Firm size and CEO overconfidence: the influence on acquirer returns

(1)

on acquirer returns

A European study on the short time wealth effects of acquisitions

University of Groningen

Faculty of Economics and Business

MSc Business Administration (Specialization Finance)

June 2008

(2)

Abstract

In this study the influence of acquirer size, CEO overconfidence and their interaction on acquirer announcement returns are examined, using a European sample of acquisitions from 2001 till now. Where previous studies found evidence for both a size effect and an overconfidence effect, in this study strong evidence of a size effect is presented with a difference of more than 2% in cumulative abnormal returns between small and big acquirers. Small acquirers create much more value for their shareholders than large acquirers. Weak evidence is provided for the influence of CEO overconfidence on acquisitions, measured by three overconfidence proxies. It appears that acquisitions in Europe are less influenced by overconfidence than acquisitions in the Anglo-Saxon countries. The size effect appears to dominate the overconfidence effect and small acquirers outperform large acquirers in almost all occasions, regardless of the degree of CEO overconfidence at the firms. The size effect is robust to other firm and deal characteristics. With regard to the combined effect, weak evidence of a negative relationship between size and CEO overconfidence is given. The puzzle around the lower returns for big firms therefore remains unsolved.

Key words: Mergers and acquisitions; Acquirer returns; Size effect; Overconfidence; Hubris; Short-term performance

(3)

1 Introduction

A lot has been written about mergers and acquisitions in the financial literature during the last decades. But it also reaches the popular press: almost no week can pass without news about mergers and acquisitions. In several countries a huge public debate has arisen about the consequences of mergers and acquisitions. Recent studies, such as the study of Moeller, Schlingemann and Stulz (2004) find that small acquirers are outperforming large acquirers and conclude that there exists a ‘size effect’ in the takeover market. Moeller et al. find in their study that managerial overconfidence plays more of a role in large firms than in small firms, what possibly explains part of the reported size effect.

A lot of attention should be given to the motives and rationale of CEO’s in this perspective. CEO’s are the ultimate decision makers with respect to the bidding decision. Sometimes managerial overconfidence or hubris appears to exist. This means CEO’s are overconfident about their own and their companies’ capabilities and therefore are paying too much to acquire other firms. While this theory was first addressed by Roll (1986), Doukas and Petmezas (2007) find in a recent study that more overconfident CEO’s make worse acquisitions than other CEO’s do (by measuring overconfidence as a result of CEO self-attribution). They find evidence for this using different proxies for CEO overconfidence but only look at private deals. Also Malmendier and Tate (2007) prove that overconfidence leads to lower quality acquisitions. In addition, Doukas and Petmezas also find that overconfident acquirers perform poor in the long run.

(5)

result of their interaction is present. The main problem statement I want to answer is therefore:

Do size and overconfidence influence European acquirer announcement returns and what is their combined effect?

In order to measure the overconfidence of CEO’s I use three measures: past stock performance of the acquiring company, high order acquisition deals and the issuance of equity of acquiring firms. The last overconfidence proxy has not been used before and is a possible contribution to existing overconfidence proxies. This thesis could also contribute to existing research because of its focus. In this thesis I use a sample of EU firms, but exclude the United Kingdom. Because the size and overconfidence effects are already researched in the United States (by Moeller et al. (2004)) and in the United Kingdom (by Doukas and Petmezas (2007)), my sample explores the size and overconfidence effects in another geographical part of the world. In the continental European countries in this sample, markets could behave different than in the Anglo-Saxon countries. In case there are differences in with respect to the size effect and the effect of CEO overconfidence between the EU countries in my sample and the US and the UK, it is an interesting question why these differences exist. The results obtained in this study could be useful for managers in judging whether a particular acquisition is likely to be value enhancing or not. In addition, the rationale of overconfident managers could be made clear for investors and investors could consider these results when judging particular mergers and acquisitions.

(6)

proxies the result is marginally significant. Using multivariate analysis, the size effect remains and only weak evidence of an overconfidence effect can be found.

When looking which effect is dominant, the size effect looks to dominate. In fact, by splitting up the sample in overconfident and non-overconfident acquirers, the size effect remains present in almost all sub-groups, while no significant results of overconfidence can be found in sub-groups of small and big acquirers. The interaction effect between size and overconfidence points at a negative relation between size and overconfidence, but this effect is also not significant. Concluding, it appears that a strong size effect is present in European acquisitions, with CEO overconfidence having at most only a very small role. The size effect in European acquisitions can not be explained by managerial overconfidence or hubris.

(7)

2 Background literature

2.1 Why do acquisitions take place?

An acquisition can be seen as a situation in which a company obtains the control rights of another company. The control rights are transferred from the board of the target firm to the board of the bidder firm. An acquisition can be made in cash, in shares or a combination of these. Besides, the acquisitions can be friendly, when the target management supports the acquisition attempt, but also hostile, when the target management resists the acquisition attempt. Jensen and Ruback (1983) define corporate control as ‘the rights to determine the management of corporate resources’.

Andrade, Mitchell and Stafford (2001) mention several reasons why acquisitions could take place: efficiency-related reasons based on arguments of economies of scale or synergies, attempts to create market power (creating monopolies or oligopolies), market discipline, in the case of an incompetent target management, empire building by bidding management and other agency costs and at last taking advantage of opportunities for diversification. They further mention that in particular time periods some of these reasons could be more influential than others, such as the diversification merger wave in the 1960’s and more recently the rise of the more market discipline driven mergers. Finally, they mention that mergers tend to occur in waves and are clustered by industries. Other important factors that could play a role in making acquisitions are bidder firm overvaluation and managerial overconfidence or hubris. A CEO thinking his firm is overvalued could make an acquisition with this overvalued shares, resulting in a cheap (in the eyes of the CEO) acquisition.

The overconfidence argument is further discussed in sections 2.6 and 2.7.

2.2 Acquisitions and abnormal returns for bidding companies

(8)

find a positive abnormal return, some find a negative, and still others find no significant abnormal returns.

In an early study Jensen and Ruback (1983) find no significant positive abnormal returns for bidders, but neither significant negative abnormal returns. Asquith, Bruner and Mullins Jr. (1983) find significant positive abnormal returns. In a European study Goergen and Renneboog (2004) also find significant positive abnormal returns for bidders. Travlos (1987) finds negative abnormal returns.

Several authors (Fuller, Netter and Stegemoller 2002) (Draper and Paudyal 2006) (Antoniou, Petmezas and Zhao 2007) (Faccio, McConnell and Stolin 2006) make a distinction between acquisitions of private and public firms. All authors find significant positive abnormal returns in the short run for acquisitions of private firms, while they find neutral to negative abnormal returns for acquisitions of public firms. Antoniou et al. (2007) find in the long run negative abnormal returns for acquisitions of private firms too. Some authors find higher abnormal returns for acquisitions with cash as method of payment (Travlos,1987) (Draper and Paudyal 1999), while others find higher abnormal returns for acquisitions with shares as method of payment (Goergen and Renneboog 2006) (Fuller et al. 2002).

According to Jarrel and Poulsen (1989) three explanations can be given for the small positive or even negative abnormal returns for bidders in acquisition. First, the complete wealth effect of the acquisition can be still unclear for shareholders at the time of acquisition. Second, competition between bidders can drive the price of the target up, leading to a zero NPV acquisition for the bidder. Third, acquisitions can be poor investment projects, possibly because managers are looking to expand their firm, rather than maximizing shareholder wealth.

(9)

Table 1

Overview of previous studies on acquisitions

In this table an overview is given of important studies done in the past about acquirer announcement returns. Besides the authors, the country, time-period, the status of the target, the event window, the sample size

and the cumulative abnormal return found in the particular study are mentioned.

Study Country Public/ Period Event- Sample CAR Private Window Size Acquirer Target

Andrade et al. US Public 1973-1998 -1,+1 4300 -0.7% (2001)

Antoniou et al. UK Both 1987-2004 -2,+2 1401 +1.26% (2007)

Asquith et al. US Both 1963-1979 -1,0 214 +0.9% (1983)

Bradley at al. US Public 1963-1984 -5,+5 236 +0.97% (1988)

Conn et al. UK1 Both 1984-1998 -1,+1 4320 +0.59% (2005)

Doukas and UK2 Private 1980-2004 -2,+2 5334 +1.18% Petmezas (2007)

Draper and UK Both 1981-2001 -20,+20 8597 +1.09% Paudyal (2006)

Faccio et al. Western- Both 1996-2001 -2,+2 735 -0.38% (2006) Europe (listed target)

3694 +1.48% (unlisted target)

Fuller et al. US Both 1990-2000 -2,+2 3135 +1.77% (2002)

Goergen and Europe Both 1993-2000 -2,+2 142 +1.18% Renneboog (2004)

Jarrell and US Both 1963-1986 -5,+5 461 +0.92% Poulsen (1989)

Jensen and US Both 1956-1981 various various non-

Ruback (1983) 3 significant

Moeller et al. US Both 1980-2001 -1,+1 12023 +1.10% (2004)

Travlos (1987) US Public 1972-1981 -10,+10 60(shares) -1.60% 100(cash) -0.13%

1_{Conn et al. (2005) use UK acquirers, but also look at cross-border acquisitions. Target firms can therefore}

come from any country.

2_{Doukas and Petmezas (2007) use UK acquirers, but also look at cross-border acquisitions. Target firms}

can therefore come from any country.

3_{Jensen and Ruback (1983) give an overview of many studies with different samples and event windows.}

(10)

2.3 Size effect of the bidder

Moeller et al. (2004) find significant positive abnormal returns for small bidders and insignificant positive abnormal returns for large bidders in the announcement period of an acquisition, using equally weighted abnormal returns and based on 21 years of US acquisitions. The abnormal returns they find are dependent on the inclusion of other possible explaining variables, such as method of payment and target firm characteristics (for example the distinction between private and public targets), but small bidders are performing much better than large bidders irrespective of the other included variables. The only acquisition form with a negative abnormal return for small bidders is the acquisition of public firms financed with shares, but large bidders perform even worse here and this difference is also statistically significant.

Moeller et al. offer different possible explanations for the existence of this size effect.

First, they argue large firms could be more overvalued than small firms. Because large firms have a high equity capitalization by definition, it could be that they are more overvalued. In testing this explanation it is difficult to distinguish this explanation from two other explanations (equity signaling explanation and growth opportunities explanation), according to the authors. The results of their study do not support the theory that bigger firms are more overvalued.

Second, equity signaling could be the case for the existence of a size effect. Moeller et al. find no support for this explanation, because by using book value instead of market value of bidders, the size effect remains present.

Third, Moeller et al. find no support for growth opportunities signaling.

(11)

acquiring companies (a higher percentage of their offers is ultimately accepted). The authors conclude that these results are in line with the hypothesis that managerial hubris is more present in large firms.

Last, small firms could have more synergy gains from acquisitions than large firms. Moeller et al. find that acquisition announcements of large firms are consistent with the presence of negative synergies, while the acquisition announcements of small firms are consistent with the presence of positive synergies.

Many other authors do include an acquirer size proxy in their multivariate analysis. For example, Doukas and Petmezas (2007) in their study prove that CEO overconfidence leads to more welfare-destroying acquisitions, controlling for other acquirer characteristics among which the logarithm of acquirer size. Doukas and Petmezas do not find any significant influence of acquirer size on announcement returns in this multivariate analysis. Malmendier and Tate (2007) do include a size proxy in their multivariate analysis, using the logarithm of acquirer assets, but because their entire sample exists of big acquiring firms, their proxy does not describe a size effect.

2.4 Other explaining variables in acquisitions

Besides the variables that I investigate some other variables can have an effect on the abnormal returns of acquiring firms. The most important variables are the method of payment (cash, shares or mixed), the status of the target (public, private or subsidiary), the form of the acquisition (tender offer, merger or proxy contest) and the nationality of acquisitions (national or cross-border).

2.4.1 Method of payment

(12)

thinks their equity is overvalued or not. Payment in shares could signal that the management thinks their shares are overvalued and therefore can buy the target firm cheaply with overvalued equity. On the other side, payment in cash could signal that the management thinks their equity has a fair value or is undervalued, making cash a more attractive method of payment than shares. This explanation presumes that managers have private information that investors do not have and therefore the stock market has no perfect information. In corporate governance this would be called information asymmetry. Another theory is that with cash acquisitions bidders use debt to finance the acquisitions and that the positive abnormal acquirer returns are in fact a leverage effect. In this case tax shields would be more optimally utilized or the agency costs of equity would be reduced. Yook (2003) provides some support for the last mentioned theory. Jensen and Ruback (1983) point out that it is very hard to identify which component is decisive in regular event study research.

However, some recent research by Fuller et al. (2002) and Goergen and Renneboog (2004) finds higher positive abnormal returns for acquiring firms which use equity than for acquiring firms using cash. Also, the status of the target could have an influence on the comparison between the use of shares and cash. These findings indicate that the method of payment effect is still unclear.

2.4.2 Status of the target

(13)

public acquisitions leads to lower abnormal returns for acquiring firms, while it leads to higher abnormal returns for acquiring firms in private acquisitions.

2.4.3 Form of the acquisition

There also exist multiple forms of acquisitions. In a merger and a tender offer, a bid on the target shares is done. The main difference is that in a merger the bid is negotiated with the target management before going to the target shareholders, whereas in a tender offer the bid is directly to the target shareholders. A proxy contest is a situation in which a (dissatisfied) group of people tries to get a majority in the board of a firm and in this way also tries to get the control of the firm. Tender offers are presumed to result in higher acquirer returns than mergers. Jensen and Ruback (1983) already conclude this in their summary of various studies, regardless of whether the acquisition is successful or not. Again, this could be the consequence of cash as a method of payment more used in tender offers compared to mergers.

2.4.4 Nationality of the acquirer and target

Mergers and acquisitions can be done both within national boundaries as between companies from different countries. Moeller and Schlingemann (2005) find for US acquirers higher returns for national acquirers than for cross-border acquirers. Aw and Chatterjee (2004) find the same for UK acquirers, where they find particular low (negative) abnormal returns for UK firms acquiring continental European firms. They conclude that the loss of control in oversea countries and the cross-cultural differences outweigh any economies of scale and skills. In another study Shimizu, Hitt, Vaidyanath and Pisano (2004) combine the findings of several studies and find mixed results with regard to acquirer abnormal returns.

2.5 Overconfidence

(14)

everywhere in every day life. Especially, when people have been successful in the past or have reached a certain (high) status, they are assumed to get more confidence and in some cases they could get so much confidence that they think they are able to manage more than they in fact can or think they have otherwise superior qualities. This is overconfidence. A synonym for overconfidence is hubris, the classic Greek word for a sort of recklessness, arrogance and overconfidence. In the Greek mythology hubris was often in place when ordinary people thought they could challenge some Gods or rules of Gods. Afterwards, they fell and got punished by these Gods. In the mythology, characteristically these people seal their own destiny because they become so blinded in their behavior that a happy ending becomes impossible.

Ben-David, Graham and Harvey (2007) use the psychological finding of miscalibration to explain behavior of corporate managers and their corporate policies. Miscalibration is often measured by which confidence intervals people attach to certain values they predict. In their study, Ben-David et al. find that CFO’s of companies tend to hugely overestimate the precision with which they would be able to predict the future stock market performance of their company. This miscalibration is the effect of both personal traits, as well as corporate characteristics. This huge miscalibration by CFO’s proves the ubiquitous overconfidence at top level corporate decision makers.

(15)

CEO’s of other firms. Miller and Ross (1975) state that overconfident people attribute past success to their own skills, while they attribute past failure to bad luck.

Malmendier and Tate (2005b) give other reasons why CEO’s could be overconfident. They point out that individuals tend to be overoptimistic about their own future and about matters which are under their control. Also, when there is high commitment with certain outcomes, more overconfidence is in place. All these premises apply to corporate CEO’s, who have to make the ultimate decisions in companies and therefore think they can greatly influence the outcomes of projects. Besides, CEO’s mostly have high commitment to the firm in the form of options and shares.

Malmendier and Tate (2005a) measure CEO overconfidence by the failure to reduce their personal exposure to company performance, measured by option holdings and company share purchases (Large option and share holdings in their own company means less diversification for the CEO than theoretically optimal. After all, besides the options and shares, also the job of the CEO is dependent on company performance). They find that overconfident managers see external funds as very costly (because they would be undervalued), while they overestimate the returns on their projects. This means overconfident CEO’s have the inclination to overinvest when there are abundant internal funds and to underinvest when external funds are necessary, in particular when they are dependent on equity. Malmendier and Tate find a strong positive relation between CEO overconfidence and the sensitivity of investment to cash flows. This leads to the conclusion that traditional share- and option pay systems fail to reduce managerial overinvestment.

(16)

2.6 Overconfidence and acquisitions

The first findings of managerial overconfidence were published by Roll (1986) with his hubris theory. Roll states that managers individually value other firms and consider an acquisition if their estimation is above that of the market. An acquisition is most likely to be successful if an individual manager values a target much higher than the average individual measured by the stock market. Because most acquiring managers do only make one or a few acquisitions in their life, they can not easily learn from past errors and therefore refrain from bidding on target firms. The overconfidence or hubris theory points out that some managers, in the absence of any real synergy gains, stay at their high valuation because they still think their valuation is correct.

It is widely believed that overconfidence can be measured by the acquisition premium that a certain acquirer pays. The higher this premium, the more overconfident a CEO could be. Hietala, Kaplan and Robinson (2003) note that directly using this premium data is theoretically not sound, because the acquisition announcement can also reveal other information about the target and bidder with respect to their respective stand-alone value and the synergy gains vary widely from acquisition to acquisition.

Hayward and Hambrick (1997) find four indicators of CEO overconfidence which are highly associated with the acquisition premium paid: recent performance of the acquiring firm, a measure of the acquiring CEO self-importance (high relative CEO compensation), the recent media appreciation for the CEO and a combination of these three factors. Using these factors as a proxy for CEO overconfidence, they find that CEO overconfidence and higher acquisition premiums are related with significant negative acquirer abnormal returns around the announcement of an acquisition. This result is strengthened by the absence of a strong board of directors. The authors mention several reasons for the possibility of the existence of CEO overconfidence in acquisitions:

(17)

- Acquisitions (especially very large acquisitions) can be very ego-involving for all parties involved. Entering new industries and greatly expanding the firm can be attractive for CEO’s.

- Acquisitions shape an environment of a winner and a loser; it gives the acquirer a chance to dominate.

These factors all can play a role in the mind of an acquiring CEO, whether he is aware of it or not.

Malmendier and Tate (2007) find that abnormal announcement returns are significantly more negative for overconfident acquirers than for non-overconfident acquirers, measuring overconfidence by CEO personal overinvestment in the company and CEO press coverage/portrayal. They also state that the chance that an overconfident CEO makes an acquisition is 65% higher than the chance that a non-overconfident CEO does. This effect is greatest for diversifying acquisitions and acquisitions that do not require external funds for financing. The authors conclude that overconfidence can create value in some instances, for example by counteracting risk-aversion, but mergers and acquisitions are not value increasing actions by overconfident CEO’s.

Ben-David et al. (2007) are linking managerial overconfidence to miscalibration. They find that overconfident managers use lower discount rates for the valuation of investment projects. This results in the fact that more overconfident managers take on projects with lower internal rates of return. Besides this, they also invest more in acquisitions and engage in more acquisitions. The announcement returns they find are negative for the overconfident managers. This could be the consequence of entering into lower IRR mergers and acquisitions because of the lower discount rates the overconfident managers use.

(18)

Petmezas find significant lower announcement returns for the overconfident acquiring CEO’s compared to other acquiring CEO’s, and also underperformance in the long term. According to the authors, a self-attribution bias is responsible for these bad returns. This is proven by the fact that high order acquirers (companies that make multiple acquisitions) see better abnormal returns at their first acquisitions than at their higher order acquisitions. The initial success makes the CEO’s believe that the success was the result of their own ability, with the consequence that they become overconfident and engage in more deals.

Ko and Huang (2007) have developed a theory that investor overconfidence in fact drives investors to overinvest in information acquisition and this leads to more efficient prices and markets, instead of less efficient prices and markets. These results are in contrast with the results of studies of CEO overconfidence with respect to investment and acquisition decisions. It remains a question why overconfident investors would be driven to overinvest in information acquisition, while overconfident CEO’s seem to prefer their own information (which they think is superior) and are not extensively searching for reliable information from outside parties.

2.7 Relationship between CEO overconfidence and company size

It is a question whether CEO overconfidence and size could party overlap. If this is the case, this would require a relationship between CEO overconfidence and the size of the firm. Bigger firms would then be led by more overconfident CEO’s than smaller firms. There are several reasons to expect this to be the case:

- Big firms could be big because of acquisitions in the past, which could be the result of overconfident CEO’s at the time. As a result, the overconfidence could be locked in the company culture.

- Big firms could be big because of good or excellent results in the past, which made the companies big. These past successes could lead to a culture where managers become overconfident about their skills and capabilities.

(19)

probably internally a very successful person and has reached a high status in the firm. This could easily lead to an overconfident feeling of the CEO.

- Big firms can exert more control in the acquisition process than small firms and can also afterwards more easily control the target company. This feeling of control of the process can lead to more overconfidence.

(20)

3 Theoretical Framework

3.1 Research objectives and hypotheses

Based on the results of previous studies, especially those of Moeller et al. (2004) and Doukas and Petmezas (2007) I point out the following three hypotheses, which should be researched to come to an answer to the main problem statement:

Hypothesis 1: Small acquiring firms have significant different abnormal returns at announcement than big acquiring firms

Hypothesis 2: Overconfident acquirers have significant different abnormal returns at announcement than non-overconfident acquirers.

Hypothesis 3: Combining the size effect and the overconfidence effect, one effect significantly dominates the other effect.

(21)

3.2 Overconfidence proxies

As mentioned before, several proxies can be used to measure acquirer CEO overconfidence. In table 2, an overview of overconfidence proxies used is given.

Table 2

Overview of CEO overconfidence proxies used in other studies

In this table an overview is given of studies in the past that used proxies for managerial overconfidence. Proxies used in these studies are mentioned.

Study Overconfidence proxy used

Ben-David et al. (2007) - CFO miscalibration based on stock market predictions and own company performance

predictions

Doukas and Petmezas (2007) - High order acquisition deals (high managerial acquisitiveness)

- Insider dealings

Hayward and Hambrick (1997) - Recent performance acquiring firm - CEO self-importance

- Recent CEO media appreciation - Combination of three above Malmendier and Tate (2005a) - CEO option holdings

- CEO share buying’s of own company Malmendier and Tate (2005b and 2007) - CEO press portrayal

- CEO personal portfolio transactions

(22)

3.2.1 Recent performance of the acquiring firm

The recent stock performance of the acquiring firm could say something about the degree of overconfidence of CEO’s. According to Hayward and Hambrick (1997) company leaders attribute recent success to themselves, although recent good stock performance could also be the result of luck or other factors. The success could lead to more self-esteem of CEO’s and higher interorganizational prestige, which can easily lead to a higher degree of CEO overconfidence. On the other hand, bad recent performance can also undermine the respect, authority and confidence of CEO’s. Therefore, firms with better recent stock performance are more likely to have an overconfident CEO than firms with bad recent stock performance. Hayward and Hambrick find that this proxy is significantly (at the 1% level) correlated with the acquisition premium, which means that the proxy is a reasonable one for measuring CEO overconfidence.

3.2.2 High order acquisition deals

The second overconfidence proxy I use in this research is high order acquisition deals. This says something of the managerial acquisitiveness of the CEO. The more acquisitive a CEO is, the more acquisitions he makes. Doukas and Petmezas (2007) point out that acquisitive CEO’s are more overconfident, because they attribute the initial success of the first acquisition to themselves and therefore engage in more acquisitions, in the presumption that they also can manage these acquisitions to a good end. High order acquisitions are therefore assumed to be led by more overconfident CEO’s than single deals or first deals of frequent acquirers.

3.2.3 Issuance of equity

(23)

overconfident CEO’s are more sensitive to cash flows when making investment decisions. This high sensitivity of corporate investment to cash flow means that CEO’s overinvest when they have abundant internal funds and underinvest when they require external funds. The CEO’s are reluctant to issue new equity in this case. From these findings of Malmendier and Tate we can conclude that overconfident CEO’s think their own company share price is undervalued and therefore dislike an issuance of equity. We would expect overconfident CEO’s therefore to make less use of equity issues when acquiring other firms compared to non-overconfident CEO’s.

3.3 Research expectations

(24)

4 Methodology

4.1 Event study

For testing my hypotheses I use the event study methodology. An event study looks at or around a certain date, when an event is happening, for company stock returns that are abnormal and therefore can be attributed to the event. In this case the announcement date of an acquisition is taken. An advantage of this methodology is that the effect of the event on the share price (and so what the opinion of the market is about the effect of the event on the firm) is measured immediately. Because of the short time period of measurement in the event study the chance of any impact of concurrent events on the share price is minimized. A disadvantage is that the long-term impact of the event on the firm performance is not measured. This could be a problem when the effects of the event are not immediately clear to the market.

(25)

where A_i,_t is the abnormal return of company i on day t, Ri,t is the actual stock return of company i at day t and R_m_,_t is the return of the market on day t.

For the whole sample, on each day in the event period, a t-value has to be computed according to the following formula:

) ( / t t A A t= σ (2)

where A is the average abnormal return on day t in the event period and t σ(At)is the standard deviation of the average abnormal return. There are n-1 degrees of freedom. A _t is computed according to the following formula:

∑

= = Nt i t i t t A N A 1 , 1 (3) where N is the number of securities with abnormal returns on day t and t Ai,t is the abnormal return of security i on day t.

) (A_t

σ is computed according to the following formula (taking into account a 5 day event-period and an estimation event-period of 145 days):

            − =

∑

=− − = 144 / ) ( ) ( 3 147 2 t t t t t A A A σ (4) where

∑

− = − = = 3 147 145 1 t t t t A A (5)

Following MacKinlay (1997) the cumulative abnormal return is computed by adding the abnormal returns of the individual days in the event period:

∑

= − = − = 2 2 2 , 2 t t t A CAR (6)

and the standard deviation of the cumulative abnormal return for a 5-day event period is then: ) ( * 5 ) (CAR ₂_,₂ σ A_t σ − = (7)

(26)

2 2 2 1 2 1 , 2 , 1 n n A A t t t σ σ ₊ − = (8)

where the under scripts 1 and 2 stand for the different sub-groups and n is the number of observations in a particular sub-group. The standard deviations used in this formula are not the same as the ones computed for testing whether abnormal returns are different from zero (as in formula 4). The standard deviations used for the differences between sub-groups are computed for the specific sub-groups on the particular event day. This is done with a statistical package (Eviews). There are min

(

n1,n2

)

−1 degrees of freedom.

For the difference in sub-groups, also a test of differences in median is done with Eviews, namely the Wilcoxon signed rank test. This test results in a z-value indicating the significance of the difference.

For all calculations trading days are used, so weekends and holidays (on which the stock exchange is closed) are not included.

4.1.1 Choice of the market index

(27)

4.1.2 Event window

In other studies different event windows are used. The event window is important because it defines how long one thinks the effect of an event on the share price takes. A too short event window does possibly not cover the entire effect of the event, while a too long event window has the risk of effects of other specific occurrences that can influence the share price. In line of many recent studies (see table 1) I choose for the -2,+2 event window. This means that besides the event date, the two days before and the two days after this date are taken into account for measuring abnormal returns. Because of the use of closing prices, some acquisitions (which are announced after trading hours) can be expected to have the most effect on the share price at day +1.

4.1.3 Non-normality, autocorrelation, variance changes and clustering

According to Brown and Warner (1985) daily stock returns could suffer from non-normality in their distribution, but they also mention that according to the Central Limit Theorem, if the abnormal returns in the cross-section of the securities are independent and are identically distributed drawings from distributions with a finite variance, then the average abnormal return (which is the important variable in an event study) converges to normality as the number of observations in the sample increases. Brown and Warner conclude in their own research that non-normality of excess returns does not influence the results in event studies using daily stock returns and that standard parametric tests for average excess returns are well specified.

(28)

A third possible problem with the event study is a sudden change in the variance of the excess return of a share. This could lead to misspecification of the model. Especially variance increases around events are worth looking at, according to Brown and Warner (1985). Such a variance increase could lead to an underestimation of the significance of excess returns around an event. In order to see whether this is a problem in my thesis, I split up the estimation period of 145 days in 2 sub-periods to see whether the variance significantly differs in the second sub-period from the first sub-period. If this would be the case, the variance of the first sub-period should be taken instead of the variance of the whole period.

A last possible distortion to the model is the clustering of event dates. If event dates are falling at the same time, the estimation parameters could be distorted. Brown and Warner state that with the market adjusted return model, the market return parameter does absorb clustering quite well and no adjustment is needed, provided that clustering within certain industries is not occurring. So when clustering within industries would take place, an adjustment for this clustering should be made.

4.2 Model specification

4.2.1 Hypotheses

(29)

difference is different from 0%. The average and cumulative abnormal returns computed are all equally weighted.

Second, I test the hypothesis whether a significant difference in abnormal returns between overconfident and non-overconfident acquirers exists. The same methodology is used as for testing the size effect; first the null hypothesis that overconfident acquirers have average/cumulative abnormal returns of 0% is tested against the alternative hypothesis that the average/cumulative abnormal returns are different from 0%. The same is done for non-overconfident acquirers. These tests are done separately for each of the three overconfidence proxies. For the difference between overconfident and non-overconfident firms the null hypothesis is that the difference in average/cumulative abnormal returns is 0%, and the alternative hypothesis is that the difference is different from 0%. Again, this is done for each of the three overconfidence proxies separately.

Last, after testing for the size effect and the effect of overconfidence, I have to test the combined effect and which effect is more dominant; the size effect or the overconfidence effect. This results in two 2*2 matrices with the size (big or small) and the measure of overconfidence (overconfident or non-overconfident) on the axes. In the first matrix the measure of overconfidence is fixed and in the overconfident/non-overconfident groups the size effect is tested (I look whether there exists still a difference between small and big firms within these groups). In the second matrix the measure of size is fixed and in the small/big firm groups the overconfidence effect is tested. All three overconfidence proxies are used and tested separately. The null hypotheses tested are that the differences in cumulative abnormal returns are 0% and the alternative hypotheses are that the differences in cumulative abnormal returns are different from 0%.

4.2.2 Measurement of company size

(30)

Euros)4. Firms with a lower market capitalization than this amount are classified as small. Also, a classification is made based on the median of the sample market capitalization. Firms above the median are classified as big, while firms below the median are classified as small.

4.2.3 Measurement of overconfidence proxies

The first proxy for managerial overconfidence is the recent performance of the acquiring firm. In line of Hayward and Hambrick (1997). I use the shareholder buy and hold returns from the 12 months prior to the acquisition announcement to measure recent performance. This means the stock price 3 days before the acquisition announcement is compared to the stock price 1 year before the acquisition announcement:

261 261 3 − = − = − = − = t t t pastyear P P P R (9)

Where R_pastyear is the buy and hold return the year before the acquisition announcement, P is the share price and the underscript is the trading day compared to the acquisition announcement (there are 261 trading days in a year).

The buy and hold returns are unadjusted for the market return, in line with the prior research of Hayward and Hambrick (The market past year return is included in the multivariate analysis, as done by Doukas and Petmezas (2007)). Firms with returns in the best half compared to the median return are labeled as overconfident, while firms with returns in the worst half are labeled as non-overconfident.

The second proxy is the high order acquisition deals of firms. In line of Doukas and Petmezas (2007) single acquirers and the first deals of multiple acquirers are labeled as non-overconfidence based acquisitions, while higher order deals of multiple acquirers are labeled as overconfidence based acquisitions.

4_{Data found at}_{www.euronext.com}_{, the document can be found at}

(31)

The last proxy used is the issuance of equity prior to making an acquisition. Companies that issue equity in the quarter of an acquisition announcement, and therefore do not see their equity as too cheap to issue, are labeled as non-overconfident acquirers, while companies that do not issue equity in the quarter before the acquisition announcement are labeled overconfident acquirers, because they think their equity is undervalued. When the number is not available (which is especially the case with some acquisitions in 2007 and 2008) the acquisition announcement is disregarded for the analysis.

4.3 Multivariate analysis

Besides the event study methodology previously discussed, a multivariate analysis is executed including a number of other possible explaining variables. For this analysis, I use of the Ordinary Least Squares (OLS) model. First I execute the analysis with only the size and one overconfidence proxy as independent variables. After that, an interaction variable is added. Then I add the control variables but exclude the interaction variable. Finally I add the interaction variable once again.

The dependent variable in the OLS analysis is the cumulative abnormal return (CAR) measured over 5 days as before. The independent variables are chosen based on previous studies on acquisitions. For some variables a dummy value is included (the overconfidence variables of high order deals and equity issuance equal 1 in case of CEO overconfidence and 0 in case not): The independent variables are as follows:

Ln(acquirer size): the natural logarithm of the market capitalization of the acquirer, the proxy for the acquirer size.

Acquirer past year return: the acquirer buy and hold share return in the year before the acquisition announcement.

(32)

Interaction variable: this variable is the product of the size variable (Ln(acquirer size)) and one of the three overconfidence variables.

Relative size: the relative size of the target compared to the acquirer. The relative size of the target is measured by taking the deal value of the acquisition and dividing this value by the size of the acquirer. The size of the acquirer is measured by taking the market capitalization of the acquirer at time of the acquisition announcement.

Cash: this variable equals 1 in case the acquisition is paid solely by cash and debt.

Shares: this variable equals 1 in case the acquisition is paid solely by shares. Mixed: this variable equals 1 in case the acquisition is paid by both cash and shares. Cross-border/National: this variable equals 1 if the acquisition is cross-border and 0 if the acquisition is national.

Public/Private: this variable equals 1 in case the target is a publicly listed firm and 0 if the target is a private firm or subsidiary.

Market to Book: this variable measures the market value of assets relative to the book value of assets. It is also called Tobin’s Q.

Leverage: this variable measures the total debt compared to the total assets of the acquiring firm.

(33)

5 Data

5.1 Data selection

For collecting data about the acquisitions, I used the database ‘Zephyr’ which gives a good overview of all the acquisitions from 1997 till now. Datastream has been used to find stock data for computing the abnormal returns and for company data about the size of the firm, the issuance of equity and the past stock performance. In order to come to a manageable sample of acquisitions, the criteria showed in table 3 have been used:

Table 3 Data criteria

In this table an overview of the variables used for the sample selection and the criteria and values to be included in the sample is given.

Variable Criterion to be included in sample

Time period - Completed between 01/01/2001 and now Percentage of stake - Initial stake 0%

- Final stake minimal 50%

Quotation - Quoted acquirer

Deal value - Minimum of 5 million Euros

Current deal status - Completed

Industries - All industries, but no financials and insurance companies

Countries (acquirer) - All countries which were member of the EU the whole time period, except the United Kingdom

(34)

The percentage of stake has to be at least 50% after the acquisition has taken place. This criterion is chosen because an acquisition of very small stakes in a company could easily have only marginal influence on the acquirer performance.

The acquirer furthermore has to be quoted on a stock exchange. Only in this way the abnormal returns of the acquirer can be measured.

For the deal value I have chosen a minimum of 5 million Euros. Again, this criterion has the goal to exclude very small deals that have only negligible impact on the acquirer.

The deal has to be completed at the moment I write this to be sure that only serious and possible successful acquisition announcements are included.

Companies in financial industries are excluded because of the special position of these firms and the regulated environment in which they operate.

(35)

5.2 Descriptive statistics

From the sample of 499 acquisitions, the main statistics with respect to acquirer size can be seen in table 4.

Table 4

Overview of acquirer size statistics

In this table some basic statistics about the acquirer size in the sample are given. Also, the classification into big and small firms based on the Euronext 100 criterion and the median market capitalization criterion

are given. Percentages of total number of acquisition announcements are included. Acquirer size statistics

Market capitalization acquirer

Mean € 5,521,606,774

Median € 1,079,520,000

Maximum € 119,650,800,000

Minimum € 2,330,000

St. Deviation € 12,737,169,992

Big/Small (based on market capitalization > 2,630 billion Euro)

Big 164 32.87 %

Small 335 67.13%

Total 499 100%

Big/Small (based on median market capitalization)

Big 250 50.10%

Small 249 49.90%

Total 499 100%

(36)

The statistics with respect to the overconfidence proxies can be seen in table 5. Table 5

Overview of overconfidence proxies’ statistics

In this table an overview of the three overconfidence proxies is given. First, for each company the number of acquisitions can be seen. Second, the number of firms that either issued equity or did not in the quarter before acquisition announcement can be seen. Last, some key numbers for the past year (the year before the

acquisition announcement) returns are given. Also percentages of the total number of acquisitions or acquisition announcements are given.

Overconfidence proxy Amount in sample Percentage

Acquirers by number of acquisitions

1 acquisition 284 77.81% 2 acquisitions 51 13.97% 3 acquisitions 17 4.66% 4 acquisitions 5 1.37% 5 acquisitions 6 1.64% 6 acquisitions 2 0.55% Total 365 100% (Number of companies) Equity issuance Yes 249 49.90% No 160 32.06% Unknown 90 18.04% Total 499 100%

Past year buy and hold returns

Mean 20.91%

Median 11.72%

Maximum 375.00%

Minimum -86.65%

St. Deviation 51.87%

(37)

Looking at the equity issue proxy, from 249 acquisition announcements the acquirer issues equity in the quarter before announcement and from 160 acquisition announcements the acquirer does not. The firms in the first group are classified as non-overconfidence based acquisition announcements and the firms in the second group are classified as overconfidence based acquisition announcements. From 90 acquisition announcements there was no data available about acquirer equity issues and these announcements are ignored when using this overconfidence proxy.

Looking at the past year buy and hold returns, the returns range from a -86.65% low to a 375.00% high. The median of 11.72% is used to classify firms as overconfident (past year return higher than 11.72%) or non-overconfident (past year return lower than 11.72%).

In appendix table A1 some other interesting sample statistics can be found, such as the method of payment, whether the deal is national or cross-border, whether the target is a public or private firm and some information about the deal values. In the appendix also more information about the distribution of acquisition announcements between different years (figure A1) and about the acquirer country of origin (figure A2) can be found.

With respect to the status of the target, it is remarkable that 90.18% of the targets are private firms, while only 9.82% are public firms. This means the results of this thesis can be quite well compared to the results of Doukas and Petmezas (2007), which use only private targets.

(38)

In figure 1, the distribution of the average abnormal returns in the estimation period can be seen.

Figure 1

Distribution average abnormal returns estimation period

In this figure an overview is given of the average abnormal returns in the announcement period. The announcement period is 145 days and ranges from t=-147 to t=-3. Also a Jarque-Bera coefficient is given,

which combines the skewness and kurtosis in a single number.

It can be seen in figure 1 by the Jarque-Bera coefficient of 1.50 that normality in average abnormal returns can not be rejected. So the parametric tests explained in the methodology section can be applied on the sample.

(39)

6 Results

6.1 Size effect

As mentioned earlier, for testing whether a size effect is present in the sample of European acquisitions, 2 different size proxies are used, namely the Euronext100 criterion (market capitalization higher than the value to be included in the Euronext100 index at the start of the index, which is 2,630 million Euros) and the median of the sample market capitalization. In table 6 the results of the t-tests are shown using the Euronext100 criterion.

Table 6

Average and cumulative abnormal returns for big and small acquirers based on the Euronext 100 criterion

In this table the average abnormal returns (AAR) for each day of the event period and the cumulative abnormal returns (CAR) are shown for the whole sample and for the sub samples of big and small acquirers

based on the Euronext 100 criterion. This means firms with a market capitalization of more than 2,630 million Euros are classified as big, while firms with a market capitalization smaller than this amount are

classified as small. All abnormal returns are percentages and are equally weighted. T-values are in parentheses. a Denotes statistical significance at the 1% level; b Denotes statistical significance at the 5%

level.

Event day All Small Big Small-Big

Acquirers Acquirers Acquirers

AAR -2 0.258b 0.369b 0.031 0.337 (2.322) (2.571) (0.210) (1.674) AAR -1 0.001 0.045 -0.088 0.133 (0.010) (0.312) (-0.586) (0.670) AAR 0 1.051a 1.420a 0.296 1.125a (9.464) (9.907) (1.972) (3.242) AAR +1 0.683a 0.850a 0.342b 0.508 (6.155) (5.931) (2.284) (1.711) AAR +2 0.135 0.178 0.046 0.133 (1.212) (1.242) (0.304) (0.572) CAR (-2,+2) 2.127a 2.862a 0.627 2.235a (8.571) (8.928) (1.871) (4.057) No. of observations 499 335 164

(40)

day 1 is significant at the 5% level and the cumulative abnormal return of 0.63% is statistically not significant.

The difference between small and big acquirers is statistically significant at the 1% level for the announcement day and also the cumulative abnormal return is highly significant with small acquirers outperforming big acquirers with 2.24% on average over the whole event period. So using the Euronext100 criterion, a size effect definitely exists. The significant positive abnormal returns for small acquirers, the insignificant abnormal returns for big acquirers and the significant difference between small and big acquirers are in line with the results of Moeller et al. (2004), although the abnormal returns in this European sample are somewhat more positive than the abnormal returns found in their US sample.

In table 7, the same tests using the median market capitalization of acquiring firms can be seen.

Table 7

Average and cumulative abnormal returns for big and small acquirers based on the median market capitalization criterion

In this table the average abnormal returns (AAR) for each day of the event period and the cumulative abnormal returns (CAR) are shown for the whole sample and for the sub-samples of big and small acquirers

based on the median market capitalization criterion. This means firms with a market capitalization larger than the sample median (which is 1,052 million Euros) are classified as big, while firms with a market capitalization smaller than the sample median are classified as small.. All abnormal returns are percentages

and are equally weighted. T-values are in parentheses. a Denotes statistical significance at the 1% level; b Denotes statistical significance at the 5% level.

Event day All Small Big Small-Big

Acquirers Acquirers Acquirers

(41)

Using the median of the market capitalization as the size criterion does not change the results very much, although big acquirers do make significant positive abnormal returns here (at t=0 and t=1 at the 1% level) and also the cumulative abnormal return is significantly positive for big acquirers. The difference between small and big acquirers is still significant however, with small acquirers outperforming big acquirers with 2.31% on average during the event period. Concluding, it can be said that the acquisition announcements of small acquirers are better received in the stock market than the acquisition announcements of big acquirers, irrespective of which size proxy is used.

6.2 Effect of CEO overconfidence

(42)

Table 8

Average and cumulative abnormal returns for firms with good recent performance compared to firms with bad recent performance

In this table the average abnormal returns (AAR) for each day of the event period and the cumulative abnormal returns (CAR) are shown for the whole sample and for the sub-samples of firms with returns higher than the median return in the sample in the year before the acquisition announcement and firms with

returns lower than the median return in the sample in the year before the acquisition announcement (where the year ends 3 days before the announcement). Outperformance of firms compared to other firms could lead to more overconfidence of CEO’s of these firms, while underperformance does not reasonably lead to

more overconfidence. All abnormal returns are percentages and are equally weighted. T-values are in parentheses. a Denotes statistical significance at the 1% level; b Denotes statistical significance at the 5%

level.

Event day All Outperforming Underperforming Difference Acquirers Acquirers Acquirers

AAR -2 0.258b 0.114 0.402 b -0.287 (2.322) (0.879) (2.260) (-1.518) AAR -1 0.001 -0.029 0.032 -0.061 (0.010) (-0.225) (0.178) (-0.744) AAR 0 1.051a 1.212a 0.888a 0.324 (9.464) (9.316) (4.998) (0.996) AAR +1 0.683a 0.883a 0.483a 0.400 (6.155) (6.784) (2.717) (1.432) AAR +2 0.135 0.111 0.158 -0.046 (1.212) (0.856) (0.888) (-0.213) CAR (-2,+2) 2.127a 2.291a 1.962a 0.329 (8.571) (7.876) (4.938) (0.626) No. of observations 499 250 249

(43)

In table 9 the results using high order acquisition deals as an overconfidence proxy can be seen.

Table 9

Average and cumulative abnormal returns for single/first deals and for high order deals

In this table the average abnormal returns (AAR) for each day of the event period and the cumulative abnormal returns (CAR) are shown for the whole sample and for the sub-samples of on the one hand single

acquirers and the first acquisitions of multiple acquirers and on the other hand high order acquisitions of multiple acquirers. A high order deal could be seen as a more overconfidence-driven deal, while a single or

a first deal could be seen as a deal where overconfidence plays less of a role. All abnormal returns are percentages and are equally weighted. T-values are in parentheses. a Denotes statistical significance at the

1% level; b Denotes statistical significance at the 5% level. Event day All High order Single/First Difference

Acquisitions Acquisitions Acquisitions (high-single/first) AAR -2 0.258b 0.169 0.290b -0.121 (2.322) (0.952) (2.332) (-0.564) AAR -1 0.001 0.073 -0.025 0.098 (0.010) (0.410) (-0.203) (0.468) AAR 0 1.051a 0.478b 1.261a -0.783b (9.464) (2.685) (10.134) (-2.142) AAR +1 0.683a 0.260 0.838a -0.578 (6.155) (1.463) (6.739) (-1.838) AAR +2 0.135 0.320 0.066 0.254 (1.212) (1.798) (0.534) (1.034) CAR (-2,+2) 2.127a 1.301a 2.431a -1.130 (8.571) (3.268) (8.737) (-1.911) No. of observations 499 134 365

In line with the results of Doukas and Petmezas (2007) high order acquisitions are associated with lower wealth effects than first (of multiple) or single deals, although the difference is only statistically significant at the 5% level on the announcement day. High order acquisitions lead to significant positive abnormal returns, but lower than single or first acquisitions. The difference of 1.13% in cumulative abnormal returns is statistically only marginally significant (at the 10% level) because of the high variance in cumulative abnormal returns. So using high order acquisition deals as a CEO overconfidence proxy, more overconfident acquirers seem to make less good acquisition decisions than their non-overconfident counterparts, but most of the times the difference is statistically not significant.

(44)

Table 10

Average and cumulative abnormal returns for firms issuing equity in the quarter before announcement compared to firms not issuing equity in the quarter before

announcement

In this table the average abnormal returns (AAR) for each day of the event period and the cumulative abnormal returns (CAR) are shown for the whole sample and for the sub-samples of firms that do not issue

equity in the quarter before the acquisition announcement compared to firms that do issue equity in the quarter before acquisition announcement. Overconfident acquirers can be thought to not issue any equity because they think their equity is undervalued. Non-overconfident acquirers can be thought to issue equity

because they think their equity is reasonably priced. From 27 firms in the sample it is unknown whether they issued equity in the acquisition year; these firms are left out of this test. All abnormal returns are percentages and are equally weighted. T-values are in parentheses. a Denotes statistical significance at the

1% level; b Denotes statistical significance at the 5% level.

Event day All Equity Equity Difference

Acquirers Non-issuers Issuers (non-issuers-issuers)

AAR -2 0.258b 0.085 0.214 -0.129 (2.322) (0.505) (1.546) (-0.618) AAR -1 0.001 -0.106 0.122 -0.227 (0.010) (-0.630) (0.880) (-1.089) AAR 0 1.051a 1.010a 0.877a 0.132 (9.464) (6.018) (6.345) (0.369) AAR +1 0.683a 1.121a 0.521a 0.600 (6.155) (6.685) (3.769) (1.895) AAR +2 0.135 0.199 0.106 0.092 (1.212) (1.183) (0.767) (0.371) CAR (-2,+2) 2.127a 2.309a 1.840a 0.468 (8.571) (6.154) (5.951) (0.774) No. of observations 499 160 249

Contrary to the expectations, the abnormal returns of equity non-issuers are higher than the abnormal returns for equity issuers, however the difference is not statistically significant at the 1% and 5% levels. An explanation could be that the overconfidence effect of these acquisitions is overshadowed by the equity signaling effect of these equity issues. Shareholders of frequent equity issuers could be foreseeing new equity issues of the acquiring firms when they make acquisition announcements.

6.3 Interaction of CEO overconfidence and size

(45)

small firms. For the size of the firm, only the median market capitalization size proxy has been used, because the 2 different proxies yielded about the same results. Subsequently, the sample has been divided in a big and a small sub-sample, again based on the median market capitalization. After this division, for each sub-sample t-tests are computed for differences in CAR of overconfident and non-overconfident firms, based on each overconfidence proxy.

If the overconfidence effect dominates the size effect, one would expect more big firms in the overconfident sub-sample (if indeed both a size effect as well as an overconfidence effect would appear), and within each sub-sample based on overconfidence no size effect would be present. If the size effect dominates the overconfidence effect, one would expect more overconfident firms in the big sub-sample (again if indeed both a size effect as well as an overconfidence effect would appear), and within each sub-sample based on size no overconfidence effect would be present.

If both effects would exist side by side, both effects should appear on their own and in the sub-samples the groups with the combined effect of having most of the ‘positive characteristics’ (small size and/or non-overconfident CEO’s) should do better than the groups with one or more ‘negative characteristics’ (large size and/or overconfident CEO’s), and the groups with one positive characteristic should do better than the ones with no positive characteristics.