Ceo overconfidence and capital structure decisions

(1)

CEO OVERCONFIDENCE AND CAPITAL STRUCTURE

DECISIONS

BACHELOR THESIS

UNIVERSITY OF AMSTERDAM

COLLEGE OF ECONOMICS AND BUSINESS

BSc Economics & Business

Bachelor Specialisation Finance and Organisation

Author:

T.J. de Jong

Student Number:

10717951

Thesis Supervisor: dr. J.J.G. Lemmen

Second Reader:

dhr. dr. J.E. Ligterink

(2)

ABSTRACT

This thesis provides quantitative evidence stating that CEO overconfidence is positively and

significantly related to capital structure for S&P 500 listed companies over the 2000-2015 period. The most reliable factors of explaining leverage - profitability (-), firm size (+), growth (-), industry conditions (+), nature of assets (+) and expected inflation (+) - are used to control results. In addition to overconfidence, all control variables are reliable determinants of book- and market- based leverage. Moreover, this thesis suggests industry mean leverage to be a reliable proxy of industry conditions. Although quantitative evidence is strong, further research to this notion is necessarily before taking decisive conclusions.

JEL classification: G02; G32

Keywords: Capital Structure; Leverage; CEO Overconfidence; Industry Median Leverage

STATEMENT OF ORIGINALITY

This document is written by Student 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.

(3)

LIST OF TABLES

TABLE I: Descriptive Statistics

TABLE I Panel A: Total Descriptive Statistics 46

TABLE I Panel B: Most Important Descriptive Statistics 26

TABLE II: Correlations

TABLE II Panel A: Total Correlation Matrix 47

TABLE II Panel B: Most Important Pairwise Correlations 28

TABLE III: A Core Model of Leverage

TABLE III Panel A: Total Debt to Book Value of Assets (BLEV) 30 TABLE III Panel B: Total Debt to Market Value of Assets (MLEV) 32 TABLE IV: Replacing Industry Median Leverage (IML) by Industry Mean Leverage (IAL) 34 TABLE V: Robustness Tests

TABLE V Panel A: Linearity Check 48

TABLE V Panel B: Durbin-Watson Statistics 49

TABLE V Panel C: Sub Periods Total Debt to Book Value of Assets (BLEV) 49 TABLE V Panel D: Sub Periods Total Debt to Market Value of Assets (MLEV) 50 TABLE V Panel E: Replacing Three-Year Nominal Treasury Bill Rates (NTBR3Y) 51

by Three-Month Nominal Treasury Bill Rates (NTBR3M)

TABLE V Panel F: Replacing Three-Year Nominal Treasury Bill Rates (NTBR3Y) 52 by Five-Year Forward Swap Rates (5YSWAP)

TABLE V Panel G: Including Fama & French Model Factor Market Return (S&P500) 53

LIST OF FIGURES

FIGURE I: Expectations Variables 12

FIGURE II: Model Specification 14

FIGURE III: Expectations Measurements 20

(6)

CHAPTER 1 Introduction

1.1 Background

A commonly made assumption in corporate finance models is that human beings are expected to be fully rational in decision making. However, a relatively new stream of research known under the name of behavioural finance relieves this assumption and examines less-than-rational behaviour from the perspective of managers (Baker and Wurgler, 2012). Questions concerning what less-than-rational behaviour implies or affects have been the focal point of research in behavioural finance. This research has viewed non-rationality as playing an essential role in corporate performance. For example, Nofsinger (2005) finds that corporate finance investment decisions can be affected by managerial characteristics (Nofsinger, 2005). In a study that examines which managerial

characteristics and abilities matter, researchers Kaplan et al. (2012) find that managerial heterogeneity is detrimental in corporate actions and performance. Managerial heterogeneity implies that people, and in specific managers, are able to distinguish themselves from others. But, what are the main

characteristics that people need to have in order to do so? And, do these proposed characteristics really influence corporate action and performance?

According to Bolton et al. (2009), empathy, team-related skills, resoluteness and overconfidence are the main characteristics which distinguish managers. With respect to

overconfidence, former studies have shown that overconfidence does indeed influence investment decisions and possibly give rise to corporate investment distortions (Malmendier and Tate, 2005a). For example, Hackbarth (2008) concludes that overconfident managers opt for higher debt levels and issue more new debt in comparison to non-overconfident managers. This indicates a potential positive relationship between managerial overconfidence and firms leverage in the capital structure. Therefore, the topic of this thesis is to find out whether managerial overconfidence genuinely affects corporate decisions on leverage. Yet, to examine this relationship, the following research question is the main focus of this thesis:

Does CEO managerial overconfidence, in the presence of control variables, positively influence firms book- and market- based leverage for U.S. (S&P 500) listed companies over the 2000-2015 period?

To address a potential relationship between managerial overconfidence and leverage, it is necessary to find appropriate control variables. Control variables are included to prevent the variables of interest, in this case managerial overconfidence, to suffer from omitted variable bias. A general condition to prevent omitted variable bias, is to include variables that are determinants of the dependent variable. Leverage is the dependent variable in this case, so to meet the condition for omitted variable bias, determinants of leverage must be known.

(7)

Up to the present, a large number of factors tend to influence leverage. However, this thesis bears on the result provided by Murray and Vidhan (2009), who, starting with a long list of factors from prior literature, came up with six reliable core factors of leverage. Out of many proposed factors, this study concludes that profitability and growth are identified as core factors that negatively affect leverage. Firm size, industry conditions, nature of assets and expected inflation are assumed to be core factors that positively affect leverage. All six core factors are reliable factors of explaining market based leverage.1 For book based leverage, the impact of firm size, growth and expected inflation is not

reliable. Although advocates of managerial overconfidence deliver proofs of overconfidence

impacting capital structure, the Murray and Vidhan (2009) study surprisingly does not investigate nor mentions any possible effect of managerial overconfidence on capital structure decisions. Since this thesis aims to provide evidence of the existence of this relationship and wishes to prevent any omitted variable biases, the abovementioned core factors of leverage are included as control variables.

Besides explaining reliable factors of leverage, Murray and Vidhan (2009) also examine the influence of replacing core factor measurements by comparables. They state that three out of the six core factor measurements can be replaced by substitutes without affecting results. Firstly, assets as a proxy of firm size can be replaced by revenues since both variables reflect a proxy of the size of a company. Secondly, direct measures of expected inflation can be replaced by nominal treasury bill rates, because expected inflation and nominal treasury bill rates are highly correlated. Thirdly, replacing tangibility by collateral as proxy for nature of assets is unlikely to matter, because the only difference between tangibility and collateral is that tangibility does not include inventories.

Remarkably, the study of Murray and Vidhan (2009) does not consider any substitutes for the core factors profitability, growth and industry conditions. Ignoring substitutes of the abovementioned core factors, and in specific of industry conditions, raises the following question: can core factors be replaced by other common factors and still adequately control for the known facts?

With respect to industry conditions, Murray and Vidhan (2009) use industry median leverage as a proxy. Industry median leverage is defined as the median of total debt to market value of assets. But, why did they specifically use median values? A possible explanation for the use of median values is called the outlier effect. According to Keller (2005), mathematical outliers cause a skewed or misleading distribution in mean measures. If samples contain outlier values, the affected mean of in this case of industry conditions, would incorrectly display a bias towards the outlier value. Median measures on the other hand, are largely unaffected by outliers and could serve as a solution to prevent such a misleading distribution. But, would this outlier effect also exist if outlier data points are excluded from statistical analyses? Would it then still make any difference to use median values instead of mean values as a proxy of industry conditions? The central limit theorem (CLT) suggests

(8)

that it doesn’t. The central limit theorem (CLT) establishes that the arithmetic mean of a sufficiently large number of variables is approximately normally distributed (Keller, 2005). Given that the sample used in this thesis meets this requirement, it is expected that industry mean leverages are not

significantly different from industry median leverages. To examine such a setting, the following sub research question is constructed:

Does replacing industry median leverage by industry mean leverage, in the presence of control variables, influence results for U.S (S&P 500) listed companies over the 2000-2015 period?

1.2 Methodology

In this thesis, a theoretical and empirical framework is provided to characterize managers as overconfident. The survey respondents are executives, and in specific CEOs. The data is obtained from several databases of The Wharton Research Data Services (WRDS). Using panel data of 1411 CEOs, a CEO overconfidence measure is constructed following Malmendier and Tate (2005a, 2005b, 2008, 2015). This measure is based on the regular tendency of executives holding stock options longer than rationality suggests. In addition to characterizing managerial overconfidence, the relationship between managerial overconfidence for S&P 500 listed American firms and capital structure decisions for the period of 2000-2015 is considered. Again, panel data from 2000-2015 is used to construct ratios of the core and dependent variables featured by Murray and Vidhan (2009). Two proxies of leverage ratios are taken into account. The first is based on market values of assets, whilst the second is based on book values of assets. As mentioned before, profitability, growth, size, industry conditions, nature of assets and expected inflation are used as control variables.

To investigate the substitution of industry condition measurements, the industry mean leverage is introduced. Company specific leverage ratios are obtained from WRDS. As well industry median leverage and industry mean leverage are separately included in the regression equation to check for differences.

Results of this thesis are obtained by performing Ordinary Least Square (OLS) regressions following Keller (2005). In analysing the results, other common problems in Econometrics besides omitted variable biases are discussed. Solving these problems ensures the results to be robust and conclusions to be internally valid.

(9)

1.3 Main Results

This thesis finds that traditional determinants of capital structure are significant for S&P 500 firms. In presenting the results, the main focus is on the influence of managerial overconfidence. With a book-based definition of leverage, the model explains approximately 29.4% of the variation in leverage, whereas the explanatory power of the model is approximately 40.6% for market based leverage. The core factors have consistent signs and statistical significance. The initial results are consistent with what has been found in the study of Murray and Vidhan (2009). Results are consistent for both measures of leverage. Profitability and growth negatively affect leverage. Tangibility, firm size, industry conditions and expected inflation positively affect leverage. Managerial overconfidence, measured by analyzing American CEOs stock options, as expected positively influences leverage. Results suggest that industry mean leverage is an appropriate proxy of industry conditions and substitute of industry median leverage. However, due to little previous research, these conclusions cannot be drawn with reasonable certainty. Further development of the interpretation of this substitution is needed, which is beyond the scope and interest of this thesis. Results are robust to a range of alternative methods and parameters.

1.4 Contribution

This thesis contributes to the behavioural corporate finance literature in the following ways. First, relevant literature about managerial overconfidence and capital structure decisions is summarized into one encompassed framework. Second, this thesis extends the present knowledge of the relationship between managerial overconfidence and capital structure by controlling whether the theory of Murray and Vidhan (2009) also applies to more recent periods or not. It is likely that corporate financing decisions change over time and whilst the study of Murray and Vidhan (2009) covers the years 1950-2003, this thesis investigates the period 2000-2015. As a result, consistent findings in some way extend the validation of their investigation to 2015. Third, the effect of replacing industry median leverage with industry mean leverage as a core factor of capital structure decisions is not examined before. The outcomes of this thesis, extend the existing knowledge of Murray and Vidhan (2009) about replacing core factors by common other factors.

1.5 Thesis Composition

The remainder of this thesis is structured as follows. The next chapter reviews the theoretical

background of this thesis, where various definitions and previous theories will be discussed. In chapter three, a brief description of the model is given. The model specifies the used dependent, independent and control variables. Chapter four presents the methodology section, whilst chapter five renders and discusses the main results of this thesis. The conclusions and limitations of this thesis are provided in

(10)

CHAPTER 2 Theoretical Framework

This chapter contains the theoretical background of this thesis. Various proposed definitions of the variable of interest, (managerial) overconfidence, are given. Post to defining managerial

overconfidence, previous findings about the relationship between managerial overconfidence and capital structure decisions are examined. To make this relationship more specific and in line with the interest of this thesis, previous literature that considers the influence of managerial overconfidence on leverage is outlined. Subchapter 2.1 concludes with the expected influence of managerial

overconfidence on leverage, which serves as starting point of this thesis.

Besides explaining managerial overconfidence and its effects, previous literature about the main determinants of leverage, which serve as control variables, is discussed. In line with subchapter 2.1, all subchapters that discuss a core variable conclude with the expected influence of this particular variable on leverage.

2.1 Overconfidence

As explained in chapter one, overconfidence turns out to be one of the distinguishing features of managers. A large amount of research in the fields of finance, economics and psychology, declares to find that people are generally overconfident (Benoît et al., 2013). Yet, it is still arbitrary what

overconfidence exactly means. Fortunately, social and experimental psychology literature has shown that self-serving biases like overconfidence are examined in detail over many decades (Malmendier and Tate, 2005a). In one of these researches, Moore and Healy (2008) mention that overconfidence can be characterized in three distinct ways. The first way to characterize overconfidence is a person that overestimates his or her own abilities or performance.

The second way to characterize overconfidence is a person that has excessive confidence in the preciseness of his/her beliefs (Moore and Healy, 2008). Overconfidence is the result of an investor who believes that his/her judgement is better than it actually is, which in the most extreme case implies that an investor thinks his/her judgement is always and precisely correct (Benos, 1998).

The third way to characterize overconfidence is by means of that a person overrates him/herself relative to others (Moore and Healy, 2008). Miller and Ross (1975) find that people’s successes are itself earned, which is why overconfident people think of themselves as being better than average. Besides that, they also state that failures are the result of bad luck, so that people frame successes and/or failures in a, for themselves, favourable way. About the better than average philosophy, Myer (1995) note that people see themselves as better than average on nearly every dimension that is subjective and socially desirable. He showed that most people find themselves more intelligent than other people, most managers value their performance as better than other managers, and most students characterize themselves as more original than their fellow students.

Although the three abovementioned forms of overconfidence briefly illustrate the concept of overconfidence, the survey of respondents in this thesis are executives. Therefore, a more specified

(11)

understanding of managerial overconfidence is required. However, as compared to overconfidence, managerial biases had a less prominent position in literature and were only examined in the fields of economics and finance. This less prominent position was the result of the notion that managers could not make biased decisions. Researchers who were critical about this notion caused a considerable growth of research in managerial biases in the last decade (Malmendier and Tate, 2005b). In advance, findings about managerial biases, and in specific managerial overconfidence, were based on criticizers as Roll (1986), Heaton (2002), Shefrin (2001) and Goel and Thakor (2000). Roll (1986) offered the Hubris hypothesis. The Hubris hypothesis states that overconfident managers complete an acquisition with no aggregated value, because they think their managing ability is great enough to succeed. Heaton (2002) concludes that actual cash flows of overconfident managers come up short with the forecasts of cash flows. He provides evidence that overconfident managers attach too much value to good outcomes and too little value to bad outcomes. Shefrin (2001) states that overconfident managers are dealing with a reduced risk perception and overestimate future profits, whereas Goel and Thakor (2000) define managerial overconfidence as underestimating asset risk.

Building on the research of Roll (1986), Goel and Thakor (2002), Shefrin (2001) and Heaton (2002), Malmendier and Tate (2005a) note that overconfident managers overestimate the return of their investment decisions. They compare managerial overconfidence to periods of overinvestment when there is a shortage of internal funds. Furthermore, they state that overconfident managers are more sensitive to investments when the availability of perceived capital is relatively cheap. In addition, Hackbarth (2008) finds that overconfident managers deal with a risk perception bias. The risk perception bias implies that overconfident managers tend to underestimate the level of risk of future earnings. In one of the most recent researches, Malmendier and Tate (2015) subdivide the managerial overconfidence bias in two forms. First, they believe that overconfident managers tend to have the opinion that the market undervalues the company’s current assets. Second, they state that overconfident managers overestimate the value of their future investments. In conclusion, they define managerial overconfidence as the overestimation of the value that is believed to be created by a manager.

2.1.1 Managerial Overconfidence and Capital Structure Decisions

Previous studies have examined the effect of managerial overconfidence on capital structure decisions multiple times.2 In one of these studies, it is suggested that the effect of overconfidence should

explicitly be modelled when analysing corporate decision-making (Ben-David et al., 2007). This entails the importance of overconfidence in corporate decision-making and suggests that these two

(12)

things are related. This suggestion is in some way confirmed by Malmendier and Tate (2005a), who by applying the traditional investment-cash flow sensitivity model of Fazzari et al. (1988), find that the investments of overconfident CEOs are more prone to internal cash flow than among rational peers. In line with the Hubris hypothesis, Malmendier and Tate (2008) find that overconfident

managers complete value-destroying mergers. They believe that overconfident managers overestimate the ability of their firm to realize transcending returns. In addition to this theory, Kolasinski and Li (2013) state that overconfident managers on average complete more acquisitions than

non-overconfident managers.

Whereas previous research mainly finds that managerial overconfidence negatively influences corporate investment decisions, Galasso and Simcoe (2011) believe that managerial overconfidence positively influences the level of innovation. This associated relation specifically holds for competitive and innovative industries. Besides that, Otto (2014) finds that CEO overconfidence can be profitable for a firm when executives’ compensation packages involve bonuses and equity-related

compensations. Overall, the previous literature regarding the relationship between managerial overconfidence and capital structure decisions show conflicting theories. Thus, no clear conclusions can be drawn about the effect of managerial overconfidence on capital structure decisions in general.

2.1.2 Managerial Overconfidence and Leverage Ratios

Although the expectation of this thesis is to find a positive relationship between managerial

overconfidence and firms’ leverage in the capital structure, Zacharakis and Stephard (2001) argue that managerial overconfidence negatively affects the amount of effort Venture Capitalists exert in their judgement which ventures to fund. They believe this increases the expected cost of financial distress, which as a result must lead to decreasing leverage ratios.

In contrast to this standout exception, Ben-David et al. (2007) state that companies with overconfident CEOs use lower discount rates, invest more and use more debt. Hackbarth (2008) reaches the same conclusion and finds that overconfident managers choose higher debt levels and issue more new debt in comparison to non-overconfident managers, indicating managerial

overconfidence and leverage are positively related. Malmendier et al. (2011) find that overconfident managers are against equity financing and issue relatively more debt. Consistent with this aversion to equity financing, Deshmukh et al. (2013) argue that there is a negative relationship between being overconfident and dividend pay-outs. This negative relationship implies that CEOs view external financing as costly and therefore also prefer debt financing over equity financing. More extremely, Shefrin (2001) believes managerial overconfidence might induce debt overhang. All the above findings indicate a positive relationship between managerial overconfidence and debt financing, such that overconfident managers prefer to issue debt rather than equity.

(13)

Although Murray and Vidhan (2009) did not provide evidence of the notion that managerial

overconfidence affects leverage, literature mostly suggests that managerial overconfidence and firms leverage are positively related. Building on this logic, the following expectation is constructed:

Expectation: Positive Relationship Between Managerial Overconfidence and Leverage

2.2 The Main Determinants of Leverage

Despite of several researches and decades of empirical tests, it remains difficult to determine the main factors that influence leverage. Previous literature provides us with several factors that claim to influence leverage ratios. It is unclear which of those proposed factors are reliable determinants of leverage. For this reason, Murray and Vidhan (2009) have considered these previous proposed factors, and investigated which factors do significantly influence leverage. They state that among those proposed factors, the following six factors intend to reliable influence capital structure:3

1. Profitability 2. Firm Size 3. Growth 4. Industry Conditions 5. Nature of Assets 6. Expected Inflation

2.2.1 Profitability

The trade-off theory states that a firms’ optimal capital structure is determined by the trade-off between the benefits and costs of debt. A beneficial example of issuing debt is an increase in tax shield, whereas bankruptcy costs can be viewed as an example of costs of debt. With respect to profitability, the trade-off theory states that profitable firms in general face lower bankruptcy cost. This implies that costs of debt are negatively related to profitability. As costs of debt decrease with profitability, the trade-off theory suggests that managers attain higher values to tax shield benefits. Thus, this tax and bankruptcy costs perspective predicts that profitable firms are able to induce higher levels of debt (Jensen, 1986).

On the contrary, the pecking order theory suggests that firms prefer internal finance relative to external finance and states that profitable firms become less levered (Myers and Majluf, 1984). Moreover, studies by Rajan and Zingales (1995) and Kayhan and Titman (2007) also argue that profitability and leverage are negatively related. Rajan and Zingales (1995) state that managers of

(14)

profitable firms avoid debt, whereas Kayhan and Titman (2007) believe this relationship holds, because firms passively add up realized profits.

Consistent with the pecking order theory, Murray and Vidhan (2009) provide quantitative evidence that proofs profitability to negatively influence book- and market- values of leverage. Bearing on the study of Murray and Vidhan (2009) the relationship between profitability and leverage is expected to be as follows:

Expectation: Negative Relationship Between Profitability and Leverage

2.2.2 Firm Size

Several researches explored that firm size is related to leverage (Titman and Wessels, 1988). Smith (1977) states that the cost of issuing equity is much higher for relatively small firms. He believes small firms prefer debt- over equity- financing. Fama and French (2002) suggest that larger firms deal less with the information asymmetry problem. Larger firms tend to prefer equity finance rather than debt finance. These arguments support the notion that firm size is negatively related to leverage.

Nevertheless, Warner (1977) states that bankruptcy costs represent a larger proportion of total firm value in cases where firms are relatively small. He also believes that firm size is positively related to the degree of diversification and assumes that relatively large companies are less sensitive to

bankruptcy. Flath and Knoeber (1980) also provide evidence that larger firms can handle higher leverage ratios because of the diversification benefit and low probability of default. On top of that, they state larger companies have less variance in their cash flows. Chittenden et al. (1996) find that larger firms are less sensitive to the problems of moral hazard and adverse selection and deal with lower monitoring costs. The above findings suggest a positive relationship between firm size and leverage (Titman and Wessels, 1988).

In line with Warner (1977), research of Murray and Vidhan (2009) provide supportive quantitative proof of the statement that larger firms tend to have higher market leverage ratios. This leads to the following expectation:

Expectation: Positive Relationship Between Firm Size and Leverage

2.2.3 Growth

According to Myers’ (1984) pecking order theory debt is preferred to equity. He, Smith and Warner (1979) state that agency costs can be reduced if firms use convertible debt. Convertible debt gives the bondholder the right to convert debt claims into equity claims. In scenarios where the riskiness of a company’s activities increase, convertible debt gives the bondholders the assurance to participate in

(15)

changes in shareholders’ value. This right to convert debt helps to control for the shareholder-bondholder conflict. Because convertibles are more valuable to shareholder-bondholders, they will be willing to pay a higher price for them. Higher prices translate into lower interest rates, providing a lower probability of default. Thus, as regards to the pecking order theory growth might be positively related to leverage.

On the contrary, Jensen and Meckling (1976) and Titman and Wessels (1988) state that high growth companies deal more with the asset substitution- and underinvestment- problem compared to low growth companies. Titman and Wessels (1988) also conclude that agency costs are likely to be higher for high growth companies. Murray and Vidhan (2003) find similar evidence implicating that growth and leverage are negatively related when firms are having a surplus in investment

opportunities. Furthermore, Adam and Goyal (2008) conclude that growth potential may be influenced by stock mispricing. Due to this mispricing in equity issuances, they state that high growth companies should reduce their leverages.

Previous explanations mostly lead to one notion: firms with high growth potential have relatively few debts. In accordance to this, Murray and Vidhan (2009) provide evidence indicating that high growth firms tend to have lower market levels of leverage, which is also the expectation in this thesis:

Expectation: Negative Relationship Between Growth and Leverage

2.2.4 Nature of Assets

Tangibility of assets is found to be a determinant of leverage (Rajan and Zingales, 1995). Titman and Wessels (1988) report a positive relationship between tangibility and financial leverage. They believe tangible assets provide better collateral for loans as compared to intangible assets. Firms that secure their debts with tangible assets can borrow at lower interest rates and thus can admit higher financial leverage. Storey (1994) prolonged the notion that bank financing depends on whether the loan can be secured by tangible assets. Stulz and Johnson (1997) argue that tangible assets encounter less

information asymmetry than intangible assets during a bankruptcy. For this reason, tangible assets are valued higher than intangible assets.

In accordance to the above theories, Murray and Vidhan (2009) conclude tangibility is positively related to both book- and market- based leverage. The expected relationship between nature of assets and leverage is therefore as follows:

(16)

2.2.5 Industry Conditions

Leverage ratios significantly differ between industries. This so-called industry effect encounters different meanings (Murray and Vidhan, 2009). A possible interpretation of the industry effect is constructed by Hull (1999), who believes that the industry median leverage is used as a benchmark to allot firms own leverage. Accordingly, Hovakimian et al. (2001) state that leverage ratios are changed towards median industry leverage. Another interpretation of the industry effect is made by

Hovakimian et al. (2004), who state that the industry effect implies a set of correlated and/or omitted variables. Therefore, including industry leverage into leverage-specific regressions, controls for omitted variable bias. Based on the theory that industries face systematic forces that influence capital structure decisions, they expect those systematic forces to be the reason of industry heterogeneity in terms of leverage decisions.

Consistent with the above findings, Murray and Vidhan (2009) state median industry leverage is positively related to both book- and market-based leverage. This leads to the following expectation:

Expectation: Positive Relationship Between Industry Conditions and Leverage

2.2.6 Expected Inflation

As the trade-off theory states, a firm’s optimal capital structure is a balance between the benefits and costs of debt. Taggart (1985) believes that the present value of tax deductibles of debt is positively related to expected inflation. This implies that costs of debt are negatively related to expected inflation. Whereas costs of debt are decreasing with respect to expected inflation, this suggests that increases in expected inflation induce higher levels of debt. The market timing theory states that capital structures are made by looking at the current conditions in both debt and equity markets. Relatively high values of expected inflation in comparison to current interest rates, result in a lower present value of interest expenses. This suggests that debt markets are a more favourable way to finance investments when expected inflation is high.

The above findings suggest a positive relationship between expected inflation and leverage (Myers, 1975). In accordance with this, Murray and Vidhan (2009) also mention a positive relationship between market leverage and expected inflation. To postpone this suggested positive relationship, the expected relationship is formulized as follows:

(17)

2.2.7 Summary

As described previously, Murray and Vidhan (2009) provide evidence that profitability, size, growth, industry conditions, nature of assets and expected inflation are reliable determinants of market values of leverage. Those results are in general consistent with previous literature. Prior to their investigation, they assumed to find robust results for several proxies of leverage. However, when investigating determinants of book values of leverage, Murray and Vidhan (2009) conclude that only industry conditions, nature of assets and profitability remain reliable variables. Barclay et al. (2006) provide a possible explanation for this remarkable finding. They argue that book values are based on past and present firm performance, whereas market values are also based on future expectations. Firm size, growth and expected inflation tend to capture future expectations, whereas this is not the case for the variables industry conditions, nature of assets and profitability. Based on this notion, it is likely to again find a difference in reliable core factors between market- and book- based leverage. In line with the evidence provided by Murray and Vidhan (2009), the expected relationships are summarized in the following table:

FIGURE I: Expectations Variables

This figure presents the expected relationship of managerial overconfidence, profitability, firm size, growth, industry conditions, nature of assets and expected inflation on book- (column 1) and market- (column 2) based leverage. Both variables are used as proxies for leverage. For definitions of all variables see Appendix A Variable Definitions. * indicate reliable factors of leverage based on the theory of Murray and Vidhan (2009).

Variables

- CONF = Managerial Overconfidence - PROF = Profitability - FSIZE = Firm Size - GROWTH = Growth - INDCOND = Industry Conditions - NOA = Nature of Assets - EXPINFL = Expected Inflation

VARIABLES BLEV MLEV

CONF Positive (+) Positive (+)

PROF Negative (-) * Negative (-) *

FSIZE Positive (+) Positive (+) *

GROWTH Negative (-) Negative (-) *

INDCOND Positive (+) * Positive (+) *

NOA Positive (+) * Positive (+) *

(18)

CHAPTER 3 – Model, Variables and Data

The third chapter focuses on the model, variables and data that is used to answer the research questions. First, an outline of the tested hypotheses is given. Furthermore, the model that is used to answer the hypotheses is explained. Finally, the applied measurement of managerial overconfidence is briefly described, followed by general explanations of the used proxies for the control variables.

3.1 Hypothesis Development

To test whether managerial overconfidence and debt are related, the following hypotheses are tested:

- Hypothesis 1 (debt): The book leverage ratios of American listed companies for the period 2000-2015 are positively affected by CEO overconfidence.

- Hypothesis 2 (debt): The market leverage ratios of American listed companies for the period 2000-2015 are positively affected by CEO overconfidence.

The following hypotheses are constructed to verify whether industry median leverage (IML) can be replaced by industry mean leverage (IAL) without altering results:

- Hypothesis 3 (industry conditions): Industry median- (IML) and industry mean- leverage (IAL) ratios of American listed companies for the period 2000-2015 are positively and in a similar way affecting book based leverage ratios.

- Hypothesis 4 (industry conditions): Industry median- (IML) and industry mean- leverage (IAL) ratios of American listed companies for the period 2000-2015 are positively and in a similar way affecting market based leverage ratios.

(19)

3.2 Model

Figure II renders the model that is used to test the proposed hypotheses of this thesis. Consistent with the proposed research question, the dependent variables of this thesis are book- and market- based leverage (BLEV; MLEV). The independent variable is CEO overconfidence (LONG). Control variables are profitability (PROF), firm size (SIZE), growth (MTBR), industry conditions (IML), nature of assets (TANG) and expected inflation (NTBR3Y).

FIGURE II: Model Specification

Figure II presents an overview of the used model. For definitions of all variables see Appendix A Variable Definitions. * indicate reliable factors of leverage. The measurements (as given in brackets) will be discussed in more detail in subchapters 3.3 and 3.4.

3.3 Measuring Overconfidence

Measuring managerial overconfidence is not straightforward and is experienced as a big challenge among many researchers (Malmendier and Tate, 2005a). However, over time, several researchers came up with possible proxies of managerial overconfidence. A possible way to measure managerial overconfidence is provided by Oliver (2005), who estimates the level of managerial overconfidence by measuring public perceptions about economic conditions. Hayward and Hambrick (1997) use

evaluations about managers to classify the level of overconfidence. Besides using opinions, they argue that managerial overconfidence is measurable by relative compensation. Moreover, they also provide evidence that the level of managerial overconfidence is related to stockholders returns. Lin et al. (2006) believe that managerial overconfidence can be measured by investigating managers’

upward-Dependent variables

Book Leverage (BLEV) Market Leverage (MLEV)

Independent variable

Managerial Overconfidence (LONG)

Control variables

Profitability (PROF) Firm Size (SIZE) Growth (MBA)

Industry Conditions (IML; IAL) Nature of Assets (TANG)

(20)

positively related to the frequency of mergers and acquisitions. After years of investigation, Malmendier and Tate (2005a, 2005b, 2008, 2011, 2015) state that managerial overconfidence is measurable by investigating the systematic tendency of executives holding stock options longer than rationality suggests.

3.3.1 Long Holder Method (LONG)

The measurement used in this thesis is based on the option exercise behaviour of executives and is explained by Malmendier and Tate (2005a, 2005b, 2008, 2011, 2015). These authors conclude that investigating executives’ compensation packages is the most common approach of measuring CEO overconfidence. They believe executives are under-diversified with respect to idiosyncratic risk. To solve for this under-diversification problem, rational executives would immediately exercise their options if possible. The optimal timing of exercising an option is determined by the executives’ level of risk-aversion, the moneyness of the option and the degree of under-diversification (Hall and Murphy, 2002). Malmendier and Tate (2015) conclude that stock options typically have a duration of ten years and are not exchangeable. Moreover, they state that stock options are only exercisable after their vesting period.4 For example, time-based vesting gives employers the right to provide stock

options over time, whilst performance-based vesting gives employers the right to only grant options after certain performance targets are met. Building on this logic, the ‘long holder’ proxy of

overconfidence is shaped. The basic idea of the ‘long holder’ method is that executives who do not diversify their portfolio by exercising fully vested options, believe that future firm performance overrules the under-diversification problem. An executive is considered to be a ‘long holder’ if vested options are held until the year of expiration, imposing the option was at least 40% in the money one year prior to expiration. The threshold of 40% ensures that out of the money options are excluded. This makes sense, because rational executives would refuse to exercise such an option. In summary, a CEO is classified as overconfident if he or she:

ever during his/her tenure as CEO exercised a vested option in the last year before expiration, imposing that; (1) the option was in the money for more than 40% exactly one year before expiration date.

Measure: Long Holder (LONG)

4 In addition to option vesting periods, a lock-up period can also be part of employee stock options. Lock-up agreements prohibit individuals

(21)

3.4 Measurements Leverage

Many different empirical definitions and several proxies of leverage have been used previously. However, the use of book or market values mainly depicts the difference between these measures. As described before, book values are based on past and present firm performance, whereas market values also include future expectations (Murray and Vidhan, 2009). Advocates of book leverage ratio prefer book leverage ratios because financial markets are believed to be volatile (Myers, 1975). However, there is no clear evidence stating that book leverage ratios are better proxies than market leverage ratios or the other way around. Hovakimian et al. (2004), for example, tend to ignore the importance of book-based leverage, whereas the studies of Oliver (2005) and Murray and Vidhan (2009) explicitly examine both book- and market- based values of leverage. With respect to the investigation of Murray and Vidhan (2009), four different ways to measure leverage are constructed. The first proxy of leverage is the ratio of total debt to book value of assets (BLEV). The second proxy of leverage is the ratio of total debt to market value of assets (MLEV). The third proxy of leverage is the ratio of long term debt to market value of assets (MLLEV) and the fourth proxy of leverage is the ratio of long-term debt to book value of assets (BLLEV). Because the main interest of this thesis is to find a significant relationship between managerial overconfidence and leverage in general, this thesis only measures leverage using the first two proxies (BLEV; MLEV):5

3.4.1 Book Value Leverage (BLEV)

Book value leverage (BLEV) can be measured as follows:

𝐵𝐿𝐸𝑉 = 𝑡𝑜𝑡𝑎𝑙 𝑑𝑒𝑏𝑡

𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓𝑎𝑠𝑠𝑒𝑡𝑠 (2)

3.4.2 Market Value Leverage (MLEV)

Market value leverage (MLEV) can be measured as follows:6

𝑀𝐿𝐸𝑉 = 𝑡𝑜𝑡𝑎𝑙 𝑑𝑒𝑏𝑡

𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓𝑎𝑠𝑠𝑒𝑡𝑠 (3)

Measures: Book- and Market- Based Leverage (BLEV; MLEV)

5A more detailed explanation is given in appendix A.

(22)

3.5 Measurements Control Variables

The control variables are measured in the following ways:

3.5.1 Profitability (PROF)

Allayannis and Weston (2001) measure profitability by means of return on assets (ROA). Return on assets is expected to significantly and negatively influence leverage (Murray and Vidhan, 2009). A firms’ profitability (PROF) can be measured by using the following formula:

𝑃𝑅𝑂𝐹 =𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝑖𝑛𝑐𝑜𝑚𝑒 𝑏𝑒𝑓𝑜𝑟𝑒 𝑑𝑒𝑝𝑟𝑒𝑐𝑖𝑎𝑡𝑖𝑜𝑛

𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (4)

Measure: Profitability (PROF)

3.5.2 Firm Size (SIZE)

In accordance to Murray and Vidhan (2009), larger firms are expected to have significantly more leverage. Yermack (1995) measures firm size by the log of total assets. Following this paper, firm size (SIZE) is measured as follows:

𝑆𝐼𝑍𝐸 = log (𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠) (5)

Measure: Firm Size (SIZE)

3.5.3 Growth (MTBR)

The market-to-book ratio is the most feasible and reliable way to estimate firms’ growth potential (Adam and Goyal, 2008). Consistent with Murray and Vidhan (2009), it is expected that firms having high market-to-book ratios tend to have low levels of leverage. Market-to-book ratio (MTBR) is measured as follows:

𝑀𝑇𝐵𝑅 =𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠

(23)

3.5.4 Nature of Assets (TANG)

Higher tangibility is expected to significantly and positively influence debt ratios. A measurement of tangibility (TANG) is formulized in the following way:

𝑇𝐴𝑁𝐺 =𝑛𝑒𝑡 𝑝𝑟𝑜𝑝𝑒𝑟𝑡𝑦,𝑝𝑙𝑎𝑛𝑡 𝑎𝑛𝑑 𝑒𝑞𝑢𝑖𝑝𝑚𝑒𝑛𝑡

Measure: Tangibility (TANG)

3.5.5 Industry Conditions (IML; IAL)

Following Murray and Vidhan (2009), it is believed that industry median leverage significantly and positively influences firms associated leverage ratio. The industry median leverage (IML) ratio is measured as follows:

𝐼𝑀𝐿 = 𝑚𝑒𝑑𝑖𝑎𝑛 𝑜𝑓 𝑡𝑜𝑡𝑎𝑙 𝑑𝑒𝑏𝑡

𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (8)

Measure: Industry Median Leverage (IML)

Besides explaining the relationship between managerial overconfidence and capital structure

decisions, this paper also aims to find whether the core factor industry median leverage (IML) can be replaced by industry mean leverage (IAL). Based on the theory of Keller (2005), it is expected that industry mean leverage significantly and positively influences leverage ratios. To come up with an appropriate measure for industry mean leverage (IAL) the method of measuring industry median leverage is used. This results in a similar, but a slightly different measure of industry mean leverage:

𝐼𝐴𝐿 = 𝑚𝑒𝑎𝑛 𝑜𝑓 𝑡𝑜𝑡𝑎𝑙 𝑑𝑒𝑏𝑡

𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (9)

(24)

3.5.6 Expected Inflation (NTBR3Y; NTBR3M; 5YSWAP)

Murray and Vidhan (2009) provide evidence that expected inflation significantly and positively influences firms’ debt. However, they provide evidence that the control variable expected inflation can easily be replaced with nominal treasury bill rates. This is because expected inflation and interest rates are highly correlated. Following this method, the following proxy of expected inflation is constructed:

𝑁𝑇𝐵𝑅3𝑌 = 𝑎𝑟𝑖𝑡ℎ𝑚𝑒𝑡𝑖𝑐 𝑞𝑢𝑎𝑟𝑡𝑒𝑟𝑙𝑦 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑜𝑓 𝑡ℎ𝑟𝑒𝑒 − 𝑦𝑒𝑎𝑟 𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑡𝑟𝑒𝑎𝑠𝑢𝑟𝑦 𝑏𝑖𝑙𝑙 𝑟𝑎𝑡𝑒𝑠 (10)

Measure: Three-Year Nominal Treasury Bill Rate (NTBR3Y)

However, nominal treasury bill rates are an indirect proxy of expected inflation. Because nominal treasury bill rates have many different durations, it might be the case that nominal treasury bill rates with different durations have different effects on leverage. To prevent such a duration-effect, a short-term nominal treasury bill rate is constructed next to the abovementioned long-short-term nominal treasury bill rate. The indirect proxy of short-term expected inflation is formulized as follows:

𝑁𝑇𝐵𝑅3𝑀 = 𝑎𝑟𝑖𝑡ℎ𝑚𝑒𝑡𝑖𝑐 𝑞𝑢𝑎𝑟𝑡𝑒𝑟𝑙𝑦 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑜𝑓 𝑡ℎ𝑟𝑒𝑒 − 𝑚𝑜𝑛𝑡ℎ 𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑡𝑟𝑒𝑎𝑠𝑢𝑟𝑦 𝑏𝑖𝑙𝑙 𝑟𝑎𝑡𝑒𝑠 (11)

Measure: Three-Month Nominal Treasury Bill Rate (NTBR3M)

Although possible duration effects now have been considered, indirect proxies remain disputed measurements of variables. Direct proxies of expected inflation would eliminate all doubts concerning this matter. Because the aim of this thesis is to provide robust results, a direct proxy of expected inflation is constructed in addition to serve as robustness test. Examples of direct proxies of expected inflation are: 1) The difference between 10-year indexed treasury bills and 10-year nominal treasury bills 2) 5 year/5 year forward swap rates. This thesis explores the effect of the latter, which is formulized as follows:

5𝑌𝑆𝑊𝐴𝑃 = 𝑎𝑟𝑖𝑡ℎ𝑚𝑒𝑡𝑖𝑐 𝑞𝑢𝑎𝑟𝑡𝑒𝑟𝑙𝑦 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑜𝑓 𝑓𝑖𝑣𝑒 − 𝑦𝑒𝑎𝑟 𝑓𝑜𝑟𝑤𝑎𝑟𝑑 𝑠𝑤𝑎𝑝 𝑟𝑎𝑡𝑒𝑠 (12)

(25)

3.6 Overview Expected Signs Coefficients

Figure III illustrates the expected sign of the abovementioned measurements in relation to leverage. Positive (+) indicates a positive relationship. Negative (-) indicates a negative relationship.

FIGURE III: Expectations Measurements

Figure III presents the expected relationship of the proxies for managerial overconfidence, profitability, firm size, growth, industry conditions, nature of assets and expected inflation on book- (column 1) and market- (column 2) based leverage. Both variables are used as proxies for leverage. For definitions of all variables see Appendix A Variable Definitions. * indicate reliable factors of le verage based on the theory of Murray and Vidhan (2009).

Variables

- CONF = Managerial Overconfidence - PROF =Profitability

- FSIZE = Firm Size - GROWTH = Growth

- INDCOND = Industry Conditions - NOA = Nature of Assets - EXPINFL = Expected Inflation

Measurements

- LONG = Long Holder - PROF = Profitability

- SIZE = Firm Size - MTBR = Market-to-book Ratio

- IML = Industry Median Leverage - IAL = Industry Mean Leverage

- TANG = Tangibility - NTBR3Y = Three-Year Nominal Treasury Bill Rate

- NTBR3M = Three-Month Nominal Treasury Bill Rate - SP500 = Market Return S&P 500

VARIABLES MEASUREMENTS BLEV MLEV

CONF LONG Positive (+) Positive (+)

PROF PROF Negative (-) * Negative (-) *

FSIZE SIZE Positive (+) Positive (+) *

GROWTH MTBR Negative (-) Negative (-) *

INDCOND

IML Positive (+) * Positive (+) *

IAL Positive (+) Positive (+)

NOA TANG Positive (+) * Positive (+) *

EXPINFL

NTBR3Y Positive (+) Positive (+) *

NTBR3M Positive (+) Positive (+)

(26)

CHAPTER 4 – Methodology

The methodology chapter briefly explains the method that is used for data gathering. Furthermore, the concept of internal validity is discussed, followed by an outline of regression specifications.

FIGURE IV: Database Relationship Model

Figure IV presents the Database Relationship Model. 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 dependent variables sample is created by combining the Compustat Monthly Updates Database, the Interest Rate Database and the website Federal Reserve Economic Data (2016). Dependent and control variables except from nominal treasury bill rates are retrieved by using the Compustat Monthly Updates Database. The Interest Rate Database is used to obtain nominal treasury bill rates. The Execucomp database is used to form the final sample. For definitions of all variables see Appendix A Variable Definitions.

4.1 Data Gathering

The Wharton Research Data Services (WRDS) consists of multiple databases and is the primary source of this thesis to obtain data. The database relationship model as depicted above visually renders the relationship 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 dependent variables sample is created by combining the Compustat Monthly Updates Database, the Interest Rate Database and the website Federal Reserve Economic Data (2016). Dependent and control variables are retrieved by using the Compustat Monthly Updates Database except for nominal treasury bill rates. The Interest Rate Database is used to obtain nominal treasury bill rates. The Execucomp Database is used to form the

(27)

4.1.1 Managerial Overconfidence Sample

The Thomson Reuters Dataset – Table 2 provides transaction-level data for both American and European options. American options are exercisable at any time before the expiration date, whilst European options are only exercisable at expiration date. From this Table 2 I extract option specific information such as transaction date, exercise date, expiration date, exercise price and ticker symbol for matching. For the variable transaction date, I use a data range criteria 2000-01 to 2015-12 to build my sample. Exercise dates and expiration dates before 2000-01 are excluded as well as incomplete records due to missing information such as ticker symbol. Furthermore, I exclude transactions without

exercise price and with expired exercise dates.

The Execucomp Database – Stock Option Grants – 1992 Format provides executive specific information. The variables executive full name, date become CEO and date left as CEO provides me the identity of the CEO as well as his/her tenure of being CEO. Ticker symbol is included to enable matching with Table 2. I use the data range 2000-01 to 2015-12 to extract data and incomplete records are excluded.

I can now start to build my sample by matching using a concatenate of ticker symbol and executive (stripped) surname. Together with the Thomson Reuters and Compustat Monthly Updates datasets, I am now able to identify transactions where American options have been exercised within one year of expiration date. This gives me the transactions that meet the first criteria of the long holder method.

The Compustat Monthly Updates Database – Security Monthly provides monthly closing share prices for North American companies. This database is needed to get the share price one year before the exercise date and check for the second criteria of the long holder method. I use a

concatenate of the ticker symbol and yearmonth to match onto my sample adding the monthly share price for the exercise date minus one year. I use this value to identify transactions where the American option was at least forty percent in the money one year before exercise date. In addition to this

database, the CRSP Monthly Stock Database is used in a similar way to extract additional stock prices that were not matched using the Compustat Monthly Updates database.

The combined data of Thomson Reuters, Compustat Monthly Updates and CRSP Monthly allow me to complete the sample and to distinguish companies’ executives in terms of being a long holder for specific periods. The binary variable long holder is coded one (1) if an executive exercised an option that meets both criteria and zero (0) otherwise.

(28)

4.1.2 The Control and Dependent Variables Sample

The Compustat Monthly Updates Database – Fundamentals Annuals provides quarterly company specific financial information for the variables: Total assets, debt in current liabilities, long-term debt,

operating income before depreciation, net property plant equipment, price close quarter, standard industry classification code, total revenues footnote and total liabilities. The extracted data range used

is 2000-01 to 2015-12. Matching takes place by means of a concatenate of ticker symbol and

yearquarter. Since this dataset contains missing and disputable records I followed the procedure of

Murray and Vidhan (2009) to exclude outliers and most of the misreported data. First, missing records of balance sheet and cash flow statement items are deleted as well as negative shareholders’ equity values. Furthermore, Financial firms (SIC code 6000-6999) and Utility firms (SIC code 4800-4999) are excluded due to their specific nature. Third, firms (Compustat footnote code AB) of which revenues increased by more than 50% due to mergers are also excluded. In contrary to Murray and Vidhan (2009), control variables are trimmed and not winsorized in both tails of the distribution at a level of 0.15% using the average value (AVG) +/- three times the standard deviation (STD). These variables are used to compute correct measures of the dependent and control variables.

The Interest Rate Database - Federal Reserve, H15 Report provides monthly interest rates for various durations. The data range used is 2000-01 to 2015-12. Monthly interest rates with durations of respectively three months and three years are retrieved and arithmetic quarterly averages are

calculated to serve as an additional control variable or robustness check. The website Federal Reserve Economic Data (2016) contains quarterly 5 year/5year forward swap rates from 2003-1 and this data is retrieved to serve as a robustness test. At this point this dataset is used to calculate median and mean industry leverages.

File: Control variables Final.xlsx

4.1.3 The Final Sample

I am now able to construct my final sample matching the mapped CEO transactions final with the set of control variables final. This final sample defines the level of managerial overconfidence and values the control- and dependent variables for every quarter from 2000-1 to 2015-4. The final sample consists of North American value firms listed on the S&P 500 market during the period 2000-2015. The data of the dependent and all control variables are quarterly. A more specific description of the variables is given in the Appendix A. The final sample contains panel data of 897 firms, 1,324 executives and 15 years. The amount of observations equals to 34,360.

(29)

4.2 Internal Validity Sample

Robustness of results is seen as extremely important in arguing reliable conclusions (Murray and Vidhan, 2009). To evaluate whether the final sample and its results are useful for answering the research questions, the concept of internal validity is taken into account. Results are declared to be internal valid if the estimated regression coefficients are unbiased and consistent, given that the standard errors yield the desired confidence levels.

According to Stock and Watson (2007), there are mainly five threats to biased regression coefficients in multiple regression analyses: omitted variable bias, misspecification of the functional form of the regression function, measurement error and errors-in variables bias, missing data and sample selection and endogeneity. For each, I discuss possible robustness test that can be executed in order to minimize this bias. The potential omitted variable bias can be eliminated by including control variables, whereas a possible test to prevent misspecification of the functional form is testing linearity. The best way to solve for errors in variables bias is by performing the regression method of

instrumental variable (IV). Missing data is considered to not introduce a bias if the sample is reduced. Several quantitative tests can be performed to eliminate the problem of sample selection. With respect to endogeneity, performing regression by the method of instrumental variables (IV) provides a

solution (Stock and Watson, 2007).

About inconsistent regression coefficients, Stock and Watson (2007) define heteroscedasticity and serial correlation as main threats. The solution to heteroscedasticity is the use of heteroskedastic-robust standard errors. Serial correlation can be eliminated by using alternative formulas for standard errors or by performing a Durbin-Watson test.

4.3 Regression Specification

The main interest of this thesis is to explore a positive relationship between managerial

overconfidence and leverage of North American listed firms for the period 2000-2015. All control variables have a quantitative nature. The dependent variable is a continuous variable, whereas the independent variable is a dummy. The most feasible way to investigate this relationship is by using Ordinary Least Square (OLS) (Keller, 2005). The basic model is constructed in the following way:

𝐵𝐿𝐸𝑉 = 𝛽0+ 𝛽1∗ 𝐿𝑂𝑁𝐺 + 𝜀𝑖

𝑀𝐿𝐸𝑉 = 𝛽0+ 𝛽1∗ 𝐿𝑂𝑁𝐺 + 𝜀𝑖

- BLEV = Book value of Leverage - MLEV = Market value of Leverage

- 𝛽0= Constant - LONG = Long Holder

(30)

𝛽0 is the intercept and 𝜀𝑖 is the error term. The relationship between book- and market- based leverage (BLEV) and managerial overconfidence (CONF) is tested using these equations. This basic model is extended with control variables to check whether the results obtained from the basic model still hold.

Adding control variables to the basic specification leads to the following equation:

𝐵𝐿𝐸𝑉 = 𝛽0+ 𝛽1∗ 𝐿𝑂𝑁𝐺 + 𝛽2∗ 𝑃𝑅𝑂𝐹 + 𝛽3∗ 𝑆𝐼𝑍𝐸 + 𝛽4∗ 𝑀𝑇𝐵𝑅 + 𝛽5∗ 𝑇𝐴𝑁𝐺 + 𝛽6∗ 𝐼𝑀𝐿 + 𝛽7 ∗ 𝑁𝑇𝐵𝑅3𝑌 + 𝜀𝑖

𝑀𝐿𝐸𝑉 = 𝛽0+ 𝛽1∗ 𝐿𝑂𝑁𝐺 + 𝛽2∗ 𝑃𝑅𝑂𝐹 + 𝛽3∗ 𝑆𝐼𝑍𝐸 + 𝛽4∗ 𝑀𝑇𝐵𝑅 + 𝛽5∗ 𝑇𝐴𝑁𝐺 + 𝛽6∗ 𝐼𝑀𝐿 + 𝛽7 ∗ 𝑁𝑇𝐵𝑅3𝑌 + 𝜀𝑖

- PROF = Profitability - SIZE = Firm Size

- MTBR = Market-to-book Ratio - TANG = Tangibility

- IML = Industry Median Leverage - NTBR3Y = Nominal Treasury Bill Rate 3 Years - 𝜀𝑖 = Error Term

To investigate whether industry mean leverage (IAL) is a reliable proxy of industry conditions, the industry median leverage (IML) is replaced by industry mean leverage (IAL). This results in the following equation:

𝐵𝐿𝐸𝑉 = 𝛽0+ 𝛽1∗ 𝐿𝑂𝑁𝐺 + 𝛽2∗ 𝑃𝑅𝑂𝐹 + 𝛽3∗ 𝑆𝐼𝑍𝐸 + 𝛽4∗ 𝑀𝑇𝐵𝑅 + 𝛽5∗ 𝑇𝐴𝑁𝐺 + 𝛽6∗ 𝐼𝐴𝐿 + 𝛽7 ∗ 𝑁𝑇𝐵𝑅3𝑌 + 𝜀𝑖

𝑀𝐿𝐸𝑉 = 𝛽0+ 𝛽1∗ 𝐿𝑂𝑁𝐺 + 𝛽2∗ 𝑃𝑅𝑂𝐹 + 𝛽3∗ 𝑆𝐼𝑍𝐸 + 𝛽4∗ 𝑀𝑇𝐵𝑅 + 𝛽5∗ 𝑇𝐴𝑁𝐺 + 𝛽6∗ 𝐼𝐴𝐿 + 𝛽7 ∗ 𝑁𝑇𝐵𝑅3𝑌 + 𝜀𝑖

- PROF = Profitability - SIZE = Firm Size

- MTBR = Market-to-book Ratio - TANG = Tangibility

- IAL = Industry Mean Leverage - NTBR3Y = Nominal Treasury Bill Rate 3 Years - 𝜀𝑖 = Error Term

(31)

CHAPTER 5 – Research Results

Chapter five discusses the main results of this thesis. First, some descriptive statistics are given. Next, the results of a performed correlation test are discussed, followed by the results of the regression analyses and robustness tests. Complete tables are listed in the Appendix.

TABLE I: Descriptive Statistics

Table I presents the most important descriptive statistics for the variables book- and market based leverage, managerial overconfidence, industry median leverage and industry mean leverage. The control and dependent variables are trimmed at the 0.15% level in both tails of the distribution before the summary statistics are calculated. The sample period is 2000-2015. Missing data, financial firms (SIC code 6000-6999), utility firms (SIC code 4800-4999) and firms involed in mergers (footnote code AB) are excluded. A definition of the variables is given in Appendix A. Number of firms 897; total number of observations = 34,360.

Panel B: Most Important Descriptive Statistics

Measurements

- LONG = Long Holder - IML = Industry Median Leverage

- IAL = Industry Mean Leverage

5.1 Descriptive Statistics

Table I represents the most important descriptive statistics of this thesis. The sample consists of 34,360 records. All non-discussed values can be interpreted in the same way as stated below.

The mean of the long holder measure (LONG) is equal to 37.2%, indicating that 37.2% of the records comprises an overconfident CEO.7 This result is very similar to Malmendier and Tate (2015),

who find that the percentage of overconfident CEOs is around 40.0%. Therefore, the overconfidence measure is suggested to be reliable even though different datasets are used. The notion that

7The mean of the long holder variable does not indicate the percentage of unique CEOs that are considered to be overconfident. The long

MEASUREMENTS N Mean SD 10th ₅₀th ₉₀th Leverage Measures BLEV 34,360 0.197 0.159 0.000 0.190 0.408 MLEV 34,360 0.120 0.124 0.000 0.089 0.294 Factors LONG 34,360 0.372 0.483 0.000 0.000 1.000 IML 34,360 0.104 0.100 0.003 0.079 0.244 IAL 34,360 0.139 0.086 0.050 0.120 0.259

(32)

approximately four out of ten CEOs are considered to be overconfident is consistent with prior discussed literature, declaring that people are generally overconfident (Benoît et al., 2013).

Book- and market- based leverage (BLEV; MLEV) averages are 19.7% and 12.0%, which implies that book- and market- values of assets differ. This finding is in line with the theory of Barclay et al. (2006), who state that book values are based on past and present firm performance, whereas market values are also based on future expectations.

Out of the 34,360 observations, industry median leverage (IML) is on average 10.4%, whereas industry mean leverage (IAL) equals to 13.9%. Standard deviations are respectively 0.100 and 0.086. The 10th_{percentiles of the distribution of industry median- and mean- leverages are 0.3% and 5.0%,}

while the sample at the 90th_{percentile has industry median- and mean- leverages of 24.4% and 25.9%.}

The fact that the descriptive statistics of both measures slightly differ, suggests that using industry mean leverage as a proxy of industry conditions will alter results. Possible explanations of this surprising finding are the existence of outliers or the fact that the sample is not normally distributed. However, decisive conclusions to this matter cannot be drawn on the basis of descriptive statistics.

The remaining factors show similar values as concluded by Murray and Vidhan (2009). Therefore, the measurements of the control variables are also suggested to be reliable estimators of the control variables.

Ceo overconfidence and capital structure decisions