DECLINING RETURNS FOR FREQUENT ACQUIRERS:
A RESULT OF MANAGERIAL OVERCONFIDENCE?
University of Groningen, Faculty of Economics and Business
Master thesis MscBA, specialization Strategy and Innovation
1
DECLINING RETURNS FOR FREQUENT ACQUIRERS:
A RESULT OF MANAGERIAL OVERCONFIDENCE?
ABSTRACT
In this paper I examine if self-bias inflicted overconfidence causes declining returns for frequent acquirers. The sample consists of 321 US firms making a total of 2531 acquisitions in the period 1992-2011. I find evidence that some acquirers have superior skills. Acquirers that generate positive announcement returns in their first deal are more likely to show positive results in subsequent deals. Further I find evidence that ‘skilled’ acquirers face declining returns. Higher announcement returns predict larger declines. This indicates that managers credit success to their own ability with overconfidence as a result. Overconfidence on its turn causes performance to decline. Next to that I find evidence that managers can learn. Bad acquirers can learn to do less bad after unsuccessful acquisitions. Skilled acquirers can reduce their overconfidence by learning.
2
Table of Content
1. Introduction ...3
2. Literature & Hypotheses ...4
2.1 Performance of Multiple acquirers...4
2.2 The concept of overconfidence ...5
2.2.1 Overconfidence ...5 2.2.2 Managerial Overconfidence ...6 2.2.3 Overconfident Acquirers. ...7 2.3 Patterns in performance ...7 2.4 Empirical Evidence ...10 2.5 Conceptual Model...12
3. Data & Methods ...12
3.1 Data ...12 3.2 Variables ...14 3.2.1 Dependent Variables ...14 3.2.2 Independent variables...16 3.2.3 Control Variables ...18 4. Results ...19 4.1 Descriptive statistics ...19 4.2 Univariate Results ...20 4.3 Multivariate Results ...23 4.3.1 Declining Returns ...24
4.3.2 Overconfidence theory tested ...25
5. Conclusion ...27
6. Limitations & implications for future research ...31
References ...33
Appendix 1: Distribution of Acquisitions over the Sample Period ...35
3
1. Introduction
Recent empirical research has revealed that firms that acquire on a frequent base face declining returns from deal to deal (Fuller et al. 2005, Conn et al 2004b, Ismail 2008, Billet & Qian, 2008). Although it is evident from different studies that frequent acquirers perform less on acquisitions positioned later in a sequence there is little consensus on what causes this phenomenon (Conn et al. 2004a, Aktas et al. 2009). Because many acquiring firms engage in multiple acquisitions (Asquith et al. 1983, Schipper and Thompson 1983, Croci & Petmezas, 2009) it is of interest to a lot of firms to know what causes these declining returns.
The focus of this study is on the managerial overconfidence (hubris) theory. According to Aktas et al. (2009) a downward trend in Cumulative Abnormal Returns (CARs) is interpreted as evidence of managerial overconfidence in most of the literature on multiple acquirers. The goal of this paper is to look if managers attribute success to their own skills, causing managerial overconfidence, which than causes declining returns for frequent acquirers. This is done by analyzing short-term stock returns of US firms that make five or more acquisitions within a three year period between 1992-2011. I look at patterns in CARs that are predicted by the hubris theory and by running regression with two proxies that measure overconfidence. I also test for firm- and deal characteristics that are known to influence acquisition performance.
The main research question of this paper is: Does managerial overconfidence cause declining returns for frequent acquirers?
This paper is complementary to works of Conn et al. (2004a,b), Billet & Qian (2008), Ismail(2008), Croci & Petmezas (2009), and Aktas et al. (2009) that also study the performances of multiple acquirers from deal to deal.
4
explained. After this, the results will be presented in section four. The paper will end with the conclusion and limitations of this study in sections five and six.
2. Literature & Hypotheses
2.1 Performance of Multiple acquirersDifferent studies (Rovit & Lemire 2003, Baker & Limmack 2001) have found that multiple acquirers outperform single acquirers. Rovit & Lemire (2003) examined 724 US firms making 7475 acquisitions between 1986 and 2001. They found that the more deals a company made the higher the excess returns were delivered to shareholders.
Baker and Limmack (2001) examined UK acquirers and found significant positive excess returns for multiple acquirers and significant negative excess returns for single acquirers. In the article of Baker & Limmack (2001 p23) it was noted that the findings that multiple acquirers outperform single acquirers suggest that acquirers learn from previous experience. This is in support of the organizational learning theory.
Simply put, the organizational learning theory expects increasing returns for multiple acquires. However, different recent studies (Fuller et al. 2005, Conn et al 2004b, Ismail 2008, Billet & Qian, 2008) have found evidence that, instead of increasing returns, multiple acquirers face a decline in their acquisition performance.
Fuller et al. (2002) found lower Cumulative Abnormal Returns (CARs) on fifth and higher bids than first bids. Looking at firms making five or more acquisition within a three year window they found that firms had a cumulative abnormal return of 2.74% on their first bid and 0.52% on their fifth and higher bids. Controlling for type of payment and type of target they found a lower CAR for fifth and higher bids for each combination of these control variables.
5
negative abnormal return of -0,99%. The authors of this article also compared returns of acquisitions with deal order 2,3,4,5,6 and 7. They found no pattern in these returns, which could be due to the low number of observations (37).
Conn et al. (2004b) compared first, second and third and fourth and above bids. They found declining returns both on the short and long run. First bids performed best. Second and third bids performed less, but still had positive returns. Fourth and higher bids even showed negative returns.
The results of the articles discussed above indicate that there is very little evidence for increasing returns and the organizational learning theory. A conclusion that was also made by Conn et al. (2004a). In this paper I will investigate if the declining returns are caused by managerial overconfidence. Before this, I will first test if performance indeed declines for the sample used in this paper.
H1: Acquisition performance declines for frequent acquirers. 2.2 The concept of overconfidence
To establish if overconfidence is the cause of declining returns it is first important to define what overconfidence exactly is. Below the idea of overconfidence in general and managerial overconfidence are reviewed.
2.2.1 Overconfidence
6
Most people are found to be too optimistic about their own future chances (Weinstein 1980). This is due to the inability to accurately estimate probabilities. People believe good things will happen to them because they want that to happen. This phenomenon is also called over-optimism in the literature.
In psychology literature self-attribution bias is documented as a common source of overconfidence. Hirshleifer (2001 p. 1549) describes the connection between overconfidence and self-attribution as follows: ‘Overconfidence and biased self-attribution are static and dynamic counterparts: self-attribution causes individuals to learn to be overconfident rather than converging to an accurate self-assessment.’ People overestimate the influence they have on good outcomes, and underestimate their influence on bad results. Bad results are too often credited to external factors. As Langer and Roth (1975 p. 951) put it: “heads I win, tails it’s chance.”
The terms over-optimism and overconfidence are used interchangeable in literature. For this study it is important to distinguish those two. Overconfidence as a result of self-attribution bias will be investigated in this study. Over-optimism will not be investigated. This is because I test declining returns to be the result of overconfidence. Overconfidence would be the result of self-bias after successful acquiring. Over-optimism is a personal trait that could already be present before the first acquisition. When present at the start of the acquisition sequence over-optimism would affect the first acquisition as much as later ones. This would not result in declining returns. Overconfidence inflicted by self bias on the other hand can arise after the first acquisition (when successful). And thereafter it can be a cause of decline. 2.2.2 Managerial Overconfidence
Managers are people too, and therefore there is no reason to believe that managers are not prone to overconfidence and self-attribution bias (Doukas & Petmezas, 2007). The idea to be ‘above average’ could even be larger for managers for several reasons.
7
therefore compare themselves to lower level managers. Malmendier & Tate (2005) argue that CEOs think they are better acquirers because they compare themselves to the average manager (which does not make many acquisitions). And they argue that CEOs have such complex jobs that it is very difficult to compare one CEOs skills to another’s. When comparison becomes more abstract the ‘above average’ effect becomes stronger (Moore & Kim, 2003).
2.2.3 Overconfident Acquirers.
According to Aktas et al. (2009) the hubris theory is the most used argument to explain declining returns. Roll (1986) was the first to introduce the hubris theory for acquirers. Based on evidence from psychology that individuals do not always make rational decisions in uncertainty Roll came up with the theory that managers are prone to hubris. Managers have too few opportunities in their career to make acquisitions to actually learn from them Roll (1986) argues. A manager will therefore convince himself that his valuation of a target is right influenced by overconfidence.
Managers can attribute the success of their initial acquisition to their own abilities. This self-attribution bias can lead to overconfidence in the next acquisition. The overconfidence can show in two ways (Doukas & Petmezas, 2009). Overconfident managers may have the feeling to be better at identifying synergies and picking targets than other managers. It can also be that due to the hubris managers engage in more acquisition. In higher order acquisitions the acquirer is not careful enough due to the hubris-effect which can lead to less careful choosing the next target or paying too much (for the deal itself or extra fees) due to less careful negotiation (Roll 1986). Second, overconfidence can lead managers to think their own company is undervalued.
2.3 Patterns in performance
8
deals. (Doukas & Petmezas 2007). Overconfidence should not affect the first acquisition if it is caused by acquisition success. This is in line with the first hypothesis.
If self-attribution bias after success causes overconfidence decline should be much more acute, or present only, for acquirers that have made a successful initial acquisition. Hence, unsuccessful acquirers will not be able to self-attribute and should thus not suffer from overconfidence. Deals following initial success may be value destructive (Roll 1986, Conn et al. 2004b, Aktas et al 2009). Aktas et al. (2009) proposes that overconfidence can be reduced by learning from failures. This would result in an initial decline in performance after a first successful acquisition. After the first successful bid firms get over-confident, but if the second is a failure they should get ‘down to earth’ again, resulting in less decline for the next acquisition in line. It might even be that performance go up after the acquirer was able to learn from his mistakes. For unsuccessful first deals there should be an increase in returns as there is no overconfidence gained from unsuccessful deals, but managers can learn from their mistakes.
H2a: Successful first acquisitions are followed by less performing acquisitions. H2b Unsuccessful first acquisitions are followed by better performing acquisitions. H2c: Successful first acquisitions are followed by acquisitions with negative returns. H2d: Decline in performance is largest after the first acquisition.
Note that finding evidence for hypothesis 2c would support the hubris theory. But not finding negative returns for deals following a successful acquisition does not contradict with this theory.
9
pattern expected under the hubris theory. Conn et al. (2004a,b) listed these theories as possible explanations for declining returns.
The Diminishing Returns Hypothesis argues that multiple acquirers take the best opportunity first, followed by the second best opportunity and so on. Under this hypothesis steady declines from deal to deal are expected. Although each acquisition in a sequence will perform less, returns should stay positive. At least until the last acquisition.
The Mean Reversion hypothesis argues that acquirers that do well at their first acquisition are unable in subsequent acquisitions to maintain the above average takeover performance. If firms are successful at the first acquisition they are more likely to continue with acquisitions as success increasing subsequent acquisition likelihood (Haleblian et al. 2006) Under this hypothesis only a decline after the first bid is expected. After this the performance should return to the mean. Returns of bids after the first successful bids in the sequence should stay steady. Firms that were unsuccessful at their first acquisition but nevertheless decided to continue to acquire should see their returns move to the mean as well.
The Indigestion Hypothesis says that an acquirer is unable to integrate an acquisition into their firm because they are still ‘digesting’ their previous acquisition. Under this hypothesis a increasing decline is expected. The more previous acquisitions a firm is digesting the harder it will get to ‘digest’ another one. Decline should also be found for firms that are unsuccessful at their first acquisition.
10 2.4 Empirical Evidence
According to Aktas et al. (2009) most literature on the performance of multiple acquirers interpret the decline in CARs as evidence of overconfidence. But declining CAR alone is not enough to assume overconfidence influences the performance of multiple acquirers. Different studies have tested the hubris theory for declining returns with different results. Some find (partial) evidence supporting this theory, whereas others find evidence that contradict with the hubris explanation of declining returns. Below a summary of the different findings.
Billet & Qian (2008) investigate if CEOs develop overconfidence as a result of previous acquisition performance. They find that first deals have returns insignificantly different from zero. Subsequent deals all have negative CARs. They also find that acquirers that have success are more likely to acquire again. And they find that acquirers that are experienced exhibit optimism in trading their companies’ stocks. All these three findings support the theory that managers that have success suffer from overconfidence in subsequent acquisitions.
Doukas & Petmezas (2007) tested for overconfidence in UK acquirers. They find that high order deals have lower wealth effects (both announcement returns and long-term performance) than first deals. They argue that this is because managers credit the success of this first deals to their own handling. This self-attribution bias makes them overconfident in later deals. They further find that managers increase their ownership around acquisition announcement because they are overconfident. And they find that trading activity increases for higher order deals indicating that there is intensified difference in opinions among investors. This is a result of overconfidence they argue, because precision of information about an investment is overestimated.
11
in later acquisitions. They also find a negative relation between acquisition frequency and performance. These findings are in line with the hubris theory. However when they test the hubris theory by examining whether success at the first acquisition leads to underperformance in the last acquisition of series they find no evidence for this. In their article Conn et al. (2004a) examine seven different theories that could explain declining returns for multiple acquirers. Based on the performance patterns they leave hubris, mean reversion and diminishing returns as possible explanations for declining returns. Conn et al. (2004a) exclude the Merger Program Announcement Hypothesis, Accounting Manipulations Hypothesis, Overvaluation Hypothesis and Indigestion Hypothesis.
In a study on frequent acquirers (at least 5 deals in 5 years) Croci & Petmezas (2009) conclude that they find not enough evidence supporting the hubris theory. They examine if a successful acquisition is correlated with a lower return in the subsequent acquisition. They find no such correlation which indicates that acquirers do not become overconfident as a result of good acquisition performance. However, when two acquisitions in a row are successful this has a negative influence on the next performance in the sequence. And after one successful acquisition the chance of a large loss-deal becomes bigger. These findings are in line with the overconfidence theory. But the chances of a large gain rises after two successful acquisitions. This finding contradicts with the overconfidence theory.
12
I argue that, although empirical research has shown that good acquisitions are not followed by bad ones, hubris can still be a cause for declining returns. It may very well be that good acquirers become overconfident after success. This can result in lesser performance in subsequent acquisitions. But this does not automatically mean that good acquirers that are ‘hubris infected’ perform less than bad acquirers in a subsequent deal. Therefore I will not only examine how self-bias affects acquisition performance, but also how self-bias affects the change in acquisition performance.
H3: Managerial overconfidence, as a result of self-attribution, is negatively related to acquisition performance.
H4: Managerial overconfidence, as a result of self-attribution, increases decline in acquisition performance.
H5: Acquirers that are successful at their first acquisition will perform better in subsequent deals than acquirers that are unsuccessful at their first acquisition.
2.5 Conceptual Model
From the hypotheses that were formulated in the previous sections I can present that what is investigated in the figure below.
H2a,b,c & H3(-) H2d & H4 (-) Figure 1: Conceptual Model
3. Data & Methods
3.1 DataThe data used to test the hypotheses is derived from the Security Data Company’s (SDC) Mergers and Acquisitions Database and Datastream. This paper looks at acquirers that make
Overconfidence
13
five or more acquisitions within a three year window between 1992-2011. The ‘three acquisition within five years’ frequency is based on the work of Fuller et al. (2002) and Doukas & Petmezas (2007) who define this as frequent acquirers.
In order to find a pattern in a sequence it is important to be sure which bid is the start of a sequence. Loderer & Martin (1990) propose a two year period in which no acquisition has been made before the first bid. Croci & Petzemas (2009) also use a two year period studying performance of multiple acquirers. To be sure that the first bid in the sample is really the first bid after a period of two years, the first two years of the sample are used as a buffer. Acquisition sequences with first bids found in this periods are excluded from the sample. A sequence in this sample starts earliest on 01-01-1992. I check if no acquisition was announced in a two year period between 1990-1992.
In order to avoid that multiple acquisition influence each other’s returns, acquisitions that are in a cluster are excluded. Sequences in which two announcements had less than two days between them are filtered out. Using a 3 day event study (the event day ,the day before and the day after) one day is never used to measure the abnormal returns of two different acquisitions made by the same firm.
To prevent that acquisitions where filtered out of a sequence no filter was applied on relative size, type of payment, target country and target industry. For the same reason acquirers had to acquire at least 51 percent of the target instead of another approach often used in M&A literature: filter out acquisitions where 100 percent of the target is acquired. A listed summary of the criteria that the sample of acquisitions meets :
1. The acquisition announcement falls within a three year period in which the acquirer made a total of at least five successful bids with a minimum of two days between each bid.
2. The acquisition is announced between 1992 and 2011and is completed within 1000 days.
3. No acquisition has been made in a two year period before the first bid of a sequence 4. The acquirer is a publicly listed firm from the United States
5. 51 percent or more of the targets shares are acquired in the transaction.
6. The deal value is equal or greater than $1 Million (excluding fees and expenses) and greater or equal to 1 percent of the acquirers market value.
14
The share price data on firms and the US market index needed to calculate the cumulative abnormal returns are derived from Datastream. If the data needed to calculate the CARs was not available from Datastream the sequence of acquisitions was removed from the sample. The sample of acquisitions that meet the criteria described above consists of 336 acquisition sequences containing a total of 2531 acquisitions. These acquisitions made by 321 different firms. As can be seen from the difference between the amount of sequences and the amount of acquirers some sequences are made by the same acquirer. This is possible because each sequence has a period of two years before the start in which no acquisition has been made. Some acquirers in the sample have made multiple acquisition sequences, but with a large enough period (two years or more) between them to see them as individual sequences.
3.2 Variables
In this section the different variables used to test the hypotheses are discussed. It is explained how the dependent and independent variables are measured and why I choose certain proxies. Further I shortly discuss for which firm and deal characteristics I control, based on which literature.
3.2.1 Dependent Variables
15
may not be a totally accurate measurement of an acquisition’s performance. But since no better alternative is available I will use this often used tool in my research too, accepting its limitations. The abnormal returns are calculated using the market-adjusted model.
ARi = Ri - Rm
Where ARi is abnormal return of acquirer i, Ri is the stock return on acquirer i and Rm is the
return of the market. The Cumulative Abnormal Return (CAR) is the sum of the daily abnormal returns over the event period.
Next to that I use the difference between two acquisitions as a dependent variable. This is calculated as (with n as the order number of the focal acquisition):
∆ Acquisition Performance = CARn – CARn-1
The period I will use to calculate the cumulative abnormal returns is a 3-day period (-1, 1) around the announcement date where the announcement date is day 0. In literature there is no standard length for the event period. I have chosen the shortest possible period in order to preserve as much data as possible (I excluded clustered acquisitions). Next to that it provides some continuity since the 3-day period was used in similar studies of Conn et al. (2004b) and Billett & Qian (2008). Because my sample exists of frequent acquirers I have not estimated market parameters based on a time period before each bid. According to Fuller et al. (2002) and Aktas et al. (2006) the beta estimations are less meaningful for frequent acquirers since previous acquisition announcement ‘contaminate’ the estimation period. Next to that, in Brown and Warner’s (1980) study it was shown that for short-window event studies the OLS market model does not significantly improve estimation .
The Standard and Poor’s composite 500 index (S&P 500) is used to calculate the return of the market. All firms in the sample are listed on this index.
16
After the univatiate analyses I will perform multivariate analyses to further test hypothesis 1 and to test hypothesis 3. In the following section the independent variables and control variables that are used in the regressions are described.
3.2.2 Independent variables
The independent variables for hypothesis 1 are bid order and bid proportion. Bid order is the number if an acquisition within a sequence of acquisitions made by a firm. The first acquisition in the sequence is given number 1, the next number 2 etc. Next to bid order I will also use bid proportion to test for hypothesis 1. Bid proportion is the bid order of an acquisition divided by the total number of acquisitions in the sequence (Ismail 2008, Conn et al. 2004a,b). This second measure is used for robustness, because the number of acquisitions in a sequence can influence acquisition performance (Conn et al. 2004a).
The independent variable for hypothesis 3 is managerial overconfidence. Managerial overconfidence is a personal trait that is not directly measurable. According to Malmendier & Tate (2005b) “the biggest challenge for the analysis of overconfidence is to construct a plausible measure of overconfidence” (p. 652).
17
Two proxies for which data is available for every acquisition in the sample are recent performance and acquisition frequency. Acquisition frequency is used in articles of Billet & Qian (2008) and Doukas & Petmezas (2007). According to these authors acquisition frequency measures overconfidence because acquiring too much firms in a short time is a bad investment strategy caused by overconfidence.
The first overconfidence proxy I will use is the number of days between the focal announcement (n) and the previous announcement (n-1). From now on I will refer to this as TBD (time between deals) like was done by Aktas et al. (2009). It must be noted that Conn et al. (2004a) argue that acquisition frequency can also be a measurement tool for indigestion. Therefore I will use another proxy to check for overconfidence
The second proxy of overconfidence that I will use is recent performance. In different studies (Hayward & Hambrick 1997, Croci & Petmezas 2009) this was used to test for hubris. It is argued that it is likely that managers attribute recent success to their own qualities too much (self-attribution bias). This makes them overconfident.
Hayward & Hambrick (1997) used the stockholders return over the year before the acquisition as a measurement tool. They argue that success leads managers to increasingly believe in their capabilities. The more successful the organization was in the past the more overconfident the manager will be. Croci & Petmezas (2009) tested if positive returns in previous acquisitions decreases the performance in later acquisitions.
Instead of creating a dummy for positive returns I will use the performance (CAR) of the acquisition prior (n-1) to acquisition n. The higher the abnormal return on acquisition n-1 the higher the overconfidence of the manager when making acquisition n. By using this measurement I make the same assumption as Hayward & Hambrick (1997) that overconfidence increases as recent performance increases.
18
When decline is a result of overconfidence, this should mean that overconfidence was not influencing the performance of the first acquisition. Rather, success causes overconfidence which on its turn causes declining performance.
To test if acquirers with success on their first acquisition have superior skills I include the dummy variable ‘initial success’ in the regression. This variable is 1 when the first acquisition in the sequence has generated a positive CAR and zero otherwise.
3.2.3 Control Variables
I will control for a number of characteristics (both firm- and deal-) that are known to influence acquisition performance. Below these control variables are listed with a short explanation.
Firm size. Moeller et al. (2004) found that small firms are better acquirers than large firms. Firm size is measured as the acquirers market value at the beginning of the year in which the acquisition was made. The data to measure firm size was derived from Datastream.
Deal size. Moeller et al. (2004, 2005) showed that large deals underperform. Deal size is the value of transaction as reported by SDC.
Domestic vs Foreign. Fuller, Netter & Stegemoller (2002) showed that acquiring foreign targets generates lower returns than national acquisitions. The dummy ‘domestic/foreign’ will be 1 when the target is foreign and 0 when the target is from the US.
Acquirer to target similarity, Morck, Shleifer, and Vishny (1990) show that acquirers
19
4. Results
Below the results of the statistical analyses are described. This section is divided in three sections. First the descriptive statistics are presented. Hereafter the results of univariate analysis will be discussed. The third section will finish this results chapter with the multivariate results.
4.1 Descriptive statistics
Appendix A provides statistics of the distribution of the acquisitions in the sample over the period of this study. The most acquisition activity is in the period 1996-2000 where the number of acquisitions in the sample exceeds 200 for each year. The number of first acquisitions is highest in the period 1995-1997. Between 1997 and 2000 the number of acquisitions with deal order six and above are higher than other years.
Table 1: Descriptive Statistics
Number Min Max Mean Median
Deals 2531 Acquirers 321 Sequences 336 TBD (days) 5 1289 149,68 102 Firm Size ($m) 3,74 312939,00 9331,34 756,03 Deal Size ($m) 1,00 59515,24 319,63 32,00
Number % of total acq.
Foreign 395 15,61 Domestic 2136 84,39 Diversified 1593 62,94 Non-Diversified 938 37,06 Initial Success 202 60,12 Initial Failure 134 39,88 In Table 1 additional descriptive statistics are provided.
The sample contains 336 acquisition sequences made by 321 acquirers. In total 2531 deals are looked at. Of the 336 acquisitions 202 started with a positive CAR on their first
20
considerable higher figure than the median. This indicates there are a few extreme large values skewing the mean upward. To prevent these large values from influencing the regression results to much I use the natural log of these three control variables.
Another interesting fact that is worth noticing is that a large percentage of the acquisitions that were made concerned domestic targets.
4.2 Univariate Results
Table 2: Mean performance per deal order
The Cumulative abnormal returns are calculated over the event window (−1, 1) where day 0 is the announcement day. Abnormal returns for a firm on a certain day is ARi = Ri - Rm where ARi is the acquirer i’s abnormal return, Ri is the stock
return on acquirer i and Rm is the return of the market (S&P 500). CARs are the sum of the abnormal returns over the event
window. Successful 1st acquirers are those firms that had a positive return on the first acquisition in the sequence.
Unsuccessful 1st acquirers had a negative return on the first acquisition
Deal Order 1st 2nd 3rd 4th 5th Higher >1 All All Acquisitions 2,40*** 0,58 1,10** 0,80* 0,45 -0,17 0,39* 0,65*** (336) (336) (336) (336) (336) (851) (2195) (2531) Successf. 1st Acq. 7,54*** 1,59*** 2,49*** 2,14*** 1,33** 0,89** 1,54*** 2,37*** (202) (202) (202) (202) (202) (439) (1247) (1449) Unsuc. 1st Acq. -5,36*** -0,94 -1,01 -1,19 -0,89 -1,30** -1,13*** -1,65*** (134) (134) (134) (134) (134) (412) (948) (1082)
***,**,* denote significant difference from zero at respectively 1,5 and 10% levels using a two tailed test
In table 2 the mean CARs for each deal order are presented for all acquisitions and for the firms that where either successful or unsuccessful at their first acquisition.For the total sample I find a positive mean abnormal return of 0,65% that is significantly different from zero. This means that on average frequent acquirers gain from acquisitions in terms of stock returns.
21
The CARs stay positive after the first acquisition until the fifth acquisitions in the sequence. So the possible negative CARs expected based on the original hubris theory of Roll (1986) are not found. Hypothesis 2c is thus not supported by these results. This finding is similar to most studies on multiple acquisitions (Croci & Petmezas 2009, Doukas & Petmezas 2007, Conn et al. 2004b). It contradicts with findings of Billett & Qian (2008) that did find negative returns after the first deal.
Looking at the CARs from deal to deal it is found that there is a decline from the first to the second deal. The difference between the first and second acquisition is the largest decline in the sequence with. The difference is 1,82 (significant at a five percent level). This is the only difference between subsequent deals that is significant. The third deal in the sequence is performing better than the second deal. After the third deal the CARs are declining again. But the difference between the third and fourth is not as large as that between the first and second acquisition. These findings support hypothesis 2d. This is in line with the theory of (Aktas et al. 2009) that success causes overconfidence (and thus a decline in returns), but managers are also able to learn (resulting in increasing returns for the third deal, and less strong decline after the success of the third deal).
22
Looking at the returns of acquirers that have success at their first acquisition I find a similar pattern as for the complete sample. From the second row in table 2 it can be seen that performance significantly declines after the first success (7,54% vs 1,54%)(t=9,46). This supports hypothesis 2a. For each deal order the CARs stay positive and significant different from zero. This is similar to findings of Ismail (2008). Different from the findings of Ismail (2008) is that after the second acquisition the performance goes up again. After the third there is again a decline. This also supports the theory of (Aktas et al. 2009) that success causes overconfidence (and thus a decline in returns), but managers are also able to learn (resulting in increasing returns for the third deal, and less strong decline after the success of the third deal).
Unsuccessful first acquisitions are followed by significant better performance (5,36% vs -1,13%)(t=4,03). This finding supports hypothesis 2b and is in line with the theory that managers can learn from mistakes in previous deals. However, firms that are unsuccessful at their first acquisition also have negative returns on subsequent deals, although not significant different from zero. This is different from findings of Ismail (2008) who found significant positive results after an unsuccessful first deal. Thus, firms seem to be learning after an unsuccessful first deal, but not enough to generate positive returns.
Interesting to see is that acquirers that are successful at their first acquisition keep performing better than acquirers that are not successful at first. For all deal orders successful first acquirers have CARs that are higher than unsuccessful first acquirers. This finding is in line with the Croci & Petmezas’ (2009) conclusion that some managers have superior acquisition skills. Hypothesis 5 is supported by these findings.
23
acquisition. I find decreasing decline however. Next to that I find increasing performance for acquirers with negative returns on their first deal. The indigestion theory also expects decline after unsuccessful acquisitions.
The acquisition program announcement hypothesis expect only significant returns after the first and maybe the second announcement. For deals with a higher bid order I still find significant positive announcement returns.
To summarize: the univeriate results point towards the overconfidence theory as a cause for decline and contradict with a number of other theories. The results also indicate that managers can learn from failures.
4.3 Multivariate Results
Below the results of ordinary least squares regressions are presented. First it is established whether frequent acquirers face declining returns. Thereafter it is tested if the overconfidence theory can explain the decline in announcement returns.
Before discussing the results of the regressions I will shortly discuss the correlation table presented in Appendix 2. The correlation table presents the correlations between the independent variables used in the regression. No unexpected correlations where found. The largest correlation observed is that between bid order and bid proportion (0,69). This was expected since these are similar measures. Bid proportion is calculated by dividing bid order by the number of bids in a sequence.
24 4.3.1 Declining Returns
Table 3: Regressions of acquisition performance on bid order and bid proportion.
The Cumulative abnormal returns are calculated over the event window (−1, 1) where day 0 is the announcement day. Abnormal returns for a firm on a certain day is ARi = Ri - Rm where ARi is the acquirer i’s abnormal return, Ri is the stock
return on acquirer i and Rm is the return of the market (S&P 500). CARs are the sum of the abnormal returns over the event
window. ‘Firm Size’ is the natural log of the market capitalization ($m) of the acquirer. ‘Deal size’ is the natural log of the deal value ($m). ‘Foreign’ is a dummy that takes value 1 if the target firm is not from the United States. ‘Target Similarity’
takes value 1 if the 4 digit SIC code of the acquirer and target are the same. ‘Bid order’ is the numerical order of an acquisition in a sequence. ‘Bid proportion’ is the bid order divided by the total number of acquisitions in a sequence.
(1) (2)
Intercept 0,82 1,18
Firm Size (log) 0,01 -0,1
Deal Size (log) 0,20 0,21
Foreign -0,02 -0,03
Target Similarity 0,38 0,39
Bid Order -0,20***
Bid Proportion -1,38*
R2 0,003 0,005
***,**,* denote statistical significance at respectively a 1%, 5% and 10% level.
In table 3 the first hypothesis is tested. Two regressions are run with acquisition performance (announcement returns) as the dependent variable. Bid order (model 1) and bid proportion (model 2) are the predicting variables of interest here. Both variables are significantly negative. With this the univariate results are confirmed. Announcement returns (CAR) are lower for deals positioned later in a sequence. For the four control variables I do not find significant results.
My findings are in line with the existing M&A literature (Fuller et al. 2005, Conn et al 2004a,b, Ismail 2008, Billet & Qian, 2008). In particular my findings are very similar to that of Conn et al. (2004b). For bid order I find a beta of -0,20 where Conn found -0,24. For bid proportion I find a beta of -1,38 where Conn found -1,47.
25 4.3.2 Overconfidence theory tested
Table 4: Regressions of acquisition performance on overconfidence and initial success
The dependent variable (CAR) and control variable are as explained in table 3. ‘Previous Deal Performance’ is the CAR of deal n-1. ‘TBD’ (Time Between Deals) is the natural log of the number of days between deal n and deal n-1. ‘Initial success’
is a dummy that takes value one if the first acquisition of the sequence generated a positive announcement return.
(1) (2) (3) (4)
Intercept -0,67 -0,02 -0,81 -0,51
Firm Size (log) 0,02 -0,03 0,04 0,04
Deal Size (log) 0,15 0,25* 0,22 0,14
Foreign -0,13 -0,18 -0,03 0,07
Target
Similarity 0,32 0,4 0,37 0,31
Prev. Deal Perf. 0,38*** 0,37***
TBD (log) -0,1 -0,06
Initial Success 1,33*** 0,542***
R2 0,15 0,003 0,02 0,15
***,**,* denote statistical significance at respectively a 1%, 5% and 10% level.
In model 1 I test how the performance of the previous deal influences the performance of the focal deal. I find a significant positive relation here. This means that a good performance is likely to be followed by a good performance. Recent performance thus can predict the performance of an acquisition. This finding does however not support hypothesis 3. I expected to see a negative relation between the CAR of the previous acquisition and that of the focal deal. Hereby it was assumed that higher CAR leads to higher overconfidence and higher overconfidence should lead to worse performance. The findings from model 1 suggest the opposite.
26
time between deals is relative low for all acquisitions in the sample. Thus a difference between the TBD’s in this sample might not reveal a different level of overconfidence. The univariate results discussed earlier provided evidence for Croci & Petmezas’ (1999) theory that some managers have superior skills. The findings from model 1 are in line with this findings. To further test this theory I included the variable ‘initial success’ test how success in the first acquisition relates to acquisition performance. A significant positive relationship is found between a positive CAR in the first acquisition and the acquisition performance. As was proposed in hypothesis 5 a good initial performance predicts good performance later on in the sequence compared to unsuccessful first acquirers. This again is in line with the superior skills theory.
The superior skills theory however does not explain the declining returns for frequent acquirers. But having established that acquisition skills play an important role in acquisition performance I argue that overconfidence can still exist. Good acquirers may suffer from overconfidence and perform less after having success as a result. However, an acquisition performance of a skilled acquirer that is negatively influenced by overconfidence may still be better that that of a unskilled acquirer. A less skilled acquirer should not face a decline as high as a skilled acquirer because an less skilled acquirer performs less which results in less overconfidence. To test this I run a regression with the same proxies of overconfidence, but this time with the ∆CAR as dependent variable.
Table 5 Regression of ∆ Acquisition Performance on overconfidence.
The dependent variable is the difference in acquisition performance between the focal acquisition and the acquisition before it. This is calculated as CARn - CARn-1 where n is the deal number of the focal acquisition. A decline in performance will thus give a negative difference and an increase a positive difference. The calculation of the CARs and the independent variables are explained at table 3. The independent variables are explained at table 4. In Model 1,2 and 3 the regression is run using the compete sample. Model 4 includes only the deals that follow a good deal (positive CAR). Model 5 includes
only the deals that follow a bad deal.
(1) (2) (3) (4) (5)
Intercept -0,67 -1,01 -0,40 3,33*** 3,41**
Firm Size (log) 0,02 0,09 0,02 -0,33** -0,23
Deal Size (log) 0,15 0,01 0,15 0,17 0,12
Foreign -0,13 -0,06 -0,13 -0,42 0,17
Target Similarity 0,32 0,19 0,32 -0,72 0,98
Prev. Deal Perf -0,62*** -0,62*** -0,97*** -0,26***
TBD (log) 0,01 -0,06 -0,04 0,14
R2 0,33 0,00 0,33 0,44 0,07
27
When ∆ Acquisition Performance is the dependent variable I find previous deal performance to be negative in the regression. This indicates that the higher the CAR of an acquisition the larger the decline that follows. And the other way around: the more negative a performance the larger the increase in performance. I run two separate regression to test if the negative relationship is indeed present for both acquisitions following a good deal (model 4) and acquisitions following a bad deal (model 5). The results show that this is the case. This supports the self-bias inflicted overconfidence theory. Higher abnormal returns lead to higher overconfidence. Higher overconfidence leads to larger declines. Bad performers are not ‘hubris infected’ and thus do not have declining returns. They see their performance go up instead. This increasing performance after unsuccessful deals can be explained by the theory of Aktas et al. (2009) that acquirers can learn from their mistakes.
Time between deals (TBD) again shows to have no impact on the dependent variable. Like after the first regression it can be concluded that either overconfidence does not influence the decline or that TBD is not a good measurement tool for overconfidence for this sample or in general.
Interesting to see is that in model 4 the control variable firm size becomes significant negative. Larger firms thus face larger declines after a successful acquisition. Moeller et al. (2004) found that the size of an acquirer can influence acquisition performance. Moeller et al. (2004) also argue that it is likely that manager of larger firms are more prone to overconfidence. This because they have made the firm large or because they know/think they have to overcome more obstacles than manager of smaller firms. Thus the negative relation between firm size and ∆ Acquisition Performance that I find is in line with the overconfidence theory.
5. Conclusion
28
declining returns. Something which has been shown in earlier literature multiple times (Fuller et al. 2005, Conn et al 2004a,b, Ismail 2008, Billet & Qian, 2008).
Further, both the univariate and multivariate results indicate that acquirers can be grouped into two groups: good (skilled) and bad (unskilled) acquirers. Acquirers that generate positive CARs at the start of their acquisition sequence show to be successful in subsequent deals as well. Unsuccessful first deals on their turn are likely to be followed by negative announcement returns in the deals thereafter. This is similar to what Croci & Petmezas (2009) found in their research. They call this superior managerial acquisition skill. Firms that have success at the first acquisition do however see a decline in their performance in subsequent deals. As Ismail argues (2008): Success at the first deal can cause managers to make worse subsequent acquisitions because they become overconfident. Unsuccessful first time acquirers see a increase in their performance. Ismail (2008) sees this as evidence for the learning theory. My findings are also in line with the hubris and learning theory of Aktas et al. (2009) who argues that hubris causes decline, but managers are able to learn too. By learning from mistakes managers are able to decrease their overconfidence.
Besides overconfidence Conn et al. (2004a) listed two other possible causes for declining returns: diminishing returns and mean reversion. Based on the patterns I find in the univariate results, I reject both possibilities. Diminishing returns expects steady decline, where I find increasing returns from the second to the third acquisition. Next to that the percentage of acquirers that generate negative returns on their first acquisition is 40%. This makes a theory that assumes that acquirers make rational decisions unlikely. The Mean Reversion Theory expects that acquirers are not able to keep up their good performance of the first acquisition. I find however that the difference between ‘good’ and ‘bad’ acquirers remains significant later on in the sequence. Thus good acquirers keep performing above the mean whereas bad acquirers keep performing below the mean.
29
discuss what can be concluded from the multivariate results interpreting them separately and thereafter what I conclude taking all the results together.
The performance in the previous acquisition is positively related to performance in the focal acquisition. Thus success follows success. This seems to contradict with the overconfidence theory. Croci & Petmezas (2009) who found similar results also draw this conclusion, because overconfidence theory expects hubris infected managers to perform less than ‘rational’ managers.
I argue however that these findings do not automatically have to rule out overconfidence. Several researchers have tried to explain the declining returns for frequent acquirers. To do so they tested how overconfidence influence acquisition performance. But as far as I know how overconfidence influences decline. When I test how recent performance influences the change in performance (decline or increase, measuring the difference between two deals) I find that acquirers with higher announcement returns (higher overconfidence) face larger decline towards the next acquisition.
Looking solely at the results with recent performance as a overconfidence proxy I can explain the declining returns as follows: Some managers have superior skills. These managers will generate positive announcement returns at their first acquisition. This group of managers is prone to overconfidence that is inflicted by self-bias. Managers that generate positive results attribute this to their own skills to much. More success inflicts more overconfidence. More overconfidence results in larger decline. However the skills are not completely eliminated by the overconfidence. Therefore the skilled acquirers will perform less compared to earlier acquisition as a result of hubris, but still generate positive returns. Unskilled acquirers increase their performance, which indicates learning, but they will not learn enough to generate positive returns. Thus unskilled acquirers will never catch up. This explains the positive relation between recent performance and the announcement returns of the focal acquisition that I have found.
30
On the other hand TBD might turn out not be a good proxy for overconfidence. Not for the specific sample I used or maybe not in general even. It might be that, because my study focused on frequent acquirers, there was too little difference in the TBD’s in the used sample. The measure TBD may reveal overconfidence when comparing frequent acquirers to less frequent acquirers. But in that case: is more frequent acquiring a result of overconfidence or does it measure overconfidence when frequent acquirers are actually performing less than less frequent acquirers for another reason?
Acquisition frequency may be a good measure for existing overconfidence (over-optimism) but not for growing overconfidence (self-bias).
The main research question of this paper is: Does managerial overconfidence cause declining returns for frequent acquirers? Overall I find the explanation that good acquirers are influenced by self-bias inflicted overconfidence resulting in declining returns the most plausible. All results except the regressions with TBD point to this. With that I draw the same conclusion as Doukas & Petmezas (2007) and Billet & Qian (2008) who also point to self-bias as the cause of declining returns. I thus answer the research question with a ‘yes’, but with a note that I assume that TBD is not a good overconfidence proxy. At least in this particular research, and maybe in general. TBD may be a better measurement tool for the indigestion theory as Conn et al. (2004a) also pointed out. Adjusting the conceptual model (see figure 1) based on my finding results in the model that is depicted below.
+
+ -
Figure 2: Recent Performance
Recent performance is a good indicator for the performance of the focal acquisition. Good acquirers keep having positive returns and bad acquirers keep having negative returns. However overconfidence causes good performers to have less positive returns in subsequent deals. Next to that I find that both good and bad acquirers can learn from their mistakes.
31
Good acquirers can learn to be less overconfident. Bad acquirers can learn to do less bad in subsequent deals.
In the regressions I controlled for different variables. This provided evidence that after a successful acquisition larger firms show larger declines to their next acquisition. This could be because managers of large firms are more prone to overconfidence. Moeller et al. (2004) argue that managers of large firms have to overcome more obstacles resulting in more overconfidence once they managed to do so. Otherwise the control variables showed no significant effect on performance.
6. Limitations & implications for future research
In research it is always important to explicate what are the limitations of a study. My focus was on frequent acquirers. I choose to use a sample that included only very frequent acquirers. It cannot automatically concluded that what was found for frequent acquirers also counts for multiple acquirers that acquirer on a less frequent base. Further I have only used announcement returns to measure acquisition performance. Zollo & Meier (2008) plead to always accompany short term measures with long term measure, because they argue that announcement returns may not accurately report acquisition performance because it relies on market sentiment. A long term measure in this study was however not very useful because the acquisitions were too close to each other that they would infect the measurement of each other.
My results indicate that learning influences acquisition performance and moderates overconfidence. I did not further test this in this paper. The mean reason for this is that I did not find a good measurement tool for which enough data could be gathered within a limited time frame and would fit the measurement tools and sample data I choose to use in order to test for overconfidence. This might be interesting for future research since learning has always been a large topic in M&A literature.
32
though. Doing this research again using other proxies and data, for example on a quality rather than a quantity base, could support or contradict my findings.
33
References
Aktas, N., de Bodt, E. & Roll, R. 2009. Learning, hubris and corporate serial acquisitions. Journal of Corporate
Finance, 15(5): 543-561.
Asquith, P., Bruner, R. F., & Mullins Jr., D. W. 1983. The gains to bidding firms from merger. Journal Of Financial
Economics. 11(1-4):121-139.
Baker, R. & Limmack, R. 2001, U.K. takeovers and acquiring company wealth changes: The impact of
survivorship and other potential selection biases on post outcome performance. Working Paper. University of Sterling.
Billett, M. T., & Qian, Y. 2008. Are Overconfident CEOs Born or Made? Evidence of Self-Attribution Bias from Frequent Acquirers. Management Science, 54(6): 1037-1051.
Brettel, M, Kasch, M. & Mueller, A. 2008. CFO Overconfidence, Optimism, and Corportate Financing. Working Paper. Aachen University.
Brown, S.J. & Warner, J.B. 1980. Measuring Security Price Performance, Journal of Financial Economics, 8(3):205-258.
Brown, S J. & Warner, J.B., 1985. Using daily stock returns, the case of event studies. Journal of
Financial Economics 14(1): 3-31.
Camerer, C. & Lovallo, D. 1999. Overconfidence and excess entry: an experimental approach.
American Economic Review, 89(3):306–18.
Conn, R.L., Cosh, A., Guest, P.M., Hughes, A., 2004(a). Why must all good things come to an end? The performance of multiple acquirers. Working Paper, Miami University
Conn, R.L., Cosh, A., Guest, P.M., Hugues, A., 2004(b). Why must all good things come to an end? The
performance of multiple acquirers. in Academy of Management(eds.) Creating actionable knowledge: Academy of Management Best Paper Proceedings (64th), 6–11.
Croci, E. & Petmezas, D. 2009. Why do managers make serial acquisitions? An investigation of performance predictability in serial acquisitions. Working Paper, University of Milan-Bicocca
Doukas, J.A. & Petmezas, D. 2007. Acquisitions, overconfident managers and self-attribution bias. European
Financial Management, 13(3):531–577
Fuller, K., Netter J., & Stegemoller, M. 2002. What Do Returns to Acquiring Firms Tell Us? Evidence from Firms That Make Many Acquisitions. The Journal of Finance, 57(4): 1763-1793
Ismail, A. 2008. Which Acquirers Gain More, Single or Multiple? Recent Evidence from the USA Market. Global
Finance Journal, 19(1):72-84
Laamanen, T., & Keil, T. 2008. Performance of serial acquirers: Toward an acquisition program perspective.
Strategic Management Journal, 29(6): 663-672.
Haleblian, J., Kim, J. Y. J., & Rajagopalan, N. 2006. The influence of acquisition experience and performance on acquisition behavior: Evidence from the US commercial banking industry. Academy of Management Journal, 49(2): 357-370.
34
Hirshleifer, D. 2001. Investor psychology and asset pricing. Journal of Finance. 56(4):1533–97.
Langer, E. J. & Roth, J. 1975. Heads I win, tail it’s chance: the illusion of control as a function of the sequence of outcomes in a purely chance task. Journal of Personality and Social Psychology, 32(6):951–55.
Loderer, C. & Martin, K. 1990. Corporate Acquisitions by listed firms: The experience of a comprehensive sample. Financial Management. 19(4):17-33.
Malmendier, U. & Tate, G. 2005. CEO overconfidence and corporate investment. The Journal of Finance. 60(6):2661-2700.
Malmendier, U. & Tate, G. 2005b. Does overconfidence affect corporate investment? CEO overconfidence measures revisited. European Financial Management. 11(5):649–659.
Moeller, S. B., Schlingemann, F. P., & Stulz, R. M. 2004. Firm size and the gains from acquisitions. Journal of
Financial Economics, 73(2): 201-228.
Moeller, S. B., Schlingemann, F. P., & Stulz, R. M. 2005. Wealth Destruction on a Massive Scale? A Study of Acquiring-Firm Returns in the Recent Merger Wave. The Journal of Finance, 60(2):757-782
Moore, D. A. and Kim, T. G. 2003. Myopic social prediction and the solo comparison effect.
Journal of Personality and Social Psychology, 85(6): 1121–35.
Morck R., Shleifer, A. & Vishny, W. 1990. Do managerial objective drive bad acquisitions? The Journal of
Finance, 45(1):31-48.
Roll R. 1986. The hubris hypothesis of corporate takeovers, The Journal of Business, 59(2):197-216. Rovit, S., & Lemire, C. 2003. Your Best M&A Strategy. Harvard Business Review, 81(3):16-17.
Schipper, K. & Thompson, R., 1983. Evidence on the capitalized value of merger activity for acquiring firms. J. Financ. Econ. 11(1-4):85–120.
Svenson, O. 1981. Are we all less risky and more skillful than our fellow drivers? Acta Psychologica, 47(2):143– 48.
Weinstein, N. D. 1980. Unrealistic optimism about future life events. Journal of Personality and Social
Psychology, 39(5):806–820.
35
Appendix 1: Distribution of Acquisitions over the Sample Period
This table presents the number of acquisitions per year for each acquisition order.
Year
Acquisition Order
1st Acq. 2nd Acq 3rd Acq 4th Acq 5th Acq 6th + Acq All
36
Appendix 2: Correlation Table of Independent Variables
This table presents the correlations between the independent variables that are used in the regressions. FS DS F TS BO BP PDP TBD IS Firm Size (FS) 1 Deal Size (DS) 0,58 1 Foreign (F) 0,16 -0,01 1 Target Similarity (TS) -0,03 0,05 0,06 1
Bid Order (BO) 0,32 0,17 0,09 0,02 1
Bid Proportion (BP) 0,15 0,11 0,07 0,00 0,69 1
Previous Deal Performance (PDP) 0,01 0,03 0,01 0,01 -0,04 -0,01 1
