CEO overconfidence and the method of payment in mergers and acquisitions

(1)

MSc Business Economics, Finance track

Master thesis

CEO overconfidence and the method of payment in Mergers and Acquisitions

Bunnik, Sebastiaan

July 2015

Thesis supervisor: Dr. Florian Peters

Abstract

In this thesis EPS guidance data is used to construct an overconfidence measure, which in

turn is used to research the effect of CEO overconfidence on the method of payment in

M&A deals in the US between 2002 and 2013. A significant relationship is found that shows

an increased likelihood of a deal financed with cash only when the acquiring CEO is

overconfident.

(2)

2

Statement of Originality

This document is written by Sebastiaan Bunnik who declares to take full responsibility for

the contents of this document.

I declare that the text and 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)

3

Mergers and acquisitions (M&A), transactions where a company buys another company, are events of great economic importance and, as a consequence, are one of the most extensively studied phenomena in economic literature. They are of great importance for two reasons: first, they are large in magnitude, which means a large shift in both the financial and organizational structure of the companies involved. Second, they are the most public of investments done by a company, which means that investors can more easily observe preferences and behaviour of the management of the companies. One such preference is observed when the acquiring firm chooses the method of payment. Because of the size of a typical M&A transaction, this is an important choice and has been shown to significantly impact the ultimate value of the deal. Several authors find significantly better returns when a deal is financed with cash for both targets (Wansley, Lane, and Yang, 1983) and acquirers (Travlos, 1987). The most common methods of payment are cash and stock (Eckbo and Langohr, 1989), where a combination of the two is used as well. Because of the consequences of this choice, it is important for investors to understand the forces that drive this decision.

The current literature on the method of payment in M&A determines two major factors that play a role in the choice for cash, stock, or a mix of both. The first is tax-related; when receiving a cash offer, the shareholders of the target firm are liable to pay capital gain tax. When receiving a stock offer, on the other hand, they are only liable at the point in time where they sell that stock, in other words the tax liability is deferred into the future and thus has a lower present value. Several studies find this to be an important advantage of a stock offer (Eckbo and Langohr (1989), Huang and Walkling (1987), Martin (1996), Mayer and Walker (1996), among others). The second major factor found in the literature is the loss of control when a stock deal is completed. Because the shareholders of the target firm receive a share in the acquiring firm, the original acquiring

shareholders face a dilution of their voting power and claim on profits. This makes a cash offer more attractive to the acquirer, evidence for which has been found by Huang and Walkling (1987), Faccio and Masulis (2005), Mayer and Walker (1996), and several others. Other significant variables that influence the method of payment are the level of cash and debt the acquiring firm possesses, the size of the deal relative to the acquirer, and the stock volatility of the acquirer. A more in-depth explanation on these variables will be given in parts 3 and 4. In general, the pecking order theory (Myers, 1984) predicts that managers prefer to finance their investments with internal means rather than external means, and prefer safe external means (debt) over risky external means (equity).

Apart from the reasonably obvious explanations for the choice for financing an M&A deal with cash, stock, or a mix listed above, research done by Ben-David, Graham, and Harvey (2007) and

(5)

5 Brown and Sarma (2007) suggests that overconfidence plays a role in the M&A process as well. Overconfidence is defined as the tendency of economic players to underestimate the variance of a potential outcome. Ben-David et al. and Brown and Sarma find that overconfident CEOs make more acquisitions and are rewarded more poorly for it by the market. Among other authors who find that overconfidence influences business decisions are Malmendier and Tate (2008), incorporating the level of internal financing available to CEOs into the M&A research by Ben-David et al. and Brown and Sarma. Malmendier and Tate find that overconfident managers have a strong tendency to prefer internal financing, which they explain by the theory that overconfident managers overestimate the value of their company’s stock and thus prefer not to use it for investments. These results confirm the pecking order theory and would suggest that overconfident managers also prefer to finance their acquisitions with cash rather than with equity, and potentially that this effect is stronger for overconfident managers than for the total population of managers.

This paper researches the relationship between the level of CEO overconfidence and the method of payment in M&A. As the pecking order is expected to have more influence on

overconfident CEOs, CEO overconfidence is expected to increase the chance of cash financing. Although research has been done on the link between CEO overconfidence and the frequency of doing M&A deals, and between CEO overconfidence and the financing preferences for investments in general, this particular link has not been studied directly, to the best of my knowledge. If the expected relationship turns out to be correct, this would give investors more reason to take CEO overconfidence into account. If it doesn’t, this might be an indication that the pecking order theory of finance doesn’t apply more strongly to overconfident managers than to other managers.

Using data on earnings per share (EPS) guidance given by CEOs of US firms between 2002 and 2014, this research constructs a proxy for overconfidence based on how narrow this guidance is and links it to the name of the CEO. Next, it is matched to a list of M&A deals, along with data on company financials and the stock price of the acquiring company. Logit regressions are run where the dependent variable is a dummy that is 1 if the deal was financed with cash only, and where the overconfidence proxy is the main explanatory variable. If the coefficient on the overconfidence proxy is significantly negative, we can say that wider forecast ranges lead to a lower probability of cash financing, which is analogous with the conclusion that overconfidence increases the probability of cash financing. I find that this is indeed the case.

The rest of this paper is structured as follows. Part 2 discusses the relevant current literature on the topics of overconfidence and the method of payment in M&A. It goes on to present the main hypothesis. Part 3 explains the econometric methodology used to test this hypothesis. Part 4 goes

(6)

6 through the process of data collection and data preparation, and ends with the descriptive statistics of the sample and the coefficients of interest. Part 5 presents the main results through which the hypothesis is confirmed or rejected. Part 6 presents robustness checks that are performed to see if the results depends on the characteristics of the sample or the form of the dependent variable. Part 7 concludes the paper.

2. Literature review

2.1 Mergers and acquisitions

Mergers and acquisitions (M&A) are events where one company buys another, or where two companies merge into one. Few studies make a distinction between mergers on one side and acquisitions on the other; most treat both events as being of the same kind, as does this study. The most commonly used argument in favour of both mergers and acquisitions is that they could result in cost and/or revenue synergies. However, many empirical studies do not find a significant positive impact of mergers and acquisitions (see, among many others, Moeller, Schlingeman, and Stulz (2005), Servaes (1991), and Kaplan and Weisbach (1992)). At the same time, global quarterly M&A volume has never been lower than $500 billion since 2009, with the number of deals never dropping below 7.000 (Bloomberg finance, 2014). If empirical evidence doesn’t seem to support the

expectation, or hope, of M&A adding value to businesses, why then do companies keep spending trillions of dollars each year buying other companies? This intuitive contradiction suggests there might be other forces at play. Behavioral distortions such as overconfidence, overoptimism, and loss aversion might lead to non-standard beliefs and preferences, or indeed even irrational decision making such as consistently undertaking unprofitable mergers and acquisitions.

2.2 Overconfidence

According to Prospect Theory (Kahneman and Tversky, 1979) economic players regularly have non-standard beliefs, one of which is overconfidence. Overconfidence in a financial setting means that economic players underestimate the variance of a potential outcome, fall for the illusion of control, and/or too often think they are better than average. This is different from overoptimism, where people overestimate the expected return of a potential outcome but not necessarily underestimate the variance. In fact, a person can be overoptimistic and at the same time overestimate the variance, which would make her ‘underconfident’. Ben-David, Graham, and Harvey (2013) call the part where the variance of a potential outcome is underestimated miscalibration, and argue that miscalibration

(7)

7 can be split into two things; the overestimation of a person’s ability to predict the future and the underestimation of the volatility of random events. This study will mainly focus on the first aspect of miscalibration rather than the illusion of control or the better-than-average effect, however, for simplicity this will still be called overconfidence.

With the emergence of behavioural finance as a field of interest to scholars, evidence of the impact of overconfidence in the financial world is growing. Ben-David et al. (2013) find that

overconfidence is positively related to the willingness of managers to invest, as well as to their tolerance of debt. Therefore, the level of overconfidence of the CEO of a company directly influences the risk profile of that company and thus should be of interest to investors. Roll (1986) first linked the notion of overconfidence to M&A behaviour and finds that CEO overconfidence is at least as important in M&A decisions as other factors as taxes, synergies, and inefficient target management. In a separate article, Ben-David, Graham, and Harvey (2007) find that overconfident managers tend to do more acquisitions than non-overconfident managers, which is in line with their finding that overconfident managers are more willing to invest, but which is more relevant to this study. This finding is confirmed by Brown and Sarma (2007). Interestingly, Ben-David et al. (2007) find that overconfident managers achieve lower abnormal announcement returns when undertaking M&A activity. This seems to indicate that investors don’t see overconfidence as a good thing in an M&A context. Doukas and Petmezas (2007) reach the same conclusion not only when looking at abnormal short-term returns, but also when looking at long-term performance.

In a study with hypotheses similar to those tested by Ben-David et al. (2007), Malmendier and Tate (2008) find that the effect of overconfidence on merger frequency is ambiguous. According to their findings, if overconfident CEOs have internal financing available, they unambiguously do more mergers than non-overconfident CEOs, but they are reluctant to issue external financing to fund investment so if they don’t have internal financing available they might do less mergers than non-overconfident CEOs. They, too, find that market reactions to merger announcements by

overconfident CEOs are significantly more negative than those by non-overconfident CEOs: they find abnormal returns for the bidding firm in the three days around the merger announcement of -0.90% for overconfident CEOs and -0.12% for non-overconfident CEOs. Malmendier and Tate reason that overconfident managers earn more negative abnormal returns because their overconfidence leads them to overpay and/or overinvest in negative NPV projects because they think they have more control than they actually do (the illusion of control) on the one hand, and on the other hand that they don’t invest in positive NPV projects because they are reluctant to issue external financing even when it would have been beneficial for the company.

(8)

8 Other decisions influenced by overconfidence in a corporate policy setting include the payment of dividends, which overconfident managers are less likely to do, and share repurchases, which overconfident managers are more likely to do (Ben-David et al., 2007). The studies listed above show that overconfidence, might have a significant impact on several aspects of corporate decision making. Arguably the biggest decisions in corporate finance are made when a company performs M&A transactions. As shown by Malmendier and Tate (2008), Ben-David et al. (2007), Doukas and Petmezas (2007), and Brown and Sarma (2007), among others, decisions on M&A transactions are not immune to managerial overconfidence. A rationale behind this is given by Malmendier and Tate (2005), who reason that individuals are more prone to overestimate outcomes, or underestimate variances, when they believe they are in control of the outcome and when they are highly committed to the outcome. CEOs in general satisfy both conditions, especially so when making big decisions, because they get the ultimate say and therefore think they have more control than they actually do, and because their compensation and reputation depends largely on the outcomes of the big decisions they take. Moreover, CEOs receive low-frequent and noisy feedback, which also accommodates overconfidence.

A challenge in any study on overconfidence is how to quantify such a personal characteristic. Because overconfidence is not a publicly available statistic, a proxy needs to be used. In the sense that overconfidence is the tendency to underestimate the variance of a potential outcome, the ideal proxy would be a forecast range made by a manager. Ben-David, Graham, and Harvey (2007 and 2013) have created such a proxy by conducting a survey that asks CFOs for their 80% confidence interval of the value on the S&P 500 one year from now. They find that CFOs are severely

miscalibrated; they underestimate the S&P’s variance to such an extent that they are right only 36% of the time when giving an 80% confidence interval. The only drawback to this methodology is that the data the authors have used is not easily available, and might be subject to a form of self -selection because managers have the choice to fill out the survey or not.

In absence of the data available to Ben-David et al., Malmendier and Tate (2005a, 2005b, and 2008) use two alternative measures of overconfidence. First, they construct a measure based on the option exercising behaviour shown by CEOs, where holding on to options that are in-the-money and that can be sold is said to be a sign of overconfidence. Secondly, they quantify a CEO’s press portrayals so that CEOs who are portrayed in the financial press as optimistic or confident more often than others are said to be overconfident. There are imperfections in both Malmendier and Tate measures, however. In the first measure, the reasons a CEO has for holding on to her options can extend beyond her beliefs or preferences; she might have inside information (although the authors find that overconfident managers don’t outperform the index which might indicate that

(9)

9 inside information is no reason for holding on to the options), a desire to send a signal to investors, a tolerance for risk, or other reasons unrelated to her level of confidence. Similarly, a standard level of confidence isn’t necessarily the only reason to sell such options; this may be done for liquidity or diversification reasons as well. Malmendier and Tate’s second measure has the drawback that it is based on a perception of outsiders. CEOs may have an interest to give a perception of optimism or confidence, even if they are not. Even if this is ignored, the measure is still subjective and too dependent on the journalists’ and researchers’ opinions on optimism and confidence.

In what can be seen as a test of Malmendier and Tate’s press-based measure of

overconfidence, Hribar and Yang (2010 and 2015) find that positive portrayals in the financial press have a significant relationship with an overconfidence measure that more accurately fits the

definition of underestimating the variance of an outcome. Hribar and Yang argue that one of the few instances in which CEOs publicly submit forecast ranges is when they announce their earnings per share (EPS) forecast for the next year. Testing if this can in fact be used as a reliable overconfidence measure, they find that CEOs who are portrayed positively in the press are more likely to miss their EPS forecasts, submit forecast ranges that are more narrow, and are more likely to submit point forecasts rather than range forecasts. Additionally, they find that overconfident CEOs are more likely to manage their earnings in the sense that they boost earnings artificially. Because of these findings, it can be assumed that EPS forecast ranges can reliably be used to construct an overconfidence measure. As will be explained in part 3, this is what will be done in this article.

2.3 Method of payment

When submitting an offer for a target firm, one of the decisions the acquiring CEO needs to make is how to finance the deal. In general, she has the choice between paying for the target firm with cash, stock, or a mix of the two. This is a crucial choice both for the target’s investors and for the

acquirer’s investors. The target’s investors are obviously impacted greatly because they sell their full stake in the company and get cash and/or stock of the acquiring company in return. For the

acquirer’s investors, on the other hand, the choice is important as well due to a potential loss of control (in case of a stock offer) and potential increased bankruptcy risk (in case of a cash offer). Furthermore, the choice of payment impacts the cash flows and the financial structure of the acquiring firm (Faccio and Masulis, 2005). The specific reasons for choosing cash or stock as the method of payment are discussed in this part.

(10)

10 One of the most crucial and material factors in the choice for the method of payment is the capital gain tax (see e.g. Eckbo and Langohr (1989), Huang and Walkling (1987), Martin (1996), Mayer and Walker (1996)). When target shareholders sell their stakes in return for cash, they generally realize a capital gain and thus have to pay capital gain tax. When they sell their stakes in return for stock, on the other hand, the capital gain is not realized immediately and the liability to pay tax is postponed. Thus, all else equal, stock offers are more attractive for target shareholders than cash offers by the difference between the current capital gain and the discounted future capital gain in case of a stock offer, multiplied by the tax rate. Because acquiring managers are aware of this, they know that this is a factor that makes it more likely that their stock offer will be accepted, compared to their cash offer.

A major reason to finance a deal with cash for a CEO who represents her shareholders, on the other hand, is the main drawback of a stock offer; a loss of control of the acquiring firm (Huang and Walkling (1987), Faccio and Masulis (2005), Mayer and Walker (1996)). Whenever a deal is financed with stock, the voting power of the current shareholders will decrease as the shareholders of the target get newly issued equity. Faccio and Masulis argue that this argument is especially powerful when the acquiring company has a majority shareholder who has between 20 and 60 percent of the voting power. Another advantage of financing a deal with cash is that it avoids an agency problem resulting from holding too much cash; because managers have the tendency to build empires, they are inclined to use excess cash to invest in negative NPV projects rather than distributing it to shareholders (Jensen, 1986). Regulatory requirements can be a reason to finance a deal with cash as well. Huang and Walkling (1987) argue that stock offers are subject to more regulatory issues, and therefore take longer to complete. This increases uncertainty in the form of the fluctuating valuation of the target firm, and threat of third parties getting involved in the bidding process. Because acquiring CEOs have to deal with less regulation when they make an offer with cash, this enables them to complete the deal quicker. Despite these reasons to pay with cash, Travlos (1987) argues that the most important reason is signalling; if a deal is financed with stock rather than cash, this might give investors the signal that the stock is overpriced and might lead them to sell the stock. Therefore, acquiring a company using cash as the method of payment is said to be a positive signal. A final reason for paying with cash for companies with a high fraction of tangible assets is that acquiring debt can be relatively cheap and provides a tax shield (Faccio and Masulis, 2005).

Apart from the capital gain tax, an important reason to pay with stock is the higher costs of financial distress if debt needs to be raised in order to finance the deal with cash. The more debt a company acquires, the more likely a default will be and the more expensive debt becomes. From a

(11)

11 target CEO’s point of view, stock offers have the advantage that they will receive voting power in the acquiring firm. Ghosh and Ruland (1998) find that CEOs of target firms who own large stakes in their company are more likely to accept stock offers. Furthermore, they find that target executives retain their jobs more often when the deal is financed with stock, which increases the likelihood of the offer being accepted.

Considering these reasons to finance a deal with either cash or stock (or both), one might wonder which reasons investors value more. Ample research has been done to the short-term and long-term acquisition announcement abnormal returns, to find if the method of payment has any effect on the value an acquisition adds or destroys. Most authors find that, from either a target, an acquirer, or a combined firm point of view, acquisitions paid with cash are rewarded more positively than acquisitions paid with stock (Eckbo and Langorh (1989), Huang and Walkling (1987), Wansley, Lane, and Yang (1983), and Travlos (1987)). Wansley et al., for instance, find that target firms earn cumulative abnormal returns of 34% in the 4 months surrounding an acquisition announcement when the deal is to be financed with cash, and 17% when it is to be financed with stock. Looking at acquirers rather than targets, Travlos finds statistically significant cumulative abnormal returns of -4% in the 11 days surrounding an announcement when the deal is to be financed with stock, versus zero abnormal returns when the deal is to be financed with cash.

2.4 M&A, overconfidence, and the method of payment

In an attempt to create a formal theory of the choices managers make when deciding on their firms’ capital structure, Myers (1984) first used the term ‘pecking order’ to explain that managers prefer to finance investments with internal resources. If these are not available, they prefer the next safest source of financing, being debt. Least preferable is the issuing of new equity, because it gives investors a signal that the equity is overpriced and because it dilutes the current investors’ voting power. For the overconfident CEO, there is a third reason not to issue new equity: because she underestimates the risk of her company, she overestimates the (risk-adjusted) value of her company, at least compared to the market’s valuation. Therefore, she believes the stock of her company is underpriced and will not want to issue equity.

If overconfident CEOs are more likely to finance investments with cash, this effect will be even more obvious when deciding on the method of payment for an acquisition; these are generally the largest investments CEOs make and thus would require the largest issue of equity, if this method of payment is chosen. The main hypothesis of this article is therefore that CEO overconfidence

(12)

12 positively influences the probability of a deal being financed with cash. However, as found by Doukas and Petmezas (2007) and Ben-David et al. (2007), overconfident CEOs earn lower abnormal returns when announcing an acquisition, whereas Wansley et al. (1983) and Travlos (1987) find that cash-financed acquisition announcement earn higher abnormal returns. If the hypothesis is confirmed, meaning the alternative hypothesis that CEO overconfidence has a zero or a negative effect on the probability of a deal being financed with cash is rejected, this would suggest that the difference in abnormal returns between cash-financed and stock-financed deals is even larger than found in previous literature, when CEO overconfidence is accounted for.

3. Methodology

The main hypothesis is that CEO overconfidence positively influences the probability of a deal being financed with cash. To test this hypothesis, a logistic regression is executed. The main dependent variable in the regressions is Cash, which is a dummy variable taking the value of 1 if the deal in question is financed purely with cash, and 0 otherwise. The main right-hand variable to test the hypothesis is Forecast range, which is the size of the EPS forecast range relative to the share price at the time of the guidance announcement. In subsequent tests the main right-hand variables Forecast accuracy and Forecast error are used. Forecast accuracy is a dummy which takes the value of 1 if the actual EPS fell within the forecast range, or was equal to the point forecast if one was given, and 0 otherwise. If the forecast was accurate, Forecast error takes on the value of zero, and if the forecast was inaccurate it is the distance between the actual EPS and the nearest border of the forecast range, divided by the share price at the time of the guidance announcement.

In order to test on causality instead of mere correlation, control variables are added. I have attempted to add all control variables that the literature predicts have an influence on the method of payment, and are possibly related to the level of CEO overconfidence. Main control variables included in the regressions are the level of cash the acquiring company possesses and the acquiring firm’s size, which are both expected to increase the chance of cash financing, and the level of debt of the acquirer, the size of the proposed deal relative to the acquirer’s market share, and the acquirer’s stock volatility, all three of which are expected to decrease the chance of cash financing. The level of cash held by the acquiring firm and the relative deal size are expected to have the largest direct effect on the method of payment; the pecking order theory predicts that CEOs should use cash as the method of payment as long as there is enough cash available, and the acquirer’s cash level and the relative deal size are expected to most significantly impact this. Additional control variables and summary statistics on all variables can be found in part 4.4. To control for potential fluctuations over

(13)

13 time, year-fixed effects are added to the model. The same is done with industry-fixed effects to control for potential fluctuations by industry. Data limitations prevent me from adding the level of majority shareholder control in acquiring and target firms, which the literature finds to be a significant influence. However, with the assumption that CEO overconfidence doesn’t impact the level of majority shareholder control and vice versa, the results obtained should be relevant nonetheless.

To test the significance of the variables, the following logistic model with heteroskedasticity-robust standard errors is estimated:

𝐶𝑎𝑠ℎ = 𝛽

₀

+ 𝛽

₁

𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡_𝑟𝑎𝑛𝑔𝑒 + 𝛽

₂

𝑐𝑎𝑠ℎ_𝑙𝑒𝑣𝑒𝑙 + 𝛽

₃

𝑑𝑒𝑏𝑡_𝑙𝑒𝑣𝑒𝑙 + 𝛽

₄

𝑎𝑐𝑞𝑢𝑖𝑟𝑒𝑟_𝑠𝑖𝑧𝑒

+ 𝛽

₅

𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒_𝑑𝑒𝑎𝑙𝑠𝑖𝑧𝑒 + 𝛽

₆

𝑣𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦 + 𝛽

₇

𝑑𝑒𝑎𝑙𝑣𝑎𝑙𝑢𝑒 + 𝛽

₈

𝑄 + 𝛽

₉

𝑆𝐸𝑄

+ 𝛽

₁₀

𝑎𝑔𝑒 + 𝛽

₁₁

𝑎𝑐𝑐𝑢𝑟𝑎𝑐𝑦 + 𝛽

₁₂

𝜑

_𝑡

+ 𝜀

where Cash is a dummy variable with the value of 1 if the acquisition is financed with cash only and 0 otherwise, forecast_range is the size of the forecast range relative to the acquirer’s stock price, cash_level is the acquirer’s level of cash, debt_level is the acquirer’s level of debt, acquirer_size is the log of the acquirer’s total assets, relative_dealsize is the deal size relative to the acquirer’s market cap, volatility is the acquirer’s stock volatility, dealvalue is the log of the dollar deal value, Q is the acquirer’s market-to-book ratio, SEQ is the log of the acquirer’s shareholders’ equity, age is the log of the acquiring CEO’s age, accuracy is a dummy that is equal to 1 if the forecast was accurate, 𝜑_𝑡 includes a dummy for each year in the sample where t is 2002-2013, and 𝛽0 and 𝜀 are the constant

and error term, respectively. The same model is tested again with accuracy as main right-hand side variable and without forecast_range, and once more with forecast_error, which is the relative forecast error, as main right-hand side variable and without accuracy and forecast_range.

The coefficient of major interest is 𝛽₁. This coefficient is expected to be significantly negative, because this would mean that smaller forecast ranges, which are explained by

overconfidence, increase the probability of a deal being financed with cash. If it is not, this could mean that either overconfidence has no significant effect on the method of payment in M&A, or that the limitations of the research obscure the actual effect. As long as overconfidence can only be estimated through proxies and is not a publicly available, undisputable metric, there is always a possibility that the proxy is actually reflecting other variables than overconfidence alone.

Furthermore, no dataset is complete so a failure to confirm the main hypothesis might have to do with the limited amount of data available for this particular study. That being said, I have attempted to construct a methodology that is as objective and general as possible, exclusively using public, numerical data provided by professional institutes. The scope of the data in terms of time and

(14)

cross-14 section is as broad as possible, only excluding time periods for which the data is not available and firms for which the data is expected to be noisy.

Apart from 𝛽₁, no coefficients directly confirm or reject the main hypothesis. However, the sign and magnitude on the other coefficients, most importantly 𝛽₂ and 𝛽₅, do matter as they might point to potential data issues. Because the variables to which these coefficients apply are theorized to have a large and unambiguous effect on the probability of cash financing, counter-intuitive results give a solid indication that the data is less than perfect. For this reason the sign and magnitude of 𝛽₂ and 𝛽5 are important for the credibility of the results. In part 4.4 summary statistics on these

three important variables are presented, and the coefficients will be shown in part 5.1.

4. Data and descriptive statistics

4.1 Mergers & Acquisitions

This study investigates the relationship between managerial overconfidence and the method of payment in mergers and acquisitions. Therefore, the starting point from a data point of view is a sample of mergers and acquisitions, or deals. A list of deals is downloaded from the Thomson One database. The event of interest is the initial announcement of the bid. In cases in which the initial bid is revised, the initial bid is still focused on because the acquiring CEO’s behaviour is the variable of interest, and I believe the initial bid (before negotiations) best reflects this behaviour. Because the data for the overconfidence measure (see 4.2) is available from 2002 until 2014, deals announced before 2002 are excluded from the sample. Furthermore, pending and failed deals and deals with a bid value of less than $1 million are excluded, as are deals in which the acquirer is based outside of the US. The data downloaded according to these restrictions contains details on the method of payment, in particular the percentage of the total (proposed) deal paid with cash and stock. There are two minor payment categories being ‘other’ and ‘unknown’. As the factor ‘cash’ is of interest in this study, these are grouped with ‘stock’ and the group is called ‘non-cash’. A dummy variable ‘Cash’ is created which takes the value of 1 if the deal is completely financed with cash and 0 otherwise. This will be the dependent variable in the main regressions as shown in the Results section. To ignore erroneous data, deals where the percentages add up to more than 100 are removed. Deals where the acquirer owns less than 50% of the target’s shares after the deal are removed as well.

(15)

15 4.2 Overconfidence measures, CEO data, company financials

The measure for overconfidence is derived from company guidance data. Usually firms provide forecasts for their next year’s earnings per share (EPS) when they present their annual results. In the vast majority of events, this forecast is a range rather than a point estimate. The EPS forecast details are downloaded from the I/B/E/S Guidance database. The companies in this database are by

definition public companies. As in the Thomson One data, only US-based companies are used. Often firms present revised EPS forecasts when they present quarterly results. These revised forecasts, as well as any forecasts that are not annual EPS forecasts, are removed. This leaves 15.097 EPS forecasts, 2.591 (17,2%) of which were point estimates and 12.506 (82,8%) of which were ranges. The actual EPS data for the companies in this subsample are downloaded from the I/B/E/S Actuals database and merged into the I/B/E/S Guidance file using the I/B/E/S ticker.

Data for the main control variables, such as acquirer firm size, cash reserves, and debt level, comes from the CRSP/Compustat merged database. Acquirer firm size is defined as the natural logarithm of the acquiring company’s assets (ACT) in the year of the bid announcement, cash reserves are defined as the natural logarithm of the cash and short-term investments (CHE) of the acquiring company in the year of the bid announcement, and debt level is defined as the natural logarithm of the sum of the acquirer’s debt in current liabilities (DLC) and the acquirer’s long-term debt (DLTT) in the year of the bid announcement. Acquirer shareholder equity is defined as the natural logarithm of the acquiring company’s shareholder equity (SEQ), the acquiring company’s market-to-book ratio (Q) is defined as the market capitalization 40 business days before the deal announcement (to avoid the effect of the run-up of the stock price in the period before the deal announcement, as suggested by Betton, Eckbo, and Thorburn (2008)) divided by the book value of equity in the year of the deal announcement, and, finally, the stock volatility is defined as the standard deviation of the acquiring company’s monthly stock returns in the 24 months ending 40 business days before the deal announcement. The data is merged into the I/B/E/S file using the company’s gvkey.

To link EPS forecasts to individual CEOs, the name, age, and period in office of each CEO is downloaded from the Execucomp database and merged into the I/B/E/S file using the company’s gvkey, matching the date of the EPS forecast announcement with the details of the CEO who was in office on that day. Because not all firms in the I/B/E/S file were also present in the Execucomp database, some observations are lost. Finally, the company’s split-adjusted stock price on the day of the EPS forecast announcement is downloaded from the CRSP database and merged into the file

(16)

16 using the company’s permno. Observations where the firm was a penny stock, defined as having a share price of less than $5, are removed. The sample now consists of 10.686 EPS forecasts.

Two measures of overconfidence are constructed based on the EPS forecast data; a forecast range and an accuracy factor. The forecast range is calculated as:

𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡_𝑟𝑎𝑛𝑔𝑒 =𝑓𝑐𝑠𝑡

𝑢𝑝𝑝𝑒𝑟 _{− 𝑓𝑐𝑠𝑡}𝑙𝑜𝑤𝑒𝑟

𝑝𝑟𝑖𝑐𝑒 ∗ 100%

Where 𝑓𝑐𝑠𝑡𝑢𝑝𝑝𝑒𝑟 and 𝑓𝑐𝑠𝑡𝑙𝑜𝑤𝑒𝑟 are the upper and lower limits of the EPS forecast range,

respectively, and price is the firm’s stock price on the date of the EPS forecast announcement. When an EPS forecast is a point estimate rather than a range, the forecast range is zero. The accuracy factor tells if, ex post, the forecast was correct in the sense that the actual EPS (adjusted for stock splits performed between the forecast announcement and the announcement of the actual EPS) lay within the EPS forecast range or was equal to the EPS forecast point estimate, and thus is given by:

𝑎𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 1(𝑎𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 𝑤𝑖𝑡ℎ𝑖𝑛 𝐸𝑃𝑆 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡)

So that it takes a value of 1 if the forecast was correct and 0 otherwise. A forecast error relative to the firm’s stock price at the time of the forecast is specified, and calculated as:

𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡_𝑒𝑟𝑟𝑜𝑟 =𝑎𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆− 𝑓𝑐𝑠𝑡𝑢𝑝𝑝𝑒𝑟 𝑝𝑟𝑖𝑐𝑒 if 𝑎𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 − 𝑓𝑐𝑠𝑡 𝑢𝑝𝑝𝑒𝑟 _{> 0,} 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡_𝑒𝑟𝑟𝑜𝑟 =𝑓𝑐𝑠𝑡𝑙𝑜𝑤𝑒𝑟−𝑎𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 𝑝𝑟𝑖𝑐𝑒 if 𝑓𝑐𝑠𝑡 𝑙𝑜𝑤𝑒𝑟_{− 𝑎𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 > 0,} 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡_𝑒𝑟𝑟𝑜𝑟 =|𝑎𝑐𝑡𝑢𝑎𝑙 𝐸𝑃𝑆 − 𝑓𝑐𝑠𝑡𝑝𝑜𝑖𝑛𝑡| 𝑝𝑟𝑖𝑐𝑒 if −∞ < 𝑓𝑐𝑠𝑡 𝑝𝑜𝑖𝑛𝑡 _{< ∞,}

Where 𝑓𝑐𝑠𝑡𝑢𝑝𝑝𝑒𝑟 and 𝑓𝑐𝑠𝑡𝑙𝑜𝑤𝑒𝑟are defined as above, and 𝑓𝑐𝑠𝑡𝑝𝑜𝑖𝑛𝑡 is the point forecast, if there is one. This can be compared with the accuracy factor in a later stage to see if the extent of inaccuracy matters. In addition, average forecast range and average accuracy by CEO are calculated. Ben-David, Graham, and Harvey (2007) argue that overconfidence is time-persistent, and if this is true the average forecast range and accuracy factors might give a more balanced view of a CEO’s

overconfidence. Unreported tests show, however, that the average forecast range and accuracy factors yield results that are more blurred than the tests using the latest forecast range and accuracy factors.

(17)

17 4.3 Linking M&A data with CEO data

To finalize the data preparation, the deals are linked to the corresponding CEO data, including the overconfidence factors and the company financials, using the acquirer’s CUSIP. The details for the targets are also merged into the Thomson One file, however it is noted that the full range of data is only available for a very small minority of targets. The acquirer’s market capitalization 40 days prior to the bid announcement, as suggested by Betton, Eckbo, and Thorburn (2008), is added and used to remove all deals in which the bid value is less than 1% of the acquirer’s market capitalization. These deals are believed to be too small to be influenced by a CEO’s non-standard beliefs and preferences. The total sample consists of 1.929 deals.

4.4 Summary statistics

Of the 1.929 deals in the sample, 981 are financed with cash only, 582 are financed with stock only, and 366 are financed with a mix of cash and stock. The fact that the vast majority of deals are financed with either 100% cash or 100% stock justifies the decision to only compare purely cash-financed deals with all other deals. Table 1 shows the breakdown of the deals by year and by method of payment. Deals are assigned to a year based on the initial acquisition announcement. Apart from 2002, the number of deals per year is relatively stable at an average of 173 deals per year. There are no large shifts in the fraction of cash-financed deals; apart from 2002 the number of deals financed with stock never exceeds the number of deals financed with cash. In total, 51% of the deals are financed with cash only and 49% of the deals are financed otherwise, giving me a balanced sample in this respect. The low fraction of mixed deals allows for having a dummy that takes a value of 1 if the deal is financed with cash only and a value of 0 otherwise as dependent variable in later analysis.

As seen in table 2, the average deal size for all deals is just over $610 million, or 12,6% of the acquirer’s market capitalization 40 business days before the deal announcement. The reason for the relatively large average dollar value is the requirement of all data being available, including details on company financials and CEO characteristics. The firms that disclose all this information are generally larger firms, and are thus able to complete larger mergers and acquisitions. Table 2 shows the breakdown of average deal size by method of payment. Deals paid with cash only are on average smaller in both dollar value and fraction of the acquirer’s market cap than other deals. This was expected as the deal size tends to increase with the portion of the deal financed with stock. The

(18)

18 average deal size of cash-financed deals is $479 million, or 10,3% of the acquirer’s market cap, where for other deals it is $747 million, or 15,0% of the acquirer’s market cap.

The summary statistics on the different overconfidence measures are shown in table 3. Overall, the average forecast range is 0,32% of the stock price at the time of the guidance, with a large standard deviation of 0,39%. The average forecast range for cash deals is smaller than for non-cash deals (0,30% vs. 0,34%) as was expected; a smaller forecast range is associated with more overconfidence, and this in turn is hypothesized to increase the chance of a deal financed with cash. The forecasts are correct in 26,3% of the time, as can be seen in the accuracy statistics. Here the difference between cash and non-cash deals is small; 26,3% for cash deals vs. 26,4% for non-cash deals. The average forecast error, including the 26,3% of observations where the forecast is correct and the error is thus 0, is somewhat larger for non-cash deals than for cash deals; 0,69% vs. 0,60%, respectively, with a full sample average of 0,65%. This means that the average error in the EPS forecast is 0,65% of the stock price at the time of guidance, or roughly twice the forecast range.

Finally, table 4 shows the summary statistics for the control variables used in the

regressions, split by method of payment. The median level of cash reserves is substantially higher for cash deals than for non-cash deals, and the firms who do cash deals are substantially larger than firms who do non-cash deals, as shown by the median firm size which is $704 million for firms that do cash deals vs. $611 million for firms that do non-cash deals. The other variables that have not yet been discussed, the debt level, the market-to-book level, and the CEO’s age, are similar between the methods of payment. The median debt level is $492 million, the average market-to-book level is 3,79, and the average age of the CEO is 55 years. Finally, the stock volatility averages 0,095 for the full sample, with a slightly higher average for the deals that are not financed with cash only; 0,098 vs. 0,093 for cash deals.

5. Results

5.1 Forecast range

Table 5 presents the results for the regressions with a dummy for Cash as the left-hand side variable, which takes on the value of 1 if the deal is paid with cash only and 0 otherwise, and the forecast range in percentage points as the main left-hand side variable. The results are for the full sample of deals. Regression 1 is a logit regression with heteroskedasticity-robust standard errors and with just the forecast range and a constant on the right-hand side of the equation. Regression 2 is the same but with the most important control variables: the log of the acquirer’s cash and debt level, and the

(19)

19 deal size relative to the acquirer’s market value. Regression 3 adds more control variables to take into account the log of the dollar deal value, the log of the acquirer’s total assets, the acquirer’s market-to-book ratio, the log of the acquirer’s shareholder’s equity, the log of the acquiring CEO’s age, the stock volatility, and the forecast accuracy. Regression 4 is the same as regression 3 but with year-fixed effects included, to control for potential variations by year in the sample. The year-fixed effects are dummies for each year in which 1 or more deal announcement has been made (2002-2013). Regression 5 has the same control variables as regression 3, but now includes industry-fixed effects, based on the acquirer’s standardized industry code (SIC). Regression 6 is the same as regression 5, with year-fixed effects.

From the results shown in regression 1, we can say that a 1% increase in the forecast range increases the odds of a deal being financed with cash only with 𝑒𝑥𝑝(−0,211) = 0,81 – 1 = −0,19 = −19,0%. That is, it decreases the odds with 19,0%. In terms of probability, this means that a 1% decrease in the forecast range increases the odds of a deal being financed with cash only with 5,25%. This result is statistically significant at the 5%-level with a z-statistic of -1,78. Because the forecast range decreases with overconfidence, we can say that overconfidence does indeed increase the probability of a deal being financed with cash only, as predicted in the main hypothesis, at least in a simple logit model with no controls or fixed effects.

In table 5, regression 2 we see that the coefficient on the forecast range is still negative, but has decreased in both absolute magnitude and significance; it is now -0,166 and has a z-statistic of -1,34. The odds of the deal being financed with cash only increase with 𝑒𝑥𝑝(−0,166) =

0,847 – 1 = −0,153 = −15,3%. In probabilities, this means a 1% decrease in forecast range leads to a 4,14% higher chance of a deal financed with cash only. The control variables that are added all have a significant effect, with the level of cash and the relative deal size having the largest impact. This was expected as these two most directly influence the acquirer’s ability to finance the investment with cash. In regression 3, the absolute value of the coefficient on the forecast range has increased to -0,328, which is significant with a z-statistic of -2,12, larger than in regression 1 and 2. This means that the odds of cash only, when control variables are added, increase with

𝑒𝑥𝑝(−0,328) = 0,72 – 1 = −0,28 = −28,0% with each 1% increase of the forecast range. Or, in probabilities, a 1% decrease in the forecast range leads to a 8,14% higher chance of a deal financed with cash only. We also see that coefficient on the cash level the acquiring firm possesses is

significantly positive. This means that deals where the acquiring firm possesses more cash are more likely to be financed with cash, as was expected by intuition and found earlier by Martin (1996). The acquiring firm’s debt level has no significant effect on the probability of cash financing. Economic

(20)

20 intuition would predict that the more debt a firm has, the harder it will be to acquire new debt with which to finance an acquisition, thus lowering the probability of cash financing. These results show no relationship, possibly indicating that the acquiring firms in this sample generally have not fully used their debt capacity. The fact that the firms in this sample are larger than average might explain this. Moving down in table 5, we see a significantly negative coefficient on the relative deal size, meaning that larger deals are less likely to be financed with cash. The intuition behind this effect is that larger deals require more cash which the acquiring firm will less probably have available, thus forcing the acquiring firm to finance the deal (at least partly) with equity. Martin (1996) also finds this effect. Probably because the relative deal size is included as a control variable, there is no significant effect of the absolute value of the deal. The firm size, measured by the log of the total assets, does have a significant effect. Because larger firms would be expected to have larger cash reserves, the fact that this effect is negative is somewhat surprising, because it means that larger firms are less likely to finance their acquisitions with cash only. The same goes for the acquiring firm’s market-to-book value, or Q. This was also found by Martin (1996) and can be explained by the assumption that firms with a higher market-to-book ratio are more likely to have overvalued equity, which in turn makes it more attractive to finance investments with that equity. Both the level of acquirer’s shareholder’s equity and the acquiring CEO’s age have no significant effect on the dependent variable. The acquiring firm’s stock volatility, however, does have a large and significant effect. In theory, the more volatile a firms stock is, the more likely it is that the stock is over- or underpriced. At the point in time when such a company makes a large investment, the stock is more likely to be overpriced then underpriced. Therefore, given the fact that the company has announced to make a large investment, in this case an acquisition, stock volatility is an indicator of overpriced equity. Overpriced equity in turn makes it more attractive to finance the acquisition with equity. This theory is confirmed by the results as the coefficient on the acquiring firm’s stock volatility in the two years ending 40 business days before the acquisition announcement is significantly negative. Finally, the forecast accuracy has a weakly significant, positive effect on the chance of cash financing.

When the year of the acquisition announcement is added as a control variable in the form of a dummy for each possible year, the results in regression 4 are obtained. The coefficient on the forecast range is again significantly negative at -0,381 with a z-statistic of -2,30, both of which are larger than in regressions 1 and 2. The coefficient means that the odds of an acquisition financed with cash only increase with 𝑒𝑥𝑝(−0,381) = 0,683 – 1 = −0,317 = −31,7% with each 1% increase in the forecast range. In other words, a 1% decrease in the forecast range leads to a 9,42% higher chance of the acquisition being financed with cash only. The coefficients on the control variables are very similar to those in regression 3, with the only notable change being the fact that

(21)

21 the market-to-book ratio (Q) is not significant in regression 4 while it is significant in regression 3. Regressions 5 and 6 show that the significance of the forecast range on the method of payment disappears when industry-fixed effects are added. Surprisingly, the z-statistics of the cash level drop considerably, while the effect of the debt level is positive and strongly significant in these

regressions where it was not in regressions 1-4. Both results are counter-intuitive, and might have come about by a dispersion of companies across many different industries; with a limited amount of acquiring companies the industry effects might give a distorted picture. As there are 131 different industries represented in the sample, and the majority of these include less than 10 deals, robust conclusions are hard to draw from a regression where they are included as control variable.

Summarizing the results for the regressions with the forecast range as the main right-hand side variable, we see that smaller forecast ranges increase the chance of a deal financed with cash. This is consistent for a regression without controls, one with controls, and one with controls and year-fixed effects. When industry-fixed effects are added, the same conclusion is not drawn, but the reason for this might be that the number of different acquiring companies is too low. The

coefficients on the forecast range are not only statistically but also economically significant in regressions 1-4, with the change in probability of a cash financing ranging from 4,14% to 9,42%. Because small forecast ranges are a proxy for overconfidence in this article, these results confirm the hypothesis that overconfidence increases the probability of an acquisition being financed with cash.

5.2 Forecast accuracy

Another potential measure of overconfidence is the accuracy of a CEO’s EPS forecasts, as suggested by Hribar and Yang (2010 & 2015). Because some firms have more stable earnings than others, narrow forecasts might not be an indication of overconfidence if they are correct. Table 5 shows that when the forecast accuracy is controlled for, the effect of the forecast range on the method of payment is still significant. In fact, the effect is stronger than in unreported tests where the forecast accuracy was removed as control variable. Table 6 shows results for logit regressions on the full sample with the dummy for cash financed deals as the left-hand side variable and with the forecast accuracy as main right-hand variable. The same approach to standard errors and control variables is used as in table 5.

Regression 1 shows the results without control variables. Here we see no significant effect of the accuracy of the forecast on the method of payment and a very small decrease in the likelihood of a deal being financed with cash if the forecast was accurate. In regression 2 the results for the

(22)

22 regression with main control variables are shown. Here, again, there is no significant coefficient on the forecast accuracy. Regression 3 has added controls, and again the coefficient on the forecast accuracy is insignificant, however the coefficient is larger than in regression 1 and the z-statistic has increased to 1,34. Contrary to expectations, an accurate forecast increases the chance of cash financing with 4,17% in regression 3. A potential reason for this might be financial rather than behavioural; when a CEO realizes her forecasted EPS, the overall financial situation might be more positive and stable than when the forecast is missed. Even if the forecast is exceeded, this means a less stable financial situation and more uncertainty, which makes stock financed acquisitions more attractive. The control variables show the same direction and similar magnitude as in regression 3 of table 5. When year-fixed effects are added, the results in regression 4 are obtained. Here the

coefficient on the forecast accuracy is somewhat smaller than in regression 3 with a similar z-statistic of 1,24, meaning that the effect is still insignificantly positive. The increase in probability of a deal financed with cash is similar as well, at 3,9%. The same potential explanation as to why the effect is positive rather than the ex-ante expected negative effect applies as in regression 3. Regressions 5 and 6 show a weaker effect of the forecast accuracy, lower significance of the cash level, and an unexpected direction and higher significance of the debt level, just as in table 5.

Table 7 presents a third angle to the EPS forecasts; the relative forecast error. Because the analysis on the forecast accuracy only takes into account if the forecast was accurate or not, and doesn’t look at the magnitude of the error if it was not, the results might be biased. The forecast accuracy is the difference between the actual EPS and the nearest border of the EPS forecast range or point, divided by the share price at the time of the forecast announcement. Table 7 shows small and insignificant effects of the magnitude of the forecast error. In regression 1, which doesn’t include control variables, the z-statistic is -1,09 and the decrease in probability of a cash financing is 0,56% per 1% increase in the forecast error. Regression 2 shows a coefficient of the same direction and magnitude when main control variables are added. In regression 3, more variables are

controlled for and the coefficient is of the same magnitude as in regressions 1 and 2, however it is now positive. Here the z-statistic is 0,49 and the increase in probability of a cash financing is 0,49% per 1% increase in forecast error. Regression 4 includes yearfixed effects and has a zstatistic of -0,50, with an increase in probability of a cash financing of 0,52% per 1% increase in the forecast error. Finally, regressions 5 and 6 add industry-fixed effects to regressions 3 and 4, respectively. The effect of the forecast error increases in magnitude and significance, which is surprising because table 6 showed a different result. As in tables 5 and 6, industry-fixed effects seem to diminish the

importance of the level of cash and increase the importance of the level of debt held by the acquirer.

(23)

23 5.3 Summarized main results

Tables 5 through 7 show the main results of this article. The most important conclusion that can be drawn from them is that the size of the EPS forecast range a CEO provides is negatively related to the probability of an acquisition being financed with cash; in other words, CEOs who provide more narrow forecast ranges are more likely to finance their acquisitions with cash. Depending on the level of control, each 1% decrease in the size of the forecast range increases the chance of cash being the method of payment with 4,14%-9,42%. This confirms the expected relationship between overconfidence and the method of payment in M&A, and shows that CEO overconfidence has a significant impact on the financial structure of a company, which is affected by M&A decisions. Forecast accuracy, on the other hand, seems to have no significant impact on the method of payment in this sample. This goes for both pure accuracy, where there is only a distinction between accurate forecasts and inaccurate forecasts, and for relative accuracy, where the magnitude of the forecast error is taken into account.

6. Robustness checks

6.1 Percentage paid with cash

To ensure the robustness of the results shown in table 5, as well as to see which conclusions would have been drawn had I used different data samples, a series of additional tests is performed and discussed in this section. First of all, table 8 shows the results of a simple OLS regression with heteroskedasticity-robust standard errors with the forecast range and several levels of control on the right-hand side, and the percentage of the total deal value that is financed with cash on the left-hand side. Note that this is the same analysis as shown in table 5, except for the change from a dummy dependent variable to a continuous dependent variable. Also note that the analysis as shown in table 5 is actually more strict than the robustness check performed here; in the main analysis, a deal where a CEO is overconfident that is financed with 99% cash is counted as non-cash, whereas the overconfidence might in fact have influenced the amount of cash used positively. The only scenario in which the main analysis would give significant results and this robustness check would not is if overconfident managers are more likely to finance their deals with either 100% cash or a very low percentage of cash. Table 8 shows that this is not the case. For the four regressions, regression 1 having no controls, regression 2 having only main controls, regression 3 having controls but no fixed effects, and regression 4 having both controls and year-fixed effects, the coefficient on Forecast range is negative, and for regressions 1, 3, and 4 it is significant. As in table 5, the

(24)

24 confidence levels are 90%, 95%, and 95%, respectively. For regressions 5 and 6, much as in table 5, the coefficients on the Forecast range are negative, but not significant. Here the significance of the cash level completely disappears, and the effect of the debt level increases and becomes

significantly positive. Again, this is surprising and might be due to the limits of the sample.

For the regression with no controls I find a coefficient of -4,664, which means that each 1% increase in the EPS forecast range decreases the percentage of cash used in the deal with 4,66%. This relationship is weakly significant with a t-statistic of -1,77. For regression 2, where the same controls are added as in regression 2 of table 5, the relationship between CEO overconfidence and the amount of cash used is somewhat weaker. Here each 1% increase in the size of the forecast range decreases the percentage of cash with 3,76%. In regression 3 this impact increases to 7,07%, and the coefficient is not only larger than in regression 1 and 2, it is also statistically significant with a t-statistic of -2,19. Regression 4 shows results for the same analysis as in regression 3, but with added year-fixed effects. We see that the effect is even more pronounced, with a coefficient of -8,208 corresponding to a decrease in the percentage of cash used of -8,21% with each 1% increase in the size of the forecast range. Note that these percentages for regressions 1-4 are in the same order of magnitude as the decreased probability of a cash-only financing when the forecast range increases, which was -5,25%, -4,14%, -8,14%, and -9,42% for regression 1, 2, 3, and 4 in table 5, respectively. The unexpected results in regressions 5 and 6 mirror those in table 5 as well. As for the signs and significance of the control variables, the results in table 8 are close to those in table 5, with the only exception that the coefficient on the log of the dollar deal value is significantly negative in table 8 whereas it was not significant in table 5. Given the results in the main analysis, all of these findings were expected.

In table 9, the same approach is taken as in table 8, but now with the forecast accuracy as main right-hand variable. As in part 5.2, Forecast accuracy is a dummy that equals 1 if the EPS forecast was accurate, and 0 otherwise. Table 9 shows that when the dependent variable is the percentage of cash with which the acquisition is financed, the forecast range still has no significant effect. Regressions 1 and 2 show that the effect is negligibly small when no controls or only the main controls are included in the analysis; the coefficient and the t-statistic are very close to zero in both regressions. Regressions 3 and 4 suggest that CEOs who have given an accurate forecast use about 2,9% more cash, but the numbers are not significantly different from zero with t-statistics of 1,09 and 1,10. Regressions 5 and 6 give slightly smaller coefficients for the Forecast accuracy, and the t-statistics go down as well. The control variables have coefficients of the same sign and magnitude as in table 8, which is no surprise as table 9 is basically the same analysis without the forecast range as main right-hand variable.

(25)

25 To further check the impact of the accuracy of the forecast on the percentage of cash used, the relative forecast error is used as the main right-hand side variable instead of the dummy for accuracy. The forecast error is calculated as in part 5.3. Table 10 shows the results for the OLS regressions. As was the case with the forecast accuracy, the forecast error does not have a significant impact on the percentage of cash used to finance the deal, at least when the control variables are added. As in table 9, the coefficient is positive, suggesting a 1% increase in the forecast error leads to 0,38% more cash (when year-fixed effects are controlled for), but the impact is not significantly different from zero. The control variables show direction and magnitude analogous to the regression in table 9.

6.2 Sample splits

In this section I investigate if the conclusions drawn from the analysis in part 5.1 would have been different if another sample, or only a certain part of the current sample, would have been used. Specifically, I identify a cut-off point for the relative deal size of 10%, the median of the acquirer’s market-to-book ratio, the median of the log of the acquirer firm size, and the median of the acquirer’s stock volatility. Then the same regressions are run as in table 5, meaning with a dummy for cash-financed deals as dependent variable and the forecast range as the main right-hand side variable, for the following sample segments: (1) the full sample, (2) & (3) all deals in which the deal is worth more than respectively less than or equal to 10% of the acquirer’s market capitalization, (4) & (5) all deals in which the acquirer’s market-to-book ratio (Q) is above respectively below or equal to the median Q, (6) and (7) all deals in which the acquirer’s firm size is above respectively below or equal to the median, and (8) and (9) all deals in which the acquirer’s stock volatility is above

respectively below or equal to the median. The coefficients and z-statistics on the Forecast range for each sample split are shown in table 11.

Looking at the relative deal size, the results show that the effect of overconfidence on the method of payment is statistically stronger for small deals (<=10% of the acquirer’s market cap) than for large deals. Even though the coefficients are of the same sign and magnitude, the standard errors are much smaller in the regression on small deals, resulting in a significantly negative coefficient on Forecast range. The test on equality between both coefficients shows that it can’t be concluded that they differ significantly, though. We see that, for small deals when controls and year-fixed effects are added, a 1% increase in the size of the forecast range implies a change in odds of cash payment of 𝑒𝑥𝑝(−0,416) = 0,66 – 1 = −0,34 = −34,0%, corresponding with a decreased probability of

(26)

26 10,2%. The number is similar, even somewhat larger in absolute terms at -11,2%, for large deals, but there is more variation which results in an insignificant effect.

Rows (4) and (5) show results for a sample split along the median market-to-book ratio (Q). Here it shows that CEO overconfidence is much more important in firms with a high Q than in firms with a low Q; in deals in which the acquirer’s Q is above the median, a 1% increase in the forecast range leads to a 15,7% lower probability of cash-financing when control variables and year-fixed effects are added. This result is highly significant as evidenced by the z-statistic of -2,83. For firms with a low Q, on the other hand, the conclusion is different. Here a 1% increase in the forecast range corresponds with a 5,8% decreased probability of cash-financing, however this is not significantly distinguishable from zero with a z-statistic of -0,98. A potential explanation for this is that overconfidence is a more important force in firms with a high Q, because they have more

investment opportunities and are less stable in terms of stock price and financials. Again, however, tests on equality between the two coefficients show that they don’t differ significantly.

The acquirer firm size is an important factor as well. When the acquirer is larger than the median in terms of log of total assets, the forecast range significantly reduces the chance of cash financing. When controls and year-fixed effects are added, a 1% increase in the forecast range leads to a significant 10,7% lower probability of cash financing in deals in which the acquirer is large. For deals in which the acquirer is small, however, this number drops to 5,8% and is not statistically significant anymore. A potential explanation is that in small firms, the choice for the method of payment is more likely to depend on other factors than CEO overconfidence, such as financial restrictions. Finally, the results for deals in which the acquirer has a high or low stock volatility are quite similar. In both, the effect of a wider forecast range is negative and weakly significant or not significant at all. Here too, the coefficients can’t be said to differ significantly.

Summarizing these robustness checks we can see that the effect of CEO overconfidence is of the same direction regardless of how the sample is split. The magnitude and significance of the coefficient differs; the size of the forecast range is most important in deals in which the deal is small, the acquirer is large, and the acquirer’s market-to-book ratio is high. However, tests on equality of the coefficient on Forecast range show no significant difference between the several sample splits.

Table 12 shows the same sample splits for OLS regressions with the percentage of cash used to finance the deal as dependent variable, and the forecast range again as main right-hand side variable. The same conclusions can be drawn from this table as from table 11; apart from the fact that CEO overconfidence influences the method of payment for the full sample, it is significant in

CEO overconfidence and the method of payment in mergers and acquisitions

MSc Business Economics, Finance track

Master thesis