DEBT MATURITY CEO OVERCONFIDENCE AND CORPORATE

CEO

OVERCONFIDENCE

AND

CORPORATE

DEBT

MATURITY

Master’s thesis Finance Supervised by: dr. Peter Smid JEL-classifications: G30; J16 Amount of words: 12.654

Abstract

In this paper we investigate whether and to what extent CEOs affect corporate debt maturity decisions for listed US firms. We hypothesized that firms with overconfident CEOs tend to have a shorter debt maturity structure than unbiased CEOs due to using a higher proportion of short-term debt. We find weak evidence suggesting that the relation is dependent on the level of liquidity risk and the level of growth opportunities. We find that overconfident CEOs use more short-term debt when the level of liquidity risk is low and the degree of growth

opportunities is high. Though, these results are not robust as they depend on the proxy that is used. Our results indicate that overconfident CEOs trade-off the increase in underinvestment issues against the increase in liquidity risk when making the decision to issue short-term debt. This relation seems not to be driven by overconfidence, but mainly by rational reasons.

Key words: Overconfidence, Optimism, debt maturity, liquidity risk Jeroen Rozema

j.rozema.7@student.rug.nl S2767368

1

1.Introduction

Our understanding of the determinants of corporate debt maturity structures stems mainly from firm-level and industry-level evidence based on traditional finance theories (e.g. Myers, 1977; Flannery, 1986; Stohs and Mauer, 1996; Johnson, 2003) which is well established. Therefore, recent studies (Huang et al. 2016; Ataullah et al. 2018) focus on personal-level evidence in relation to information asymmetry between executives and investors and agency issues between shareholders and debtholders. These studies investigate the relationship between managerial overconfidence and debt maturity. As stated by Ataullah et al. (2018), there is limited empirical support that investigate the effects of personal characteristics, such as overconfidence, on debt maturity. Therefore, this paper focuses on whether and to what extent United States (US) Chief Executive Officers (CEOs) make different debt maturity decisions based on the concept of overconfidence.

2 (moderately) overconfident CEO and vice versa by investigating the debt maturity structure of United States listed firms.

The main evidence regarding debt maturity and overconfidence comes from three

different studies that all investigated a different country. Two results are conflicting. Landier and Thesmar (2008) shows, for a sample of small unlisted French firms and by using a survey approach, that an overconfident manager self-selects short-term debt contracts. Short-term debt is selected to benefit from their optimistic beliefs regarding future firm performance. In their model, they show that optimistic entrepreneurs believe that the short-term interest rate is lower than the long-term rate and that they never default. The finance costs are likely to be lower when the short-term debt is refinanced against better terms in the future. Huang et al. (2016) extended this literature towards listed US firms by finding that overconfidence also plays a role in debt maturity decisions of CEOs. Their main finding is that overconfident CEOs use more short-term debt than CEOs which are not (classified as) overconfident. They classify CEOs as overconfident when they hold their firm share options to the final year of maturity despite being deeply in the money based on an extended information asymmetry model of Flannery (1986). They find that CEOs with a higher exposure to firm specific risk persistently use more short-term debt. They also find that liquidity risk has no effect on this relationship, which could imply that overconfidence has a negative impact on firms, as more risk might result in higher bankruptcy costs. In contrast, Ataullah et al. (2018) investigated listed firms in the United Kingdom (UK) and find that overconfident CEOs are associated with a longer maturity of corporate debt. They rely on the concept of agency theory,

implementing the theoretical model of Hackbarth (2009), which shows that overconfidence can also lead to a longer maturity of corporate debt. Ataullah et al. (2018) find that

overconfidence reduces the underinvestment problem, by mitigating the agency costs of long-term debt. Their main argument is that overconfident managers are likely to underestimate the time value of money. By investing within a shorter timeframe compared to rational managers, overconfident CEOs can create value for debt holders in firms that are not in distress by implementing positive net present value projects. This could imply that hiring an

overconfident manager can be useful in this situation, as the presence of an overconfident manager provides more freedom to choose their debt maturity structure compared to relying on short-term debt.

3 structures of US firms affected by CEOs that are classified as overconfident? Does this

relationship depend on the degree of liquidity risk and growth opportunities? We investigate the effects of overconfidence on debt maturity for firms in the US over the 2010 – 2017 period using both pooled OLS and 2SLS regressions.

This paper is structured as follows. Section two lists the relevant literature and ends with our hypothesis. In section three we describe both methods and data. The results are given in section four. Section five gives robustness tests. This paper ends with the conclusion in section six.

2.Literature review

This section explains, both theoretically and empirically, possible differences in corporate debt maturity. A distinction is made between traditional finance which investigates firms and industries and behavioral finance that uses a more managerial perspective.

2.1 Traditional view

As stated by Ataullah et al. (2018), traditional finance literature focuses mainly on firm characteristics (such as firm size, growth opportunities, cash flow, earnings volatility and tax rate) and market conditions (e.g. interest rate volatility and credit ratings) to help explain differences in debt maturity structure.

Within the perfect market framework of Modigliani and Miller (1958), variations in debt maturity have no impact on the value of the firm. However, when market imperfections, such as agency problems and information asymmetries emerge, short-term debt might become optimal for certain companies. Myers (1977) explains that companies with high growth opportunities are likely to prefer short-term debt, because long-term debt creates agency problems.

One of the agency problems is underinvestment, also known as debt overhang. This occurs when positive net present value projects are not financed due to too high pre-existing debt levels (Myers, 1977). For example, if managers undertake positive net present value investments financed by long-term debt, part of the benefits will go to existing long-term debt holders. When the level of pre-existing debt is too high, debt-holders will not finance

4 issue between bondholders and shareholders. If debt matures before growth options cease, the underinvestment problem is mitigated. This finding is supported by many studies, for example by Johnson (2003), who find that optimal leverage is low for firms with a high degree of growth opportunities, measured by the market-to-book ratio. Another agency problem is asset substitution. Jensen and Meckling (1976) find that shareholders have a stimulus to confiscate debt holder wealth by substituting into riskier investments. Brockman et al. (2010) complement the findings of Johnson (2003) by showing that short-term debt can also prevent the asset substitution problem related to executive compensation structures. Therefore, according to the traditional view, agency costs between shareholders and debtholders can be mitigated by reducing debt maturity. Myers (1977) also states that asset substitution problem can be reduced by matching the maturity of assets with that of the liabilities as this reduces the risk of financial distress. A longer maturity is associated with more long-term debt.

Another theory that might have an impact on debt maturity is signaling theory, driven by asymmetry of information. Flannery (1986) argues that high quality firms want to signal high quality by issuing short-term debt. Since low quality firms are unable to afford the additional costs associated with the switch to short-term debt, only high quality firms can signal positive private information. Brockman et al. (2010) argue that abnormal earnings can be used to measure firm quality where more (abnormal) earnings imply a higher quality.

Regarding firm quality, the Z-Score of Altman (1977) is also a used measure of credit quality. Firms with a higher Z-score are associated with a higher quality as a higher score is associated with less financial distress. Therefore, firms with a high Z-score are able to increase the maturity of debt by issuing long-term debt. For the same reasoning, a company credit rating on existing long-term debt is also incorporated as a measure for firm quality in various studies (e.g. Johnson, 2003; Brockman et al. 2010).

5 Trade-off theory might also explain some debt maturity decisions. According to

Diamond (1991), leverage is positively related to debt maturity, as firms with a high degree of leverage might prefer to hold long-term debt in order to prevent sub-optimal liquidation. Johnson (2003) find that the degree of long-term debt is positively related to the level of liquidity risk. When the volatility of cash flow increases, the probability of repaying debt decreases. Therefore, long-term debt might be preferred over short-term debt when there is a high degree of cash flow risk. However, as explained by Myers (1977), this might lead to underinvestment. Since short-term debt reduces underinvestment, Johnson (2003) argues that firms trade-off the decrease in underinvestment with the increase in liquidity risk when deciding to issue short-term debt.

2.2 Behavioral view

The behavioral view focuses more on managerial characteristics compared to the firm-level and industry-firm-level perspective of the traditional view. Several studies argue that

overconfidence and optimism can play a role in corporate finance decisions, such as the choice of debt maturity. Weinstein (1980) argues that overconfident managers overestimate future outcomes on which they have direct influence. This study describes that there are three factors that drive overconfidence. The first factor is the illusion of control (the CEO choses the project, which makes this person belief that the outcome will be positive). The second factor is the high degree of commitment to good outcomes (the CEO receives company stock and stock options which increase CEO wealth, especially with rising stock prices). The last factor lies in abstract reference points (there are numerous reasons for project failure and not all can be linked to CEO performance). This also makes it hard to compare performance across individuals. Ataullah et al. (2018) link overconfidence to self-attribution bias, which is the tendency to attribute company achievements to own skills and company failure to outside factors (Miller and Ross, 1975).

On the one hand, overconfident CEOs tend to overestimate future firm profitability of the firm. On the other hand, they tend to underestimate risk. This can have numerous

implications. Hackbarth (2009) explicitly makes a distinction between “overconfidence” and “optimism”. In the model of Hackbarth, optimism is defined as “the subjective belief that favorable future events are more likely than they actually are, the better-than-average effect” (Hackbarth, 2009 p.2).

6 firm specific risk is underestimated (Hackbarth, 2009). Optimism is an overestimation of the growth rate of earnings (growth perception bias) while overconfidence underestimates the riskiness of the earnings (risk perception bias). Hackbarth (2009) theoretically models the implications for overconfident and optimistic CEOs where all other parties (e.g. shareholders and debtholders) are rational. Rational in this sense implies that market participants can anticipate on the irrationalities of the manager. Within this situation, Hackbarth (2009) shows that managers with growth perception bias are likely to interpret external financing as very costly, as they believe that their securities are undervalued. This implies that these managers prefer the standard pecking order of finance. As outlined by Myers and Majluf (1984), this means that internal funding is preferred over external financing and debt is preferred over equity. In contrast to growth rate bias, managers with risk reception bias will follow an opposite pecking order according to Hackbarth’s model. As these CEOs underestimate the risks of company earnings, they view debt as undervalued by the market. At the same time, given the convexity between equity and debt, they view equity as overvalued. Therefore, when external finance is needed managers with risk perception bias will issue equity over debt. This provides a challenge to the standard pecking order paradigm. According to the trade off-model, rational investors will adjust the market price, caused by the issued shares, in such a way that the market value of equity does not change. Therefore, in both situations, biased managers make sub-optimal issues of debt. CEOs with growth perception bias believe their company is more profitable than in it in reality is and therefore more likely to experience financial distress. CEOs with risk perception bias believe their company is less risky than it is in reality and therefore less likely to prone to financial distress. However, both issue debt more often and use more debt compared to rational managers.

Malmendier and Tate (2005) add by finding that overconfident CEOs overinvest when internal funds are available and there is a lack of governance and capital market discipline to keep them from overinvesting. When internal financing is insufficient and external financing is required, they curb investment.

7 Huang et al. (2016) extended the theoretical model of Flannery (1986) which shows that firms with positive private information use short-term debt to signal the market that their stock is undervalued. The extension in the model of Huang et al. (2016) shows that

overconfident managers also prefer to finance with short-term debt when there is no asymmetric information between managers and shareholders in reality, implying that overconfidence is the main driver for this association.

In contrast, Hackbarth (2009) argues that managerial overconfidence can also lead to a longer maturity of corporate debt. Within his framework, he shows that biased managers invest larger amounts than rational managers due to a lower perceived uncertainty of new projects. This is referred to as the “leverage effect”. More leverage leads to more

underinvestment when holding everything else constant. However, he also argues that biased managers invest earlier than rational managers, which is referred to as the “timing effect”. By investing sooner, the underinvestment problem is reduced. They state that for mild levels of “managerial bias”, the costs of having more leverage is likely to be lower than the benefits of investing within a shorter timeframe. If so, shareholder wealth increases as the agency costs of debt overhang decrease. The intuition in this case is that it might be beneficial for firms to hire an overconfident CEO when there are large growth opportunities as this provides more freedom to choose the debt maturity structure. Furthermore, when the level of pre-existing debt is relatively high, there is no debt maturity decision to be made, as there is no possibility to issue additional debt (Brockman et al. 2010; Ataullah et al. 2018). Furthermore, when there are no growth opportunities, there is also no reason to hire an overconfident manager as investing sooner does not add value for long-term debt holders. However, when there are growth options, than overconfident managers act in the interest of long-term debtholders by preventing underinvestment. Another possible explanation is given by Hirshleifer et al. (2012), who find that optimistic CEOs are better innovators in innovative industries, implying that overconfident managers are better in exploiting growth opportunities.

2.3 Empirical evidence

A survey of Graham and Harvey (2001) shows that only 3% of American CFOs believe that their stocks are overvalued, which would imply that at least some managers express signs of overconfidence. Empirical research uses the term overconfidence and optimism

8 Hirshleifer, 2001) provide empirical evidence that this is likely the case. Furthermore, as outlined by Hackbarth (2008), the implications of growth rate bias (optimism) and risk perception bias (overconfidence) indicate that both use suboptimal debt levels. Nevertheless, there are two studies which find that the maturity of debt is affected in opposite directions due to CEO overconfidence.

Huang et al. (2016) use an extended model of Flannery (1986) and find that CEOs who are classified as overconfident use more short-term debt compared to non-overconfident CEOs. They state that overconfident CEOs believe to have positive private information with respect to future firm performance. Therefore, biased CEOs believe that they can refinance debt against lower costs in the future, after positive news arrives at the markets. They also show that firms with overconfident CEOs have higher leverage compared to

non-overconfident CEOs and they argue that the level of liquidity risk does not alter this relationship. Custódio et al. (2013) report a similar finding. They find that US firms have issued much more short-term debt. In their sample from 1976 to 2008, they state that the median percentage of debt maturing after three years decreased from 53% to only 6%. They argue that the decrease in the 1980s and 1990s was mainly caused by firms with a high degree of information asymmetry and by new firms issuing equity. They state that this might lead to a higher exposure to credit and liquidity shocks.

In contrast, Ataullah et al. (2018) use time varying measures of overconfidence and find that managerial overconfidence is positively related to debt maturity for a sample of CEOs in the UK, using the reasoning of Hackbarth (2009). In their sample from 2000 to 2010, the timing effect was stronger than the leverage effect. The findings indicated a stronger positive relation between debt maturity and overconfidence for companies with high growth opportunities and a weaker relationship for companies with a high degree of leverage.

9 managerial risk taking is discourages when the sensitivity of the CEO compensation to the underlying stock price increases. In contrast, the risk taking preferences of CEOs are

encouraged when the stocks volatility increases, implying that the maturity of debt is reduced. Interestingly, they show that this relationship almost disappears when the firm uses more short-term debt. Their result suggests that creditors take the risk seeking behavior of CEOs into account and adjust their yields accordingly. This would imply that short-term debt might reduce agency costs of debt associated with managerial compensation incentives.

Datta et al. (2005) further find that the degree of alignment between shareholders and managers can be measured by controlling for managerial ownership. They empirically find that ownership is inversely related to debt maturity. When a self-interested CEOs holds zero company shares, they likely prefer to finance with long-term debt as short-term has to be refinanced annually. Long-term debt is associated with less monitoring and therefore preferred. When a self-interested CEO holds more company shares, the interest of the CEO aligns with the shareholder, implying that this agency problem is alleviated according to Datta et al. (2005).

Market conditions are also associated with the length of debt maturity. Brick and Ravid (1985) find that firm value increases when the term structure of interest rates is upward sloping. When the interest rate of long-term debt increases in line with maturity, the benefits of the tax shield increase as well. Therefore, they argue that the slope of the term structure of interest rates is positively related to the maturity of debt.

2.4 Hypothesis

All in all, it seems likely that managerial overconfidence can affect the choice of debt maturity. From a risk perception perspective, an overconfident CEO might self-select short-term debt to limit the effects of perceived share price mispricing. From the traditional finance point of view, short-term debt should mitigate agency problems of long-term debt (e.g. Jensen and Meckling, 1976; Myers, 1979). Brockman et al. (2010) add by finding that CEO

compensation risk is also mitigated by using short-term debt. Therefore, we hypothesize that the behavioral finance literature and traditional finance literature come to the same

conclusion. We hypothesize that an overconfident CEO, prefers a shorter debt maturity structure, primarily driven by an underestimation of firm specific risk.

10

3 Research method and data

3.1 Model

To test our main hypothesis, we use a pooled cross-sectional panel model similar to Huang et al. (2016). We start with pooled OLS and follow with 2SLS due to possible

endogeneity concerns between leverage and debt maturity. There is heteroscedasticity present in our OLS model according to an unreported Breusch-Pagan test. This test shows that we have to reject the null-hypothesis that the error terms are independent. Furthermore, an unreported Hausman test shows that a model with time fixed effects is superior to random effects. We therefore focus on the results that use fixed effects. Our main model is given in equation 1. 𝑫𝒆𝒃𝒕 𝒎𝒂𝒕𝒖𝒓𝒊𝒕𝒚𝒊𝒕 (𝑺𝑻) = = 𝜶_𝟎 + 𝜶_𝟏 𝑳𝑶𝑵𝑮𝑯𝑶𝑳𝑫𝑬𝑹_𝒊𝒕+ 𝜶_𝟐 𝑳𝒅𝒆𝒍𝒕𝒂_𝒊𝒕+ 𝜶_𝟑𝑳𝒗𝒆𝒈𝒂_𝒊𝒕 + 𝜶₄𝑶𝑾𝑵_𝒊𝒕 + 𝜶_𝟓𝑳𝑬𝑽_𝒊𝒕+ 𝜶₆𝑳𝒔𝒊𝒛𝒆_𝒊𝒕 + 𝜶_𝟕𝑳𝒔𝒊𝒛𝒆𝟐_𝒊𝒕+ 𝜶_𝟖𝑨𝒎𝒂𝒕_𝒊𝒕+ 𝜶₉𝑨𝒗𝒐𝒍_𝒊𝒕 + 𝜶10𝑨𝒆𝒑𝒔𝒊𝒕+ 𝜶𝟏𝟏𝑴𝑻𝑩𝒊𝒕+ 𝜶𝟏𝟐𝑻𝑬𝑹𝑴𝒊𝒕 + 𝜶𝟏𝟑𝑹𝒂𝒕𝒆𝑫𝒊𝒕 + 𝜶_𝟏𝟒𝒁𝑺𝑪𝑶𝑹𝑬𝒅_𝒊𝒕+ 𝜶_𝟏𝟓𝑳𝑰𝑸_𝒊𝒕+ 𝜶_𝟏𝟔𝑴𝑨𝑳𝑬_𝒊𝒕 + 𝜺𝒊 (𝟏)

where all firm variables are measured for firm

𝒊

at time

t

at fiscal year-end except for 𝑨𝒗𝒐𝒍_𝒊𝒕 which is measured over the fiscal year. 𝜺𝒊 represents the error term. The definitions are given in Table 1.

Prior literature mainly uses the extent of debt due within a certain amount of years as a measure for debt maturity. Huang et al. (2016) describe that most studies use a timeframe of debt due within one to five years divided by long-term debt. Ataullah et al. (2018) use both the ratio of long-term debt due within one year to total debt and the ratio due in more than five years to total debt. In contrast, Huang et al. (2016) use six different time frames with respect to short-term debt over total debt and report that there are no particular reasons to believe that one measure is superior to the other. Therefore, we will measure debt maturity as the debt due within one year over total debt. As a robustness check, we also use the long-term debt due within five years over total debt.

11 and Tate (2005). Since the CEO receives a large part of their remuneration in company stocks and stock options, it is highly exposed to the idiosyncratic risks of the firm. Firms prohibit perfect hedging by not allowing to short-sell company stocks, which implies that in-the-money stock options should be exercised shortly after the vesting period, i.e. the period before options are (at least partly) exercisable, in order to reduce of exposure to firm specific risk. Hall and Murphy (2002) argue that the threshold to exercise depends on many factors, such as risk aversion, current wealth and remaining duration of the stock option. Malmendier and Tate (2005) explain that if these variables need to be taken into account to establish a threshold for rational option exercise, the sample is greatly reduced due to a lack of data availability. This threshold is not required for the LONGHOLDER measure, since the determination of whether a CEO is overconfident is based on whether the CEO holds the option until the final year of maturity. Therefore, overconfidence is measured by using a dummy variable. When the CEO decides, at least once in the sample period, not to exercise the stock option before its final year of maturity and the option is at least 40%1 _{in the money when it entered the final year of} maturity, the CEO is classified as overconfident for all of its years in the sample. This

threshold is conservative as the mean (median) percentage that options are in the money entering its final year of maturity for this sample is 352% (228%).

Managerial options typically have a maturity of 10 years and the vesting period lasts 5 years (Core and Guay, 2002). Therefore, in order to measure the degree of overconfidence of CEOs, this approach requires that CEOs have a relatively long tenure within the same firm. As explained by Huang et al. (2016), this has implications for the use of fixed effects, which controls for unobserved characteristics over time. Huang et al. (2016) explain that the option based approach leaves (too) little time-series variation for the use of firm-fixed effects. Therefore, following Huang et al. (2016) we use industry fixed effects for the

LONGOLHDER regressions based on the Fama and French 12 Industry Classification. All

errors are clustered at the firm level and we use time fixed effects to control for macroeconomic events.

Our other measure of overconfidence, 𝑵𝑷𝑹𝒊𝒕, represents the degree of overconfidence based on inside trading behavior used in Ataullah et al. (2018)2. According to Ataullah et al.

1 _{Based on the model of Hall and Murphy (2002) and following Huang et al. (2016). This choice is line with a}

constant relative risk aversion coefficient assumption of 3 and assumes that 67% of CEO wealth is invested in their own company stock.

2 _{Ataullah et al. (2018) measure overconfidence using multiple time-varying measures. The volume based NPR}

12 (2018), CEOs learn to become overconfident over time because of self-attribution bias,

implying that the CEO attributes success based on own ability and failure to external factors. They state that a static measure of overconfidence is misleading and imprecise. Therefore, in contrast to Malmendier and Tate (2005) and Huang et al. (2016), they annually determine whether the CEO expresses signs of overconfidence. To discriminate whether this measure alters results, we also adopt their share volume based measure. By doing so, there is sufficient time series variation for the use of firm fixed effects as we require only two years of CEO tenure within the same firm. The NPR lies in the interval of [-1,1] where a higher score implies a stronger degree of managerial overconfidence. This also implies that the NPR ratio is constructed under the assumption that the degree of overconfidence can vary per CEO.

Following Brockman et al. (2010), we control for agency costs associated with

managerial compensation by computing: 1) the sensitivity of the value of the CEO portfolio to a change in the stock price (𝑳𝒅𝒆𝒍𝒕𝒂𝒊𝒕), and 2) the sensitivity of the value to a change in the volatility of the underlying stock (𝑳𝒗𝒆𝒈𝒂𝒊𝒕). These measures require us to determine the total share value and option value of the compensation portfolio of the CEOs. We follow the same steps as given in the appendix of Brockman et al. (2010). This approach requires us to

measure the Black and Scholes (BS) values of all CEO options based on the option pricing model of Black and Scholes (1973) that is adjusted for dividends by Merton (1973). All six inputs needed to value newly granted options during a year can be extracted from

ExecuComp, making this calculation fairly straight forward. However, the BS values for in-the-money options that are either exercisable but not exercised or not exercisable need to be approximated as the exercise price3 and time to maturity4 are not provided. Therefore, these values have to approximated. We use the method of Core and Guay (2002) following Brockman et al. (2010) and Huang et al. (2016), which is said to explain 99% of total variation in the value of the CEO portfolio.

3 _{For the exercise price, we use the estimated aggregate realizable values of the exercisable and unexercisable}

options (OPT_EX_EXER_EST_VAL and OPT_UNEX_EXER_EST_VAL within ExecuComp) and the number of exercisable and unexercisable options (OPT_EX_EXER_NUM and OPT_UNEX_EXER_NUM) to calculate back the average exercise price. The realizable value of CEO options is divided by the number of options to investigate to what extent, on average, the stock price of the options are above the exercise price. This value is than subtracted by the stock price to find the exercise price.

4 _{With respect to maturity, Core and Guay (2002) assume that the time to maturity of unexercisable options is}

13 The key input of BS options however, is the volatility measure, as the impact of

volatility on the value of options is severe. The volatility measure within ExecuComp is calculated over the 60 months preceding fiscal year end. However, since the Vega variable is based on annual volatility, we use the annualized stock return volatility retrieved from CRSP. Once the delta (∆) and vega (v) is calculated for each option partition, we add the values and calculate the increase in value due to a 1% in stock price and annual volatility respectively. The calculation is given in equation 2 below and equal to the equation outlined in Brockman et al. (2010).

𝑫𝒆𝒍𝒕𝒂 = 𝑺

𝟏𝟎𝟎 (∆𝒏𝒈𝑵𝒏𝒈 + ∆ 𝒆𝒙𝑵𝒆𝒙 + ∆𝒖𝒏𝒆𝒙𝑵𝒖𝒏𝒆𝒙 + 𝑵𝑺𝑻𝑶𝑪𝑲)

𝑽𝒆𝒈𝒂 =_𝟏𝟎𝟎𝟏 (𝒗 𝒏𝒈𝑵𝒏𝒈 + 𝒗 𝒑𝒈𝒆𝒙𝑵𝒑𝒈𝒆𝒙 + 𝒗 𝒖𝒏𝒆𝒙𝑵𝒖𝒏𝒆𝒙 (𝟐)

where S denotes the stock price and N the amount of options or stocks in hundreds of thousands. The subscripts NG, EX, UNEX and STOCK denote newly granted options,

previously granted options that are exercisable, previously granted options not exercisable and total stock ownership, respectively.

Following Datta et al. (2005), we control for agency issues between shareholders and managers by incorporating managerial ownership in terms of shares (𝑶𝑾𝑵𝒊𝒕) as they argue that managerial ownership is inversely related to debt maturity. Another control for agency costs is liquidity (𝑳𝑰𝑸_𝒊𝒕), measured as current assets minus current liabilities following Ataullah et al. (2018). For leverage (𝑳𝑬𝑽𝒊𝒕), we use long-term debt divided by the market value of the firm as our measure for our OLS regressions, following previous studies (e.g. Diamond, 1991; Brockman et al. 2010; and Huang et al. 2016). As a robustness check, we also measure leverage by using total debt over total assets used in Ataullah et al. (2018).

Since the relation between debt maturity and size might be non-linear (Diamond, 1991), we control for both firm size (𝑳𝒔𝒊𝒛𝒆𝒊𝒕) and firm size squared (𝑳𝒔𝒊𝒛𝒆𝟐𝒊𝒕), using the natural logarithm of the market value of the firm. Whether the company is able to attract long-term debt is measured using dummy variables (whether the company has a S&P credit rating on long term debt (𝑹𝒂𝒕𝒆𝑫𝒊𝒕) and whether the Z-score (𝒁𝑺𝑪𝑶𝑹𝑬𝒅𝒊𝒕) is greater than 1.80). A score below 1.80 indicates that the company is likely facing financial difficulties.

14 (1986) explained that investors will not be able to distinguish the quality of the firm when information asymmetry exists. Therefore, they place an average default premium on all firms. In order words, the default premium of high (low) quality firms is overestimated

(underestimated). High quality firms signal their quality to the market by issuing short-term debt. By doing so, they prevent that investor wealth is transferred to other firms. This measure is forward looking, implying that we measure (𝑨𝒆𝒑𝒔𝒊𝒕) as the difference between next year and this year earnings per share. We expect that CEOs who believe that the earnings per share at t+1 will be relatively higher than the earnings at time t signal the market by issuing short-term debt at time t following previous studies (e.g. Johnson, 2003; Datta et al. 2005; and Brockman et al. 2010). Malmendier and Tate (2005) explain that the key difference between (inside) information and overconfidence lies in persistence. They argue that overconfidence in terms of underestimating firm risk is a fixed effect while private information is transitory. They explain that the likelihood of a CEO having positive private information for numerous years in a row is small. However, they find that overconfident CEOs persistently expose themselves to company risk. Malmendier and Tate (2005) therefore state that once CEOs are they will remain overconfident.

Huang et al. (2016) state that they do not include other personal characteristics (e.g. gender, tenure, age) in their debt maturity model as there is no theory that suggest that these characteristics alter debt maturity decisions. However, psychology literature is quite extensive in the observation that males express a higher degree of overconfidence then females (Barber and Odean, 2001). For example, Bengtsson et al. (2005) show that males are more inclined to aim for higher grades at Universities. There is also evidence which suggest that females are less overconfident than males with respect to making financing decisions. For example, Barber and Odean (2001) show that men trade 45% more than woman, reducing annual

returns more excessively compared to their female counterparts. A similar finding is presented by Huang and Kisgen (2012), that directly investigated whether woman are less overconfident than men. They find that men take on more acquisitions and debt than their female

counterparts. They also show that woman issue stock options sooner than men, which supports previous argument that males might be relatively more often overconfident that females. Therefore, in contrast to Huang et al. (2016) we include a gender dummy in our model (𝑴𝑨𝑳𝑬_𝒊𝒕) denoting 1 if the CEO is a male and 0 if the CEO is female5_.

3.1.2 Endogeniety concerns

15 As stated by Barclay et al. (2003), leverage and debt maturity are endogenous and therefore jointly determined. They explain that a change of an exogenous variable might both directly and indirectly influence endogenous variables within a simultaneous system of equations. For example, when a firm alters its investment policy it will have a direct and indirect effect on the choice of capital structure. The direct effect is that an increase in growth opportunities implies an increase in the use of short-term debt. However, there is a separate, indirect effect on leverage that occurs at the same time. More growth opportunities is associated with less leverage, which might change the desired use of short-term debt. To control for endogeneity, we use two staged least squares (2SLS) regressions. The 2SLS

method approach requires instruments correlated with leverage but not with debt maturity. We will follow prior literature (e.g., Barclay et al., 2003; Johnson, 2003; Brockman et al. 2010; and Huang et al. 2016) and use four different instruments for leverage. Brockman et al. (2010) state that theoretical studies provide no evidence for an effect of the profitability measure return on assets, fixed asset ratio (when controlled for asset maturity) and marginal tax rate on debt maturity. Therefore, these variables are treated as orthogonal to the error term and

excluded in the second stage6.

6 _{The equation is likely to be biased if these assumptions do not hold. To address this problem, we will test the}

reliability of this assumption in the result section.

Table 1

Defintions of variables, predicted signs and datasource Datasource Predicted sign Dependent variable

Debt maturity Debt maturity ratio, measured in two alternative ways

1 ST1 Debt in current liabilities / total debt (sum of debt in current liabilities and long-term debt) (Huang et al. 2016) Compustat Annual industrial 2 ST5 The proportion of long-term debt maturing within 5 years (including debt in current liabilties) over total debt (Brockman et al. 2010) Compustat Annual industrial

Key independent variables

1 Longholder Dummy variable taking the value of unity if the CEO ever held an option to the final year of the maturity during the sample period

and the option is at least 40% in-the-money entering its last year, zero otherwise (Malmendier and Tate, 2005) ExeuComp - Outstanding Equity Awards + 2 NPR (Difference between the annual insider share purchases and sales) / (the sum of CEO insider purchases and sales) in terms of

share volume (Ataullah et al. 2018) ExecuComp-- Annual Compensation +

Control variables

Ldelta The natural logarithm of one plus CEO's portfolio price sensitivity, defined as the change in the value of CEO stock and option

portfolio responding to a 1% increase in the share price of company common stock (Brockman et al. 2010)

-Lvega The natural logarithm of one plus CEO's portfolio volatility sensitivity, defined as the change in the value of CEO stock and option

portfolio responding to a 1% increase in the annualized standard deviation of company common stock (Brockman et al. 2010) ExecuComp - Outstanding Equity Awards + LIQ Liquidity. The ratio of current assets to current liabilties (Ataullah et al. 2018)

OWN The number of shares (excluding options) owned by the CEO divided by total common shares outstanding at the end of the fiscal

year (Datta et al. 2005) ExecuComp-- Annual Compensation +

Aeps Abnormal earnings. (Earnings per share in year t+1 - earnings per share in year t) / fiscal year-end stock price (Flannery, 1986) Compustat Annual Industrial file Amat Bookvalue weighthed average of the maturities of PPE and current assets, computed as (gross property, plant and equipment /

total assets ) * (gross PPE / depreciation expense) + (current assets / total assets) * (currents assets / cost of goods sold) Compustat Annual Industrial file -LEV Market leverage. Measured as Long term debt / market value of the firm (Diamond, 1991) Compustat Annual Industrial file -LEV2 Book leverage. Measured as total debt as defined above over total assets (Ataullah et al. 2018)

Lsize Natural logarithm of the market value of total assets, computed as (stock price at fiscal year end * common stock used to

calculate eps) + book value of total assets - book value of equity (Diamond, 1991) Compustat Annual Industrial file

-Lsize2 Square of Lsize (Diamond, 1991) Compustat Annual Industrial file +

MTB Market value of the firm as defined above / book value of total assets (Myers, 1977) Compustat Annual Industrial file + RateD Dummy variable taking the value of unity if the firm has a S&P credit rating on long-term debt and zero otherwise (Johnson, 2003) Compustat Annual Industrial file -Avol

Monthly stock standard deviation of the assets during the fiscal year. Measured as stock return standard deviation * (market value of equity / market value of the firm). The returns are from CRSP monthly file and the financial accounting data is from Compustat

Annual Industrial file. Return deviation of debt is assumed to be equal to zero. (Brockman et al. (2010). +/-TERM Yield on 10-year US government bonds substracted from the the yield on 6-month rate on government bonds at fiscal year end FRED of Federal Reserve Bank of St. Louis -ZSCOREd Dummy variable taking the value of unity if the firm has an Altman's Zscore in excess of 1.80 and zero otherwise Compustat Annual Industrial file

-Instrumental variables for 2SLS regressions

-16 The instruments for leverage are fixed assets ratio (𝑭𝑨_𝒊𝒕), which is measured as fixed assets over total assets, net operating tax loss (𝑵𝑶𝑳𝒊𝒕), which is a dummy denoting 1 if there is a net operating tax loss carry forward and 0 otherwise, return on total assets (𝑹𝑶𝑨𝒊𝒕), measured as operating income over total assets and investment tax credit (𝑰𝑻𝑪_𝒊𝒕), which is a dummy variable denoting one if there is a non-zero investment tax credit and zero otherwise. Brockman et al. (2010) describe that relatively more fixed assets (FA) is associated with less asset substitution problems, implying that the capital structure is more efficient. Furthermore, fixed assets ratio also increase debt capacity as fixed assets can be used as collateral for loans, thereby increasing liquidation value and with it leverage. As outlined by Myers (1984), return on assets can be used as a proxy for profitability since more profitable firms likely issue less debt, consistent with pecking order theory. However, Jensen (1986) describes that it is also possible that firms deliberately choose for higher debt levels to pay out larger levels of free cash flows. As outlined by Brockman et al. (2010), the benefits of leverage are reduced when alternative tax shields are present. Therefore, it is expected that 𝑰𝑻𝑪_𝒊𝒕 and 𝑵𝑶𝑳_𝒊𝒕 are

associated with lower leverage. The first stage of our 2SLS regressions is given in equation 3 and equal to the 2SLS regressions used in Huang et al. (2016). 7

𝑳𝑬𝑽 = 𝜶_𝟎 + 𝜷_𝟏𝑹𝑶𝑨_𝒊𝒕 + 𝜷_𝟐𝑵𝑶𝑳_𝒊𝒕 + 𝜷_𝟑𝑰𝑻𝑪_𝒊𝒕 + 𝜷_𝟒𝑭𝑨_𝒊𝒕 + 𝜺𝒊 (𝟑)

where 𝜺𝒊 represents the error term. The measurements, predicted signs and precise definitions are given in Table 1.

3.2 Data sources

Main data requirements are that we need to have personal stock and option data of CEOs which can be matched with the financial accounting data of firms. From the Wharthon Research Data Services (WRDS) database, we can retrieve this data for US S&P 1500 firms. Therefore, we limited our sample to the S&P 1500 firms, following Huang et al. (2016) and Malmendier and Tate (2005). We follow the same steps as Huang et al. (2016) to construct the sample by

7 _{In Appendix B, we also use the full set of simultaneous 2SLS equations used in Brockman et al. (2010)}

17 matching three databases within WRDS using thicker codes. First, from ExecuComp, we retrieve the personal compensation data needed to calculate the portfolios of CEOs8. Next, monthly stock prices returns are obtained from the Center for Research in Security Prices (CRSP). These datasets are combined with the Annual Industrial file form Compustat, from which we retrieve all relevant financial and accounting information. For the term structure of interest rates variable we need yields of long-term and short-term government bonds. These yield are obtained from the St. Louis Federal Reserve Bank website9.

We limit our analysis to industrial firms by only incorporating companies with SIC codes between 2000 and 5999 following previous studies (e.g. Datta et al. 2005; Brockman et al. 2010; Huang et al. 2016). By doing so, we exclude financial firms and regulated firms which use leverage for different reasons than industrial firms. This drops our maximum number of unique S&P 1500 firms to 1.134. According to Core and Guay (2002), we need up to 10 years of historical data to perfectly identify the needed characteristics of CEO option portfolios for the use of Black and Scholes valuations. Therefore, our initial dataset incorporates firms from the year 2000 onwards. That dataset contained 13.289 firm-year observations for 1.134 unique firms between 2000 and 2018. Furthermore, non-categorical variables are winsorized by the 1th and 99th percentile to avoid problems with outliers. Following Brockman et al. (2010) and Huang et al. (2016), observations that deviate from sensible bounds (e.g. less than 0% debt or above 100%) are excluded as well. For the measure of abnormal returns (𝑨𝒆𝒑𝒔_𝒊𝒕) we require 1

year ahead price earnings data to proxy for asymmetric information. Therefore, we limit our sample to firms in 2017. We therefore use a time period of 2010-2017, which is in line with the comparable study of Huang et al. (2016) who used a sample for the period 2006-2012. After merging the datasets we end up with our final sample which is an unbalanced panel with 4.802 firm-year observations for 739 unique firms. The loss in observations appears to be random. The difference between the samples is given in table A.1 of Appendix A. Most observations are in the wholesale sector (21%), manufacturing (19%) and non-durable goods sector (12%). Our sample is slightly smaller in terms of firms compared to Huang et al. (2016), which had 4.309 firm-year observations for 944 different firms.

8 _{We note that ExecuComp is also comprised from multiple different databases. From the database Outstanding}

Equity Awards, we retrieve the exercise prices and time to maturity of new granted options that we use to determine whether the CEO is overconfident. This information is merged with the other information obtained from the database Annual Compensation within ExecuComp (e.g. number of exercisable options).

18

4 Results

4.1 Descriptive statistics

Table 2 provides the descriptive statistics for the firm variables. Panel A of Table 2 contains the firm specific variables and shows that 12.5% of total firm debt matures within one year (ST1) on average and 51.1% (ST5) matures within 5 years. These numbers are much lower than Huang et al. (2016) who reported that 17.4% (+4.9%) of total debt had a maturity of less than one year and 59.4% (+7.3%) had a maturity of debt maturing within 5 years. Therefore, it appears that the average amount of short-term debt has decreased with 28.2% (4.9/17.4 =) over a 10 year period. We find more differences compared to Huang et al. (2016). The main difference is that 92.7% percent of all sample firms have a credit rating on long-term debt, compared to 63.7% (-29%) in Huang et al. (2016). There are multiple possible explanations for this increase. It is possible that this increase is due to our larger sample in terms of size. The average market value of the firm in our sample is $ 17 billion, which is $ 1.5 billion (10%) higher compared to Huang et al. (2016). Perhaps a more benign explanation is the relatively low interest rate which arguably makes it easier for a firm to meet the credit rating requirements and also makes lending cheaper. Faulklander and Petersen (2006) state that credit ratings provide the access to the credit market. It appears that much more listed US firms have access to the debt market compared to a decade ago.

Panel B provides the differences between firms that either have an overconfident or unbiased CEO, separated by using the means of the measure LONGHOLDER. Interestingly, and in contrast to Huang et al. (2016), we observe that overconfident CEOs on average hold more long-term debt, but the difference is insignificant for ST1 and only significant at the 10% level for ST5 in terms of mean. However, we note that this observation is the opposite compared to expectations. We also find that overconfident CEOs more often have a credit rating on long-term debt while Huang et al. (2016) observe a significant difference in opposite direction. Consistent with Huang et al. (2016), we observe that firms with an overconfident CEO have on average a lower maturity of the assets (AMAT), more growth opportunities

(MTB) and lower debt levels (LEV). With respect to leverage, this suggest that overconfident

CEOs do not seem to issue more debt as argued by Hackbarth (2008).

Table 3 provides the descriptive statistics in terms of CEO characteristics. In panel A, we present that 24.8% of CEOs hold on to their options until the final year of maturity

19 (median) sensitivity of the CEO portfolio (Delta) is $ 545.482 ($ 115.619) which is a fraction lower compared to Brockman et al. (2010) and Huang et al. (2016) in terms of mean but much

Table 2

Descriptive statistics for the firm characteristics

Panel A: Pooled sample (N = 4.802) for period 2010-2017

mean median Stdev max min skew kurt

ST1 0.126 0.039 0.212 1.000 0.000 2.627 9.975 ST5 0.512 0.489 0.366 1.010 0.000 0.015 1.618 Size ($m) 16.984 4.063 41.859 430.213 0.105 0.005 0.035 Lsize 8.423 8.316 1.570 12.972 4.656 0.331 2.774 Lsize2 73.407 69.157 27.446 168.274 21.677 0.784 3.405 LEV 0.157 0.138 0.132 0.600 0.000 0.917 3.531 LEV2 0.409 0.396 0.182 1.000 0.078 0.539 3.279 MTB 1.914 1.620 1.004 7.033 0.730 2.017 7.907 Aeps 0.006 0.004 0.099 0.840 -0.562 1.389 20.550 Avol 0.055 0.049 0.028 0.171 0.009 1.169 4.585 Amat 9.862 6.569 9.696 51.075 0.272 1.791 6.175 LIQ 2.337 1.993 1.396 8.457 0.422 1.717 6.523 RateD 0.927 1.000 0.260 1.000 0.000 -3.292 11.838 ZSCOREd 0.676 1.000 0.468 1.000 0.000 -0.754 1.569 TERM 1.697 1.580 0.721 3.406 0.386 0.451 2.139 ITCd 0.327 0.000 0.469 1.000 0.000 0.737 1.543 ROA 0.143 0.133 0.073 0.397 -0.074 0.669 4.077 NOLd 0.617 1.000 0.486 1.000 0.000 -0.481 1.231 FA 0.261 0.205 0.197 0.933 0.001 1.136 3.644

Panel B: difference in firm characteristics due to overconfidence

Mean Median Stdev Mean Median Stdev

ST1 0.124 0.041 0.207 0.129 0.034 0.221 ST5 0.506 0.484 0.364 0.525 * 0.503 0.371 Size ($m) 18.290 3.840 56.596 19.336 *** 4.752 54.904 LEV 0.164 0.147 0.135 0.143 *** 0.121 0.125 LEV2 0.416 0.401 0.185 0.393 *** 0.387 0.173 MTB 1.873 1.580 0.986 2.004 *** 1.718 1.038 Aeps 0.008 0.004 0.108 0.003 0.004 0.074 Avol 0.055 0.048 0.029 0.054 0.050 0.026 Amat 10.634 7.045 10.367 8.182 *** 5.645 7.789 LIQ 2.310 1.965 1.388 2.394 * 2.048 1.413 RateD 0.914 1.000 0.280 0.956 *** 1.000 0.204 ZSCOREd 0.685 1.000 0.465 0.657 * 1.000 0.475 TERM 1.697 1.580 0.725 1.697 1.580 0.713 ITCd 0.309 0.000 0.462 0.366 *** 0.000 0.482 ROA 0.141 0.129 0.074 0.148 *** 0.138 0.070 NOLd 0.619 1.000 0.486 0.611 1.000 0.488 FA 0.266 0.212 0.196 0.251 ** 0.192 0.199

* Indicates significance at the 10% level in terms of mean

** Indicates significance at the 5% level in terms of mean *** Indicates significance at the 1% level in terms of mean

Non-overconfident CEOs Overconfident CEOs

20 lower in terms of median, as Huang et al. (2016) reported a median of $ 221.959, which is almost twice as high. In terms of the volatility sensitivity (Vega) we find a similar pattern. The mean (median) Vega is $145.731 ($37.191). It appears that the sensitivity in terms of volatility is lower compared to the study of Huang et al. (2016) which might be because our research period contains no major crisis. However, then we would also expect to find a higher delta since the US stocks prices tripled in the past decade, which should imply that options are more in the money on average. However, since we start measuring our sample from 2000 onwards we also have Delta and Vega values for the period 2006 to 2012. Our median Delta in this period was $ 160.333 and median Vega as $ 41.582 which are both higher than for our sample for the 2010-2017 period. Nevertheless, these medians are still significantly lower than Huang et al. (2016) despite using the same research period. Therefore, it appears that we either have a different sample or that our Delta and Vega variables are measured differently than in Huang et al. (2016). The difference is perhaps driven by the fact that 1.449 out of 4.830 firm-year observations have a Vega of zero. We further observe that CEOs hold on average 1.7% percent of total firm shares and average total CEO compensation amounts to $6.3 million.

In panel B we separate the sample into overconfident and non-overconfident in terms of LONGHOLDER. We report that, in contrast to Huang et al. (2016), overconfident CEOs do not significantly hold more shares (OWN). We also find that overconfident CEOs do not buy more shares than that they sell (NPR) in contrast to Ataullah et al (2018). We do observe that overconfident CEOs have much higher scores in terms of Vega and Delta and tenure, similar to Hirschleifer et al. (2016) and Huang et al. (2016). We also observe that males are more often classified as overconfident compared to females.

21 measure for asset maturity (AMAT) is highly positively correlated (0.82) with one of the instruments for leverage, fixed asset ratio (FA), which suggest that both variables measure the same thing. Furthermore, the correlation between FA and STI and between AMAT and ST1 is close to equal ( both -0.05). Therefore, it seems that both variables can proxy for each other. Since we expected that asset maturity is negatively related to debt maturity and the literature (e.g. Malmendier and Tate, 2005; Brockman et al. 2010; Huang et al. 2016) find no evidence that asset maturity has an impact on leverage, we measure asset maturity according to these studies and threat asset maturity as orthogonal to the error term in the leverage equation implying that we set the coefficient to zero10. With respect to the other proxies for leverage, we find that all other instruments are correlated less with debt maturity than with leverage which is according to expectations.

10 _{In a robustness test, we show that using both variables provides a more reliable result.}

Table 3

Descriptive statistics of the CEOs characteristics

Panel A: Pooled sample (740 firms) for period 2010-2017 for 1.172 CEOs

mean median Stdev max min skew kurt

LONGHOLDER 0.315 0.000 0.465 1.000 0.000 0.797 1.636 NPR 0.039 0.010 0.223 1.000 -1.000 -0.414 10.873 Delta ($ x1.000) 543.685 114.924 1270.538 11678.870 0.000 4.964 33.753 Vega ($ x1.000) 146.260 37.225 258.827 2009.503 0.000 2.985 13.740 Ldelta 4.245 4.753 2.537 9.366 0.000 -0.361 2.049 Lvega 3.094 3.643 2.413 7.606 0.000 -0.130 1.552 OWN 0.018 0.003 0.043 0.261 0.000 3.793 17.553 Age 56.900 57.000 6.930 93.000 30.000 0.441 4.180 Male 0.962 1.000 0.192 1.000 0.000 -4.810 24.138 Tenure 7.184 5.000 7.171 54.000 0.000 1.957 8.360 Total compensation ($ x1.000) 6330.939 4622.573 7037.426 156077.900 0.000 7.337 114.504 Panel B: overconfident and non-overconfident CEOs (N = 4.802)

Mean Median Stdev Mean Median Stdev

NPR 0.039 0.010 0.223 0.043 0.017 0.196 Delta ($ x1.000) 543.685 114.924 1270.538 692.693 210.639 1451.269 Vega ($ x1.000) 146.260 37.225 258.827 205.485 *** 74.384 320.058 OWN 0.018 0.003 0.043 0.018 *** 0.004 0.044 Age 56.900 57.000 6.930 56.771 56.000 6.992 Male 0.962 1.000 0.192 0.979 *** 1.000 0.142 Tenure 7.184 5.000 7.171 8.226 *** 7.000 6.928 Total compensation ($ x1.000) 6330.939 4622.573 7037.426 6330.776 4747.879 5838.116 * Indicates significance at the 10% level in terms of mean

** Indicates significance at the 5% level in terms of mean *** Indicates significance at the 1% level in terms of mean

Non-overconfident CEOs Overconfident CEOs

22

4.2 Initial OLS results

Table 4 provides our initial OLS results using ST1 as our dependent variable in model [1] up to model [6] and ST5 as dependent variable in models [7] and [8]. The effects with respect to debt maturity are presented using no controls, all controls and all controls with either firm fixed effects as suggested by Ataullah et al. (2018) or industry fixed effects

suggested by Huang et al. (2016). According to our hypothesis, we expect that overconfidence is associated with relatively more short-term debt. In contrast to this hypothesis, both our overconfidence variables are insignificant in all models. In unreported results, we find that the amount of short-term debt increased compared to 2010 as we find that all year dummies except 2013 are significantly lower than that of 2010 which is left out in all regressions. We do not interpret the controls at this stage as there likely is endogeniety present in this model which could imply that the results in Table 4 are inconsistent, as OLS does not control for endogeniety.

This table reports the initial results using OLS and fixed effects. The definitions of all variables are given in Table 1. The dependent variables of debt maturity are proxied by ST1 and ST5. The findings with respect to overconfidence measure LONGHOLDER are given in the odd model numbers and the findings regarding NPR are given in the even model numbers. The regressions in model [5] onwards all use fixed effects. Model [5] and model [7] use industry fixed effects based on the Fama and French 48 industry classification. The industry group non-durable goods is left out to avoid the dummy trap in these models. Models [6] and [8] use firm fixed effects. RateD is excluded in these models as this variable barely fluctuates over time within our research period. The sample period is from 2010 to 2017. The year 2010 is left out to avoid the dummy trap. The stars *, ** and *** denote significance at the 10%, 5% and 1% level respectively.

Table 4

Debt maturity and CEO overconfidence using pooled OLS and fixed effects

ST1 ST1 ST1 ST1 ST1 ST1 ST5 ST5 Dependent variable [1] [2] [3] [4] [5] [6] [7] [8] LONGHOLDER 0.005 (0.007) 0.007 (0.011) 0.005 (0.011) 0.021 (0.019) NPR -0.022 (0.014) -0.011 (0.013) 0.001 (0.011) 0.004 (0.020) Male -0.063 ** (0.029) -0.062 ** (0.029) -0.062 ** (0.030) -0.044 (0.032) 0.002 (0.043) 0.051 (0.077) OWN 0.089 (0.143) 0.091 (0.145) 0.051 (0.144) -0.142 (0.177) -0.196 (0.256) -0.048 (0.366) Ldelta -0.003 (0.002) -0.003 (0.002) -0.002 (0.002) 0.003 ** (0.002) -0.003 (0.003) 0.001 (0.003) Lvega 0.001 (0.002) 0.001 (0.002) 0.001 (0.002) 0.000 (0.002) -0.001 (0.003) -0.004 (0.003) TERM 0.004 (0.004) 0.003 (0.004) -0.024 *** (0.008) -0.005 (0.008) -0.009 (0.013) 0.003 (0.011) LEV -0.352 *** (0.066) -0.354 *** (0.066) -0.333 *** (0.066) -0.504 *** (0.074) LEV2 -0.189 *** (0.052) -0.190 ** (0.083) MTB 0.008 (0.005) 0.007 (0.005) 0.009 * (0.005) 0.000 (0.010) -0.010 (0.009) 0.008 (0.018) Lsize -0.132 *** (0.031) -0.132 *** (0.031) -0.121 *** (0.031) -0.129 ** (0.060) -0.102 * (0.053) -0.090 (0.093) Lsize2 0.006 *** (0.002) 0.006 *** (0.002) 0.005 *** (0.002) 0.005 (0.003) 0.002 (0.003) 0.000 (0.005) Aeps 0.007 (0.024) 0.006 (0.024) 0.008 (0.018) -0.015 (0.022) 0.025 (0.037) -0.003 (0.031) Avol 0.060 (0.171) 0.054 (0.172) 0.138 (0.179) -0.233 (0.177) -0.141 (0.346) 0.100 (0.274) Amat -0.001 *** (0.000) -0.001 *** (0.000) -0.002 *** (0.001) 0.000 (0.001) -0.004 *** (0.001) -0.001 (0.002) LIQ -0.020 *** (0.004) -0.020 *** (0.004) -0.019 *** (0.004) -0.055 *** (0.008) -0.052 *** (0.009) -0.046 *** (0.012) RateD 0.043 *** (0.014) 0.043 *** (0.014) 0.043 *** 0.081 *** (0.030) ZSCOREd 0.184 *** (0.012) 0.184 *** (0.012) 0.184 *** (0.012) 0.168 *** (0.016) 0.242 *** (0.022) 0.161 *** (0.026) CONSTANT 0.124 *** (0.004) 0.126 *** (0.003) 0.799 *** (0.140) 0.801 *** (0.140) 0.753 *** (0.148) 0.994 *** (0.268) 1.275 *** (0.252) 1.230 *** (0.418)

Year fixed effects no no no no yes yes yes yes

Industry fixed effects no no no no yes no yes no

Company fixed effects no no no no no yes no yes

F-statistic 0.7 2.5 17.6 *** 17.6 *** 10.2 *** 11.0 *** 13.2 *** 3.8 ***

Adjusted R^2 0.000 0.001 0.173 0.173 0.178 0.140 0.158 0.121

23

4.3 Results using 2SLS

To explore whether the results alter when we control for endogeniety between

leverage and debt maturity, we will use 2SLS regressions suggested by Datta et al. (2005) and Brockman et al. (2010). The regression results are presented in Table 5. We perform a

Houseman test for endogeniety for the instruments in the second stage for both LEV and

LEV2. The null hypothesis of the Houseman test is that all variables are jointly exogenous,

implying that the OLS result in Table 4 is efficient. The Hausman test for endogeniety is only significant for LEV2 using the 4 instruments (test value of 0.2 for market leverage and 7.6 for book leverage) and shows that OLS estimates without controlling for endogeniety is not consistent for book leverage. Therefore, the instrumental variable approach is only justified for book leverage. Next, we perform a test for weak instruments based on the first stage F-statistic. At the first stage, the endogenous explanatory variable leverage is regressed on the 4 instruments as outlined in equation 3. If the test value for weak instruments is above the critical value of 24.58 than they are considered as appropriate instruments for using 4

instruments. We find a test value of 23.2 [model 3] for market leverage (LEV) and a test value of 38.4 [model 4] when using book leverage (LEV2). Therefore, we conclude that our

instruments for market leverage are weak for but not weak for book leverage11, implying that we use LEV2 from Models [5] onwards. Nevertheless, the findings for our overconfidence measures are similar to the OLS results. Both LONGHOLDER and NPR are insignificant in all models12.

After controlling for endogeniety between leverage and debt maturity, we find that most estimates of the coefficients of our control variables have the predicted sign but not all are significant and most depend on the proxy that is used. We find some support for the finding of Brockman et al. (2010) that CEOs with a high degree of price sensitivity use more long-term debt, though we do not find that CEOs with a high volatility sensitivity use more short-term debt.

We find significant results for a positive relationship between debt maturity and liquidity (LIQ) in most models which is consistent with Ataullah et al. (2018). The results partly support the matching hypothesis of Myers (1977) which implies that companies match debt maturity with asset maturity. We also find that companies with higher growth opportunities use

11 _{Since market leverage is considered to have weak instruments, we solely use book leverage throughout the}

24 more short-term debt to reduce the underinvestment problem (Myers, 1977), as MTB is significant in all ST1 models. Similar to Huang et al. (2016), we find some support for the argument that firms use long-debt to increase tax benefits though TERM is only significant for the ST1 regression in model [4]. Regarding firm quality, we find support for the non-linear association of credit quality according to Diamond (1991), as Lsize2 and Lsize are significant in most models that use ST1. Interestingly, and inconsistent with Johnson (2003) and Huang et al. (2016), having a credit rating on long-term debt is associated with more short-term debt which is the opposite of expectations. The same implies for companies with high Z-scores as these are positively significant in all models. The results suggest that high quality firms are associated with relatively more short-term debt. We find no significant results with respect to information asymmetry (Aeps) similar to Huang et al. (2016). These results suggest that CEOs do not use more short-term debt to signal the market positive private information in our sample period (Flannery, 1986). We also find no indication that the amount of stock ownership increases debt maturity, as OWN is never significant.

This table reports the results using 2SLS. The definitions of all variables are given in Table 1. The dependent variables of debt maturity are proxied by ST1 and ST5. Models [1] and [2] present the first stage for leverage using the four instruments presented in equation 3. Models [3] up to [6] present the results for LONGHOLDER. Models [7] and [8] present the results for NPR. The sample period is from 2010 to 2017. All second stage regressions contain year fixed effects where 2010 is left out. All models with LONGHOLDER use industry fixed effects where the industry group non-durable goods is left out. Models [7] and [8] use firm fixed effects. The stars *, ** and *** denote significance at the 10%, 5% and 1% level respectively.

Table 5

Debt maturity and the relation with managerial overconfidence using 2sls

ST1 ST1 ST5 ST1 ST5 Dependent variable [1] [2] [3] [4] [5] [6] [7] LONGHOLDER 0.005 (0.011) 0.007 (0.011) 0.027 (0.020) NPR 0.000 (0.011) 0.005 (0.024) OWN 0.065 (0.145) 0.127 (0.159) -0.057 (0.282) -0.064 (0.169) -0.496 (0.369) Male -0.062 ** (0.031) -0.060 * (0.033) -0.007 (0.047) -0.021 (0.026) 0.009 (0.056) Ldelta -0.003 (0.002) -0.004 ** (0.002) -0.003 (0.003) 0.001 (0.002) 0.003 (0.004) Lvega 0.001 (0.002) 0.004 (0.003) 0.000 (0.004) -0.001 (0.002) 0.002 (0.005) TERM -0.024 *** (0.008) -0.023 *** (0.009) -0.007 (0.014) -0.004 (0.007) -0.001 (0.015) LEV (predicted) -0.244 (0.278) LEV2 (predicted) 0.130 (0.251) 0.299 (0.330) -1.129 * (0.635) 3.052 ** (1.380) MTB 0.013 (0.012) 0.031 * (0.016) -0.012 (0.009) 0.027 *** (0.006) 0.010 (0.014) Lsize -0.128 *** (0.039) -0.154 *** (0.034) -0.150 ** (0.062) -0.160 *** (0.043) -0.168 * (0.094) Lsize2 0.006 (0.002) 0.007 *** (0.002) 0.005 (0.003) 0.006 ** (0.003) 0.004 (0.006) Aeps -0.001 (0.031) -0.036 (0.027) -0.031 (0.051) 0.055 (0.055) -0.262 ** (0.119) Avol 0.220 (0.316) 0.556 * (0.284) 0.388 (0.495) -0.302 (0.230) 1.013 ** (0.501) Amat -0.002 *** (0.001) -0.002 *** (0.001) -0.004 *** (0.001) -0.001 (0.002) 0.005 (0.003) LIQ -0.018 *** (0.004) -0.014 ** (0.007) -0.028 (0.019) -0.083 *** (0.019) 0.044 (0.041) RateD 0.044 *** (0.015) 0.050 *** (0.016) 0.093 *** (0.032) ZSCOREd 0.178 *** (0.021) 0.156 *** (0.017) 0.210 *** (0.035) 0.194 *** (0.023) 0.054 (0.051) FA 0.149 *** (0.010) 0.098 *** (0.009) NOLd 0.018 *** (0.004) 0.026 *** (0.004) ITCd -0.024 *** (0.004) -0.042 *** (0.004) ROA -0.585 *** (0.024) -0.751 *** (0.024) CONSTANT 0.200 *** (0.006) 0.324 *** (0.006) 0.846 *** (0.144) 0.822 *** (0.159) 1.212 *** (0.260) 1.568 *** (0.293) 0.072 (0.638)

Year fixed effect no no yes yes yes yes yes

Industry fixed effect no no yes yes yes no no

Company fixed effect no no no no no yes yes

Company clustered errors yes yes yes yes yes yes yes

F-statistic 221.3 *** 50.1 *** 9.8 *** 9.4 *** 13.5 *** 12.8 *** 4.9 ***

Adjusted R^2 0.335 0.443 0.170 0.128 0.160 0.127 0.098

Observations 4,802 4,802 4,802 4,802 4,802 4,802 4,802

Hausman test (second stage) 0.2 7.6 *** 116.5 ***

Weak instrument test 23.2 38.4 38.4 3.1 3.4

25 So far, both of our overconfidence measures are insignificant. Huang et al. (2016) and Malmendier and Tate (2005) argue that using firm fixed effects mainly controls for multiple short-tenured CEOs which can lead to sample bias. Furthermore, the NPR ratio was never significant1314, implying that we do not find indication that CEOs learn to become

overconfident over time because of self-attribution bias suggested by Ataullah et al. (2018). Therefore, we focus on industry and year fixed effects for only our LONGHOLDER variable. As outlined in the previous section, we find that our instruments are not weak using book leverage. Therefore, throughout the rest of this paper, we will now focus solely on our results using LEV2.

4.4. Overconfidence and liquidity risk, leverage and growth opportunities

One of the main findings of Huang et al. (2016) was that their strong association between CEO overconfidence and the use of short-term debt was not deterred by liquidity risk. They state that overconfident CEOs need to make a trade-off between the associated benefits, in the form of an expected increase in firm value, to the costs, which is primarily a higher level of liquidity risk. If overconfident CEOs do not take liquidity risk into account in their decision to issue short-term debt, than it would imply that there is no relation between overconfidence and liquidity. When the trade-off between more short-term debt and more liquidity is made, however, it could imply that only overconfident CEOs with a relatively low level of pre-existing liquidity risk will issue more short-term debt. If this is the case, than the relation between overconfidence and liquidity risk is concentrated in companies with a relatively low level of liquidity risk. We follow Huang et al. (2016) to test whether we find different results. If the decision of an overconfident CEO to issue short-term debt is

counteracted by an increase in liquidity risk, than we should find a negative significant coefficient between the interaction of liquidity risk and overconfidence.

Three different measures are used for liquidity risk, from which two are identical to Huang et al (2016), namely leverage (LEV) and asset volatility (Avol) suggested by Johnson (2003). We also use the level of liquidity (LIQ) as proxy for liquidity risk as short-term debt decreases liquidity. The results are presented in panel A of Table 6. Interestingly, the

interaction of leverage and overconfidence is negatively significant in model [1], though at

13 _{In unreported results, we find that using industry fixed effects also does not lead to significant results for the}

NPR ratio.

14 _{LONGHOLDER is also never significant up to this point, but since this measure is used most often in the}

26 the 10% level and only in this model. Furthermore, LONGHOLDER is now positively

significant at the 10% level in model 1. The estimated positive coefficient of LONGHOLDER indicates that firms with an overconfident CEO have 16% more debt that matures within one year compared to non-biased CEOs. When we do not take leverage into account it would imply that the average amount of debt that matures within 12 months (ST1) is highly

economically significant since the average short-term debt is 12.6%, as outlined in Table 2. It indicates that overconfident CEOs with a low leverage hold a maximum of (16% / 12.6 = ) 127.0% more short-term debt on average holding everything else constant. However, the size of the interaction term is much stronger than the size of LONGHOLDER. Taken together, overconfident CEOs on average hold (16.0 – 35.5 =) 19.5% less short-term debt that matures within one year compared to CEOs not classified as overconfident. There is only weak evidence that this is the case, as this is only significant at the 10% level in 1 of the 6 models. Nevertheless, this evidence suggest that the action of CEOs to borrow short-term debt is affected by the level of liquidity risk which is in contrast to Huang et al. (2016). The evidence suggest that overconfident CEOs take liquidity risk into account in their debt maturity

decision

In Table 6, we also test whether the results alter by testing the agency cost of debt hypothesis used in Ataullah et al. (2018). Firms with more growth opportunities have more agency problems in the form of debt overhang and asset substitution which can be reduced with the use of short-term debt. As Ataullah et al. (2018) in their sample find that

overconfidence is linked to more long-term debt, they show that overconfident CEOs have more underinvestment problems in firms with a high degree of growth opportunities,

measured by M2B. To see whether overconfident CEOs in our sample are affected by growth opportunities, we use an interaction term between MTB and LONGHOLDER. The results are presented in models [7] and [8] of Table 6. The interaction term is not significant in both models. If anything, we find that LONGHOLDER is positively significant in model [8] at the 10% level. Overconfident CEOs on average have (6.2 / 12.6= ) = 49.2% more debt that