CEO overconfidence and risk-taking behavior

CEO Overconfidence and Risk-Taking

Behavior

Master’s Thesis 2016

Marco van der Lee 10667636

2

Statement of Originality

This document is written by student Marco van der Lee 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

This thesis explored the relationship between CEO overconfidence and risk-taking behavior of a firm. Overconfident CEOs only see the upsides of an opportunity and fail to see the negative aspects of this opportunity. I use data of the CEOs of the S&P 500 (2002-2014) to research the overconfidence. Robust regression analysis shows that using the CEO overconfidence dummy of Malmendier and Tate (2005) that

overconfident CEOs have an impact on the risk-taking behavior of the firm they run. Additionally this paper concludes that when counting age as an interaction variable that CEO overconfidence no longer has effect on the risk-taking behavior.

4

Table of Content

1.3 Motivation and Relevance ... 6

1.4 Structure ... 7

2. Literature Review and Hypothesis ... 8

2.1 Managerial Effects ... 8

2.1.1 Upper Echelons Theory ... 8

2.1.2 Empirical evidence of Upper Echelons Theory ... 9

2.2 Overconfidence ... 10

2.3 Risk-taking behavior ... 12

2.3.1 Age and risk ... 14

2.4 CEO overconfidence and Risk ... 16

3. Method ... 18 3.1 Sample Selection ... 18 3.2 Empirical models ... 19 3.3 Control variables ... 22 4. Results ... 23 4.1 Descriptive statistics ... 23 4.2 Multivariate analysis ... 25 4.3 Robust Regression ... 29 5. Conclusion ... 33 References ... 35 Appendix ... 40

1. Introduction

Since the governance and the ownership of firms has been separated executive managers have been running the firms on a daily basis while the owners have become the residual claimants of the firms. Much of the prior literature has focused on the effects of

company practices, such as the balanced scorecard adoption in a company. The premise here was that on the basis of their characteristics, such as their degree of diversification, certain firms were more inclined than other to adapt specific practices. More recent literature has focused on the managerial effects on a company. Bertrand and Schoar (2003) showed that a significant degree of the strategical and financial decisions made can be explained by so-called manager effects. Likewise Geiger and North (2006) found that the presence of specific CFOs can impact the financial reporting style of a firm. Most of the literature that focuses on the managerial effects is based on the Upper Echelons theory (Hambrick & Mason, 1984; Hambrick, 2007).

The Upper Echelons theory suggests that the individual characteristics of a manager affect the choices they make in different situations. Building on this theory researchers for example have found that an organization can be affected by CEO experience (Custódio and Metzger, 2013; Huang, 2014; Hu and Liu, 2015; Huang, 2014). Bertrand & Schoar (2003) also looked into the question whether observable characteristics explain the differences in style across managers. They find that older CEOs on average tend to be more conservative while managers with a MBA-degree follow more aggressive strategies. This paper will focus on one specific CEO attribute, namely overconfidence.

CEOs in general are prone to becoming overconfident relative to others (Camerer and Lovallo, 1999; Kidd, 1970). This is because highly skilled people are more

overconfident than others (Camerer and Lovallo, 1999). The amount of research performed on this topic is limited but it has shown that CEO overconfidence has an effect on the organization they work for. They pursue innovation and invest more in research and development (Galasso and Simcoe, 2011; Hirschleifer et al, 2012) and give lower dividends to shareholders (Deshmukh et al., 2013).

Another stream of literature has investigated the risk-taking behavior of managers. Starting from the common assumption than human subjects are risk-averse, research has

6 found substantial variation in the degree in which subjects are more or less inclined to take risks. For example, male executives are more susceptible to risk-taking behavior than female executives (Waldron, et al., 2005). Moreover, firms are documented to promote the risk-taking behavior of their executives through the provision of stock options (Coles et al. 2006). Serfling (2014) has shown that older managers tend to be more risk averse than their younger counterparts. The research will also study whether there is interaction between age and CEO overconfidence.

No studies have, to my knowledge, examined how CEO overconfidence may increase the risk-taking behavior of the firm, as the CEO exerts much influence on the firm and has a major influence on the strategic decisions of a firm.

1.2 Research Question

This thesis will research whether CEO overconfidence has an effect on the risk-taking behavior of a firm. In order to conduct this research the following question will be investigated:

Does the overconfidence of a CEO influence the firm’s risk-taking behavior?

1.3 Motivation and Relevance

This research contributes to the recently emerged body of literature regarding

managerial effects. Bertrand and Schoar (2003) document the importance of individual managing styles in explaining the behavior of firms. This has prompted a new line of research that tries to explain which managerial characteristics specifically explain firm behavior, one such characteristic is overconfidence.

A limited amount of research has already been conducted in the area of CEO overconfidence. The topic of risk-taking behavior has for example been researched by looking at the incentives provided to managers. However little research has been conducted in establishing a connection between managerial traits and risk-taking behavior. So, my research on the relation between CEO overconfidence and risk-taking behavior adds to this limited pool of knowledge and addresses the possibility that the risk-taking behavior of firms is not only limited to the provision of stock option,

incentive compensation of executives, but also by the characteristics of the managers. From a societal point of view research on risk is important considering the recent scandals regarding the risks managers have taken which has resulted in organizations’ downfall or that organizations end up being saved by the government with tax money. Therefore a better understanding of why executives take excessive risk (from a societal point of view) is an important first step in efforts focused on mitigating excessive risk-taking behavior of firms.

1.4 Structure

The paper will be structured as followed: Section two will discuss prior literature and develop the hypothesis. Section 3 will discuss the research methodology explaining the sample selection, empirical models and the variable measurement. Section 4 will provide an overview of the empirical results and the paper will conclude with section 5 which will include the conclusion and research limitations.

8

2. Literature Review and Hypothesis

2.1 Managerial Effects

Most of the studies conducted on firm behavior have focused on the effects of company practices, yet there is little consideration for the role a manager might play in explaining the firm’s behavior. This means that under this point of view it doesn’t matter what kind of manager is in the company, all managers are homogenous and can perfectly replace each other without affecting the firm in any way. However recent literature has been focusing more on the managerial effects on an organization.

Bertrand and Schoar (2003) provide evidence that the presence of another manger can influence the behavior of a firm. They investigated this by measuring commonalities in corporate level decisions. CEOs, CFOs and other managers in top position have an effect on the investment decisions of a firm, the cash flow and the financial policies, such as dividend payout and interest coverage, of a company. Geiger and North (2006) show that CFO changes influence the discretionary accruals. Ge et al. (2011)

investigated the effects of CFOs on the accounting practices of an organization and found that the CFO style explains a significant amount of the heterogeneity in the accounting choices of a firm. The CFO style is more reflected in the accounting choices when the job demands and job discretion are high.

2.1.1 Upper Echelons Theory

The Upper Echelons theory (Hambrick & Mason, 1984; Hambrick, 2007) provides another view on firm behavior and suggests that the individual characteristics of a managers can impact the decision making process. The Upper Echelons theory suggest an idea with 2 interconnected parts: first, the managers act on their own interpretations of the situations they are facing. Second, these personal interpretations come from the managers’ experiences, values and personalities (Hambrick, 2007). This theory is built on the idea that complex or unclear situations cannot be objectively “knowable” and can only be interpreted (Mischel, 1977).

2.1.2 Empirical evidence of Upper Echelons Theory

Experience in general is a valuable thing to have, CEOs with more experience affect the firm in various ways. For instance CEOs with more industrial experience have more success when undertaking merger than CEOs who lack this experience, this is because the experienced manager is able to capture a bigger portion of the merger surplus (Custódio and Metzger, 2013). A CEO with diverse experience has easier access to external financial sources, this is because variety of experiences brings connection which makes the attraction of external financing easier (Hu and Liu, 2015). Lastly experienced CEOs increase the firm value by dispossessing divisions where they have no experience so they can focus on the industries they are specialized in (Huang, 2014).

The educational background of a CEO also makes a difference. Over the years the professional and educational background of a CEO has changed. In the early 60’s firm specific experience was important as a CEO, however this importance has dropped in favor of more general management skills and more CEOs have become MBA graduates (Frydman, 2007). Bertrand (2009) explains that the firms in recent years have more complex financial arrangements and strategies which are needed to remain competitive. The educational background of the CEOs provides the knowledge to understand the complex situations, hence CEOs have MBA degrees in order act accordingly on those complex situations. This change in knowledge also leads to an increase in externally recruited CEOs as the internal candidates lack the general management knowledge (Frydman, 2007). Bertrand and Schoar (2003) find that CEOs with an MBA

qualification exhibit greater aggressiveness in their choice of strategy, as indicated by higher investment levels, greater responsiveness to growth opportunities, smaller dividend payouts, and a greater degree of diversification.

Older CEOs are more accurate with their financial reports than their younger counterparts (Huang, Rose-Green & Lee, 2012). Also older CEOs are less likely to restate their financial statements. Finally, they also find that older CEOs are less like to beat or meet analyst earnings forecast. Murdrack (1989) finds that older CEOs are more ethical than their younger counterparts. Terpstra et al. (1993) find that younger CEOs are more prone in engaging insider trading which confirms the argument that older CEOs are more ethical. Davidson et al. (2007) show that older CEOs often cut investment in long term projects in favor of short term projects, the effect is particularly strong for

10 CEOs closing retirement which shows that they acting in their own interest in order to make as much profit before they retire. Older CEOs, once fired, have more troubles finding a similar position in another firm than their younger counterparts (Ward et al., 1995)

Adams et al. (2005) find that firms with more powerful CEOs have a more variable firm performance because of the influence the CEO has over the board. The firm performance varies from being the worst performing firms in their sample but also the best in their sample. Successors to powerful CEOs tend to have a lower level of power (Ting, 2013) and when a successor is appointed the firm performance of the firm becomes less volatile. Research on CFO effects show that when an organization hires a new CFO the discretionary accruals of that organization dropped significantly following the appointment of the new CFO (Geiger and North, 2006). These results are greater when the firm hires a CFO from outside the organization, this is because outsiders bring new perspectives to financial reporting in the country which in turn leads to new ways of reporting.

2.2 Overconfidence

As has been established in the previous paragraph the influence of a manager can affect firm behavior. The main focus of this thesis will be on the overconfidence of CEOs. According to Langer (1975) overconfidence can be described as an overestimation of one’s own abilities, this overestimation is related to events in one’s own personal situation. Weinstein (1980) stated that there are three factors that contribute to overconfidence: The illusion of control, a big intensity of commitment to positive outcomes and abstract notes that makes it difficult to compare performances of different individuals. According to Kidd (1970) executives managers are believe that they will perform better than others, which shows that they are prone to show overconfidence. This overconfidence is stronger because CEOs are highly skilled individuals (Camerer and Lovallo, 1999).

Malmendier and Tate (2005, 2008) and Malmendier et al. (2011) started the focus on CEO overconfidence and their work has led to more research in this area, they deem a manager to be overconfident when he overinvests his own funds in the company he runs. Malmendier and Tate (2008) argue that there are two types of managers: Rational

managers and overconfident managers and claim that the two have a different kind of behavior. According to Malmendier and Tate (2005) overconfident CEOs tend to

systematically overestimate their return on investments. In their research they found that overconfident CEOs increase the cash-flow and the investment sensitivity. This is because they overestimate the value of an investment and they don’t finance from external sources. They also don’t issue new shares as they believe that the market undervalues the stock and therefore they believe that shareholders will pay less for these stocks than they are worth. Since overconfident CEOs overvalue the stocks and see external financing as costly they pay lower dividends in order to gain more financing for future investments (Deshmukh et al., 2013). Overconfident CEOs also invest more in research and development, pursue innovations and obtain more patents and patent citations than their non-overconfident counterparts (Galasso and Simcoe, 2011; Hirschleifer et al, 2012). Since research and development is considered to be a risky investment (Coles et al, 2006) there is an indication that overconfident CEOs do engage is more risky behavior.

Overconfident CEOs are also more prone to engaging in mergers (Malmendier and Tate, 2008; Brown and Sarma, 2007). According to Malmendier and Tate (2008) the CEOs have a tendency to overestimate the values of synergies and their ability to generate returns. This leads to them paying more for companies than they are actually worth as they overvalue them. The market reactions to a merger is stronger with an overconfident CEO as the share value of an organizations drops more with than with their non-overconfident counterparts. This is a sign that an overconfident CEO is prone to take risky initiatives.

According to Chen, Crossland and Luo (2015) overconfident CEOs are less responsive to corrective feedback as opposed to their non-overconfident counterparts. Overconfident CEOs follow their own way and tend to make the same mistakes as before since they do not listen to other people’s feedback and they believe they know it better. This behavior leads to more risk for the firm they lead as repeated mistakes have a negative effect on the firm. Yilmaz (2010) shows that when a firm already has an overconfident CEO the firm tends to choose another overconfident CEO as a successor more often than a non-overconfident successor. Hiller and Hambrick (2005) have concluded that overconfident CEOs tend to make decisions faster, less comprehensive and more centralized. The CEOs believe that what do is correct and therefore make the

12 decision based more on intuition than on information.

Andriosopoulos et al. (2013) have researched the relationship between CEO overconfidence and share buyback completion rates. Their research conclude that CEO characteristics serves a strong signal for a firm to intend to complete a buyback program. Overconfident CEOs in particular have shown to have a tendency to complete buyback programs. After a completed buyback program these firms are likely to conduct another buyback program. They also concluded that overconfident CEOs disclose more

information to the public.

Qin, Mohan & Kuang (2015) investigated the relationship between CEO overconfidence and cost stickiness. Their results show that firms with overconfident CEOs have greater cost stickiness compared to others. This is because overconfident CEOs have an optimistic bias and believe that they are capable to restore sales when the situation occurs that sales decline. As a result of this the overconfident CEOs keep excessive administrative, selling and general resources when which in turn leads to greater cost stickiness.

Hribar and Yang (2016) show that overconfident CEOs affects management forecasts. First, overconfident CEOs are more likely to issue a forecast as they are optimistic about the forecast. They are optimistic about future performance and are willing to commit to it by issuing a forecast (Libby & Rennekamp, 2012). Secondly, overconfident CEOs, when issuing a forecast, will be positively biased in their forecast. This is in line with the findings of Malmendier and Tate (2005, 2008). Lastly,

overconfident CEOs are narrower with their predictions compared to others. This is because they believe that they can forecast the future with greater precision.

2.3 Risk-taking behavior

Risk is the chance that an investment’s actual return will be different than expected, the bigger the difference is from the expectation the bigger the risk will be. For example 2 investments both have an expected value of €1.000: Investment A has a 50% chance of €1.500 profit and 50% chance of having €500 profit, investment B has a 50% chance of €3.000 profit and 50% chance of having a €1.000 loss. The dispersion of the outcome relative to the expected return is much greater for Investment B than investment A’s and that is risk. Taking risks can be rewarding but at the same time these risk can have opposite effects as well. There a different ways to look at risk. On one hand one must

take risk in order to increase the value (Carey and Stulz, 2005). On the other hand risk while risk can improve the financial situation it can damage its value (Merton, 1993). There are certain types of people who are more willing to take risk than other. The risk preference of a manager affects the firm in various ways, for example the financial policies of firms are influenced by the risk preference of both the CFO and the CEO. A CEO’s risk-preference affects the leverage and cash-holding policies while the CFO’s risk preferences affect the leverage and accrual decisions (Chava and Purnanandam, 2010). Firms may provide specific equity based compensation in order to increase the risk appetite of a manager. Specifically, firms may provide stock options as a means of increasing the risk-appetite of managers as the managers will enjoy the upside potential while they face no downward potential. Coles et al. (2006) show that managerial compensation provides a strong incentive to take risk, specifically when it comes to the investment policies, such as research and development and properties, and debt of an organization.. The act of rewarding managers with equity based pay incentives to take risks is a popular practice, however this has recently come under fire as the managers take excessive risk to obtain the incentives with dire consequences such as the scandals with WorldCom, Enron and others. However the risk-taking behavior of the managers cannot be fully explained by the incentives-individual characteristics and traits also play a role in this risk-taking behavior.

The performance of a manager explains some of the risk-taking behavior. An underperforming manager is more likely to increase the relative risk of a company as their performance affect the firm in a negative way. Therefore because of this they are more likely to be fired. The lower the risk of employment becomes the lower the relative risk becomes. (Hu et al., 2011). This can be explained through the fear of losing one’s job which makes a person more willing to take risk in order to maintain their job. Female managers are more risk conservative than their male counterparts (Mallin & Farag, 2016). Female asset-managers consider themselves to be more risk-averse than their male counterparts and the female manager possess a more passive portfolio than their male counterparts. The female managers do not strive to be the ultimate top performer and therefore do not engage in risky situations (Rebeggiani, 2007). Ammann and Verhoven (2009) show that when a fund manager does well in the first half of a year the manager will take more risks in the second half of the year. Inexperienced manager take more risk than their experienced counterparts. According to Menkhoff et al. (2006)

14 fund managers that are inexperienced take more risk than the experienced

fund-managers, this also leads to the conclusion that inexperienced managers yield higher returns as well. Recent literature has shown that managers are willing to misreport when they have invested in the firm they run. This happens because a negative report would affect the company’s stock and with it the manager’s investment (Armstrong et al., 2013). This behavior is not restricted to investment but also to stock options. This misreporting happens in order to prevent a negative impact in the mangers’ personal wealth. Managerial risk preferences can also be influenced by the job security and tenure (Berger, Ofek & Yermack, 1997). CEOs who have been in the firm for a longer time show less risk as the leverage for the firms controlled by the CEO is lower. A manager who has low job security show higher leverage for the firm he runs. Wright et al. explain that inside ownership also plays a role in the risk-taking behavior of a company. The higher level of inside ownership the more risk the manager is willing to take in order to gain more profit from these shares. However when there is a point where the level of inside ownership is too high and the managers are less willing to take risks, this effect is called ownership concentration. Pedersen and Thomsen (2003) state that ownership concentration will affect the risk appetite since large shareholders have a less diverse portfolio and have therefore spread their risk less. This in turn will make these shareholders pursue lower risk projects in order retain lower risk in their portfolio.

2.3.1 Age and risk

The age of a person also has influence on the decision making process of a person, also on the risk of that process. Prior literature show mixed results on how people with different ages react to risk. Part of the prior literature that show that younger managers are less risk averse than their older counterparts. Younger CEOs face career concerns hence they are less prone in engaging in risky projects. (Hirshleifer and Thakor, 1992). Since these CEOs do not have reputation as a good manager the backlash of bad decisions will affect them more as it can decrease the chance that they will get a manager position in the future. (Zwiebel, 1995). Hong et al. (2000) state in their research that younger analysts are more like to follow other people’s predictions as they are punished more harshly when they make a mistake. By not deviating from the general forecast they take less risk by not

showing a bold forecast. Avery and Chevalier (1999) find that younger CEOs are less prone to taking risk as they are less secure about their executorial skills. Therefore the young CEOs are more afraid of making mistakes. According the findings of March and Shapira (1987) older CEOs have the confidence to engage in bold initiatives since they have the experience to do so.

On the other hand you have literature that show that younger managers show a tendency to take more risk than their older counterparts by engaging into riskier investments. Orens & Reheul (2013) claim that older CEOs are more risk averse than younger CEOs. Therefore, in line with the agency theory (Jensen & Meckling, 1976) the older CEOs have a tendency to engage in corporate decisions which are not aligned with the interests of shareholders.

Prendergast and Stole (1996) show that younger managers try to signal the market that they are managers with better abilities compared to other managers. They do this by pursuing more risky and more aggressive investments. They also claim that younger managers exaggerate their talents towards other in order to look more talented. Older managers are not willing to change their investment strategies as by doing so they might show that their previous investment decisions were not correct. Also according to Beber and Fabri (2012) younger CEOs engage in more short term objectives in order to try and build a good reputation, therefore they take more risks.

Hambrick and Mason (1984) show 3 different reason on why younger managers are prone to more risk-taking behavior. First, older managers have reached a point in their lives where they focus on stability, therefore they focus on having a stable financial situation and they focus on job security. Engaging in risky projects could jeopardize this stability. Secondly, older managers focus more on maintaining the status quo of the firm. Risky projects can affect the firm in a negative or positive way and therefore change the status quo. Lastly, older managers have less stamina, both mentally and physically. Because of that they are less able to learn new behavior and create new ideas. In turn they are less able to take and more focused on maintaining balance. Taylor (1975) show that older managers take more time to make a decision. This is because they first gather information and then evaluate this information in-depth. According to Serfling (2006) older manager are more risk-averse than their younger counterparts. Having an older CEO in charge in a firm results in lower stock return volatility, which means they take less risk. Furthermore older CEOs engage in less risky investment policies. They invest less in

16 research and development (Barker & Mueller, 2013), they manage firms with more

diversified operations and make more diverse acquisitions therefore lowering the risk by spreading it. Older CEOs maintain a lower operating leverage. They also show that when the next most influential manager after the CEO is old as well that the risk is even lower. When a young CEO’s next most influential manager is also young the risk that a firm takes is higher. Graham, Harvey & Puri (2013) find that younger CEOs are more risk-tolerant opposed to their older counterparts. The younger CEOs usually run fast growth companies since these companies engage in risky projects in order to grow fast. Yim (2013) finds that younger CEOs engage in corporate acquisitions more often.

2.4 CEO overconfidence and Risk

It has been established that CEO overconfidence and managerial traits in general have an impact on firm behavior, the question remains whether CEO overconfidence affects the risk-taking behavior of a firm and if so, how. An overconfident person tends to

overweight the positive aspects of certain situations and underweight the negative aspects. Therefore when a CEO has an opportunity the CEO will mainly focus on the positive aspects of this opportunity and neglect the negative aspects it brings or in other words the CEO will fail to see the risk associated with the decision. In sum,

overconfidence may lead CEOs to adopt riskier projects as they on the one hand under weigh or complete ignore the lower and possibly even negative outcomes and on the other hand overweigh the more positive outcomes associated with investment projects. Because of this overconfidence, CEOs may, in some cases unintentionally, increase the risk a firm takes.

The greater risk that an overconfident manager may take can further be increased if they are awarded equity incentives as a means for interest alignment. Malmendier and Tate (2005, 2008) and Malmendier et al. (2011) consider managers to be overconfident when they overinvests in their own company, this means that they will try to gain profit from the overinvestment through their investment strategies. While according to the agency theory (Jensen and Meckling, 1976) equity incentives are intended to align the interests of the principals with the managers’ goals, they now can also prompt managers to take more risks in order to get more out of their personal investments.

risk-taking behavior.As established by Galasso & Simcoe (2011) and Hirschleifer et al (2012) Overconfident CEOs invest more in research and development. Research and development expenditures are typically viewed as an investment with a higher rate of risk compared to capital expenditures such as property plant and equipment (Bhagat and Welch, 1995). Menkhoff et al (2006) theorizes that young fund managers engage in high risk projects because of a higher degree of overconfidence, less herding behavior, or a lower degree of risk aversion.

In sum, I argue that overconfident CEOs will take more risk than their non-overconfident counterparts. Knowingly they will increase the risk in order to maximize their overinvestments, but unknowingly because they systematically underestimate dangers and overestimate their ability to control the outcomes of future event.

Overconfident CEOs will engage in risky projects because they believe that there is less risk than there actually is because their judgment is clouded by their overconfidence. Alongside the dummy used by Malmendier and Tate I will also scale the overconfidence data in order took at CEO overconfidence using all stock option data opposed to only the highest variable. To conclude the following hypothesis will be tested:

- H1a: CEO overconfidence measured by dummy has a positive effect and the risk that a company takes

- H1b: CEO overconfidence measured by scale has a positive effect and the risk that a company takes

As age has proven to be have an effect on the risk-taking behavior of a person since older people are more risk-averse than younger people with the measures of risk in this research, I also plan to research the overconfidence of a CEO while using the age as an interaction variable. Using age as an interaction variable I believe that the

overconfidence will have an effect on the risk-taking behavior that a firm takes. For this I will use the following hypothesis.

-H2a: CEO overconfidence measured by dummy and the age of a CEO have a positive effect and the risk that a company takes

H2b: CEO overconfidence measured by scale and the age of a CEO have a positive effect and the risk that a company takes

18

3. Method

3.1 Sample Selection

The initial sample will be retrieved from the ExcecuComp database and will afterwards be filtered on the S&P 500 firms. The sample will cover the period 2012-2014 as that is the latest data available. The initial sample taken from ExcecuComp contained 70160 observations, however a lot of these observations were observations that had managers who were not CEOs and /or were not CEO of an S&P 500 firm, removing these observations has led to the removal of 62355 observations. Furthermore there were observations that were missing data on the incentives of CEOs, 2458 lacked these data and were therefore removed from the sample. Firms within the finance industry (SIC code between 6000 and 6999) and firms from utilities (SIC code between 4990 and 4999) were removed as well as they are heavily subjected to regulations and therefore the manager might manage the earnings differently. (Cheng & Warfield, 2005). The removal of these firms led to an elimination of 1371 observations. The annual

fundamentals were taken from Compustat and the monthly stock data to calculate the stock return volatility was obtained from CRSP as the amount of firms with daily data available was too low to use for this research. Missing data on these subjects has led a removal of 1709 observations. In the end the various stock option packages were merged for each CEO in a way that every CEO has one combined package per fiscal year. This merger turned 1691 stock option package observation to 317 observation. Finally 2 observations were dropped for being outliers. The final sample which will be used in this research consists of 315 observations

Table 1: Sample Selection Number of Observations

ExcecuComp sample 70160

Minus: Observation that were not CEO’s at that time 51755 Minus: Observations that were not S&P 500 10700

Minus: Observations with missing data on incentive data 2458

Minus: Observations from SIC 6000-6999 and 4900-4999 1371

Minus: Observations missing data on fundamentals

and observations missing data on stock options 1724 Minus: Observation merged on creation longholder value 1374

Minus: Outliers 2

Final sample 315

3.2 Empirical models

In order to measure CEO overconfidence I will use the longholder indicator as used by Malmendier and Tate (2005). If a CEO holds an option till the expiration date at least once and if the option is at least 40% in the money then he is considered to be a longholder. The 40% threshold is based on the model of Hall and Murphy (2002). A rational manager would exercise the option at an earlier stage in order to avoid downward risks. An overconfident CEO believes he can still raise the value of the options and will therefore not exercise them in order to increase the value of the options. This is a noisy indicator as CEOs with an option close to expiration date and 39% in the money can be overconfident as well since the option might have exceeded the 40% threshold. Even though they might be overconfident in this thesis I will still consider them to be rational. The dummy used to represent the longholder variable is called LHDUMMY

In addition to the model Malmendier and Tate (2005) provide I will a scaled average longholder value. In order to create this I will calculate the overconfidence for each package just as Malmendier and Tate did, however instead of considering a CEO overconfident when a CEO is 40% in the money I will take all option packages for each fiscal year and weigh them using the market value of the total assets as used by

Ertugrul, Sezer and Srimans (2008). For example a CEO has 3 different option packages one consists of 1000 options which is 5% in the money on consists of 500 options

20 which are 2 % out of the money and one package of 50 options which is 45% in the

money. Using Malmendier and Tate’s dummy this CEO will be considered overconfident due to the 45% package. However with the scaled variant the CEO overconfidence ratio is considered to be 4.03% in the money as it gives a greater weight to the larger packages.

This method will create a scaled average value of stock option and will nullify the situation where the CEO has a big package that is not over 40% in the money and a small package that is over 40% in the money as the big option package has a larger weight . This also creates a scale of overconfidence as opposed to a dummy where the CEO is either overconfident or not which should provide more accurate results as the noisiness in the dummy is reduced There is however noise involved in this calculation as the variable includes stock option packages which are not in the money; the question is whether these packages have influence on the decision making behavior of a CEO. The variable used to represent this way of calculating CEO overconfidence is

LHSCALE.

In order to measure the risk-taking behavior of a firm I will use various proxies. The first measure focuses on volatility of stock returns (VOLATILITY) consistent with Core and Guay (1999). Also, as suggested by Serfling (1999), I will use research and development costs as a proxy for risk-taking behavior, this will be measured using the annual research and development expenses scaled by sales (R&D). Finally I will use Altman’s Z-score (Altman, 1968) to calculate the probability that the firm will go bankrupt. The lower this is the more risk there is within this firm. The Z-score is calculated as follows:

Z = 1.2X1 + 1.4X2 + 3.3X3 + 0.6X4 + 1.0X5.

Where

X1= Working capital / Total assets X2= Retained Earnings / Total assets

X3= Earnings before interest and taxes / Total assets X4= Market value equity / Book value of total debt X5= Sales / Total assets

Since a lower value of Z_SCORE shows a higher degree of risk I created an alternate variable Z_SCORET which is Z_SCORET = -1 * Z_SCORE so that a higher value shows more risk while the same values and differences remain. This way the Z-Score fits the empirical model as for all risk proxies a higher number indicates a higher degree of risk. The first hypothesis of this research is tested with the following empirical models: H1a: RISK_TAKING PROXIES = β0 + β1 LHDUMMY + Controls

H1b: RISK_TAKING PROXIES = β0 + β1 LHSCALE + Controls Where:

LHSCALE = The overconfidence of a CEO scaled by sales

LHDUMMY = The overconfidence of CEO where it equals 1 if the CEO is overconfident and equals 0 is its not

RISK_TAKING PROXIES = The risk taking proxies used in this research which are -Z_SCORET = Altman’s Z-score transformed

-VOLATILITY = The stock return volatility calculated on a monthly base -R&D = Research and development costs scaled by sales

The coefficient β1 gives the relationship between CEO Overconfidence and risk which is measured using the Z-score, Stock return volatility and the research and development costs. β2 Gives the relationship between overconfidence and risk using the dummy as used by Malmendier and Tate (2005). On basis of my hypothesis I expect that a higher degree of overconfidence will result in more risk, therefore I expect that β1+ β2>0 as for all the risk proxies a higher value is associated with more risk.

The second hypothesis will be tested using the following empirical model H2a: Z_SCORET + VOLATILITY + R&D = β0+ β1 LHDUMMY + β2 AGE + β3

(LHDUMMY *AGE) + Controls

H2b: Z_SCORET + VOLATILITY + R&D = β0+ β1 LHSCALE + β2 AGE + β3 (LHSCALE

22 Where:

LHSCALE = The overconfidence of a CEO scaled by sales

LHDUMMY = The overconfidence of CEO where it equals 1 if the CEO is overconfident and equals 0 is its not

AGE = The age of the CEO at the end of the fiscal year.

RISK_TAKING PROXIES = The risk taking proxies used in this research which are Z_SCORET = Altman’s Z-score transformed

VOLATILITY = The stock return volatility calculated on a monthly base R&D = Research and development costs scaled by sales

In this model the interaction between the CEO overconfidence measures and age will be interpreted. Based on previous research I expect that age has a negative effect on risk, therefore I believe that β2<0. Just like in my first hypothesis I will still predict that CEO overconfidence has a positive effect on risk. Finally I believe that even with age as an interaction variable that CEO overconfidence has a positive effect on the risk a firm takes.

3.3 Control variables

In this model several variables will be used in both empirical models. Consistent with prior literature performance, firm size, growth and leverage will be used as control variables as they have an effect on CEO compensation and/or risk. Performance will be controlled for by using the variable PERFOR. This is measured by calculating the yearly ROA, this is added as compensation is usually linked to the performance of a person (Core et al., 1999). To account for firm size the control variable SIZE will be used, which is the natural

logarithm of assets; prior literature shows that bigger firms need more talented people which in turn require more compensation (Core et al., 1999). GROWTH OPP represents the growth opportunities of a firm and is calculated by dividing the market value of equity by the book value of equity, this is added as control variable as prior literature shows that growth has a positive association with CEO compensation (Smith and Watts, 1992). LEVERAGE is calculated by dividing the total liabilities by the book value of equity since the compensation policy plays a role in dealing with financial distress (Kostiuk, 1990). Lastly one control variable for managerial characteristics will be added in the form of

GENDER. The control variable GENDER is equal to 1 if the gender is male and equal to 0 if the gender is female.

4. Results

4.1 Descriptive statistics

Table 2 contains the descriptive statistics for the full sample. Panel A shows the statistics for the dependent and independent variables of this research. On a weighted average the stock option packages of CEO’s are not in the money as they are valued at 70% of their price on average. However 43 of the 315 (13.7%) observations are deemed overconfident if you look at a CEO’s highest valued stock option package. Comparing this to the sample that Malmendier and Tate used (2008) you will see a decrease in overconfident CEOs as their sample (1984-1994) is 21.2% overconfident. This low variable can be explained since the S&P 500 index has suffered a decrease in value after January 2008. The firms in this sample spend 6.5% of their sales on research and

development. The average transformed Z-score is -3.961 which translates to a Z-score of 3.961 which is the safe zone, however at least 10% of the firms are in the gray zone which starts at a value of 2.6. The average volatility of the firms is 0.058. The average age of the CEO in this sample is 55 years

Panel B shows the descriptive statistics for the control variables used in this research. The size of the company (ln assets) has an average value of 9.511 which means that the average size of the sample is about 38 billion dollars. The average performance of the sample is calculated through the ROA which is 7.8%. The Growth, calculated with the book to market ratio, is on average 34.9% in the sample and the average leverage, calculated through the debt/equity ratio, is 2.075. Finally 7.3% of the CEOs in this sample is female.

24 Table 2

Descriptive statistics

N Mean Median Sd P10 P25 P75 P90

Panel A: (In)dependent Variables

LHSCALE 315 0,677 0,636 0,326 0,382 0,497 0,783 0,971

AGE 315 55.14 55 5.56 48 52 59 62

R&D 315 0,065 0,031 0,073 0,005 0,015 0,085 0,190 Z_SCORET 315 -3.961 -3.669 2.055 -5.686 -4.585 -2.784 -2.048 VOLATILITY 315 0,058 0,050 0,027 0,034 0,041 0,066 0,092

Panel B: Control Variables

SIZE 315 9,511 9,474 1,116 8,083 8,613 10,33 11,07 PERFOR 315 0,078 0,078 0,042 0,029 0,052 0,104 0,131 GROWTH OPP 315 0,349 0,303 0,223 0,153 0,206 0,421 0,606 LEVERAGE 315 2,075 1,301 3,011 0,578 0,904 2,037 4,016

Table 3 contains the Pearson correlation matrix for this research’s sample, bold correlations are significant at 10% level. The variables for CEO overconfidence are closely related to each other, this makes sense as both variables essentially use the same data, however the dummy uses the highest value and the scale uses the weighted

average. Age is negatively related to risk and overconfidence which is in line with the hypothesis as a higher age is associated with less risk opposed to overconfidence. The risk proxies all have a positive relation with the overconfidence proxies which is in line with the empirical model.

Table 3

Pearson Correlation Matrix

LHDUMMY LHSCALE AGE R&D VOLATILITY Z_SCORET

LHDUMMY 1 LHSCALE 0,5224 1 AGE -0,0202 -0,0184 1 R&D 0,0907 0,0938 -0,0174 1 VOLATILITY 0,1900 0,1661 -0,0955 0,1084 1 Z_SCORET 0,2404 0,2758 -0,0982 -0,2075 -0.0723 1

4.2 Multivariate analysis

Table 4 contains the results of the multivariate regression for the first hypothesis. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. the results are divided in 2 panels in order to research if a scaled CEO overconfidence variable shows different results. Panel A shows the results for the dummy analysis. The results are not consistent with the first hypothesis. The dummy analysis reports that overconfident CEOs are only positively associated with higher research and

development costs (p<0.01), this is in line with the findings of Deshmukh et al (2013) who also reported that overconfident CEOs have higher research and development costs. For the Volatility of stock returns and Altman’s Z-score the model does not show that CEO overconfidence has a significant effect on those.

Panel B shows the results for the scaled analysis. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Unlike the dummy, this variable shows to have a positive effect on R&D (p<0.01) and a positive effect on ZSCORET (p<0.05), which means that a higher degree of overconfidence has a relation to a high Z-score. The variable shows no significant relation with the volatility of stock returns.

Looking at the control variables we can see that Age has only a negative effect on the Z-score. Firm size has an effect on all risk proxies. The effect is negative on (R&D p<0.05), p<0.01, Volatility p<0.01) and positive on Z_SCORET. The gender variable is not significantly associated with any of the risk proxies which is not consistent with the findings of Rebeggiani (2007), however this sample does not contain many female CEOs. Performance only has a negative effect on the transformed Z-score (p<0.01) and leverage affects the research and development costs negatively (p<0.01) and the

transformed Z-score positively (p<0.01). Finally growth opportunities has a negative effect the R&D (p<0.01) and a positive effect volatility (p<0.01).

The models as a whole are highly significant as shown by the F-value which all have p-values below 1%. The dummy model explains about 12% of the R&D variance, 38% of the transformed Z-score variance and 14% of the stock return volatility

variance. The scaled model explains about 13% of the R&D variance, 38% of the transformed Z-score variance and 14% of the stock return volatility variance.

26 Table 4

Multivariate Regression H1 Panel A: Dummy analysis

R&D Z_SCORET VOLATILITY

Coef t-stat p-value Coef t-stat p-value Coef t-stat p-value

Β0 Intercept 0.17*** 3.14 0.002 -6.34*** -4.88 0.000 0,12*** 5.81 0.000 β1 LHDUMMY 0.04*** 3,37 0.001 0.42 1.42 0.155 0.01 1,64 0.103 AGE -0.00 -0.11 0.910 -0.03* -1.79 0.074 -0.00 -0.83 0.408 GENDER 0.02 1.00 0.318 0.05 0.13 0.900 -0.00 -0,16 0.875 SIZE -0.01** -2.34 0.020 0.51*** 5.75 0.000 -0.01*** -4.52 0.000 PERFOR -0.01 -0.09 0.928 -19.07*** -7.47 0.000 -0.00 -0,00 0.998 GROWTH OPP -0.06*** -3.28 0.001 0.80 1.45 0.148 0.03*** 3,84 0.000 LEVERAGE -0.17*** -4.17 0.000 0.13*** 3.72 0.000 0.00 0,46 0.644

Observations R-Square F-value P-value R&D 315 0.1187 5.909 0.000 Z_SCORET 315 0.3807 26.961 0.000 VOLATILITY 315 0.1430 7.321 0.000

Panel B: Scale analysis

R&D Z_SCORET VOLATILITY

Coef t-stat p-value Coef t-stat p-value Coef t-stat p-value

Β0 Intercept 0.14*** 2.63 0,009 -6.74** -5.19 0,000 0.11*** 5.56 0.000 β1 LHSCALE 0.05*** 3.73 0,000 0.76** 2.42 0,016 0.01 1,41 0.161 AGE -0.00 -0.08 0.939 -0.03* -1.80 0.073 -0.00 -0.81 0.421 GENDER 0.02 0.99 0,321 0.04 0.11 0.909 -0.00 -0,16 0.874 SIZE -0.01** -2.30 0,022 0.51*** 5.83 0,000 -0.01*** -4,50 0.000 PERFOR -0.02 -0.18 0,854 -19.11*** -7.54 0,000 -0.00 -0,06 0.892 GROWTH OPP -0.08*** -3.37 0,001 0.62 1.15 0,251 0.03*** 3.97 0.000 LEVERAGE -0.01*** -4.55 0,000 0.11*** 3.24 0,002 0.00 0,33 0.771

Observations R-Square F-value P-value R&D 315 0.1256 6.301 0.000 Z_SCORET 315 0.3882 27.833 0.000 VOLATILITY 315 0.1411 7.205 0.000

Table 5 contains the results of the multivariate regression for the second hypothesis. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels,

respectively. Panel A’s results show that the overconfidence dummy is not significantly associated with any of the proxies which means that the hypothesis is rejected.

Furthermore age only has a negative effect (p<0.1) on the transformed Z-score, which means that higher age is associated with a higher z-score which means less risk. Also the age of the CEO has no effect on the other risk proxies. Since AgeXLHD is not significantly associated by any of the variable it shows that overconfidence is not affected by the age of a CEO.

Panel B’s results show the multivariate regression H2b. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Panel B’s results show, in comparison to panel A’s results, that CEO overconfidence does have a positive effect on the research and development costs of a firm (p<0.05). Age has as a negative effect (p<0.1) on the Z-score transformed, which means that higher age is associated with a higher z-score which once again means less risk. The interaction between age and overconfidence is positively significant for the research and development costs (p<0.01) which that there is interaction between the two variables for the research and development costs.

Looking at the control variables we can see that firm size has an effect on all risk proxies. For the dummy variable the effect is negative on (R&D p<0.05) and Volatility p<0.01) and positive on Z_SCORET (p<0.01). For the scaled version the the firm size significantly affects the research and development costs negatively at a lower level (p<0.10). The gender variable is not significantly associated with any of the risk proxies which is once again not consistent with the findings of Rebeggiani (2007), however this sample does not contain many female CEOs. Performance only has a negative effect on the transformed Z-score (p<0.01) and leverage affects the research and development costs negatively (p<0.01) and the transformed Z-score positively (p<0.01). Finally growth opportunities has a negative effect the R&D (p<0.01) and a positive effect volatility (p<0.01). The models as a whole are highly significant as shown by the F-value which all have p-F-values below 1%.

The dummy model explains about 12% of the R&D variance, 58% of the Z-score variance and 14% of the stock return volatility variance. The scaled model explains

28 about 14% of the R&D variance, 39% of the Z-score variance and 14% of the stock

return volatility variance. Table 5

Multivariate Regression H2 Panel A: Dummy analysis

R&D Z_SCORET VOLATILITY Coef

t-stat

p-value Coef t-stat p-value Coef t-stat p-value Β0Intercept 0,17*** 3,03 0,003 -6.38*** -4.83 0,000 0,11*** 5,54 0,000 β1LHDUMMY 0.10 0,63 0,526 0.28 1,25 0,779 0,08 1,25 0,213 Β2 AGE -0,00 -0,02 0,986 -1.69* -2,01 0,091 -0,00 -2,01 0,598 Β3 AgeXLHD -0,00 -0,37 0,710 -0,17 -1,80 0,865 -0,00 -1,12 0,262 GENDER 0,02 1,00 0,320 0,12 0,09 0,901 -0,00 -0,16 0,869 SIZE -0,01** -2,30 0,022 5.74*** 5,14 0,000 -0,01*** -4,43 0,000 PERFOR -0,01 -0,13 0,899 -7.43*** -10,46 0,000 -0,00 -0,12 0,907 GROWTH OPP -0,08*** -3,27 0,001 1.36 4,69 0,173 0,03*** 3,45 0,001 LEVERAGE -0,01*** -4,18 0,000 -4.83*** 6,30 0,000 0,00 0,36 0,720

Observations R-Square F-value P-value R&D 315 0,1191 5,173 0,000 Z_SCORET 315 0,5806 52,945 0,000 VOLATILITY 315 0,1466 6,569 0,000

Panel B: Scale analysis

R&D Z_SCORET VOLATILITY

Coef t-stat p-value Coef t-stat p-value Coef t-stat p-value

Β0 Intercept 0,13 2,26 0,24 -6,25*** -4,56 0,000 0,11*** 5,21 0,000 β1LHSCALE 0,04** 2,50 0,013 0,01 0,03 0,976 0,00 0,87 0,386 Β2AGE -0,00 -0,46 0,649 -0,04** -2,26 0,025 -0,00 -0,94 0,347 Β3 AGEXLH 0,00*** 3,47 0,001 0,01** 2,51 0,013 0,00 1,36 0,173 GENDER 0,02 1,34 0,182 0,04 0,11 0,909 -0,00 -0,04 0,970 SIZE -0,01* -1,88 0,061 0,52*** 5,75 0,000 -0,01*** -4,29 0,000 PERFOR -0,01 -0,10 0,918 -19,10*** -7,52 0,000 -0,00 -0,03 0,977 GROWTH OPP -0,07*** -3,08 0,002 0,62 1,14 0,254 0,03*** 4,06 0,000 LEVERAGE -0,01*** -4,37 0,000 0,11*** 3,16 0,002 0,11*** 5,21 0,000 Observations R-Square F-value P-value

R&D 315 0,1416 6,310 0,000 Z_SCORET 315 0,3893 24,383 0,000 VOLATILITY 315 0,1435 6,407 0,000

4.3 Robust Regression

The robust regression investigates the sensitivity of the results by giving each variable a weight depending on how they behave. If an observation is less likely to be an outlier the observation will be given a bigger weight. An outlier is an observation that has different behavior from the rest of the data (Barnett & Lewis, 1994). Since it is unlikely that any data in the sample is a result of data entry errors there is no reason to remove these outliers. By putting a weight on them you remove influence of the outliers while keeping them in the sample. This results into more accurate results. The robust

regression used in the research rely on the rreg command in Stata.

Table 6 contains the results of the robust regressions for the first hypothesis. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. In line with the hypothesis the results of the robust regression shows that the dummy variable has a positively significant effect on all the risk proxies (R&D p<0.01, Z_SCORET p<0.10, VOLATILITY p<0.10). This means that the sample was

influenced by outliers as the results in the multivariate regression show that only R&D is significantly affected by the dummy.

Panel B shows the results of the second hypothesis. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. In line to the hypothesis that the variable LHSCALE has a significantly positive effect on the R&D (p<0.01) and stock return volatility (p<0.01), the results show that LHSCALE has no significant effect on the Z-score of a firm. The noisiness in this variable make the out of the money packages more influential as the mean for the value of stock options

packages is low. One could question whether a stock option package that is not in the money has any effect on a CEO’s decision. The R&D results are consistent with the findings of Deshmukh et al (2013) who concluded that R&D was influenced by CEO overconfidence.

Looking at the control variables you will see that age and gender no not affect the risk-proxies. Size is negatively significant for the research and development costs (p<0.01) and the volatility of stock returns (p<0.01) but Firm size is positively significant for the transformed Z-score (p<0.01). Performance has a positively significant effect on the research and development costs (p<0.10) and it significantly associated in a negative way for the transformed Z-score (p<0.01). Growth

30 opportunities is positively associated with both the transformed Z-score (p<0.05) and the volatility of stock returns (p<0.01). Finally the leverage is negative associated with the research and development costs at different significance levels for the scaled

(p<0.05) and dummy variables (p<0.01), Leverage is also positively associated with the transformed Z-score (p<0.01). The models as a whole are highly significant as shown by the f-value which all have p-values below 1%.

Table 7 contains the results of the robust regression for the second hypothesis. Table a shows the results for H2a. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Not in line with the hypothesis the model shows that the CEO overconfidence variable is only significant for the volatility of stock returns (p<0.05). This means that there is no evidence that suggests that overconfidence affects risk as a whole, therefore the hypothesis is rejected. There is a negative

interaction between overconfidence and age (p<0.05). Panel B shows the robust regression analysis results for hypothesis 2b. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. This hypothesis is also rejected as there are no significant results with the overconfidence variable and the risk proxies. There is also no significant relation between age and risk and there is no interaction between age and overconfidence.

Looking at the control variables we can see that gender once again has no effect on the risk proxies used in this research. Size is negatively significant for the research and development costs (p<0.01) and the volatility of stock returns (p<0.01). Firm size is positively significant with the transformed Z-score (p<0.01). Performance is only positively significant with the research and development costs for the dummy analysis (p<0.10), the results show no relation between the scaled overconfidence variable and performance for the robust regression. Performance is negatively significant for the transformed Z-score (p<0.01). Growth opportunities is positively associated with the research and development costs at different significance levels for the scaled (p<0.05) and dummy variables (p<0.01) and it has a significantly positive effect on the volatility of stock returns (p<0.01). Finally the leverage significantly affects the research and development costs in a negative way (p<0.05) and leverage has a significantly positive effect on the transformed Z-score (p.0.01) The models as a whole are highly significant as shown by the f-value which all have p-values below 1%.

Table 6

Robust Regression H1 Panel A: Dummy Analysis

R&D Z_SCORET VOLATILITY

Coef t-stat p-value Coef t-stat p-value Coef t-stat p-value

Β0 Intercept 0,09** 2,56 0,011 -3,89*** -5,40 0,000 0,09*** 6,80 0,000 Β1 LHDUMMY 0,03*** 3,19 0,002 0,31* 1,85 0,066 0,01* 1,70 0,089 AGE 0,00 0,91 0,363 -0,01 -1,22 0,222 -0,00 -0,67 0,503 GENDER -0,01 -0,57 0,567 -0,25 -1,22 0,223 0,00 0,05 0,964 SIZE -0,01*** -3,18 0,002 0,21*** 4,43 0,000 -0,00*** -4,93 0,000 PERFOR 0,13* 1,95 0,052 -17,89*** -12,63 0,000 -0,03 -1,06 0,292 GROWTH OPP -0,01 -0,39 0,698 0,63** 2,07 0,039 0,04*** 6,88 0,000 LEVERAGE -0,00** -2,25 0,025 0,07*** 4,00 0,000 0,00 0,99 0,325

Observations F-value P-value R&D 315 4,46 0,000 Z_SCORET 315 52,77 0,000 VOLATILITY 315 17,23 0,000

Panel B: Scale analysis

R&D Z_SCORET VOLATILITY

Coef t-stat p-value Coef t-stat p-value Coef t-stat p-value

Β0 Intercept 0,08** 2,25 0,025 -4,24*** -6,32 0,000 0,08*** 7,03 0,000 Β0 LHSCALE 0,03*** 2,98 0,003 0,12 0,70 0,487 0,01*** 2,88 0,004 AGE 0,00 0,86 0,390 -0,01 -1,16 0,248 -0,00 -0,57 0,570 GENDER 0,01 0,48 0,635 0,27 1,30 0,193 -0,00 -0,04 0,965 SIZE -0,01*** -3,28 0,001 0,22*** 4,37 0,000 -0,00*** -5,08 0,000 PERFOR 0,13* 1,86 0,063 -18,10*** -12,74 0,000 -0,03 -1,13 0,261 GROWTH OPP -0,00 -0,10 0,919 0,79** 2,59 0,010 0,03*** 6,36 0,000 LEVERAGE -0,00*** -2,61 0,009 0,08*** 3,99 0,000 0,00 0,58 0,564

Observations F-value P-value R&D 315 4,46 0,000 Z_SCORET 315 52,62 0,000 VOLATILITY 315 17,54 0,000

32 Table 7

Robust Regression H2 Panel A: Dummy analysis

R&D Z_SCORE VOLATILITY

Coef t-stat p-value Coef t-stat p-value Coef

t-stat p-value Intercept 0,08** 0,13 0,011 -4,15*** -6,13 0,000 0,09*** 7,10 0,000 LHDUMMY 0,01 0,85 0,893 0,44 0,21 0,836 0,08** 2,22 0,027 Age 0,00 0,12 0,395 -0,01 -1,16 0,249 -0,00 -0,33 0,744 AgexLHD 0,00 0,57 0,908 -0,00 -0,06 0,949 -0,00** -2,09 0,038 GENDER 0,01 -3,16 0,572 0,25 1,22 0,223 -0,00 -0,01 0,991 SIZE -0,01*** 1,94 0,002 0,22*** 4,41 0,000 -0,00*** -4,75 0,000 PERFOR 0,13* -0,36 0,053 -17,91*** -12,55 0,000 -0,03 -1,30 0,196 GROWTH OPP -0,01 -2,22 0,723 0,62 1,98 0,049 0,03*** 5,87 0,000 LEVERAGE -0,00** 2,57 0,027 0,07*** 3,96 0,000 0,00 0,38 0,704

Observations F-value P-value R&D 315 3,88 0,000 Z_SCORE 315 45,98 0,000 VOLATILITY 315 15,20 0,000

Panel B: Scale analysis

R&D Z_SCORE VOLATILITY

Coef t-stat p-value Coef t-stat p-value Coef

t-stat p-value Β0 Intercept 0,01 0,11 0,909 -3,31** -2,52 0,012 0,11*** 4,50 0,000 Β1 LHSCALE 0,13 1,51 0,132 -1,31 -0,76 0,449 -0,02 -0,77 0,439 Β2 AGE 0,00 1,49 0,138 -0,03 -1,23 0,218 -0,00 -1,21 0,228 Β3 AGEXLHS -0,00 -1,20 0,231 0,03 0,83 0,406 0,00 1,04 0,299 GENDER 0,01 0,58 0,561 0,28 1,33 0,184 -0,00 -0,00 0,997 SIZE -0,01*** -3,37 0,001 0,21*** 4,33 0,000 -0,00*** -5,05 0,000 PERFOR 0,13 1,88 0,061 -18,09*** -12,71 0,000 -0,02 -1,03 0,306 GROWTH OPP -0,00 -0,23 0,817 0,82*** 2,70 0,007 0,03*** 6,65 0,000 LEVERAGE -0,00** -2,38 0,018 0,07*** 3,81 0,000 0,00 0,50 0,614

Observations F-value P-value R&D 315 4,29 0,000 Z_SCORE 315 46,09 0,000 VOLATILITY 315 15,37 0,000

5. Conclusion

This papers researchers whether the overconfidence of a CEO has an effect on the risk that a firm takes. Based on prior literature I predict that CEO Overconfidence has a positive effect on the risk a firm takes. I predict that a CEO sees the positive outcomes of an action and ignores, or even fails to see the negative effects of this action. A CEO is considered overconfident when the CEO has a stock option package available that is 40% in the money but the CEO decides not to execute it as he believes that he can raise the stock price even more. Furthermore this research a variant on CEO overconfidence that shows overconfidence in a numerical scale by using a weighted average of a stock option packages instead of a dummy. This way it uses all the stock options available to a CEO instead of only the highest one

The results of the research are not consistent with the projections when

performing a multivariate regression, however when performing a robust regression in order to limit the influence of outliers the results show that CEO overconfidence does have an effect on the risk-taking behavior of a firm. Using a sample of 315 unique observations over a period of 2012-2014 robust regression analysis shows a significant relation between CEO overconfidence and risk using the dummy variable presented by Malmendier and Tate (2005). Using a weighted average of all the stock option packages possessed by the CEO shows different results as there seems to be no significant effect on the Z-score. However this method has shown to be noisy as the packages have a mean of 0.677 which means that on average the stock option packages are not in the money. Logically speaking it is doubtful that a package that is this far out of the money has an influence on the decision making behavior of a CEO. When accounting for interaction with age the research has shown CEO overconfidence has no significant impact on the risk-taking behavior of a firm. Lastly this research has shown that the ratio of overconfident CEOs has dropped from 21.2% to 13.7% which shows a decrease in overconfident CEOs. Interestingly enough the results show no significant effect between age and risk-taking behavior. Which is in contrast to most of the prior literature.

This research answers the calls of prior literature that state that managerial effects and CEO overconfidence needs to be researched more often. As far as I know this research is the first to explore the relation between CEO overconfidence and risk taking

34 behavior. Furthermore prior literature has explored the effects of overconfidence on

periods before the financial crisis of 2007/2008 where this research is performed on sample from a period after the 2007/2008 crisis.

Results of the study are subjected to the following limitations. The data used for this research is generated from the firms in the S&P 500. By this means the findings of this research cannot be transferred to: non-American firms, small and mid-sized firms and privately held firms. Further research can investigate the overconfidence in a different setting. Secondly the proxies used for overconfidence are subjected to certain issues, the overconfidence dummy presented by Malmendier and Tate has the issue that CEOs who have stock option packages which are 39% in the money at the end of the year are not considered overconfident as the option might have exceeded 40% during the fiscal year. For the scaled variant the issue is that options that are not in the money have a heavy influence on the scale and it is questionable whether option packages that are out of the money have influence on the CEO. Further research can investigate whether there is a variable that can remove or reduce the issues described above.

Lastly the sample period of the research was limited as data in the period after the crisis of 2007/2008 is limited. Further research can investigate the effects of the

2007/2008 crisis and how it has an effect on CEO overconfidence and the risk of older CEOs as this research has shown that there have been changes. Future research on the effect should not only be limited to CEO overconfidence and age but it should also explore other characteristics.

References

Adams, R., Almeida, H., & Ferreira, D. (2005). Powerful CEOs and Their Impact on Corporate Performance. The Review of Financial Studies, 18(4), 1403-1432. Altman, E. (1968). Predicting Financial Distress of Companies. Journal of Finance, 1,

1-54.

Ammann, M., & Verhofen, M. (2009). The impact of prior performance on the risk-taking of mutual fund managers. Annals of Finance, 5(1).

Andriosopoulos, D. A. (2013). Information disclosure, CEO overconfidence, and share buyback completion rates. Journal of Banking & Finance, 37, 5486-5499. Armstong, C., Larcker, D., Ormazabal, G., & Taylor, D. (2013). The relation between

equity incentives and misreporting: The role of risk-taking incentives. Journal of Financial Economics, 109(2).

Avery, C. &. (1999). Herding over the career. Economics letter, 53, 327-333. Barnett, V. &. (1994). Outliers in Statistical Data. John Wiley & Sons.

Beber, A. a. (2012). Who Times the Foreign Exchange Market? Corporate Speculation and CEO Characteristics. Journal of Corporate Finance, 18(5), 1065-1087.

Berger, P. O. (1997). Managerial entrenchment and capital structure decisions. Journal of Finance, 52, 1411-1438.

Bertrand, M., & Schoar, A. (2003). Managing with style: The effect of managers. The quarterly journal of economics.

Bhagat, S., & Welch, I. (1995). Corporate research and development investments: International comparisons. Journal of Accounting and Economics, 19, 443-470. Camarer, C., & Lovallo, D. (1999). Overconfidence and Excess Entry: An Experimental

Approach. American Economic Review, 89(1), 306-318. Carey, M. &. (2005). The risks of financial institutions.

Chava, S., & Purnanandam, A. (2010). CEOs versus CFOs: Incentives and corporate policies. Journal of Financial Economics, 97, 263-178.

Chen, G. C. (2015). Making the same mistake all over again: CEO overconfidence and corporate resistance to corrective feedback. Strategic Management Journal, 36(10), 1513-1535.

Cheng, Q. &. (2005). Equity Incentives and Earnings Management. Accounting Review, 80(2), 448.

36 Coles, J., Daniel, N., & Naveen, L. (2006). Managerial incentives and risk-taking. Journal

of Financial Economics, 79(2), 431-468.

Core, J. &. (1999). The use of equity grants to manage optimal equity incentive levels. Journal of Accounting and Economics, 28(2), 151-184.

Core, J. H. (1999). Corporate governance, chief executive officer compensation, and Þrm performance. Journal of Financial Economics, 51, 371-406.

Custódio, C., & Metzger, D. (2014). Financial expert CEOs: CEO׳s work experience and firm׳s financial policies. Journal of Financial Economics, 114(1), 125-154.

Davidson, W. X. (2007). The influence of executive age, career horizon and incentives on pre-turnover earnings management. Journal of Management and Governance, 11, 45-60.

Denis, D., & Denis, D. (1995). Dismissals, Performance Changes Following Top Management. The Journal of Finance, 50(4).

Deshmuk, S., Goel, A., & Howe, K. (2013). CEO overconfidence and dividend policy. Journal of Financial Intermediation, 22(3), 440-463.

Ertugrul, M. S. (2008). Financial Leverage, CEO Compensation, and Corporate Hedging: Evidence from Real Estate Investment Trusts. J Real Estate Finan Econ, 36, 57. Farag, H. &. (2016). The influence of CEO demographic characteristics on corporate

risk-taking: evidence from Chinese IPOs. The European Journal of Finance, 1-30. Frydman, C. (2007). Rising through the ranks: the evolution of the market for corporate

executives 1936-2003. Working Pap.

Galasso, A., & Simcoe, T. (2011). CEO Overconfidence and innovation. Management Science, 57(8), 1469-1484.

Ge, W., Matsumoto, D., & Zhang, J. (2011). Do CFOs Have Style? An Empirical Investigation of the Effect of Individual CFOs on Accounting Practices. Contemporary Accounting Research, 28(4).

Geiger, M., & North, D. (2006). Does Hiring a New CFO Change Things? An

Investigation of Changes in Discretionary Accruals. The Accounting Review, 81(4), 781-809.

Geiger, M., North, D., & O'Connell, B. (2005). The auditor-to-client revolving door and earnings managerment. Jornal of Accounting, Auditing and Finance, 20(1). Graham, J. R. (2013). Managerial Attitudes and Corporate Actions. Journal of Financial