Firm size and organizational innovation : the influencing impact of CEO overconfidence?

(1)

FIRM SIZE AND ORGANIZATIONAL INNOVATION: THE

INFLUENCING IMPACT OF CEO OVERCONFIDENCE?

MASTER THESIS

UNIVERSITY OF AMSTERDAM

GRADUATE SCHOOL OF BUSINESS

MSc Business Administration – Strategy Track

Author:

T.J. de Jong

Student Number:

10717951

Thesis Supervisor: dr. M. Stienstra

Date of Submission: 20-06-2018

(2)

ABSTRACT

There is still an ongoing debate over whether small or large organizations are more successful at innovation. Academic research reports positive, non-significant, and negative relationships between firm size and organizational innovation. In an attempt to deepen the understanding of these mixed findings, this thesis tests for both a moderating and a mediating influence of CEO overconfidence on the relationship between firm size and organizational innovation. Secondary data are collected for S&P listed companies over the 2002-2015 period. Regression results provide evidence for a significant positive moderating impact of CEO overconfidence on the size-innovation relationship. Moreover, the results indicate a significant partial mediating impact of CEO overconfidence on the size-innovation relationship. While controlling for the impact of CEO overconfidence, this thesis continuously provides quantitative evidence stating that the degree of firm size is positively and significantly related to the degree of R&D expenditures for S&P listed companies over the 2002-2015 period. Although quantitative evidence is relatively strong, further research into these notions is necessarily before taking decisive conclusions. Further research is recommended to use multiple proxies to measure CEO overconfidence and organizational innovation and to use the first-order derivative of the proxy for organizational innovation. Future research is also recommended to examine other managerial characteristics that might influence the size-innovation relationship as well, such as empathy, team-related skills, and resoluteness.

JEL Classifications:

M10; M12

(3)

ACKNOWLEDGEMENTS

I would like to offer my sincerest gratitude to my thesis advisor dr. Marten Stienstra of the Graduate School of Business at the University of Amsterdam, who consistently allowed this paper to be my own work, but steered me in the right direction and advised me whenever he thought I needed it. Moreover, I would like to express my special thanks to my father, who guided me through the process of data collection and helped me out whenever I ran into any technical issues. Finally, I must express my very profound gratitude to my parents, family and close friends for providing me with unfailing support and continuous encouragement throughout my years of study and through the process of researching and writing this thesis. This accomplishment would not have been possible without them.

Thank you.

Timothy Joël de Jong

STATEMENT OF ORIGINALITY

This document is written by Timothy Joël de Jong, 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.

(4)

LIST OF TABLES

TABLE I: Descriptive Statistics 36

TABLE II: Industry and Firm Size Distribution of the Sample 57

TABLE III: Correlations 38

TABLE IV: Moderating Impact Managerial Overconfidence 40

TABLE V: Mediating Impact Managerial Overconfidence TABLE V Panel A: Mediating Impact Managerial Overconfidence 43

TABLE V Panel B: Sobel Test 57

TABLE VI: Hypotheses Testing Overview 45

TABLE VII: Robustness Tests TABLE VII Panel A: Normality Check 58

TABLE VII Panel B: Serial Correlation Check 59

LIST OF FIGURES

FIGURE I: Conceptual Path Diagram for Moderating Effect 20

FIGURE II: Conceptual Path Diagram for Mediating Effect 20

FIGURE III: Variables Overview 24

FIGURE IV: Conceptual Databases Diagram 31

(6)

CHAPTER 1 – Introduction

1.1 Background

Explaining performance differentials among firms is a major, perhaps even the most important, research topic in the field of strategy up to date. Answers to the question why certain firms have a competitive advantage over others are said to be extremely important since they provide explanations how to achieve economic profits and a sustained competitive advantage (Porter, 1985; Rumelt, 1984). The core of explaining a competitive advantage lies in relaxing assumptions of the traditional

neoclassical economic model. A commonly made assumption in traditional neoclassical economic models is that markets are expected to be static. Dynamic theories relax this assumption and provide additional explanations of why performance differentials among firms do exist if we consider markets to be dynamic. Schumpeter’s idea about creative destruction is an example of such a dynamic theory. According to Schumpeter’s theory (1943), creative destruction is the major reason for increased investments and business fluctuations. Following his theory, creative destruction – or a firm’s ability to innovate – then becomes crucial for organizational survival. The question that then arises is: How can firms best organize themselves to facilitate strategic renewal and adaptation through innovating?

Questions concerning what drives or affects innovation was and still is a focal point of research in strategy. Many researchers argue that firm size is one of the major determinants of organizational innovation (Damanpour, 1992). Despite the fact that there are numerous investigations about the relationship between firm size and organizational innovation, there is still an ongoing debate over whether small or large organizations are more successful at the adoption of innovations.

Academic research reports positive, non-significant, and negative relationships between firm size and organizational innovation (Dewar and Dutton, 1986; Jervis, 1975; Hage, 1980). Therefore, it can be said that new additional research could improve the current understanding of the size-innovation relationship. In the abovementioned papers, the researchers argue that firm size only directly facilitates or hinders organizational innovation since they explore the expected causal relationship using direct causal effect hypotheses. However, Wegener and Fabrigar (2000) state that a complementary approach

(7)

to explore causal hypotheses entails the use of moderating and mediating causal effect hypotheses. Methodologists often present moderation and mediation effects concurrently because they are two competing causal theories about the mechanism through which a third variable operates between a cause and an effect (Frazier et al., 2004). Following this reasoning, it is argued that testing for both a moderating and a mediating impact will further strengthen the association between two causally related variables. Hence, in order to obtain a better understanding of the relationship between firm size and organizational innovation, it is necessary to consider moderating and mediating factors that potentially influence this relationship.

When considering prior research testing for moderating and mediating impacts on the size-innovation relationship, Damanpour’s meta-analysis (1992) incorporates the impacts of moderators such as type of innovation, type of organization, measure of size, stage of adoption, and scope of adoption. With respect to mediators, his analysis briefly discusses the impacts of mediators such as structural complexity and centralization. However, for both scenario’s, a potential but forgotten organizational factor may be CEO overconfidence. But, what is the reason to believe that CEO overconfidence in general impacts corporate actions? And, more specifically, on the basis of what theories could CEO overconfidence moderate and/or mediate the size-innovation relationship?

1.2 Research Gap

The reason to believe that CEO overconfidence impacts corporate actions in general is as follows: Corporate models very often assume rational decision making of human beings. However, as we know, people have emotions and fully rational behavior is not always the case in practice. Therefore, the research stream of behavioral finance relaxes the rationality assumption and examines the impact of irrational behavior from the perspective of managers on corporate actions and decision making (Baker and Wurgler, 2012). Up to date, this research convincingly concludes that non-rationality indeed plays an essential role in corporate performance. Nofsinger (2005), for example, finds that managerial characteristics as managerial heterogeneity can substantially impact corporate financial investment decisions. This basically suggests that managers, who have the ability to distinguish themselves from others, are able to influence corporate actions. In essence, there are four core

(8)

characteristics that enable a manager to distinguish him/herself. These are empathy, team-related skills, resoluteness and overconfidence (Bolton et al., 2009). As explained above, managerial

overconfidence showcases to be one of the distinguishing features of a manager to influence corporate actions and decision making.

Taken that into account, the reason to believe that CEO overconfidence might moderate the size-innovation relationship is as follows: Individuals are defined as overconfident when they undertake irrational decisions and by doing so, possibly overestimate the return of their investment decisions. Overconfident CEO’s deal with a risk perception bias, which implies that they tend to underestimate the level of risk of future earnings. Innovative projects as applying new business methods, developing new technologies, or offering new products and services, are said to be risky and challenging (Hackbarth, 2008). This theory suggests that overconfident CEO’s undertake more risky actions, of which innovation is one. Hirshleifer et al. (2012) confirm this notion and find that

overconfident CEO’s invest more in innovation than non-overconfident CEO’s. Clearly managerial overconfidence triggers and directly facilitates innovation. But, would the strength of the relationship between firm size and organizational innovation also depend on whether a CEO is overconfident or not? It might indeed be the case that CEO overconfidence also explains the strength and/or direction of the relationship between firm size and organizational innovation. Therefore, CEO overconfidence is introduced as a potential moderator of the size-innovation relationship.

The reason to believe that CEO overconfidence might mediate the size-innovation relationship is as follows: Managers of larger firms tend to earn higher salaries, have more prestige and garner greater publicity than managers of smaller firms (Malmendier and Tate, 2005a). These higher salaries, higher levels of prestige, and greater publicity could result in higher levels of overconfidence.

Moreover, as a firm grows, the leading CEO of that company could become more overconfident about his/her own abilities as growing in terms of size is on average associated with good and effective leadership. The bigger a firm gets, the more overconfident a CEO could possibly get as well. Based on this notion, it could be argued that a CEO’s level of overconfidence is caused by the size of the firm he/she runs, and in turn, this causes the degree of organizational innovation. Based on this motivation,

(9)

the variable CEO overconfidence is also introduced as a potential mediator of the size-innovation relationship.

In sum, the topics of this thesis are to find out whether CEO overconfidence genuinely has a moderating and/or mediating impact on the size-innovation relationship. Yet, to examine these relationships, the following research question is the main focus of this thesis:

How do firm size and CEO overconfidence, both separately and jointly, influence organizational innovation?

1.3 Contributions

The contributions of this thesis to theory and practice are expected to be as follows: Firstly, this thesis aims to summarize relevant literature about the concepts and relations of and between CEO

overconfidence, firm size and organizational innovation into one encompassed framework. Secondly, this thesis tries to extend the present knowledge about the relationship between firm size and

organizational innovation, since it explores the relationship in a new and more recent time period. Where markets are expected to change over time, it is plausible that the effects of drivers on organizational innovation also change over time. Therefore, results of this thesis could serve as a robustness test to check whether previously made conclusions still hold in a different and more recent time period. Thirdly, this thesis aims to extend the knowledge of previous work of dynamic and irrational choice theories by analyzing the moderating and/or mediating impact of CEO

overconfidence on the relationship between firm size and organizational innovation. By testing the moderating and/or mediating impact of CEO overconfidence, this thesis intends to create new insights of how managerial characteristics affect the relationship between firm size and organizational

innovation. In sum, the main expected contribution of this thesis is that the combined individual contributions all together will help in creating a more unambiguous interpretation of the size-innovation relationship.

(10)

1.4 Thesis Overview

The remainder of this thesis will be structured as follows: Chapter two will provide an overview of the relevant literature, tested hypotheses and used models. Chapter three entails the methodology section. This chapter mainly explains the type of research, sample, and method of data gathering. Moreover, this chapter discusses the measurements of all included variables, concept of internal validity, and data analysis. In chapter four, the main results of the constructed empirical research are provided. Finally, chapter five discusses and concludes what the implications of the findings are. Also, the actual contributions and limitations of this research together with potential recommendations for future research are given in this chapter.

(11)

CHAPTER 2 – Theoretical Framework

Chapter two covers the theoretical background of this thesis. First of all, various proposed definitions of the concept of organizational innovation are given. Besides that, previous findings about the relationship between firm size and organizational innovation are examined. Subchapter 2.2 introduces the concept of CEO overconfidence. Post to defining CEO overconfidence, the existing theoretical content about the relationship between CEO overconfidence and organizational innovation will be highlighted. Subchapters 2.3 and 2.4 individually discuss the moderating and mediating impact of CEO overconfidence on the size-innovation relationship, whereas subchapter 2.5 gives an overview of the conceptual models that are used to test the proposed hypotheses of this thesis.

2.1 Organizational Innovation

As mentioned earlier, a firm’s ability to innovate impacts investments and business fluctuations and therefore is crucial for organizational survival (Schumpeter, 1943). Companies innovate for two reasons. The first reason is to respond to changes in internal and external environments. The second reason is because companies have the desire to influence and shape the internal and/or external environment they are embedded in themselves (Damanpour, 1992). As innovativeness seems to be such an important determinant for corporate performance, it is helpful to have a clear definition of what exactly defines organizational innovation. Fortunately, there is extensive literature available regarding the concept of organizational innovation. Innovation is studied from several perspectives, which subsequently leads to various distinctive definitions of the concept. Some researchers examine the concept of innovation by looking at the adoption and diffusion of innovations, whereas others examine innovations at different levels of analysis, being individuals, organizations, or even

populations of organizations (Kimberly, 1981; Van de Ven and Rogers, 1988). This research examines the concept of innovation by focusing on adoptions of innovations executed by organizations. The goal of an adopted innovation is that it ultimately contributes to the performance and/or effectiveness of the entire organization. In such a setting, an innovation is defined as the adoption of an idea or behavior. Examples of ideas or behaviors are systems, policies, programs, devices, processes, products or services that are new to the adopting organization (Damanpour and Evan, 1984). Relying on this

(12)

definition, organizational innovation concerns all value chains of the organization and all aspects of its operations. This implies that there is no distinction made between the adopted type of innovation. An innovation is said to be adopted if it satisfies three common stages: initiation, development, and implementation (Damanpour, 1992). Once an organization develops or buys an innovation that completes these three stages, the innovation is said to be adopted.

Firm Size and Organizational Innovation

Firm size is said to be one of the drivers of organizational innovation (Damanpour, 1992). However, conflicting relationships between these two variables have been established over time. Some

organizational scholars argue that larger firms do not necessary have an association with greater innovativeness. Relatively small organizations can be more innovative because they are more flexible in nature. This is because small organizations have a higher ability to adapt and improve. Moreover, they experience less issues to accept and implement innovations (Hage, 1988). In addition to that, Mintzberg (1979) states that innovations require combining and integrating different parts of an organization. Due to their flexibility, small organizations tend to more easily achieve this compared to larger organizations. Although some organizational scholars argue that the relationship between firm size and organizational innovation is negative, the majority of researchers argue that firm size facilitates innovation. The idea is that larger organizations have more complex and diverse facilities. These facilities support the adoption of innovations (Nord and Tucker, 1987). Two concrete examples of such facilities are resources and human capital. A large organization with a large amount of resources can easier be innovative because they can tolerate potential losses of unsuccessful innovations. A large organization that is run by more professional and skilled workers has more potential to be innovative because it generally has more technical knowledge (Damanpour, 1992). In essence, the argumentation is that large organizations can more efficiently accumulate and combine resources. This ability, consequently, puts them in the forefront of technological developments.

The abovementioned conflicting theories indicate that there is still an ongoing debate over the direction of the size-innovation relationship. Potential factors that cause this diversity in research findings are conceptual and methodological factors. The use of different research designs, control

(13)

variables, measurements, and moderators potentially impact the established relationship. In order to further strengthen and specify the relationship between firm size and organizational innovation, Damanpour (1992) systematically reviews the size-innovation relationship through a meta-analysis. Essentially, his research incorporates and combines all previous researches to analyze the relationship between firm size and organizational innovation. By observing their correlations, he finds that there is a positive relationship between firm size and organizational innovation. Based on Damanpour’s (1992) research and the fact that the majority of researchers support the notion that the relationship between firm size and organizational innovation is positive, the following hypothesis is established:

Hypothesis 1: The degree of firm size is positively related to the degree of organizational innovation.

2.2 CEO Overconfidence

Although the self-serving bias of overconfidence has been examined in detail by social and experimental psychology over many decades now, it is still arbitrary to precisely define what

overconfidence exactly means (Malmendier and Tate, 2005a). Therefore, this thesis proposes several definitions of the concept of overconfidence. Obviously, there might be some overlap between these definitions, but there are certainly some differences visible as well. The goal is that the combined presented theories and associated definitions give a reliable overview into the concept of

overconfidence.

The first definition of the concept of overconfidence is given by Moore and Healy (2008). According to them, overconfidence can be characterized in three different ways. First of all, they characterize someone as an overconfident person if they overestimate their own abilities or

performance. Secondly, they state that someone is characterized as overconfident if that person has excessive confidence in the preciseness of his/her beliefs. Overconfident people believe that their judgement is consistently better than it in reality is. In the most extreme case, this implies that a person believes that their judgment is always and precisely correct (Benos, 1998). Thirdly, a person is

characterized as overconfident if he/she overvalues him/herself relative to others (Moore and Healy, 2008). Overconfident people have a perception of being better than average because they believe that

(14)

their successes are self-earned, whereas failures are only the result of bad luck (Miller and Ross, 1975). Basically, this implies that people overrate themselves because they frame their successes and/or failures in a favorable way. In addition to that, Myers (1975) concludes that this better than average philosophy holds for nearly every dimension that is socially desirable and subjective. Overconfident people think that they are more intelligent than others. Managers believe that they outperform their co-managers. And students find themselves more original than their classmates.

The abovementioned theory introduces the concept of overconfidence in a general meaning. However, the people of interest in this thesis are managers, and in specific CEO’s. That said, a better understanding of in this case managerial overconfidence is needed. Unfortunately, managerial biases like overconfidence have not been examined that much as compared to biases in general. This is because rational and unbiased behavior has been a long-used assumption in corporate models. The literature regarding managerial overconfidence is limited to a handful of researches. One of the first criticizers of the managerial rationality assumption is Roll (1986). He establishes the hubris

hypothesis, which implies that an overconfident manager completes a non-added value acquisition deal because of the believe that their managing abilities are just great enough to make it a success. Another criticizer of the managerial rationality assumption is Heaton (2002), who finds that overconfident managers consistently overvalue their forecasted cash flows. As a result, actual cash flows were not able to reach the forecasted cash flows levels. Overconfident managers overvalue these forecasted cash flows because they assign too much value on good outcomes and too little on bad ones. Besides Roll (1986) and Heaton (2002), Goel and Thakor (2000) also provide evidence of irrational managerial behavior. They state that overconfident managers overestimate future profits and deal with a reduced risk perception bias. Implicitly, this implies that overconfident managers are said to underestimate asset risk (Shefrin, 2001). In addition, Hackbarth (2008) also concludes that an overconfident manager deals with a risk perception bias as overconfident managers tend to underestimate the level of risk of future earnings. The most recent research regarding managerial biases is done by Malmendier and Tate (2005a). According to them, managers are characterized as overconfident if they overestimate the return of their investment decisions. They associate managerial overconfidence to periods of overinvestment when there is a lack of internal capital. Additionally, they

(15)

state that an overconfident manager is more sensitive to invest if the needed capital can be acquired relatively cheap and is available on a large scale. In one of their most recent researches, Malmendier and Tate (2015) define a manager as overconfident if he/she meets the following two criteria. The first criteria is that a manager must have the opinion that the market undervalues the company’s current assets. The second criteria is that a manager overestimates the value of his/her future investments. The total overestimation that is believed to be created by a manager then theoretically defines his/her level of managerial overconfidence.

CEO Overconfidence and Organizational Innovation

Prior research into organizational innovation examines whether overconfidence is needed to realize entrepreneurial and innovative behavior (Astebro, 2003). Psychologists have provided evidence for the notion that individuals generally over-estimate their own abilities (Benoît et al., 2013). As mentioned in the previous paragraph, people, managers, and even students have the tendency to be overconfident. CEO’s are particularly subjected to this bias because overconfidence is stronger among highly skilled individuals who can handle complex situations (Moore and Kim, 2003). As the research and

development stages of products and services involve uncertainty and complexity, overconfidence might play an important role in the innovation process. Galasso and Simcoe (2010) studied the relationship between a CEO’s beliefs about future performance and standard measures of

organizational innovation. They find that overconfident CEO’s underestimate the probability of failure and more likely innovate. Basically, this result suggests that overconfident CEO’s indeed bring their firms into newer technological directions than non-overconfident CEO’s. Hirshleifer et al. (2012) also conclude that there is a positive relationship between CEO overconfidence and organizational

innovation. They argue that firms with overconfident CEO’s have greater return volatility and invest more in innovation. Additionally, they conclude that firms with overconfident CEO’s obtain more patents, patent citations, and achieve greater innovative success for given research and development expenditures. The abovementioned findings indicate a positive relationship between managerial overconfidence and organizational innovation, such that overconfident managers are expected to be

(16)

relatively more innovative than non-overconfident managers. Building on this logic, the following hypothesis is established:

Hypothesis 2: The degree of CEO overconfidence is positively related to the degree of organizational

innovation.

2.3 Moderating Impact CEO Overconfidence

The first two hypotheses explore direct causal relationships between two separate variables. Hypothesis one tests for the direct relationship between firm size and organizational innovation, whereas hypothesis two tests for the direct relationship between CEO overconfidence and

organizational innovation. As mentioned before, it is suggested that testing for moderating and/or mediating effects will further strengthen the relationship between two causally related variables, in this case being firm size and organizational innovation. In accordance with that idea, several researchers have used moderating factors to explore the size-innovation relationship in more detail. Kimberly and Evanisko (1981), for example, argue that the impact of firm size on organizational innovation may depend on the type of innovation. They introduce type of innovation as a moderator because they believe that certain types of innovations only occur at certain firm size levels.

Administrative innovations are examples of innovations that only become necessary when an organization becomes larger and subsequently grows into a more complex and differentiated organization. In addition to type of innovation, other previous examined potential moderators of the size-innovation relationship are type of organization, measure of size, stage of adoption, and scope of adoption (Damanpour, 1992). Although the relationship between CEO overconfidence and

organizational innovation has been previously tested on a direct basis, previous studies have surprisingly not investigated nor mentioned anything about a possible moderating impact of CEO overconfidence on this relationship.

As the first goal of this thesis is to test for a potential moderating impact of CEO

overconfidence on the size-innovation relationship, a better understanding of moderators in general is necessary. A moderator can be defined as a qualitative or quantitative variable that affects the

(17)

direction and/or strength of the relation between an independent and dependent variable (Baron and Kenny, 1986). A simple analogy for a moderator is a gear box that adjusts the strength of a gas pedal on the motor of a vehicle. In order to test for a moderating impact, a causal theory and design behind the data is necessary. This implies that the moderator must have a causal theoretical relationship with the dependent variable. Fortunately, there is a causal theory available that explains the relationship between CEO overconfidence and organizational innovation. The theory is as follows: Overconfident people, including CEO’s, deal with a risk perception bias. This implies that they are easier inclined to underestimate the level of risk of potential future earnings. The adoption of organizational innovations involves certain risks and challenges. As this theory expects that overconfident CEO’s undertake more risky actions, it can also be expected that overconfident CEO’s generally adopt more innovations compared to non-overconfident CEO’s. As already mentioned in the second hypothesis, it is expected that the direct relationship between CEO overconfidence and organizational innovation is positive. Building on the established hypotheses one and two, which both expect to find positive relationships, this thesis proposes and expects that the moderating impact of CEO overconfidence on the size-innovation relationship is also going to be positive. In line with this idea, the following hypothesis is constructed:

Hypothesis 3: The degree of CEO overconfidence positively moderates the size-innovation

relationship, such that higher levels of CEO overconfidence amplify the relationship between firm size and organizational innovation

2.4 Mediating Impact CEO Overconfidence

The second goal of this thesis is to test for a potential mediating effect of CEO overconfidence on the size-innovation relationship. Where previous research indicates quite extensive exploration of moderating factors, this is not the case for mediating factors. Previous studies only mention the potential mediating effects of structural complexity and centralization but have not quantitively investigated possible mediating impacts on the size-innovation relationship yet (Damanpour, 1992). The main difference between a moderator and a mediator is as follows: as we now know, a moderator

(18)

impacts the strength and/or direction of a causal relationship. The absence of a moderator will not lead to losing the initial found relationship between the dependent and independent variable. A mediator, on the other hand, is caused by the independent variable, and in turn, causes the dependent variable. Without a mediator, the initial relationship between the dependent and independent variable could potentially disappear (Frazier et al., 2004). The relationship is then said to be completely mediated. When the mediator decreases the direct causal effect between the dependent and independent variable, the relationship is said to be partly mediated (Wu and Zumbo, 2008). A simple analogy for a mediator is a line of dominos, where the first domino knocks over the second and the second domino knocks over the third. Without the second domino, the first domino would not indirectly knock over the third domino. Here, the second domino is a complete mediator. If the mediating effect of the second domino is tested for, the relationship between three domino pairs needs to be established. The first domino pair relationship that needs to be established is between domino 1 and 3. Those dominos are the

independent and dependent variable, being firm size and organizational innovation. Fortunately, this relationship is already established and constructed in expectation 1, which expects firm size to be positively related to organizational innovation. The second domino pair relationship is between domino 2 and 3. Those dominos are the mediating and dependent variable CEO overconfidence and organizational innovation. This relationship is also already established in expectation two, which expects CEO overconfidence and organizational innovation to be positively related. The third domino pair relationship is between domino 1 and 2. Those dominos are the independent and mediating variable, respectively firm size and CEO overconfidence. Unfortunately, there is no available study that investigates how firm size impacts the level of CEO overconfidence. Nevertheless, there are theories that examine this relationship the other way around. The concept of empire building, for example, states that managers irrationally overinvest and increase the size of their firm because they have the desire to run larger firms rather than small ones. Another reason for CEO’s to increase the size of a firm by over-investing is because they are overconfident (Malmendier and Tate, 2005a). Therefore, it could be argued that overconfident CEO’s are relatively easier inclined to increase their firm size compared to non-overconfident CEO’s. Assuming that CEO overconfidence and firm size are positively related, it could also be argued that this positive relationship holds the other way around.

(19)

According to Malmendier and Tate (2005a), managers of large firms tend to earn high salaries, have more prestige and garner greater publicity than managers of small firms. Higher salaries, more prestige and greater publicity could then result in higher levels of overconfidence. Moreover, if a firm grows, the leading CEO of that company could become more overconfident about his/her own abilities as growing in terms of size is generally associated with good and effective leadership. The bigger a firm gets, the more overconfident a CEO could possibly get as well. Based on this notion, it could also be argued that CEO overconfidence mediates the size-innovation relationship. Therefore, the

following two hypotheses are constructed:

Hypothesis 4: The degree of firm size is positively related to the degree of CEO overconfidence.

Hypothesis 5: The degree of CEO overconfidence mediates the size-innovation relationship.

2.5 Models

Figure I and figure II give an overview of the conceptual models that are used to test the proposed hypotheses of this thesis. In accordance with the proposed research question, organizational

innovation is the dependent variable, firm size is the independent variable, and CEO overconfidence is used both as a moderator and a mediator. Figure I gives an overview of the conceptual path diagram to test for the moderating effect of CEO overconfidence on the size-innovation relationship, whereas figure II outlines the mediating conceptual path diagram.

(20)

FIGURE I: Conceptual Path Diagram for Moderating Effect

Figure I presents the conceptual path diagram for the moderating effect of CEO overconfidence. For definitions of all variables see Appendix A – Variable Definitions. The measurements will be discussed in more detail in subchapter 3.3.

FIGURE II: Conceptual Path Diagram for Mediating Effect

Figure II presents the conceptual path diagram for the mediating effect of CEO overconfidence. For definitions of all variables see Appendix A – Variable Definitions. The measurements will be discussed in more detail in subchapter 3.3.

Assumptions and Boundary Conditions

The neoclassical model of perfect competition has five basic assumptions. These are: the large number assumption, the homogeneity assumption, the mobility assumption, the rationality assumption, and the transaction cost assumption (Smith, 2003). In order to explain performance differentials through innovation among firms, this thesis relaxes three basic assumptions of the neoclassical model of

(21)

perfect competition: the homogeneity, mobility, and rationality assumption. Because of that, the models in this thesis rely only on the large number- and transaction cost assumption. The models relax the basic rationality assumption as this thesis examines the impact of CEO overconfidence, a core characteristic of managerial heterogeneity, which potentially can cause irrational decision making. Additionally, the models of this thesis relax the homogeneity- and mobility assumption. A general though in neoclassical economic theory is that markets are expected to be static. However, this thesis examines the concept of organizational innovation, which can only take place if markets are

considered dynamic. Assuming markets are dynamic, directly leads to supporting the low church stream of the resource based-view of which basic assumptions are resource heterogeneity and

imperfectly mobile resources (Barney, 1991). Finally, the models in this thesis rely on the shareholder value perspective that is normally adopted in neoclassical economic theory. The shareholder value perspective emphasizes profitability over responsibility and sees organizations primarily as instruments of its owners (Smith, 2003).

To practically examine the moderating and mediating impact of CEO overconfidence on the size-innovation relationship, the models are confronted with three general boundary conditions. The first boundary conditions is that the scope of research is limited to S&P listed firms for the period of 2002-2015. The second boundary condition of the models has to do with the measurement of

organizational innovation. This model solely relies on resource input into innovation when measuring organizational innovation and does not consider resource output into innovation as a proxy for organizational innovation. The third boundary conditions of the models is that the used measurement to proxy CEO overconfidence only functions for American stock options. The models, therefore, are only bound to American options and do not consider European options when determining the degree of CEO overconfidence.

(22)

CHAPTER 3 – Methodology

The methodology chapter explains the applied approach to answer the research question. First, the type of research will be discussed. Second, the approach to get from a population to sample is

outlined. Third, a specification of the variables and measurements is given. Fourth, a brief overview of the data gathering method and final sample will be presented. Fifth, the concept of internal validity is discussed, followed by an outline of the data analysis.

3.1 Type of Research

This thesis uses a deductive approach of research, meaning that reasoning occurs from the particular to the general. The hypotheses are developed based on existing theory, and from there, a research

strategy is designed to test these hypotheses. Additionally, this thesis uses the positivist approach of social research. The main goal of this thesis is to quantitatively test whether CEO overconfidence has a significant moderating and/or mediating impact on the size-innovation relationship. Hence, the overall design of this study will be quantitative empirical work. The needed statistical data is retrieved from existing databases. Moreover, this secondary data will have a longitudinal nature involving measurements over time. This so-called panel data contains observations of multiple variables obtained over multiple time periods for the same firms and individuals.

3.2 Population vs Sample

The people of interest in this thesis are executives, and in specific CEO’s. Moreover, this research focuses on CEO’s of S&P listed American firms. The time period of this research is 2002-2015, a total of fourteen years. In addition to characterizing managerial overconfidence, the dependent, independent and control variables are retrieved for the same S&P listed American firms and for the same time-period of 2002-2015. The total CEO’s sample consists of 6,820 unique CEO’s, whereas the total dependent, independent and control variable sample amounts to approximately 200,000 unique observations. The number of unique CEO’s is relatively small because CEO related data is provided on a yearly basis and CEO’s obviously stay at their respective company for periods longer than a year. Besides that, missing data in measuring overconfidence leads to additional losses in the number of

(23)

available observations. The amount of observations for the dependent, independent and control variable sample is relatively large since this information is easier to retrieve. However, missing values of balance sheet and cash flow statement items are deleted. The final combined sample defines the level of managerial overconfidence and values the control- and (in)dependent variables for every year from 2002 to 2015. The final sample comprises panel data of 450 firms and 760 executives. The amount of observations is equal to 3,416. The procedure of Murray and Vidhan (2009) is followed to create this final dataset. The total amount of observations decreases because certain steps in the process of creating this final data set lead to data loss. First of all, the final sample is constructed by matching both datasets together. Missing data and/or misreported values in one of the abovementioned variables will lead to one less observation in general. Both financial- (SIC code 6000-6999) and utility firms (SIC code 4800-4999) are excluded from the sample. Secondly, outliers are excluded by a method of winsorizing. This implies that all variables are trimmed in both tails of the distribution at a 0.15% level using the average value +– three times standard deviation approach. As a result, the final sample classifies CEO’s as overconfident and values the control, dependent and independent variables for every year from 2002 to 2015.

(24)

3.3 Variables and Measurements

Figure III gives an overview of the variables that are used in this thesis. The measurements of the dependent, independent, moderator/mediating, and control variables are outlined below.

FIGURE III: Variables Overview

Figure III gives an overview of the variables that are used in this research. For definitions of all variables see Appendix A – Variable Definitions.

3.3.1 Dependent Variable – Organizational Innovation

Several proxies of organizational innovation have been previously used to measure organizational innovation. The two most common ways to measure organizational innovation are based on

quantitative information about either patents or R&D investments. The main difference between these two proxies is that the patent based proxy measures resource output into innovation, whereas the R&D investment proxy measures resource input into innovation (Hirshleifer et al., 2012). Hall et al. (2005) measure resource output into innovation by exploring patent counts and patent citations, whereas the research of Hirshleifer et al. (2012) uses the number of patent applications by a firm and the total number of citations ultimately received by the patents applied for as proxies. Data related to patents are constructed using the NBER patent database. However, a current pitfall of the patent-oriented measures of innovation is that the latest year in the NBER database is 2006. Since this thesis focuses on the time period of 2002-2015, patent-oriented measurements are not applicable for this

(25)

investigation. The remaining measure of innovation, resource input into innovation, has long been measured by total R&D effort. As total R&D effort has long been viewed as a key determinant of technological progressiveness in firms, the use of this measure would be logic. However, the use of this absolute measure is criticized. Basically, the critique is that absolute values don’t give a good overview of technological progressiveness at all. Therefore, researchers have increasingly appreciated the importance of analyzing relative compositions of R&D efforts within industries. Firm value, for example, plays a major role in conditioning the composition of R&D efforts within industries. This relative composition is measured by scaling R&D investments to book value of assets (Opler et al., 1999). Imagine that two firms have complete different total asset values, but the same levels of total R&D efforts. In this case, it obviously would be too simplistic and incorrect to conclude that these two firms are equally innovative only because the levels of total R&D investments for these firms are the same. Following this method, a measurement of input-oriented innovation is formulized as follows:

𝑅𝐼𝐼𝐼 = $&& ()*(+,-./0(1

2334 567/( 38 611(.1 (1)

Measure: Resource Input Into Innovation (RIII)

3.3.2 Independent Variable – Firm Size

As explained in chapter two, larger firms are expected to significantly adopt more innovations (Damanpour, 1992). According to Hall and Ziedones (2001), firm size is measured as the natural logarithm of sales. Following this paper, firm size is measured as follows:

𝑆𝐼𝑍𝐸 = log (𝑡𝑜𝑡𝑎𝑙 𝑠𝑎𝑙𝑒𝑠) (2)

(26)

3.3.3 Moderator and Mediator – CEO Overconfidence

There are several methods to determine whether a CEO is classified as overconfident. Managerial overconfidence is challenging to measure because it is an abstract concept without a predetermined quantitative value. Despite the issues that arise regarding measuring managerial overconfidence, several researchers have come up with possible methods to do so. Oliver (2005), for example measures managerial overconfidence by looking at public and more specific CEO’s opinions about general economic conditions. Another method to measure managerial overconfidence is provided by Hayward and Hambrick (1997), who use CEO-related assessments to determine a CEO’s level of overconfidence. Where these measurements mainly focus on opinions of either CEO’s or examiners, managerial overconfidence is also measurable through the exploration of relative compensations. More specific, Hayward and Hambrick (1997) opt that the level of managerial overconfidence has a relation with the return on shares that are gained by stockholders. In addition to that, Doukas and Petmezas (2007) state that there is a positive relationship between managerial overconfidence and the frequency that a firm is involved in a merger and/or acquisition. Although the abovementioned proxies of managerial overconfidence have been used in previous research and are well known, the majority of research prefers Malmendier and Tate’s (2005b, 2008) methods to measure overconfidence.

Malmendier and Tate (2005b, 2008) state that managerial overconfidence is measurable in two ways. The first way to measure managerial overconfidence is based on press announcements. The press-based measurement of CEO overconfidence essentially records the number of times a CEO announcement includes words that suggest overconfidence and links this to the total number of announcements. The second way to measure overconfidence is the so called long holder method. The long holder method investigates the systematic tendency of executives to hold stock options for a longer period than rationality would suggest. Since the overall research design of this thesis is quantitative empirical work, the long holder method is the solely used proxy to measure managerial overconfidence.

(27)

Long Holder Method

Malmendier and Tate (2005b, 2008, 2015) conclude that investigating CEO compensation packages is the most reliable and common approach to measure CEO overconfidence. Their theory starts with the assumption that executives are exposed to excess idiosyncratic risk because they hold

under-diversified portfolios. That said, a rational CEO would immediately exercise a vested option to solve for this under-diversification problem. Choosing when to exercise an option depends on the

executives’ level of risk-aversion, the moneyness of the option and the extent to which the CEO’s portfolio is under-diversified (Hall and Murphy, 2002). Stock options typically have a duration of ten years, are not exchangeable, and can’t be exercised prior to the vesting period (Malmendier and Tate, 2015). Based on the abovementioned assumptions and theory, the ‘long holder’ proxy of

overconfidence came into existence. Basically, the long holder method states that a CEO is expected to be overconfident if he/she doesn’t exercise a fully vested option. The reasoning is as follows: Not exercising a vested option suggest that a CEO believes that the future performance of the firm will outstrip the potential losses that are associated with an under diversified portfolio. The long holder method classifies an executive as ‘long holder’ if he/she holds a vested option that was at least 40% in the money one year prior to expiration until the year of expiration. The method only considers in the money options, because rational executives would refuse to exercise options that are out of the money. Overall, a CEO is classified as overconfident if:

He/she ever during his/her tenure as CEO exercised a vested option in the last year before expiration,

imposing that; (3) the option was in the money for more than 40% exactly one year before expiration date.

(28)

3.3.4 Control Variables

Including control variables is of significant importance to adequately address the associated size-innovation relationship and the possible moderating and/or mediating impact of CEO overconfidence on that relationship. Because of that, the following question arises: What are the other determinants of organizational innovation besides the previous mentioned variable firm size? Fortunately, several researchers have come up with possible variables that tend to influence a firm’s ability to innovate. Although some of these studies do investigate the potential direct impact of CEO overconfidence on organizational innovation, they surprisingly do not investigate nor mention a possible moderating and/or mediating impact of CEO overconfidence on the size-innovation relationship. To examine such a setting adequately and prevent the possibility that omitted variable bias arises, this thesis includes the core factors, industry, gender, capital intensity, and stock return as control variables. The reasoning for including a control variable and a proxy to measure that control variable is listed below.

Industry

In addition to firm size, market structure plays an important role in conditioning the composition of R&D efforts across industries (Cohen and Klepper, 1996). Greater product complexity, for example, leads to increasing fractional efforts dedicated to innovation. Companies in the petroleum industry may need a relatively higher fraction of R&D expenditures to achieve similar innovativeness than companies in the consumer-goods industry. According to Frenkel et al. (2001), high-tech industries have relatively high expenditures on R&D investments compared to the traditional industries. To control for that, the average innovativeness of an industry is measured by scaling average R&D expenditures to the average book value of assets of that industry. The inclusion of this ratio controls for the objectiveness of the organizational innovation measure and subsequently allows for comparing levels of innovation within and between industries. Hence, the expected relationship between industry values and innovation is expected to be positive. This thesis follows the standard industry

(29)

classification (SIC) to classify industries (WRDS, 2018)1_{. Average industry innovativeness is}

measured as follows:

𝐼𝑁𝐷 = 65(06I( -+,/1.0J $&& ()*(+,-./0(1

65(06I( -+,/1.0J 2334 567/( 38 611(.1 (4)

Measure: Average Industry Innovativeness (IND)

Gender

According to Charness and Gneezy (2012), women invest less, and thus appear to be more financially risk averse than men. Therefore, it is expected that gender is positively related to the degree of organizational innovation. To control for this effect, the sampled CEO’s are classified as either male or female. The reasoning to only look at the potential gender effect for CEO’s is based on two arguments. The first reason is that CEO’s are the main respondents in this research. Therefore, it would be logic to consistently analyze the effect of gender on this specific group of people. Secondly, CEO’s are said to significantly impact firm performance and survival at all levels and in different ways (Rosen, 1990). Barnard (1938), for example, argues that leaders create a collective purpose that binds all participants in the organization, whereas Tichy and Cohen (1997) argue that top leaders are crucial in deciding an organization’s course of action. Based on the abovementioned argumentations, gender is formulized as follows:

A binary variable, coded 1 for male CEO’s and 0 otherwise. (5)

Measure: Gender (GENDER)

1_{The SIC classifies 12 different industries, being: Agriculture, Forestry and Fishing (SIC 0100-0999), Mining (SIC}

1000-1499), Construction (SIC 1500-1799), Not Used (1800-1999), Manufacturing (SIC 2000-3999), Transportation,

Communications, Electric, Gas and Sanitary Service (SIC 4000-4999), Wholesale Trade (5000-5199), Retail Trade (5200-5999), Finance, Insurance and Real Estate (SIC 6000-6799), Services (SIC 7000-8999), Public Administration (SIC

(30)

9100-Capital Intensity

Following the procedure of Hirshleifer et al. (2012), capital intensity is expected to be positively related to resource input into innovation. Capital intensity is the amount of available capital resources in relation to other resources of production, especially labor. Hall and Ziedones (2001), state that capital intensive firms, who have large sunk costs in complex manufacturing facilities seem to have the largest incentives to adopt innovations. This is because the adoption innovation protects them against the threat of costly litigation and enables them to negotiate about external technologies on more favorable terms. In line with this theory, capital intensity is measured as follows:

𝐶𝐼 = 𝑙𝑜𝑔(+(. *03*3(0.J, *76+. 6+, (N/-*O(+.

+/O2(0 38 (O*73J((1 ) (6)

Measure: Capital Intensity (CI)

Stock Return

Following the procedure of Hirshleifer et al. (2012), this thesis includes stock returns as a control variable. Their results indicate that stock returns negatively impact a firm’s resource input into innovation. Stock returns are measured by the buy-and-hold return of the fiscal year:

𝑆𝑇𝑂𝐶𝐾𝑅 =1S60( *0-T( (.)U 1S60( *0-T( (.UV)_{1S60( *0-T( (.UV)} (7)

(31)

3.4 Data Collection

The Wharton Research Data Services (WRDS) website comprises multiple databases and is the primary source used to obtain data. Figure IV renders the relationships between the used databases. The managerial overconfidence sample is created by combining the Thomson Reuters Dataset, Execucomp Database, Compustat Monthly Updates Database, and CRSP Monthly Stock Database. The control and (in)dependent variables sample is created by combining the Compustat Monthly Updates Database and the CRSP Monthly Stock Database. Dependent, independent and control variables are retrieved by using the Compustat Monthly Updates Database except for the variable stock returns.

FIGURE IV: Conceptual Databases Diagram

Figure IV presents the Conceptual Databases Diagram. For definitions of all variables see Appendix A – Variable Definitions.

3.4.1 Managerial Overconfidence Data Collection

The Thomson Reuters Dataset consists of transactional data regarding American- and European options. American- and European options differ from one another in that American options are

exercisable at any time before expiration, whereas European option can be only exercised at expiration date. The Thomson Reuter Dataset is used to obtain option specific information such as the

transaction date, exercise date, expiration date, exercise price, and ticker symbol. Both exercise- and

(32)

holds true for incomplete records due to missing information as, for example, transactions without exercise price and expired exercise dates.

The Execucomp Database consists of executive specific information, such as the executive full

name, gender, date become CEO, and date left as CEO. Based on these variables, the identity of a

CEO as well as his/her tenure of being at the head of the company can be identified. To enable matching with the Thomson Reuters Dataset, the variable ticker symbol is included in the dataset as well. Both databases are matched using a concatenate of ticker symbol and executive surname. The combined dataset enables it to identify transactions where American options have been exercised within one year of expiration and results in transactions that meet the first criteria of the long holder method.

The Compustat Monthly Updates Database comprises monthly share prices for North

American Companies. From this database, share prices are retrieved one year before the exercise date to check for the second criteria of the long holder method. Matching occurs by using a concatenate of

ticker symbol and yearmonth. With this data, transactions where American options were at least forty

percent in the money one year before exercise date can be identified. Following this procedure, the CRSP Monthly Stock Database is used as an additional source to extract stock price information that was not available in the Compustat Monthly Updates Database.

The combination of the Thomson Reuter, Execucomp, Compustat Monthly Update, and CRSP Monthly Updates Database allows for completing the Managerial Overconfidence Data Collection and distinguishing executives in terms of being a long holder in a specific period. An executive is

classified as a long holder and coded one (1) if (s)he exercised an option that meets both criteria and zero (0) otherwise.

3.4.2 Control and (In)dependent Variables Data Collection

The Compustat Monthly Updates Database provides yearly company specific financial information for the variables: R&D expenditures, book value of assets, total sales, net property plant and equipment,

number of employees, and standard industry classification code. These variables are used to compute

(33)

innovativeness. The data regarding the control variable gender is retrieved from the managerial overconfidence sample, whereas stock returns are retrieved from the CRSPS Monthly Stock Database. Average industry innovation levels are simply obtained by calculating the average innovation levels of separate SIC. Matching occurs by means of a concatenate of ticker symbol and yearquarter.

3.4.3 Final Data Collection

The final sample is constructed by matching the CEO Overconfidence file with the to Control Variables file. The final sample classifies CEO’s as overconfident and values the control, dependent and independent variables for every year from 2002 to 2015. The sample consists of North American listed firms on the S&P market. All data related to the dependent and control variables are yearly. The final sample comprises panel data of 450 firms, 760 executives and a total of fourteen years. The amount of observations is equal to 3,416.

3.5 Reliability and Validity

Regression results are said to be internally valid when the estimated regression coefficients are unbiased and consistent and when the standard errors yield the desired confidence levels (Stock and Watson, 2007). This thesis carries out several robustness tests to aim for as reliable and valid results as possible. The performed robustness tests aim to solve for internal validity threats as heteroscedasticity, omitted variable bias, non-normal distribution, serial correlation, fixed time effect, and missing data and selection biases.

First of all, heteroscedasticity problems are eliminated by using and reporting robust standard errors. Secondly, the possibility that the omitted variable bias arises is eliminated through the

inclusion of several control factors. More specifically, this thesis controls for industry, gender, capital intensity, and stock returns. Moreover, the normal distribution of the variables is tested in several ways. Although some tests have been performed for all variables, the main goal regarding normality is to provide evidence for the notion that normality of the dependent variable organizational innovation can be assumed. Normal Q-Q plots and graphs are analyzed in addition to the performance of

(34)

output of organizational innovation has a lower confidence interval of 0.579 and an upper confidence interval of 0.743. This test supports the notion that the dependent variable is assumed to be normally distributed. This is because the empirical criteria of skewness outputs states that acceptable values lay between the values -1 and 1. A conflicting result is found with respect to the Kolmogorov Smirnov Output test. As all statistics are significantly different from zero, all scores of this test lead to the rejection of the null hypothesis of a normal distribution. A reason for this conflicting finding is the fact that the Kolmogorov Smirnov Output test is relatively powerful and therefore easily leads to rejection of the null hypotheses. Taken this into account and considering that the majority of the normality tests do suggest normality of the dependent variable, this thesis assumes the variable organizational

innovation to be fairly normally distributed (see Appendix Table VII). In addition to checking for normality, the Durbin-Watson test result is used to check for serial correlation. The Durbin-Watson statistic has a value of 1.891, which is relatively close to the aspired level of 2. As the Durbin-Watson statistics of the organizational innovation model has a value close to 2, this indicates no similarity between observations as a function of the time lag between them. Subsequently, this thesis controls for fixed time effects by including yearly dummy variables in the regression specification. Lastly, the possibility that missing data points and/or selection criteria lead to biases is considered as well. The missing data points bias is not assumed to arise because the exclusion of data points takes place at random. The survivorship bias is an example of a selection bias. The fact that the sample also contains defunct companies corrects for this survivorship bias.

Although this thesis carries out a number of robustness tests to ensure an internally valid investigation, results can’t be declared perfectly robust. This is because this thesis doesn’t control for some important potential threats to biased and/or inconsistent coefficients. Issues regarding possible measurement errors and errors in variables biases, misspecifications of the functional form, and endogeneity are considered to be beyond the scope of this thesis. Therefore, results are considered to be fairly, but not perfectly robust.

(35)

3.6 Data Analysis

The needed statistical data and regression coefficients are generated by performing multiple ordinary least square (OLS) regressions (Keller, 2005). The Statistical Package for the Social Sciences (SPSS) is the used software package to perform these regression analyses. All the obtained OLS results use unstandardized beta coefficients, robust standard errors and adjusted R-square values.

The moderating impact of CEO overconfidence on the size-innovation relationship is

measured with the first model of the Process V.3 method of Hayes (2017). The numerical independent variable is mean centered before running the analysis in process. According to this method,

moderating impacts are measured following a three-step process. The first step is to perform a multiple regression model that solely includes control variables. In step two, a multiple regression model that includes both the control variables as the independent variables is presented. The final step is to present a multiple regression model that includes the interaction term between firm size and managerial overconfidence.

As the fourth model of the Process method of Hayes (2017) offers a model to perform mediating analysis, it would make sense to use this method for the mediating analysis as well. However, Hayes’ (2017) method unfortunately doesn’t allow for a dichotomous mediating variable when performing mediating analysis. As CEO overconfidence is characterized as a dichotomous binary variable, this method can’t be used to measure this specific mediating impact. In the process of considering alternative design frameworks to test for a mediation effect, the Kenny approach shows to be the default paradigm to model mediation (Spencer et al., 2005). Hence, this thesis relies on the method of Baron and Kenny (1986) to establish a potential mediation effect. Their work is summarized into a four-step data analytic method. The first step is to show that the independent variable is related to the dependent variable. The second step is to show that the independent variable is correlated with the mediator. Here, the mediator is defined as the dependent variable and is

predicted by the independent variable. Step three is about showing that the mediator affects the dependent variable, while controlling for the independent variable. The final step is to compare the unstandardized 𝛽_V coefficients of step 1 and step 3. The extent to which the mediator accounts for the overall direct effect as measured in step 1 is calculated by simply subtracting 𝛽 − 𝛽 .

(36)

CHAPTER 4 – Research Results

Chapter four presents the main results of this thesis. At first, an outline of relevant descriptive statistics is given. Secondly, a correlation matrix is presented to discuss the relationships between variables. Finally, the results of the moderating and mediating regression analyses are discussed. The complete tables are listed in the Appendix.

TABLE I: Descriptive Statistics

Table I presents the descriptive statistics for the variables organizational innovation, firm size, managerial overconfidence, industry, capital intensity, stock return, and gender. The control and dependent variables are trimmed at the 0.15% level in both tails of the distribution before the statistics are calculated. The sample period is 2002-2015. Missing data, financial firms (SIC code 6000-6999), and utility firms (SIC code 4800-4999) are excluded. A definition of the variables is given in Appendix A. Number of firms 450; number of executives 760; total number of observations = 3,416.

Panel A: Relevant Descriptive Statistics

4.1 Descriptive Statistics

Table I gives an overview of relevant descriptive statistics of this thesis. The descriptive statistics provide information about the final sample that is used to perform multiple regression analyses. This section will compare and/or analyze the obtained results of the descriptive statistics in order to briefly

MEASUREMENTS N Mean SD MIN MAX

Dependent And Independent Variable RIII SIZE 3,416 3,416 0.036 3.170 0.274 0.836 0.000 0.621 0.010 5.637 Factors LONG 3,416 0.380 0.486 1.000 1.000 IND 3,416 0.057 0.014 0.000 0.070 CI 3,416 4.711 0.386 3.207 6.600 STOCKR 3,416 0.085 0.436 -0.998 1.985 GENDER 3,416 0.980 0.148 0.000 1.000