variable Overconf equals 1. The money-ness is calculated as the fiscal year closing stock price divided by the average strike price of stock options. So money-ness of 180 percent means that the stock price is 1.8 times as high as the average strike price.. In the next column, the p-value is shown. The lower this value, the more significant the coefficient of the DDD-estimator is. This coefficient estimates the effect of SOX for overconfident CEOs for different levels of overconfidence. Firstly, it is shown that for lower levels of overconfidence, the effect of SOX is insignificant. For CEOs who hold options lower than 160 percent in the money on average, there is no effect found from SOX. The higher the money-ness, the higher the value and significance of the coefficient. Then, at a certain point of high levels of money-ness, starting at 220 percent, the significance drops again and continues to do so as money-ness increases.
The conclusions from this table are twofold. First, it is an indication that the average money-ness of stock option holdings is a decent proxy for overconfidence, even if this value is
overestimated due to the assumption of all options being in the money. Higher levels of money-ness are associated with higher significance of the effect, as is predicted by economic theory. The
28 percent. The second implication of the table is that the effect of SOX on overconfident CEOs is non-monotonic. It increases from low, to moderate levels of money-ness, after which it decreases for higher levels. This can be an indication that, for high levels of overconfidence, the introduction of SOX did not have a significant effect. Therefore, there might be an optimal level of overconfidence for which the effect of SOX is strongest. Perhaps, because CEOs that are overly overconfident are less affected by the changes in corporate governance. This could be interpreted in two ways. It could either mean that CEOS with high levels of overconfidence continue to do more acquisitions than moderately overconfident CEOs after SOX. Possibly because the policy changes are not enough to curb their acquisition activity. On the other hand, it can imply that highly overconfident CEOs did less acquisitions before SOX than the moderately overconfident CEOs. It is possible that CEOs who are more overconfident, are more known to be overconfident and thus monitored better, even for firms with lower corporate governance. For moderate levels of overconfidence, this is perhaps less known and they could have more ease doing acquisitions.
Probit Probit Probit Probit
(1) (2) (3) (4) Size 0,3352 0,3513 0,4547 (3,71)*** (3,86)*** (3,87)*** Cashflow -0,0061 0,0031 0,0079 (-0,05) (0,03) (0,06) Tobin's Q 0,2802 0,0271 0,0362 (0,46) (0,44) (0,58) Options Vested 0,0006 0,0006 0,0006 (2,62)*** (2,64)** (2,57)*** TopConf 0,1011 -0,0754 -0,0827 -0,0671 (1,41) (-0,71) (-0,438) (-0,62) SOX -0,5304 -0,4164 (-1,09) (-0,81) Treat -0,0602 0,0781 (-0,42) (0,41) SOXTreat -0,4427 -0,7055 (-2,10)** (-2,02)** TopConfSOX 0,3014 (2,88)*** TopConfTreat 0,0441 (0,62) TopConfSOXTreat 0,3581 (0,93)
Industry Fixed Effects yes yes yes yes
Year Fixed Effects yes yes yes yes
Observations 3.513 3.513 3.513 3.513
