Why Do (Some) Mergers Fail?

Jinghua Yan*

First draft: February, 2006 This version: September 10, 2006

ABSTRACT

This paper presents a model that incorporates product market competition into the

standard neoclassical framework. The model explains why value-maximizing firms

conduct mergers that in effect undermine shareholder value. In a Cournot setting,

this paper demonstrates a prisoners’ dilemma for merging firms in a merger wave.

Empirically, this paper identifies a negative relation between the performance of

horizontal mergers and the clusteredness of horizontal M&A activities, which is

consistent with the theory. Moreover, this relation is independent of the method of

payment and increasing in acquirer’s managerial ownership, which rules out the

alternative interpretations from existing theories. Finally, this relation is weaker for

concentrated or durable-goods industries, which further supports the product market

competition explanation of this paper.

*The Wharton School of University of Pennsylvania, Email: jyan@wharton.upenn.edu. I thank Philip Bond, Gary Gorton, Wei Jiang, David Musto, Jozsef Molnar, Vinay Nair, Michael Roberts, Pavel Savor, Rob Stambaugh, Geoff Tate, Ding Wu, Julie Wulf, Jianfeng Yu, Paul Zurek, and seminar participants at Wharton for helpful discussions. I am especially grateful to my committee members, Franklin Allen, Chris Géczy, Itay Goldstein, João Gomes, and Andrew Metrick, for their comments, guidance, and encouragement. All errors are my own.

1

1 Introduction

Why do some types of mergers lower shareholder value?1 Agency theory attributesthe failure of mergers to principal-agent problem. Market timing theory attributesthe negative post-announcement stock performance to overdue correction of mispric-ing. This paper proposes an explanation by incorporating the role of product marketcompetition into the standard neoclassical framework in the absence of agency costand market ine¢ ciency. In a neoclassical setting where mergers facilitate technologytransfer between �rms, mergers that take place outside merger waves (hereafter, o¤-the-wave mergers) increase shareholder value due to value maximization. However,if such horizontal mergers take place in a wave that is driven by technology shocks,2

the improved technology of merging �rms and an increasingly concentrated marketstructure alters the competitive landscape for non-merging rival �rms. When com-plementarity of production strength between acquirers and targets is su¢ ciently high,the standalone �rms in an on-going merger wave may face a declining pro�t marginand a shrinking market share. The merger wave thus resembles a game of prisoners�dilemma: Each individual pair chooses to merge despite that their combined acts re-sult in a lower shareholder value relative to that prior to the merger wave. Therefore,mergers that take place in a merger wave (hereafter, on-the-wave mergers) may lowershareholder value.

On the empirical front, this paper shows that on-the-wave horizontal mergers un-derperform o¤-the-wave horizontal mergers, which is consistent with the model. Therelation between merger wave and post-merger performance is robust to a number ofperformance measures, industry classi�cations, and empirical approaches. Moreover,the underperformance of on-the-wave mergers is more pronounced in less concentratedor non-durable goods industries. And stand-alone industry rivals also underperformin horizontal merger waves. These �ndings further support the model.

While the empirical relation between performance and merger wave supports thispaper�s model, it can also be explained by several existing theories in the literature,including market timing theory and agency theory. The market timing theory, ex-empli�ed by Shleifer and Vishny (2003) and Rhodes-Kropf and Viswanathan (2004),suggests that the acquirer uses its relatively overvalued stock as currency to purchasethe target company�s stock. Such stock market driven mergers have poor long-runstock performance due to the correction of misvaluation. A central prediction of the

1See Jensen and Ruback (1983) and Andrade, Mitchell, and Sta¤ord (2001) for a survey of thisliterature.

2Several papers have found that merger waves are triggered by industry-level technology or dereg-ulation shocks, e.g., Mitchell and Mulherin (1996), Rhodes-Kropf et al (2005), and Harford (2005).

2

Theory \ Features ValueMaximizing

MarketEfficiency Merger Waves Post­merger Performance

Market Timing Theory  (SV(2003), RKV(2004)) Yes No Misvaluation

wavesMixed

(stock < cash)

Agency Theory  (Jensen (1986), GKR (2005)) No Yes Pre­emptive

wavesMixed (low < high

managerial ownership)

Neoclassical Theory  (JR(2002), MM(1996)) Yes Yes Fundamental

shocks Always non­negative

Neoclassical with Product MarketCompetition  (this paper)

Yes Yes Fundamentalshocks

Mixed(on­the­wave

< off­the­wave)

Empirical PredictionsAssumptions

market timing theory is that stock deal acquirers underperform cash deal acquirersin the long run.3 To examine the possibility of the market timing theory as an al-ternative explanation, I show in the data that the relation between performance andmerger wave is independent of the method of payment and increasing in the net salesof insider shares, which is not entirely consistent with the misvaluation explanationprovided by the market timing theory.

The agency theory of mergers, �rst proposed by Jensen (1986), suggests that value-destroying mergers are driven by the manager�s incentive to grow the �rm beyondits optimal size. More recently, Gorton, Kahl, and Rosen (2005) show that whenmanagers have private bene�t of control, fundamental shocks may trigger defensivemerger waves. One of the key predictions of agency theory is that low managerialownership in the acquirer �rm leads to poor post-merger performance.4 In the data, Ishow that the negative relation between performance and merger wave strengthens asmanagerial ownership increases, which fails to support agency theory as an alternativeexplanation.

The rest of this paper is organized as follows: Section 2 presents a neoclassicaltheory model with product market competition, which demonstrates the prisoners�dilemma faced by �rms in a merger wave. This leads to the main prediction ofthe paper: on-the-wave mergers may lower shareholder value. Section 3 tests thepredictions of the model and addresses a number of alternative interpretations of the

3See Loughran and Vijh (1997) and Rau and Vermaelen (1998) for evidence supporting thisprediction.

4See Lewellen, Loderer, and Rosenfeld (1985) for evidence supporting this prediction.

3

results from competing theories or hypotheses. Section 4 relates my �ndings to theexisting literature on mergers. Section 5 concludes.

2 The Model

As is stated in the introduction, the objective of the theory model is to demonstratethe theoretical possibility of a merger wave equilibrium where the value-maximizationprinciple is upheld and merging �rms� value is lowered. To do this, we use thesimplest possible model: a static Cournot equilibrium with one period of production.Each pair of �rms is faced with the decision whether or not to merge. The decisionto merge imposes an integration costs on the merging �rms, but also brings aboutproduction e¢ ciency gains thanks to complementarity of strength in production. Ina competitive Cournot setting, such production e¢ ciency gains may lower rival �rms�value regardless of their decision whether or not to merge. When the integrationcosts and the degree of complementarity are both su¢ ciently high, a merger wavemay leave each pair of �rms worse o¤ than under the status quo.

2.1 Model Setup

In this model, there are N �rms in the economy (N is even). Half of the �rms(N2) produce under high technology and the other half produce under low technology.

Let Ch (C l) denote the cost of production for high (low) technology �rms (Ch < C l).Each �rm is endowed with K0 units of capital stock. All �rms produce identicalgoods and strategically set the quantity of production in a Cournot setting. There isone period of production, during which a stand-alone �rm of type i produces qi = Ki

Ci

units of goods, where i = h; l.

Aggregate industry production is given by Q =NXi=1

qi. The market clearing

price for �rms�output is assumed to satisfy an inverse demand function of a constantelasticity form,

P = X Q�

where 1 is the price elasticity of demand. X describes the condition of the aggregate

economy and is deterministic.

At t = 0, high and low technology �rms form N2pairs, and each pair decides

whether or not to merge. If two �rms decide to merge, the combined �rm�s tech-nology, denoted as Chl, is 1

1

Ch+� 1

Cl

. � is interpreted as the fraction of low technology

4

�rm�s technology that is missing in the high technology �rm. For example, Adidashas higher overall productivity than Reebok, but Reebok has stronger distributionchannels in European and Asian markets than Adidas does. Or, America West hashigher productivity than US Airways does, but US Airways�strength in the Northeastwill complement America West�s strength in the West. This notion of technologysynergy is closest to Rhodes-Kropf and Robinson (2004). If � = 0, then we have thespecial case that Chl = Ch, as in Jovanovic and Rousseau (2002). That is, whenthe degree of complementarity is zero, low type �rm�s technology will be completelyreplaced by the high type �rm�s technology. If two �rms remain standalone, theircosts of production will remain unchanged, respectively.

We rule out "mergers of likes", i.e., mergers of two �rms of the same type, forpure tractability reasons. One can relax this assumption. The key intuitions andresults will remain the same.5 Moreover, in the special case C l = Ch, i.e., all �rmsare identical at t = 0, this assumption becomes irrelevant. In addition, we rule outone �rm matching with multiple �rms, because it takes a considerable amount of timefor the acquirer to conduct a thorough due diligence on the target.6

Let Ai denote the decision of pair i, where i 2 f1; 2; :::; N2 g. Ai = 1 if two �rmsmerge and Ai = 0 otherwise. Let x denote the number of pairs that decide to merge

at t = 0, i.e., x =PN

2i=1Ai. It then follows that among the remaining N � x �rms in

the economy, x �rms produce under Chl, N2� x �rms produce under Ch, and N

2� x

�rms produce under C l. I assume there is no entry in this model.

At t = 1, each �rm adjusts their capital stock level Ki such that marginal utilityof each additional unit of capital stock, e.g., dRi

dKi = k, where Ri denotes revenue of�rm i and k is the cost of capital stock supplied by an external market with perfectlyinelastic supply. Firm i�s pro�t is given by

�i(Ki; P ) = PKi

Ci� kKi

The �rm�s revenue depends on its capital stock directly, because it uses the capitalstock to produce the revenue generating good, and indirectly, because the price of theindustry good depends, partly, on the �rm�s production.

d�i

dKi=P

Ci+Ki

CidP

dKi� k = 0

5The empirical evidences on "mergers of likes" are mixed: Servaes (1991) �nds that mergers ofhigh M/B and low M/B have higher total returns. In contrast, Rhodes-Kropf and Robinson (2005)�nd that mergers typically pair together �rms with similar M/B ratios.

6Empirically, it is uncommon for one �rm to make multiple acquisitions within the same year.

5

Di¤erentiating the inverse demand function P = X Q� with respect to Ki gives,

dP

dKi=dP

dQ

dQ

dKi= � P

CiQ;

and substituting into the previous equation together with qi = Ki

Ciyields,

k = (1� qi

Q)P

Ci

which indicates that the �rm internalizes the price impact in proportion to its marketshare, q

i

Q. Since �rms�marginal valuations of capital equate, (1� qi

Q) 1Ci= (1� qj

Q) 1Cj,

for any j 2 f1; 2; :::; N � xg. Summing over �rms yields,

qi

Q=�C(x)� (1�

N�x)Ci

�C(x)(1)

subject to constraint �1�

N�x�Ci

�C(x)< 1 (2)

where�C(x) =

(N2� x)Ch + (N

2� x)C l + xChl

N � x�C denotes the equal-weighted industry average capital requirement per unit of pro-duction. We can rewrite the equation marginal value of capital equating marginal

cost of capital as k = (1�

N�x�C)P . Hence,

P (x) = k�C

1� N�x

(3)

andQ(x) = XP (x)�

1

For i 2 fh; l; hlg, we have

qi(x) = Q(x)�C � (1�

N�x)Ci

�C(x)

Ki(x) = Ciqi(x)

Ri(x) = P (x)qi(x)

�i (x) = Ri(x)� k(Ki �K0)

6

where k(Ki�K0) is the cost of capital stock investment (or proceeds from disinvest-ment, if Ki < K0) from t = 0 to t = 1. Since the external capital stock market isfrictionless in this model, i.e., there is no capital adjustment cost, initial capital stocklevel K0 becomes irrelevant to �rm value. Without loss of generality, we assumeK0 = 0 from now on.

Each pair of �rms maximize total cash �ows, revenue net of capital adjustmentcost, so the value of each pair of �rms is given by

V (x) = maxAi(x)

f�(x)g

Optimal strategy for each pair of high-low technology �rms is characterized asfollows: A�i (x) = 1 if and only if

�hl(x+ 1)� I > �h(x) + �l(x) (4)

where x 2 f0; 1; :::; N2�1g. I is the integration costs (or �xed cost savings, if I < 0).7

In this model, economic synergies and costs of a merger come from three distinctsources: (a) the combined �rm adopts high type �rm�s productivity; (b) the com-bined �rm�s productivity may be higher than high type �rm�s productivity, due tocomplementarity of strength in productivity; (c) the integration costs I.

The externalities of mergers are two folded: (a) a merger reduces the number ofcompetitors by one, hence moving the economy closer to monopoly; (b) a mergerincreases the average productivity of competitors in the economy. The former e¤ectleads to positive externality, whereas the latter leads to negative externality.

De�nition 1 Merger wave equilbrium: x� = N2is a pure-strategy Nash equilibrium if

A(N2� 1) = 1, where A�i (x) is given by (4).

Conventional event studies draw conclusions on the value change by comparingthe post-merger value, i.e., �hl(N

2)�I, with the status quo value, �h(0)+�l(0), hence

the following de�nition on the types of merger waves:

De�nition 2 A merger wave is value-creating (value-destroying) if �hl(N2) � I >

�h(0) + �l(0) (�hl(N2)� I < �h(0) + �l(0)).

7One can also assume integration costs to be proportional. The results remain unchanged.

7

2.2 Model Solution

Similar to several other theories of mergers that involve product market competition,8

this paper uses backward induction to solve the model. Production and pro�ts aredetermined by Cournot production game among the remaining �rms in the industry.Continuing from the previous section, the solution of the Cournot economy at t = 1is standard. From (3), we have

P (x) = k�C(x)

1� N�x

Inverse of demand function gives

Q(x) = XP (x)�1 = X(k

�C(x)

1� N�x

)�1

The market share of each �rm is given by (1). Hence,

qi(x) =�C(x)� (1�

N�x)Ci

�C(x)Q(x) =

�C(x)� (1� N�x)C

i

�C(x)X(k

�C(x)

1� N�x

)�1

Finally, the pro�t function of each �rm is derived as follows,

�i(x) = P (x)qi(x)� k(Ki �K0)

= X �C(x)1�1 k1�

1

�1�

N � x

� 1 �1

| {z }aggregate economic condition

1

1�

(1� N�x)C

i

�C(x)

!| {z }

market share

0BBB@1� (1�

N�x)Ci

�C(x)| {z }average pro�t margin

1CCCAwhere i = h; l; and hl.

To simplify notations, we normalize Ch to 1 and let l = Cl

Ch, � = Chl

Ch; (then �

becomes l(1�� 1)), C = Ch; and �(x) = (N�x� )

((N2�x)(1+l)+x�) : The pro�t function simpli�es

to�h = X

1

C1�

1 �(x)

1 �1k1�

1 (1� �(x))2

�l = X1

C1�

1 �(x)

1 �1k1�

1 (1� l�(x))2

8See, for example, McCardle and Viswanathan (1994), Molnar (2006), Salant, Switzer, andReynolds (1983), and Gowrisankaran (1999).

8

�hl = X1

C1�

1 �(x)

1 �1k1�

1 (1� ��(x))2

subject to the following regularity conditions,9

8x 2 [0; N2� 1] : �(x)l < 1 (5)

A �rm�s pro�t function is the product of three terms: condition of aggregateeconomy, the �rm�s market share, and the �rm�s pro�t margin. The latter two termsboth depend on the �rm�s cost of production, Ci; relative to average cost of productionin the economy, �C, and the number of remaining �rms in the economy, N�x. Whenthe number of mergers increases in the economy, there are fewer �rms remaining andthe economy moves closer to the monopoly, thus increasing rival �rms�market shareand pro�t margin. However, when the degree of complementarity between merging�rms is su¢ ciently high, the decrease in number of competitors (N�x) does not fullycompensate for the increase in average competitiveness ( �C). In this case, a mergerwill lower the standalone value of rival �rms.

As in the standard neoclassical framework, two �rms�incentive to merge are in-creasing in the di¤erence in technology (l), increasing in the degree of complementarity(�), and decreasing in the integration costs (I). Besides economic reasons, this modelalso features a strategic incentive to merge: the decision for each pair of �rms can bedependent on their rivals�decisions.

2.3 Existence of Value-Destroying Merger Waves

Depending on the value of a few key parameters, a merger wave equilibrium mayincrease or decrease shareholders�value under the status quo. This section startswith the special case � = 0 as in Jovanovic and Rousseau (2002) to examine conditionsfor value-creating and value-destroying merger waves.

Proposition 1 Merger wave equilibrium�x� = N

2

�payo¤ dominates the status quo

(x = 0), e.g., �hl(N2) � I > �h(0) + �l(0), if (i) � = 1 (or � = 0) and (ii) regularity

conditions (5) hold.

The Appendix provides a detailed proof for this proposition. Proposition 1 showsthat under constant returns to scale and the standard neoclassical assumptions on

9In the case where �(x)l > 1 for some x, low technology �rms would become unpro�table andwill be forced to exit. This set of conditions simpli�es the proof of the propositions and is by nomeans crucial.

9

technological advancement in mergers, a merger can only have positive externalitieson its rivals, i.e., the standalone values of the rivals improve as the number of mergersin the industry increases. For rival �rms, the gain from having fewer competitors(lower N�x) always outweighs the loss from facing (on average) more technologicallyadvanced competitors (lower �C). If the merger wave is an equilibrium, then each pairof �rms must be better o¤ than standing still conditional on all other pairs merging.By transitivity, merger wave equilibrium must pay-o¤ dominate the status quo. Thisstatement holds for all value of l. The following proposition demonstrates that non-zero complementarity of productive strengths is not only necessary but also su¢ cientfor a merger wave equilibrium to be payo¤ dominated by the status quo.

Proposition 2 There exists a merger wave equilibrium�x� = N

2

�that is payo¤ dom-

inated by the status quo, e.g., �hl(N2) � I < �h(0) + �l(0) if (i) � > �, (ii) I 2�

�hl(N2)� �h(0)� �l(0); �hl(N

2)� �h(N

2� 1)� �l(N

2� 1)

�, and (iii) regularity con-

ditions (5) hold, where � is given by l�

N� (N2 � )(1+l)

� 1�.

The Appendix provides a detailed proof for this proposition. When the degreeof complementarity between two �rms is su¢ ciently high, e.g., � > �, mergers bringabout negative externalities on competitors. For rivals, the gains from having fewercompetitors (lower N � x) can be outweighed by facing tougher competitions in theproduct market (lower �C). Therefore, in a merger-wave equilibrium, every pair of�rms may be worse o¤ than under the status quo, even though their strategy isprivately value-maximizing conditional on other pairs�actions. Such a merger waveis labeled as value-destroying by conventional event studies.

2.4 Comparative Statics

We have shown that a merger wave may create or destroy shareholder value depend-ing on the value of �. This section examines the comparative statics on �, theminimum threshold for a merger wave to be value-destroying, with regard to tworelevant industry characteristics.

Corollary 3 �= l

�N�

(N2 � )(1+l)� 1�is decreasing in N . Moreover, as N ! 1,

�! 0.

Corollary 3 states that the minimum degree of complementarity required for amerger wave to destroy private value (�) is decreasing in the number of �rms operating

10

in the industry (N). The intuition is as follows: The externalities of a horizontalmerger are two-fold. On the one hand, a merger improves rivals� value due tohigher oligopoly rents. On the other hand, it lowers rivals�value due to toughercompetition. Since the former e¤ect is decreasing in number of companies, thethreshold of technological synergy (�) for the latter e¤ect to dominate must also bedecreasing in the number of companies (N). Therefore, the model predicts that inconcentrated industry (low N), � is likely to be high and merger waves are less likelyto destroy value.

Corollary 4 �= l�

N� (N2 � )(1+l)

� 1�is increasing in .

Corollary 4 states that �, the minimum degree of complementarity required for amerger wave to destroy private value, is decreasing in the price elasticity of demand ofthe industry ( 1

). The intuition is that as price elasticity of demand increases, prod-

uct market competition toughens and the competition e¤ect will be more pronouncedeverything else equal. Therefore, the minimum degree of complementarity for amerger to have destructive impact on rival �rms�value is lower for highly competitiveindustries (high 1

). Klemperer (1987a, b) shows that the noncooperative equilib-

rium in oligopoly with consumer switching costs resembles the collusive outcome inan otherwise identical market without consumer switching costs. The implicationis that the switching costs in e¤ect decrease the price elasticity of demand in theindustry. Therefore, the corollary above implies that � is higher for industries withhigh consumer switching costs, ceteris paribus.10 Hence, the model predicts that inindustries with high switching costs, merger waves are less likely to destroy value.

2.5 On-the-wave and O¤-the-wave Mergers

In the baseline model, the degree of complementarity (�) and integration costs (I)are identical across �rms. I can relax this assumption by describing them using thefollowing equations:

�it = � + "�;t

I i = I + "iI

10Klemperer (1987a, b) also argue that in addition to the tacit collusion, consumer switching costsalso lead to ferocious competition for market share before consumers attach themselves to suppliers.Since this paper studies the change in �rm value following merger events, the pre-merger �rm valuealready takes this pre-attachment competition e¤ect into account.

11

where "�;t denotes the time-variant industry shock on synergy and "iI denotes theidiosyncratic variation in integration costs for the ith pair of �rms. The character-ization on synergy is the same in spirit to the standard neoclassical framework a laJovanovic and Rousseau (2002).11 Under the simple characterization of the degreeof complementarity and integration costs, we can describe the merger activities in anindustry as follows:

Under normal economic conditions, "�;t � 0. An average pair of �rms in theeconomy will not merge regardless of other �rms�actions, because ��(x) � I � 0for all x. In this case, only pairs with extremely low realizations of integrationcosts ("iI � 0) will merge.12 The proposed merger (rejected by FTC) betweenO¢ ceDepot and Staples is such an example.13 Participants of these o¤-the-wavemergers are always better o¤ than the sum of their standalone values under thestatus quo due to value maximization. This prediction has been veri�ed by existingliterature that horizontal mergers on average create values for shareholders (see, forexample, Mitchell and Mulherin (1996)). When there is an industry-wide shockon synergies ("�;t > 0) driven by a technology innovation or deregulation such that��(x) � I > 0 for all x, a merger wave will be the unique equilibrium. Such anexample can be a technology breakthrough in online payment processing that leads tohigh production synergies between a conventional bookstore and an internet retailer.For another example, a deregulation that allows telecommunication companies tooperate across di¤erent states may lead to an increase in synergies between two phonecompanies originally operating in di¤erent states. In both examples, the integrationcosts to realize these production synergies may be rather high. Proposition 2 showsthat when � and I are both su¢ ciently high, such a merger wave may leave each pairof �rms worse o¤ than under the status quo.

Therefore, the central prediction of the model is that merger waves may lowershareholder value but o¤-the-wave mergers create shareholder value. Moreover, in avalue-destroying merger wave, some matched pairs have high positive idiosyncrasiesin integration costs ("iI), thus remaining stand-alone after merger waves. Thesestand-alone �rms also absorb negative externalities brought about by value-destroyingmerger waves of their rivals. Therefore, the model also predicts that rival �rms�valuemay also fall following horizontal merger waves.

11We may also allow integration costs to vary over time, but the main implications are the same aslong as there is more time variation in synergy than in integration costs. This is not an unreasonableassumption.12The costs of both physical integration (such as computer systems) and metaphysical integration

(such as �rm culture) are idiosyncratic in nature.13In this paper�s data sample, there are no other horizontal mergers in the same 4-digit SIC code

over the event window of this merger.

12

Finally, the model also has two cross-industry predictions: Corollary 3 predictsthat industries with higher concentration (lower N) are less likely to have value-destroying merger waves; Corollary 4 predicts that industries with higher switchingcosts (higher ) are less likely to have value-destroying merger waves.

2.6 Product Market Price and Mergers

Although this is not the focus of this paper, the theory model generates predictionson the relation between mergers and product market price. A horizontal mergergenerates two countervailing e¤ects on product market prices, a market power e¤ectand a productive e¢ ciency e¤ect. When productive synergies are low, the marketpower e¤ect dominates and price rises. When productive synergies are high, theproductive e¢ ciency e¤ect prevails and product market price falls. The empiricalevidence on the impact of horizontal mergers on product market pricing is mixed butconsistent with the theory model. Using a sample of airline mergers from 1985 to1988, Kim and Singal (1993) showed that merging �rms raised airline tickets by 9%relative to the routes una¤ected by the mergers.14 In the sample used in this paper,the M&A activities in transportation industry, i.e., Fama-French industry 40, are lowduring Kim and Singal�s sample period relative to other periods, e.g., only 38 outof the total 323 mergers were announced during this 4-year window. Therefore, theincrease in product price is consistent with this paper�s theory prediction that low-synergy, o¤-the-wave mergers lead to higher product market price. More recently,Focarelli and Panetta (2003) argues that it takes time to realize productive synergies.Using a unique dataset of deposit rates of Italian banks, they show that deposit ratesfall in the short run but rise in the long run after mergers, i.e., the market powere¤ect dominates in the short run, but the productive e¢ ciency e¤ect dominates inthe long run. Thus, the competition e¤ect identi�ed in this paper is also present indata outside our sample.

3 Empirical Method and Results

3.1 Data and Methods

From Thomson Financial�s Securities Data Corporation (SDC), I obtain all domesticcompleted mergers or tender-o¤er bids from 1979 to 2004. I exclude all repurchases

14Prager and Hannan (1998) �nd that large horizontal mergers of banks substantially reducedeposit rates, but �nancial companies are excluded in the sample due to heavy regulation.

13

and leverage buyouts.15 I assign each acquirer and target to one of the Fama-French48 industry groups based on their SIC codes recorded by SDC at the time of theannouncement. If the acquirer and the target are in the same industry group, thenthe merger is identi�ed as horizontal.

For each horizontal merger, I use the number of contemporaneous horizontal merg-ers in the same industry, as a measure of "clusteredness," or concentration of mergerannouncements. That is, I sum the number of horizontal mergers in the same in-dustry announced during the announcement month, the previous 3 months, and thefollowing 3 months. I then normalize this number by the total number of mergersannounced in that industry over the entire sample period. This adjusted number ofcontemporaneous horizontal mergers will be the measure of clusteredness.

Harford (2005) calculates the highest 24-month concentration of mergers for eachindustry. He identi�es merger waves if the actual 24-month peak concentration ex-ceeds the 95th percentile of the simulated distribution. The measure of clusterednessused in this paper is di¤erent in two crucial ways due to di¤erent theoretic motiva-tions: I only focus on horizontal mergers because two �rms in the same industry aremore likely to have complementarity of strength in production. In addition, thismeasure of clusteredness is continuous because externalities from rivals�mergers arecontinuous in nature.1617

I choose the window [t-3, t+3] month because in this paper�s theory model �rmsmake their merger (and ensuing investment) decisions simultaneously, e.g., during thesame year. The results to be shown are robust to using a shorter or longer window,e.g., [t-1, t+1], [t-2, t+2], [t-6, t+6], and [t-12, t+12]. I use the total number ofmergers within each industry over the entire sample period similar to Harford (2005).This is to tease out variability of merger activities across di¤erent industries.18

[Insert Table 1a, Table 1b]

Table 1a lists the performance measure variables I use in this paper. Summary

15I do not exclude deals for which less than 100% of the target shares are acquired. To avoiddoublecounting of multiple announcements of the same merger, I keep one observation per calendaryear for each unique pair of acquirer and target. In the �nal sample, deals in which 100% of thetarget share was acquired account for over 90% of the observations.16Rosen (forthcoming) used number of mergers over the last three years as a measure of wave. I

will relate my �ndings to his in the tests.17Harford (2005) only includes deals greater than $50 million in value. Including smaller deals

reduces noise in the clusteredness measure constructed in this paper. The results to be shown arerobust to di¤erent thresholds for deal value.18The baseline results are robust to using number of companies from COMPUSTAT tape in the

given industry as the denominator. However, the number of companies is complicated by industryconcentration as well as the population of public companies in the industry.

14

statistics of the key characteristics are shown in Table 1b. I examine the stock per-formance from the day before the announcement to 365 days after the announcement.In addition, I use the change in return on assets (ROA) as the operating performancemeasure.19 In some tables, I also report the stock return and the change in operatingperformance over two years. The choice of horizon balances two o¤setting concerns.On the one hand, short-horizon event studies fail to take into account the impactfrom the horizontal mergers that are announced subsequently. On the other hand,long-horizon performance may be driven by events unrelated to the merger wave mea-sured by num3adj. The magnitude of the acquirer and the target�s announcementreturn over a short event window, e.g., 10 days, in the data sample is close to theresults reported in other recent large-sample studies, such as Andrade, Mitchell, andSta¤ord (2001), Rosen (forthcoming) and Moeller, Schlingemann, and Stulz (2004).In the entire sample, the average medium/long-run abnormal return of the acquireris 4.5% over one year and 6.2% over two years. Neither is statistically distinguish-able from zero. Same is true for operating performance measures. If these mergerannouncements are evenly distributed over the sample period of 25 years, then themean of the clusteredness measure (num3adj) should be 7 months

(26�12) months = 2:2%. Inthe data, the mean of clusteredness measure (num3adj) is signi�cantly higher (3.8%).This is consistent with the well-established fact that mergers happen in waves.

3.2 Results and Discussions

3.2.1 Horizontal Merger Wave and Post-merger Performance

The main prediction of this paper�s theory model is that on-the-wave horizontal merg-ers underperform o¤-the-wave mergers.

[Insert Table 2]

In the univariate analysis, I sort all mergers20 where the acquirer is public by themeasure of clusteredness into quintiles. In Table 2, I report by quintile the stock andoperating performance measures and deal characteristics de�ned in Table 1. Thereis a strong decreasing trend in both stock and operating performance from the leastclustered (quintile 1) to the most clustered (quintile 5). The di¤erence between theleast clustered quintile and the most clustered quintile (1-5) is striking. On average,acquirer�s stock of the most clustered horizontal mergers underperform that of the

19Results remain qualitatively the same if operating income (DATA13) is used in place for earnings(DATA18).20Univariate results in Table 2 are robust to two equal-sized subperiods: 1979-1997 and 1998 to

2004.

15

least clustered by 21% over one year and by 54% over two years. The di¤erence forchange in ROA is 5%.

[Insert Table 3a & 3b]

To verify the �ndings in the univariate analysis, I regress stock performance mea-sures on the previously de�ned measure of within industry clusteredness (num3adj ),stock dummy, the interaction between stock dummy21 and industry-level merger clus-teredness, and the measure of overall clusteredness (num3alladj). The measure ofoverall clusteredness (num3alladj) is de�ned as the number of all mergers duringthe event window [t-3m, t+3m] normalized by 10�4. This measure is analogous tothe measure of hot/cold market in Rosen (forthcoming).22 In all regression modelsused in this paper, I control for acquirer�s book to market ratio and log of acquirer�smarket capitalization, both measured at the end of calendar year t-1. I also controlfor industry dummies and year dummies (unreported). Standard errors are robustto industry and year clustering.23 Coe¢ cients and t-statistics are shown in Table 3a.The measure of within industry clusteredness is negatively associated with stock andoperating performance after the merger. In model (2), for example, a 1% increase innum3adj leads to a 5.7% decrease in cumulative abnormal return for acquirer�s stockover 365 days.24 The negative relation does not change after we control for stockdummy (-5.63 vs. -5.64). The negative and signi�cant coe¢ cients on stock dummyare consistent with prior literature that documents underperformance of stock deals.25

Comparing model (3) with model (1) and (2), including stock dummy increases R2

by 0.01 whereas including the clusteredness measure increases R2 by 0.02. Moreover,the coe¢ cient on interaction term (stock_dummy � clusteredness) is indistinguish-able from zero.26 This indicates that the underperformance of on-the-wave mergersis independent of method of payment. In all but two speci�cations, the coe¢ cienton the overall clusteredness (num3alladj) is insigni�cant. This indicates that therelation between overall merger wave and post-merger performance is weak.27 Theresults also suggest that post-merger performance is driven by changes in industry

21Stock dummy, created by SDC, equals to 1 if over 50% of the consideration is paid in stock.22The main di¤erence is that Rosen (forthcoming) used the number of mergers announced during

the 3-year period prior to the merger announcement.23Results are robust to industry-month clustered errors.24Results (untabulated) using the number of mergers announced from [t-3m, t] as an instrument

variable for num3adj are largely the same.25See, for example, Loughran and Vijh (1997).26This result is consistent with Harford (2005).27This is consistent with the �ndings in Rosen (forthcoming), who used the number of mergers

as measure of waves and did not �nd signi�cant relationship between clusteredness and long-runperformance. This result is also consistent with the conclusion in Mitchell and Sta¤ord (2000) thatthere does not exist di¤erential post-completion performance between hot- and cold-market mergers.

16

brought by horizontal merger waves. In addition, acquirer�s size seems to be neg-atively associated with post-merger performance. This is consistent with Moeller,Schlingemann, and Stulz (2004)�s �ndings over a shorter event window. In models(7) and (8), I control for �rm �xed e¤ect to examine the subsample of acquirers thatmade multiple acquisitions over the sample period. The economic magnitude of thecoe¢ cients weakens, e.g., from -5.63 to -2.11 for acquirer�s CAR over one year, butthe statistical signi�cance remains despite considerable reduction in sample size.28

Table 3b repeats the baseline regressions in Table 3a on horizontal mergers de�nedusing 4-digit SIC codes. A �ner industry classi�cation allows us to better identifyhorizontal mergers, but it also reduces sample size by 40% and makes the clustered-ness measure noisier as there are fewer horizontal mergers at the 4-digit SIC level.Nonetheless, the baseline results still hold across all performance measures. It is in-teresting noting that the coe¢ cient on stock dummy is much weaker in Table 3b thanin Table 3a: stock deals underperform by 4.5% over one year in model (1) of Table3a and by only 1.7% in model (1) of Table 3b. In the latter case, the coe¢ cientbecomes statistically insigni�cant. This is consistent with market timing theory:under �ner industry classi�cation, same-industry �rms�valuations are more likely tomove in tandem and method of payment is likely to be indicative of misvaluation.

[Insert Table 4 - baseline regression 4-digit SIC]

In Table 4, I regress operating performance, i.e., change in acquirer�s ROA,29 onclusteredness measure controlling for acquirer characteristics. The coe¢ cient on theclusteredness measure remains signi�cant for operating performance. In the entiresample, a 1% increase in clusteredness measure leads to a 0.8% decrease in changein acquirer�s return on assets (ROA) over one year and 3.3% decrease in change inacquirer�s ROA over two years. Change in acquirer�s ROA may not be an accuratemeasure of post-merger operating performance as the correct benchmark for pre-merger ROA should be the weighted average of acquirer and target�s ROA prior to

28Untabulated results show that the baseline results remain unchanged if we control for (a) thetrailing one-year abnormal return of the acquirer, (b) the number of "dormant" days since the lastacquisition made by a �rm in the given industry, a la Song and Walkling (2004), and (c) acquirer�scash holdings, a la Harford (1999).29This paper measures operating performance by de�ating earnings by book value of assets. The

common concern that book value of assets include non-operating assets is less relevant in this setting,as taking the di¤erence in ROA teases out any mismeasurement. Moreover, the choice of M&Aaccounting method (pooling vs. purchasing) is unlikely be systematically di¤erent for on-the-wavevis-a-vis o¤-the-wave mergers. Other choices of de�ators used in the literature include market value,a la Healy, Palepu, and Ruback (1992), and sales, a la Heron and Lie (2002) and Kaplan (1989). AsBarber and Lyon (1996) point out, using sales ignore change in productivity of assets whereas usingmarket value ignores time variation in discount rate and growth prospects. The disadvantages ofboth measures are obvious in the setting of this paper.

17

the announcement, also commonly known as "pro forma" ROA. We calculate thisadjusted change in ROA for deals that we have accounting information on the target.This signi�cantly reduces the sample size, as most of the targets in the sample areprivate. The results from models (3) and (4) in Table 3 show that the relationbetween operating performance and merger clusteredness is not driven by the pooroperating performance of targets during merger waves: after taking into account thetarget�s pre-merger operating performance, a 1% increase in clusteredness leads to a0.9% (6.3%) reduction in post-merger operating performance over one (two) year(s).If I run the same regressions using 4-digit SIC code classi�cations, models (5) through(8) show that the results are qualitatively the same.

[Insert Table 5]

Fama (1998) and Mitchell and Sta¤ord (2000) argue that long-term buy-and-holdanalysis su¤ers from the bad model problem.30 It is possible that the underperfor-mance of clustered mergers is due to the poor performance of asset pricing modelduring the merger wave periods. Moreover, Schultz (2003) presents pseudo mar-ket timing theory that explains the discrepancy in post-IPO performance betweenthe event-time approach and the calendar-time approach. To address both of theseconcerns, I employ a calender-time rolling window technique that several papers advo-cated.31 At the end of each quarter, I sort the acquirers of all horizontal mergers an-nounced during the quarter into 3 quantiles by the degree of clusteredness (num3adj).I then form a long-short portfolio that buys acquirers�stocks from the least clusteredquantile and short sells those from the most clustered quantile. I hold this portfoliofor 12 months starting from 3 months after the quarter end. I skip a quarter betweenthe formation period and holding period to tease out any announcement e¤ect.32 Ifa stock disappears from the CRSP within 15 months of the announcement, value-weighted market returns are used to replace the missing monthly returns. For eachmonth, the rolling window long-short portfolio return is an average of the four value-or equal-weighted long-short portfolios of acquirers from the most recent four quar-ters. I then use the Fama French 4-factor model to estimate the abnormal return of30Loughran and Ritter (2000) counter-argues that calendar-time portfolio approach su¤ers from

low-power problems. If their argument holds and the calendar-time portfolio approach indeedreduces the power of rejecting the null hypothesis, it only increases the statistical signi�cance of theresults found in this paper.31See, for example, Brav, Geczy, and Gompers (1997), Fama (1998), and Mitchell and Sta¤ord

(2000).32This is not crucial to the �ndings. Results (unreported) are qualitatively similar without

skipping a quarter.

18

this rolling window portfolio:

Racquirers of least clustered; t �Racquirers of most clustered; t= �+ � [RMt �Rft] + sSMBt + hHMLt + uUMDt + "t

where Rft is the one-month Treasury bill rate, RMt is the monthly return on a value-weight market portfolio of NYSE, Amex, and NASDAQ stocks, SMB is the di¤erencebetween the returns on portfolios of small and big stocks (below or above the NYSEmedian), HML is the di¤erence between the returns on portfolios of high- and low-BE/ME stocks, and UMD (Up Minus Down) is the average return on the two highprior return portfolios minus the average return on the two low prior return port-folios. In Table 5, I report average (monthly) returns, Fama-French 4-factor alpha(monthly), and statistical signi�cance of the alpha for 3 samples: all cash deals, dealswith non-zero stock payment (the complement of all cash deals), and all deals. Ta-ble 5 shows results for two speci�cations: value-weighted and equal-weighted. Theequal-weighted (value-weighted) long-short portfolio generates a monthly return of1.2% (1.2%) and an � of 0.6% (0.8%). Both �0s are statistically signi�cant at 10%.Moreover, the two mutually exclusive subsamples (cash deals and stock deals) havethe same equal-weighted (value-weighted) raw monthly return of 1.1% (1.0%). Con-sistent with earlier evidence, cash deal acquirers on average outperform stock dealacquirers in this data sample. Nonetheless, the relation between underperformanceand clusteredness is a distinct trend independent of the method of payment.

It is important noting that even though the calendar-time approach rejects thenull hypothesis that there is no di¤erential performance between on-the-wave ando¤-the-wave mergers, these results by no means reject market e¢ ciency because informing long-short portfolios, ex post data is used to construct the clusterednessmeasure (num3adj). Therefore, this is not an implementable trading strategy.

3.2.2 Alternative Interpretations

The baseline results indicate that on-the-wave mergers underperform o¤-the-wavemergers. This result draws several alternative interpretations as other existing the-ories and hypotheses deliver prediction of the same stylized fact. In this section, Iexamine the possibility of each of these alternatives.

Hypothesis I: On-the-wave mergers underperform o¤-the-wave mergers because ac-quirers overpay for targets during merger waves.

[Insert Table 6]

It is well documented that acquirers often pay substantial premium above targets�

19

market price.33 Roll (1986) �rst proposes that hubris may lead acquirer managersto overpay for targets. More recently, Moeller, Schlingemann, and Stulz (2004) andMalmendier and Tate (2005) provide evidences in support of the hubris hypothesis. Isit possible that acquirers�stocks perform poorly because acquirer managers overpaidfor the targets during merger waves? To answer this question, we examine thecombined return of acquirer and target for the subsample of mergers in which targetsare public. We sort the subsample into quintiles by the clusteredness measure. InTable 6a, I report for each quintile acquirer and target�s combined returns, de�nedas a value-weighted portfolio of acquirer and target over 365 days.34 This equalsto the return of a hypothetical shareholder who owns both companies. If the stockunderperformance is due to overpaying for targets, the decreasing trend shown inTable 2 should weaken or disappear when using these combined measures. On thecontrary, the underperformance of the most clustered quintile remains economicallysigni�cant: acquirer and target�s CAPM alpha over 365 days is -9.8%. Moreover, themagnitude of underperformance remains as strong as in Table 2 across all performancemeasures.35 In addition, regression results shown in Table 6b con�rm that acquirerand target�s combined return decreases signi�cantly as the clusteredness measureincreases.36 And the results are largely the same if 4-digit SIC code classi�cation isused to identify horizontal mergers (see models (5) and (6)). Finally, two additionaltests (results untabulated) failed to support Hypothesis I: If acquirer CEOs are morecon�dent during merger waves, then we should expect the merger premium paid onthe target to be higher for highly clustered mergers. In the data, however, the relationbetween clusteredness and merger premium is negative and insigni�cant. Moreover,after controlling for manager �xed e¤ect,37 the relation between clusteredness andperformance remains signi�cant, which indicates that CEO characteristics such asovercon�dence, cannot explain away the di¤erence in performance between on-the-wave and o¤-the-wave mergers.38

Hypothesis II: On-the-wave mergers underperform o¤-the-wave mergers because

33See, for example, Andrade, Mitchell, and Sta¤ord (2001).34That is to hold a value-weighted portfolio of acquirer and target from day t-1 to day t+10 and to

hold acquirer�s stock from t+11 to t+365. This is similar to the measure for synergy gain suggestedby Bradley, Desai, and Kim (1988).35To rule out the possibility of sample selection bias, untabulated results show that the negative

relation between acquirer�s CAR and clusteredness for the overall sample (used in Table 2 and Table3a) remains unchanged for the subsample of deals with public targets (used in Table 6).36Another way to control for the premium paid on the target is to examine the acquirer�s long-run

return excluding the annnouncement event window [t-1d, t+10d]. Untabulated results show thatthe relationship between performance and clusteredness remain the same.37This requires narrowing the sample down to Compustat ExecuComp.38CEO overcon�dence a la Malmendier and Tate (2005) exhibits little time variation for the same

manager.

20

merger waves are preemptive driven by managerial private bene�t of control.

[Insert Table 7]

Gorton, Kahl, and Rosen (2005) suggest value-destroying merger waves may oc-cur when private bene�t control is su¢ ciently high. Is it possible that the value-destroying merger waves are defense-driven? To control for the severity of agencycost in each merger, I obtain the acquirer CEO�s percentage ownership from ThomsonFinancial Insider Trading data.39 The criteria of non-missing ownership reduces thesample by 40%. I stratify the remaining sample by percentage of ownership by theCEO. Table 7a shows the univariate results for the subsample where percentage own-ership is greater than 0.5%.40 The pattern of relative underperformance by clusteredmergers becomes even stronger than in Table 2.41 The di¤erence in one-year (two-year) stock performance rises from 21% to 35% (54% to 75%). This indicates thatthe underperformance of clustered mergers is exacerbated by high managerial owner-ship. In addition, I use regression analysis to examine the e¤ect of CEO ownership onthe relation between merger performance and clusteredness. The coe¢ cient on theclusteredness measure is negative and signi�cant for all four performance measures.The positive coe¢ cient on shares held by the CEO in model (1) and (2) implies thathigher acquirer managerial ownership leads to better acquisitions for shareholders.This is consistent with the �ndings in Lewellen, Loderer, and Rosenfeld (1985) andsupports the agency theory of mergers. However, the coe¢ cient on the interactionterm carries a negative sign (statistically signi�cant for both stock performance mea-sures), which is opposite to what Hypothesis II predicts. Therefore, even thoughthe regression results support the agency theory in general, they do not indicate thatagency theory explains the relative underperformance of clustered mergers as Gorton,Kahl, and Rosen (2005) suggest.

Hypothesis III: On-the-wave mergers underperform o¤-the-wave mergers becausemerger waves are triggered by over-valuation waves.

[Insert Table 8]

Shleifer and Vishny (2003) and Rhodes-Kropf and Viswanathan (2004) predict

39The remaining sample contains only CEOs who traded their shares during the event window.Although not prohibited by the SEC, insider sales around merger announcement are less commondue to �rms� self-imposed restrictions. This sample selection issue should not lead to any biasone way or the other. Indeed, untabulated results verify that the main deal characteristics arethe same for the subsample with non-missing CEO ownership data from the TFN. Alternatively, Ialso obtained CEO ownership from COMPUSTAT ExecutiveComp. The results (not reported) arequalitatively similar with weaker statistical signi�cance due to much smaller sample size.400.5% is the median of the distribution of percentage ownership in the sample.41It is also stronger than the results (untabulated) using subsample with non-missing CEO own-

ership data from TFN.

21

that mergers announced during overvaluation waves tend to have poor post-mergerperformance. This alternative hypothesis may potentially drive the baseline resultsin two distinct ways. The measure of clusteredness and the method of payment arepositively correlated in the merger sample: Table 2 shows that stock deals accountfor 26% of the most clustered quintile vis-a-vis 15% of the least clustered quintile ofhorizontal merger. More importantly, merger waves may be driven by market over-valuation in the industry. Therefore, even cash deals on-the-wave may be associatedwith poor ensuing stock performance.42 To address these two issues, I �rst stratifythe entire sample by method of payment and public status of the target and focus ondeals in which 100% of the payment is made in cash and the target is private. Thesemergers are not subject to the premise of market timing theory: acquirer�s overvalua-tion or target�s undervaluation.4344 To better prevent the clusteredness measure frombeing in�uenced by valuation waves, I use the concentration of all-cash deals insteadof that of all deals of to measure clusteredness. The clusteredness of all-cash deals isde�ned as the number of all-cash deals45 over the event window [t-3m, t+3m] dividedby the total number of all-cash deals over the entire sample period. The choiceof methodology here ensures that both the sample of mergers and the clusterednessmeasure are not related to stock market valuation. The two clusteredness measure,that of all deals and that of all-cash deals, have a correlation of 0.89 in the sample.This is consistent with Harford (2005), who �nds that cash deal activity and overallmerger activity are highly correlated at the industry level. The univariate resultssorted on the new clusteredness measure are shown in Table 8a. Consistent withthe previous �ndings,46 these all-cash and private-target deals on average have betterpost-announcement performance than the overall sample: The mean of CAR over 365days is 7.4 in this subsample compared to 4.4% in the overall sample. Moreover,

42Market timing theory predicts that mergers announced during stock market overvaluation wavesshould be stock deals. The argument here allows the possibility that such deals are paid in cash forother reasons.43If one assumes that private companies can be misvaluated and clustering is due to undervalua-

tion, the market timing theory a la Shleifer and Vishny (2003) implies a positive relation betweenclusteredness and performancce: A merger wave driven by undervaluation of the target stocks leadsto overperformance of the acquirer stocks. This theoretical possibility therefore predicts the oppo-site.44Harford (1999) suggests that cash-rich �rms more likely to make value-destroying acquisitions.

It is possible that mergers on the wave involve acquirers with better access to �nancing in capitalmarket. To rule out this alternative interpretation, I also exclude all acquirers who issued debt orequity from day t-365 to day t+365.45All-cash deals are union of the following two sets of mergers: (a) mergers where the acquirer is

public, the entire consideration is paid in the form of cash, and the acquirer did not issue stocks orbonds from one year prior to the merger announcement to one year afterwards; (b) mergers wherethe acquirer is private.46See, for example, Andrade, Mitchell, and Sta¤ord (2001).

22

acquirer�s stocks in all 5 quintiles have positive CAR over 365 days. The di¤erencein stock performance between the two extreme quintiles remains economically signif-icant. Acquirer�s CAR over 365 days decreases from 11.0% for the least clusteredquintile to 2.7% for the most clustered quintile. While the di¤erence for the sub-sample of cash deals (8.3%) is smaller than the di¤erence for the whole sample shownin Table 2 (15.8%), it stays economically and statistically signi�cant. This indicatesthat market timing theory cannot entirely explain away the trend documented inTable 2. The regression approach con�rms the �nding. Comparing the coe¢ cienton clusteredness of all deals in model (1) with the coe¢ cient on clusteredness of cashdeals in model (2), the relation between performance and clusteredness is weakerfor mergers with method of payment being cash, as is suggested by the univariateanalysis.47 Nonetheless, the negative relation remains strong: for each 1% increasein the clusteredness measure, acquirer�s stock performance over 365 days drops by3.7% (see model (2)). The relation is also economically and statistically signi�cantfor operating performance. These results indicate that while market timing theory issupported in the data, it can at best explain away a fraction of the relation betweenthe clusteredness and post-merger performance.

Since the clusteredness measure using cash deals and overall deals are highly cor-related (correlation of 0.89 using the Fama-French 48 industry classi�cation), is itpossible that the relation between clusteredness of cash deals and performance is sim-ply driven by misvaluation waves, an omitted variable in the regression? Or, to whatextent is the clusteredness of cash deals a proxy for the clusteredness of stock deals,hence the market misvaluation? To address this potential omitted variable problem,we create clusteredness measure using stock deals only,48 which proxies for misvalu-ation, and regress the acquirer�s stock performance on both clusteredness measures.The market timing theory predicts that only the high clusteredness of stock dealsleads to poor post-merger performance, whereas this paper�s theory model predictsthat both coe¢ cients have negative signs, because both measures proxy for the degreeof competition e¤ect. Table 8 Panel C shows the horse-race regression results. Inmodel (4), the clusteredness of both cash and stock deals have signi�cant and negativecoe¢ cients. They weaken each other�s explanatory power (compared to coe¢ cientsin models (2) and (3)) possibly because the two measures are highly correlated (cor-relation of 0.54). The results using 4-digit SIC does, as shown in models (5) through(8), indicate that the coe¢ cients on both clusteredness measures are signi�cant andnegative. Moreover, the two variables do not weaken each other�s explanatory power,as the correlation between the two clusteredness measures becomes much lower (0.02)

47An alternative interpretation of the weakened coe¢ cient is that the measure num3cashadj isnoisier than num3adj, because there are fewer cash deals.48The stock deals are the complement of the previously de�ned cash deals.

23

in this �ner industry classi�cation. Overall, the results seem to indicate that bothclusteredness measures have similar predictive powers on post-merger performance,which is not entirely consistent with the market timing theory.

[Insert Table 9 - regression controlling for insider net sales]

It is well established that insider trading can forecast returns.49 Firms whoseshares have been intensively sold (bought) by insiders tend to underperform (over-perform) benchmarks in subsequent periods. Following Jenter (2005), I use insidertrading as a window into managers�perception of their stocks being over- or under-valued. To further study the possibility of the market timing theory as an alternativeexplanation for the baseline results, I use data collected by Thomson Financial fromthe required SEC insider trading �lings. For each merger, I sum all (split-adjusted)open market transactions for all insiders50 over [t-365d, t-1d], with sales entering pos-itively and purchases entering negatively. I create the variable dummy of net insidersales that equals to one if the net sales of all insiders during the year prior to themerger announcement is positive and zero otherwise. Loosely speaking, insider salesdummy indicates whether or not the acquirer stock is perceived to be overvalued.To examine whether the relation between clusteredness and post-merger performanceis due to market timing, I regress the post-merger stock performance on dummy ofinsider sales, merger clusteredness, and the interaction term of the two. In models(1) and (2) in Table 9, the signi�cant and negative coe¢ cient on the dummy of netinsider sales is consistent with earlier �ndings that insider trades are informative. Ifthe market timing theory explains the baseline results, the negative relation betweenperformance and clusteredness should be more pronounced for overvalued acquirers,as indicated by insiders sales dummy. However, the coe¢ cients on the interactionterm are positive and signi�cant in both models. This indicates that the relation be-tween performance and clusteredness is not explained by the misvaluation of acquirerstock during merger waves. Jin (2002) suggests that managers may diversify awayindustry risk by shorting competitors�stocks. Therefore, a better performance mea-sure to examine the strong form of market e¢ ciency is the industry-adjusted stockreturn. Model (3) shows regression coe¢ cients using acquirer�s Fama-French 48 in-dustry adjusted stock returns over 365 days. The coe¢ cient on the interaction termremains positive and signi�cant. Finally, by using the 4-digit SIC industry classi�ca-tion, I reach the same conclusion (see models (4) and (5)). Therefore, evidence frominsider trading does not provide support for the hypothesis that the market timingtheory explains the baseline �ndings.

49See Seyhun (1998) for a comprehensive review of this literature and a discussion of SEC rules,�ling requirements, and available data.50Results are the same if (a) only transactions by o¢ cers and directors are included (b) the

absolute level of insider trades (normalized by number of shares outstanding) is used.

24

Hypothesis IV: On-the-wave mergers underperform o¤-the-wave mergers becausemerger waves are driven by negative industry-wide demand shocks.

[Insert Table 10 - horizontal clusteredness and diversifying clusteredness]

One last alternative interpretation is that merger waves are procyclical, whichimplies that horizontal merger waves occur when the industry starts declining. Thisalternative neoclassical hypothesis suggests that the negative relationship betweenpost-merger performance and clusteredness is due to industry-wide demand shocks,rather than the competitive e¤ects as this paper�s model suggests. To what extentare the baseline results driven by variation in industry condition vis-a-vis competi-tion among �rms? To empirically disentangle these two complementary e¤ects ischallenging, because industry-wide demand shocks are not directly observable.

Under Hypothesis IV, we should observe both diversifying mergers and horizon-tal mergers following a negative industry-wide demand shock. On the theory front,Gomes and Livdan (2004) and Maksimovic and Phillips (2002) both suggest that itis optimal for �rms to make diversifying mergers when there is a negative shock tothe incumbent industry. In spirit of their neoclassical theory predictions, I createadditional clusteredness measures using diversifying mergers to proxy for variation inindustry condition over time. The clusteredness measure for deals where the acquirer(target) is in the given industry is de�ned as the number of diversifying mergers dur-ing the event window [t-3m, t+3m] that involve �rms in the given industry as acquirer(target) divided by the total number of diversifying mergers over the sample period(1979-2004) that involve �rms in the given industry as acquirer (target).51 Broadlyspeaking, if the clusteredness of diversifying mergers is high, the industry condi-tion is poor. In Table 10, I regress post-merger performance on horizontal mergerclusteredness, diversifying merger clusteredness (acquirer), and diversifying mergerclusteredness (target), controlling for other variables. The coe¢ cient on horizontalmerger clusteredness is negative and signi�cant. This indicates that the clusterednessof diversifying mergers negatively predicts the post-merger performance of horizontalmergers.52 Moreover, after adding the two measures of cross-industry clusteredness,the coe¢ cients of horizontal merger clusteredness are signi�cantly reduced by 1/3 to1/2 in magnitude (from -5.6 in model (1) to -3.8 in model (2) and from -8.2 in model(3) to -4.3 in model (4)), suggesting that a fraction of the explanatory power of theclusteredness measure is attributable to industry condition variation. However, the

51All mergers that involve �nancial or utility �rms are excluded.52This result can also be interpreted using the market timing theory. I also used all cash deals

to de�ne both measures of clusteredness. The cash proxies are noisier because there are fewer cashdeals. But they are better proxies for industry econmic condition, because industry-overvaluation-driven mergers are excluded. Nonetheless, the results (untabulated) using cash proxies are largelythe same.

25

coe¢ cient on the horizontal merger clusteredness remains economically and statisti-cally signi�cant.53 In addition, the cross-industry merger clusteredness do not seemto predict the post-merger operating performance. Nor do they weaken the predictivepower of horizontal merger clusteredness.

The results indicate that (a) the clusteredness measures of cross-industry mergersexplain post-merger stock performance for horizontal mergers, probably because itproxies for change in industry condition as earlier theories suggested; and (b) aftercontrolling for this change in industry condition, the original clusteredness measurestill explains (albeit to a lesser degree) post-merger performance. To the extent thatdiversifying mergers may be driven by an increase in competitiveness within the in-cumbent industry due to horizontal mergers with high productive e¢ ciency gains, ourregression results are understating the e¤ect of horizontal clusteredness (num3adj).Therefore, it is conservative to conclude that 1

2to 2

3of the explanatory power of

merger clusteredness remains after controlling for variation in industry condition.

3.2.3 Post-merger Performance of Rival Firms

[Insert Table 11 - rivals performance]

The theory model also predicts that merger waves destroy value of industry rivals,whether they participated in the wave or not, thus leading to poor performance ofthe entire industry. To test this prediction, I examine the relation between clus-teredness and market and industry performance following the merger. Model (1) inTable 11 shows that value-weighted market index does not underperform followingclustered mergers relative to unclustered mergers, which indicates that market-widemisvaluation is unlikely the driving force for these horizontal merger waves. Model(2) shows that industry returns are lower following clustered mergers: A 1% increasein clusteredness measure decreases industry return over 365 days by 4.3%. Thisis consistent with both my model and the market timing theory. Nonetheless, ifindustry-speci�c misvaluation drives the horizontal merger waves, as Rhodes-Kropfand Viswanathan (2004) suggested, the industry returns should be lower followingstock deals than cash deals. However, model (2) in Table 11 does not support thatargument: The coe¢ cient on the stock dummy is negative but insigni�cant. Weobtain similar results by using abnormal returns over 730 days.

If both the acquirer and the entire industry lose value following merger waves,does the acquirer underperform relative to the industry? The answer, according tomodels (4) and (5), is yes. An increase of 1% in clusteredness leads to a 2.4% decreasein acquirer return relative to the industry. While this result is consistent with the

53Same results (untabulated) hold if we use 4-digit SIC industry classi�cation.

26

market timing theory, it can also be consistent with an extension of this paper�stheory model: First of all, the Fama-French 48 industry portfolio include companiesthat are not direct competitors of the merging �rms in the product market. Moreimportantly, the externalities of a horizontal merger wave are not evenly distributedacross all rival �rms. In the model extension discussed in the Appendix, �rms thatparticipated in mergers are the �rms who absorb the most negative externalitiesfrom rivals�mergers.54 Consistent with this extension, the magnitude of relativeunderperformance is further reduced when we match by size and book to marketratio a la Barber and Lyon (1997).55 In model (7) where we adjust by size and B/Mmatched-�rm�s return, acquirer�s buy and hold abnormal return has an insigni�cantloading on the clusteredness measure.56 Note that the coe¢ cient on the stock dummyremains signi�cant, which is consistent with the market timing theory.

This paper proposes a new motivation for �rm-match buy-and-hold abnormalreturn method a la Barber and Lyon (1997). Controlling for �rm-characteristics,such as size and book to market, is one way to provide a proxy for the standalonevalues conditional on other �rms�merger actions. Current empirical literature hasbeen analyzing value creation or destruction for each pair of �rms as if they are iso-lated. Instead of comparing post-merger value and pre-merger value conditional onrivals�actions, the conventional event studies have been examining the unconditionalchanges in value measured by stock and operating performance following the mergerannouncement.57 While this method is adequate for periods when few mergers hap-pen, it becomes problematic during merger waves. As this paper�s model suggests,during a horizontal merger wave, standalone value of two �rms conditional on ri-vals�merger actions can be rather di¤erent than the unconditional �rm value. Morespeci�cally, when a large number of productivity-enhancing mergers take place withinan industry, standalone �rms�value may decrease signi�cantly from their pre-wavelevel. Therefore, even though mergers appear to be value-destroying using conven-tional empirical methods, it is not necessary that the value-maximization principleshould be violated. Using the industry-size-B/M-matched control �rm method, we

54The results in models (4) and (5) are also consistent with the market timing theory. Under-performance of merging �rms relative to industry rivals may be due to the correction of �rm-levelovervaluation as Rhodes-Kropf and Viswanathan (2004) suggest.55Barber and Lyon (1997) suggest that test statistics using cumulative abnormal returns, i.e.,

using the market as the reference portfolio, are subject to new listing bias. In this case, the positivenew-listing bias is likely to be stronger for mergers on the wave, because empirically stock issuesand mergers are positively correlated. Therefore, correcting such a bias may only strengthen thenegative relation between clusteredness and post-merger performance.56Same results (untabulated) hold for horizontal mergers de�ned using 4-digit SIC industry clas-

si�cation.57See, for example, Loughran and Vijh (1997) and Rau and Vermaelen (1998).

27

fail to reject the hypothesis that mergers on the wave destroy value conditionallyrelative to o¤-the-wave mergers.58

3.2.4 Cross-industry Comparison

[Insert Table 12 - HHI]

In the baseline model, there is only one industry. In a world with multiple indus-tries with di¤erent characteristics (assumed to be exogenous), the model predicts that(ex ante) highly concentrated industries are less subject to value-destroying mergerwaves driven by competition. As Corollary 3 states, ceteris paribus, the synergythreshold for merger waves to be value-destroying is higher for concentrated indus-tries with fewer number of competitors. The Bureau of Census conducts quinquennialplant-level surveys on manufacturing �rms in the US, where I draw the number ofcompanies and industry concentration measured by HHI for each 4-digit SIC manu-facturing industry. For each horizontal merger, the most recent HHI reported priorto the announcement is used as the ex ante industry concentration measure.

To examine the e¤ect of industry concentration on the relation between clustered-ness and performance, I regress the stock and operating performance measures onthe clusteredness measure, concentration measure, and the interaction term betweenthe two. This paper�s model predicts a positive sign on the interaction term. Theresults in Table 12 indeed verify the prediction. Like in previous tests, the coe¢ -cient on merger clusteredness is signi�cant and negative across all four performancemeasures. The coe¢ cient on the interaction term is signi�cantly positive for all fourperformance measures. The results indicate that while merger waves on averagelower shareholder value, �rms in highly concentrated industries have relatively betterperformance following the merger waves than those in competitive industries.59

[Insert Table 13 - Relationship]

In the baseline model, �rms compete only on quantity. If there exists a relation-ship between �rms and customers, the e¤ect of competition is likely to be weakened.As Corollary 4 states, the value-destroying merger waves are more likely to occur in

58Nonetheless, there is still a signi�cant and negative relation between clusteredness measure andthe post-merger stock return over 2 years. This result is consistent with market timing theory,as one year may not be su¢ cient for the market to correct the mispricing of acquirers�stocks. Itcan still be consistent with this paper�s theory: Using industry-size-B/M match is a crude methodto proxy for the hypothetical standalone value of the acquirer. A more approapriate empiricalapproach is to examine the endogenous selection of acquiring �rms in a given industry using allrelevant �rm characteristics.59Results remain the same if I use the number of �rms instead of HHI as a measure for industry

concentration.

28

industries with little or no relationship. Following Cremers, Nair, and Peyer (2006),I use the 2 digit SIC to characterize industries into relationship and non-relationshipindustries.60 I label industries that operate in the service sector or the durable goodssector as relationship industry. Generally speaking, industries are relationship-basedif they not only produce goods but also provide services through an ongoing rela-tionship with consumers or clients. As Klemperer (1987a, b) suggests, relationshipindustries (industries with high consumer switching costs) have lower price elasticityof demand. In the Cournot setting of this paper, low price elasticity of demand leadsto high minimum threshold of productive synergies for a merger wave to lower pri-vate value of shareholders. Therefore, the model predicts that the relation betweenclusteredness and performance is likely to be weakened in relationship industry.

To test this prediction, I regress performance on relationship industry dummy,the clusteredness measure, and the interaction term between the two. The resultsare shown in Table 13. The theory model predicts that the coe¢ cient on the inter-action term is positive. Indeed, across all four di¤erent performance measures, theinteraction term has a signi�cant and positive coe¢ cient. This implies that whilehorizontal merger waves tend to lower shareholder value, they do so to a lesser degreein relationship industries.61

4 Related Literature

Mitchell and Mulherin (1996) and Andrade, Mitchell, and Sta¤ord (2001) show thathorizontal merger waves have been a consistent feature during the past century. Anumber of papers have examined the post-announcement stock and operating per-formance of mergers to answer a variety of questions. This paper is the �rst toempirically analyze the relation between clusteredness and long-run post-merger per-formance of horizontal mergers. Harford (2005) and Rosen (forthcoming) both stud-ied the relationship between merger wave and post-merger performance. Both foundinconclusive results on the performance of on-the-wave mergers relative to o¤-the-wave mergers. In contrast, I identify a negative relation between clusteredness andpost-merger performance within the subset of horizontal mergers. Horizontal mergersare unique due to synergies in production technology, a key ingredient in this paper�s

60I thank Vinay Nair for directing me to this empirical extension.61The results are robust to excluding business services (Fama-French Industry #34) from the non-

relationship industries and/or including �nancials (Fama-French Industry #44-47) in the relationshipindustries.

29

theory model. It is, therefore, of little surprise that this paper draws di¤erent con-clusions by focusing on horizontal mergers.

Using a shorter event window, Eckbo (1983), Song and Walkling (1998), Fee andThomas (2004), and Shahrur (2005) all empirically document positive abnormal re-turns on the rivals upon announcement of horizontal mergers. By using a longer eventwindow, this paper �nds that rivals�reaction to horizontal mergers varies dependingon the clusteredness of the merger. Merger waves that are driven by technologi-cal synergies are likely to lower industry rivals� value due to productive e¢ ciencygains.62 While the previous literature found little support for productive e¢ ciencygains brought about by horizontal mergers, this paper resolves the issue on both the-oretical and empirical fronts: O¤-the-wave mergers driven by �xed cost savings arepredicted to have positive externality on rivals, which attenuate or o¤set the negativeexternality arising from on-the-wave mergers driven by technological synergies. Bysorting on clusteredness, this paper empirically identi�es mergers that are caused byproduction e¢ ciency gains. More importantly, existing productive e¢ ciency theoryhas failed to reconcile with the positive comovement between merging �rms and rivalsfound in the data.63 This paper�s model provides an explanation why high produc-tion synergies can lead to poor post-merger returns for both the merging �rms andthe rivals.

In the debate whether there is abnormal return after merger announcement,Loughran and Vijh (1997) and Rau and Vermaelen (1998) �nd that acquirers of cash(stock) o¤ers earn positive (negative) long-run abnormal returns, whereas Mitchelland Sta¤ord (2000) counter-argues that the long-run abnormal performance goesaway after using the correct statistical inference or using the calendar-time portfolioapproach suggested by Fama (1998). This paper empirically identi�es and theoreti-cally rationalizes a pattern in post-merger performance related to clusteredness. Thedi¤erence in performance between on-the-wave and o¤-the-wave mergers is robust tousing both event study and the calendar-portfolio approach that Mitchell and Sta¤ord(2000) and Fama (1998) advocated.

In the debate whether poor post-merger performance implies violation of valuemaximization, agency theory (Gorton, Kahl, and Rosen (2005)) and behavioral hy-potheses (Moeller, Schlingemann, and Stulz (2004) and Malmendier and Tate (2005))argue that the value maximization principle is violated, whereas the market timingtheory (Shleifer and Vishny (2003) and Rhodes-Kropf and Viswanathan (2004)) sug-

62This is consistent with Banerjee and Eckard (1998).63Anticipation hypothesis, a la Song and Walkling (1998) and Song and Walkling (2004), lends

solution to this puzzle. However, it is unlikely that the short-lived anticipation e¤ect can help explainthe positive comovement between merging �rms and rivals� long-run performance, as documentedin this paper.

30

gests that negative post-merger stock performance need not be interpreted as viola-tion of value maximization if mergers are driven by misvaluation in the stock market.More recently, Savor (2006) documents poor stock performance of failed stock merg-ers relative to successful stock mergers, henceforth arguing that stock mergers, suchas AOL-Time Warner, create value for acquirer�s shareholder by taking advantage ofmispricing on the market. In contrast, this paper takes a new approach to reconcilenegative post-merger performance with value maximization. Introducing productmarket competition allows neoclassical theory to explain why we observe negativepost-merger stock and operating performance for horizontal merger waves. And Ishow in the empirical section that this speci�c pattern of underperformance does notseem to be driven entirely by market timing theory. The method of payment does notpredict the post-merger operating performance, whereas the measure of clusterednessdoes. In addition, while the method of payment does predict the post-merger stockperformance, it does not strengthen or weaken the relation between clusteredness andpost-merger stock performance.

5 Conclusion

While neoclassical theory has had success in explaining what drives mergers, it hashad little success in reconciling with the mixed empirical evidence on post-mergerstock and operating performance. By introducing product market competition intoexisting neoclassical framework, this paper enriches its predictions. More speci�cally,the value-maximization assumption and negative post-merger stock performance neednot be contradictory with each other in a horizontal merger wave. Empirically, thispaper shows that on-the-wave horizontal mergers have poor post-announcement stockand operating performance. The newly identi�ed relation between performance andclusteredness cannot be explained away by existing theories in the literature. Therelation is consistent with the product market competition explanation, which isfurther supported by the following additional evidences: (a) merger waves not onlylower value of merging �rms but also lead to poor performance of industry rivals, dueto the loss of competitive edge; (b) merger waves in highly concentrated industriesdestroy less shareholder value; and (c) merger waves in relationship industries destroyless shareholder value.

6 Appendix

31

6.1 Proof for Proposition 1:

From model solution, we have

�h = X1

C1�

1 �(x)1�

1 k1�

1 (1� �(x))2

@�h

@x= X

1

C1�

1 k1�

1

@h�(x)

1 �1 (1� �(x))2

i@x

= X1

C1�

1 k1�

1 @�

@x(1� �)�

1 �2��

1

� 1�(1� �) + 2(�1)�

�= X

1

C1�

1 k1�

1 (1 + l)(N

2� )� �(N � )

((N2� x)(1 + l) + x�)2

(1� �)�1 �2

�1

� 1� �� 1

�

�Note that

1. � = 1, �(0)l = (N�x� )l((N

2�x)(1+l)+x�) =

(N� )l(N2)(1+l)

< 1 =) (N2� )l � N

2< 0 =)

(1 + l)(N2� )� (N � ) < 0 =) @�

@x< 0 for all x.

2. One can show that 1 � 1� �

�N2� 1�� 1

��N2� 1�< 0 for any l that satis�es

(2) when � = 1 as long as N � 2.

Let g(l) = 1 � 1 � �

�N2� 1�� 1

��N2� 1�= �(1 + 1

)N2+1�

(1+l)N2

� 1 + 1 . g(l) is

monotone increasing in l:

�(0)l < 1 ) l <N2

N2� . Therefore, g(l) < g

�N2

N2�

�=�N2

��1(1 + � N) < 0 for

any , if N � 2.3. Since �(x) is decreasing in x, 1

� 1� � (x)� 1

� (x) < 0 for all x.

Hence, @�h

@x> 0 for all x. Analogously, @�

l

@x> 0 for all x. =) �h(N

2� 1) + �l(N

2�

1) > �h(0) + �l(0) =) �hl(N2)� I > �h(0) + �l(0):

Note:

In the baseline model, there are only two types of �rms, high and low. If each �rmhas di¤erent technologies, e.g., fCigNi=1, Proposition 1 still holds. In other words, in aCournot setting, a merger that does not change production technologies of remaining�rm always bene�t the rivals.

32

To show this mathematically, we know from the model set-up that in the standard

Cournot setting, qi

Q2=

�C�(1� N�x )C

i

�C= 1

� Ci(N�x� )

(N�x) �C . Let j; k(Cj > Ck) be the pair

company that merged. All we need to show is that (N�x� )(N�x) �C < (N�x�1� )

(N�x�1) �C0 , where�C 0 denotes the average cost of production after the merger. This is equivalent toshowing that (N�x� )C

j

(N�x) �C0 < 1, which holds trivially from regularity condition (2).

6.2 Proof for Proposition 2:

From the previous proof, we know

@�h

@x= X

1

C1�

1 k1�

1 (1 + l)(N

2� )� �(N � )

((N2� x)(1 + l) + x�)2

(1� �)�1 �2

�1

� 1� �� 1

�

�Note that

1. � < (N2 � )(1+l)N� , =) (1 + l)(N

2� )� (N � )� > 0 =) @�

@x> 0

2. One can show 1 � 1 � �(0) � 1

�(0) < 0 for any l that satis�es (2) as long as

N � 2.Let h(l) = 1

� 1 � � (0) � 1

� (0) = �(1 + 1

) N� (1+l)N

2

� 1 + 1 . h(l) is monotone

increasing in l:

�(0)l < 1 ) l <N2

N2� . Therefore, h(l) < h

�N2

N2�

�=�N2

��1(1 + � N) < 0 for

any , if N � 2.3. Since �(x) is increasing in x, 1

� 1� � (x)� 1

� (x) < 0 for all x.

Hence, @�h

@x< 0 for all x. Analogously, @�

l

@x< 0 for all x. =) �h(N

2� 1) + �l(N

2�

1) < �h(0) + �l(0).

Therefore, there exists I 2��hl(N

2)� �h(0)� �l(0); �hl(N

2)� �h(N

2� 1)� �l(N

2� 1)

�such that �hl(N

2)� I < �h(0) + �l(0) but �hl(N

2)� I > �h(N

2� 1) + �l(N

2� 1):

Finally, we need to show that for any l that satis�es �(0)l < 1, 9�, such that� <

(N2 � )(1+l)N� and �(N

2� 1)l < 1 can both hold.

�(N2� 1)l < 1 ) � >

(N2� )l�1N2�1 . Therefore, we only need to show that (

N2� )l�1N2�1 <

(N2 � )(1+l)N� for any l that satis�es �(0)l < 1.

33

Let f(l) � (N2 � )(1+l)N� � (N

2� )l�1N2�1 . f(l) is monotone decreasing in l as N > 2.

�(0)l < 1) l <N2

N2� . ) f(l) > f(

N2

N2� ) = 0:

6.3 Extensions

6.3.1 Merger of Likes

In the baseline model, mergers of likes, e.g., mergers of two �rms of the same technol-ogy type, are ruled out for technical simplicity. In practice, pairing of merging �rmsis random. Therefore, mergers of likes are quite possible. In fact, Rhodes-Kropf andRobinson (2005) document that high M/B �rms typical acquire high M/B targets.In this subsection, we examine the scenario where �rms of the same technology typeare allowed to pair and merge with one another. As in the baseline model, there aretwo types of �rms at time 0, high technology (Ch) and low technology (C l). Allowingmerger of likes introduces 3 possible pairing of �rms: high-low, high-high, and low-low. To compare the propensity to merge for di¤erent pairs of �rms, a few additionalassumptions are needed: If two high (low) type merges, the combined �rm will havea cost of production of Chh and C ll. Without loss of generality, I normalize Ch = 1,then let l = Cl

Chand � = Chh

Ch, where l > 1 and � < 1. I also assume that the degree of

complementarity between two high types is the same as that of two low types, e.g.,� = Cll

Cl. For simplicity, I assume that integration costs to be a constant fraction of

�rm pro�t under status quo, e.g., I l = ��l and Ih = ��h.64

Lemma 5 For any �, there exists � 2 f�;��g, such that two high types do not merge,e.g. �hh � 2�h � Ih < 0 and two low types merge, e.g. �ll � 2�l � I l > 0 if = 1:

The above lemma states that ceteris paribas, low-type pairs are more likely tomerge than high-type pairs. Moreover, low type �rms also absorb more negativeexternalities from merger waves:

Lemma 6 If � >�, @�l=�l

@x< @�h=�h

@x:

The above Lemma states that when merger wave brings about negative external-ities on rivals (� > �), low-technology �rms su¤er a larger fraction of value loss than

64One can also assume I l = Ih. Additional conditions will be needed and the technical com-plexity will considerably increase. The simplied assumptions below make the main intuitions moretransparent.

34

high-technology �rms. The intuition is that low technology �rms�market share andpro�t margin shrink more than high technology �rms do, because the merger mavesmove the low types closer to being unpro�table. In the extreme case, when a mergerwave moves the low types to making almost zero pro�t, then their value loss becomes100%.

The above two lemmas show that during a value-destroying merger wave, low-type technology �rms (a) su¤er a greater amount of value loss due to competitionand (b) are more likely to merge with each other everything else equal. This leads tothe following prediction: During a merger wave, merging �rms underperform theirstand-alone rivals, which is consistent with the empirical results in Table 11.

6.3.2 Diversifying Mergers

This paper presents a model of mergers within one industry. In the data, there are twotypes of mergers: horizontal and diversifying mergers. I choose horizontal mergers asthe subject of my empirical tests for the following two reasons: (a) complementarityof production strength is mainly present in horizontal mergers; (b) competition is amore probable motivation for horizontal mergers. It is outside the scope of this paperto build an extended theory model with multiple industries where �rms are allowed tomake decisions to make diversifying and/or horizontal acquisitions. To what extentcan this paper�s empirical �ndings be applied to diversifying mergers? I regress theperformance measures of diversifying mergers on four distinct clusteredness measuresconstructed from four mutually exclusive sets of mergers: the horizontal mergers inthe acquirer�s industry, the horizontal mergers in the target�s industry, the diversi-fying mergers where the acquirer is in the same industry as the acquirer, and thediversifying mergers where the target is in the same industry as the target. Thecoe¢ cients on horizontal merger clusteredness measures are signi�cant and negativeabsent of diversifying merger clusteredness measures. After introducing diversify-ing merger clusteredness measures, however, the coe¢ cients on the horizontal mergerclusteredness measures become insigni�cant. Therefore, even though there seems toexists a similar relationship between performance and clusteredness for diversifyingmergers, it is not entirely consistent with this paper�s model. It is well possible thatproduct market competition plays a much smaller role in determining the success ofdiversifying mergers vis-a-vis horizontal mergers. Nonetheless, to theoretically char-acterize the relationship between clustering of horizontal and non-horizontal mergersremains an open question for future research.

35

References

[1] G. Andrade, M. Mitchell, and E. Sta¤ord. New evidence and perspectives onmergers. Journal of Economic Perspectives, 15:103�120, 2001.

[2] A. Banerjee and E. W. Echard. Are mega-mergers anticompetitive? Evidencefrom the �rst great merger wave. The RAND Journal of Economics, 29:803�827,1998.

[3] B. M. Barber and J. D. Lyon. Detecting abnormal operating performance: Theempirical power and speci�cation of test statistics. Journal of Financial Eco-nomics, 41:359�399, 1996.

[4] B. M. Barber and J. D. Lyon. Detecting long-run abnormal stock returns: Theempirical power and speci�cation of test statistics. Journal of Financial Eco-nomics, 43:341�372, 1997.

[5] M. Bradley, A. Desai, and E. H. Kim. Synergistic gains from corporate acquisi-tions and their division between the stockholders of target and acquiring �rms.Journal of Financial Economics, 21:3�40, 1988.

[6] A. Brav, C. Geczy, and P. Gompers. Is the abnormal return following equityissuances anomalous? Journal of Financial Economics, 56:209�249, 2000.

[7] K. M. Cremers, V. Nair, and U. Peyer. Weak shareholder rights: A productmarket rationale. Working Paper, Mar. 2006.

[8] B. E. Eckbo. Horizontal mergers, collusion, and stockholder wealth. Journal ofFinancial Economics, 11:241�273, 1983.

[9] E. Fama. Market e¢ ciency, long-term returns, and behavioral �nance. Journalof Financial Economics, 49:283�306, 1998.

[10] C. E. Fee and S. Thomas. Sources of gains in horizontal mergers: evidence fromcustomer, supplier, andrival �rms. Journal of Financial Economics, 74:423�460,2004.

[11] D. Focarelli and F. Panetta. Are mergers bene�cial to consumers? evidence fromthe market for bank deposits. American Economic Review, 93:1152�1172, 2003.

[12] J. Gomes and D. Livdan. Optimal diversi�cation: Reconciling theory and evi-dence. Journal of Finance, 59:507�535, 2004.

36

[13] G. Gorton, M. Kahl, and R. Rosen. Eat or be eaten: A theory of mergers andmerger waves. Working Paper, Apr. 2005.

[14] G. Gowrisankaran. A dynamic model of endogenous horizontal mergers. TheRAND Journal of Economics, 30:56�83, 1999.

[15] J. Harford. Corporate cash reserves and acquisitions. Journal of Finance,53:1969�1997, 1999.

[16] J. Harford. What drives merger waves. Journal of Financial Economics, 77:529�560, 2005.

[17] P. M. Healy, K. G. Palepu, and R. S. Ruback. Does corporate performanceimprove after mergers? Journal of Financial Economics, 31:135�175, 1992.

[18] R. Heron and E. Lie. Operating performance and the method of payment intakeovers. Journal of Financial and Quantitative Analysis, 37:137�156, 2002.

[19] K. Hou and D. T. Robinson. Industry concentration and average stock returns.Journal of Finance. Forthcoming.

[20] M. C. Jensen. Agency costs of free cash �ow, corporate �nance, and takeovers.American Economic Review, 76:323�329, 1986.

[21] M. C. Jensen and R. S. Ruback. The market for corporate control: The scienti�cevidence. Journal of Financial Economics, 11:5�50, 1983.

[22] D. Jenter. Market timing and managerial portfolio decisions. Journal of Finance,60:1903�1950, 2005.

[23] L. Jin. CEO compensation, diversi�cation, and incentives. Journal of FinancialEconomics, 66:29�63, 2002.

[24] B. Jovanovic and P. Rousseau. The Q-theory of mergers. American EconomicReview, 92:198�204, 2002.

[25] S. Kaplan. The e¤ects of management buyouts on operating performancce andvalue. Journal of Financial Economics, 24:581�618, 1989.

[26] E. H. Kim and V. Singal. Mergers and market power: Evidence from the airlineindustry. American Economic Review, 83:1�24, 1993.

[27] P. Klemperer. The competitiveness of markets with switching costs. The RANDJournal of Economics, 18:138�150, 1987.

37

[28] P. Klemperer. Markets with consumer switching costs. Quarterly Jounral ofEconomics, 102:375�394, 1987.

[29] W. Lewellen, C. Loderer, and A. Rosenfeld. Merger decisions and executive stockownership in acquiring �rms. Journal of Accounting and Economics, 7:209�231,1985.

[30] T. Loughran and J. Ritter. Uniformly least powerful tests of market e¢ ciency.Journal of Financial Economics, 55:361�389, 2000.

[31] T. Loughran and A. Vijh. Do long-term shareholders bene�t from corporateacquisitions? Journal of Finance, 52:1765�1790, 1997.

[32] H. Louis. The cost of using bank mergers as defensive mechanisms againsttakeover threats. Journal of Business, 77:295�310, 2004.

[33] V. Maksimovic and G. Phillips. Do conglomerate �rms allocate resources ine¢ -ciently across industries? theory and evidence. Journal of Finance, 57:721�767,2002.

[34] U. Malmendier and G. Tate. Who makes acquisitions? CEO overcon�dence andthe market�s reaction. Working Paper, Oct. 2005.

[35] K. F. McCardle and S. Viswanathan. The direct entry versus takeover decisionand stock price performance around takeovers. Journal of Business, 67:1�43,1994.

[36] M. L. Mitchell and J. H. Mulherin. The impact of industry shocks on takeoverand restructuring activity. Journal of Financial Economics, 41:193�229, 1996.

[37] M. L. Mitchell and E. Sta¤ord. Managerial decisions and long term stock priceperformance. Journal of Business, 73:287�329, 2000.

[38] S. B. Moeller, F. P. Schlingemann, and R. M. Stulz. Firm size and the gainsfrom acquisitions. Journal of Financial Economics, 73:201�228, 2004.

[39] S. B. Moeller, F. P. Schlingemann, and R. M. Stulz. Wealth destruction on amassive scale? A study of acquiring-�rm returns in the recent merger waves.Journal of Finance, 60:757�782, 2005.

[40] J. Molnar. Preemptive horizontal mergers: Theory and evidence. Working Paper,Feb. 2006.

38

[41] R. Prager and T. H. Hannan. Do substantial horizontal mergers generate signif-icant price e¤ects? Journal of Industrial Economics, 46:433�452, 1998.

[42] M. Rhodes-Kropf and D. T. Robinson. The market for mergers and the bound-aries of the �rm. Working Paper, Oct. 2005.

[43] M. Rhodes-Kropf and S. Viswanathan. Market valuation and merger waves.Journal of Finance, 59:2685�2718, 2004.

[44] R. Roll. The hubris hypothesis of corporate takeovers. Journal of Business,59:197�216, 1986.

[45] R. Rosen. Merger momentum and investor sentiment: the stock market reactionto merger announcements. Journal of Business. Forthcoming.

[46] S. W. Salant, S. Switzer, and R. J. Reynolds. Losses from horizontal merger: Thee¤ects of an exogenous change in industry structure on cournot-nash equilbrium.Quarterly Jounral of Economics, 98:185�199, 1983.

[47] P. Savor. Do stock mergers create value for acquirers? Working Paper, Jan.2006.

[48] P. Schultz. Pseudo market timing and the long-run underperformance of ipos.Journal of Finance, 58:483�517, 2003.

[49] H. Servaes. Tobin�s q and the gains from takeovers. Journal of Finance, 46:409�419, 1991.

[50] H. N. Seyhun. Investment Intelligence from Insider Trading. MIT Press, Cam-bridge, MA, 1998.

[51] H. Shahrur. Industry structure and horizontal takeovers: Analysis of wealthe¤ects on rivals, suppliers, and corporate customers. Journal of Financial Eco-nomics, 76:61�98, 2005.

[52] A. Shleifer and R. Vishny. Stock market driven acquisitions. Journal of FinancialEconomics, 70:295�311, 2003.

[53] M. H. Song and R. A. Walkling. Abnormal returns to rivals of acquisition targets:a test of the acquisition probability hypothesis. Journal of Financial Economics,55:143�171, 2000.

[54] M. H. Song and R. A. Walkling. Anticipation, acquisitions and the bidder returnpuzzle. Working Paper, July 2004.

39

[55] F. Toxvaerd. Strategic merger waves: A theory of musical chairs. Working Paper,May 2004.

40

Table 1a – Variable Definitions Variable Name (Full name &

Abbreviations) Definition

Overall merger clusteredness, num3alladj

Number of all mergers from month t-3 to month t+3 divided by 10000; t is the announcement month.

Industry merger clusteredness, num3adj

Number of horizontal mergers from month t-3 to month t+3 divided by total number of mergers in that industry over the sample period; t is the announcement month.

Stock dummy (SDC defined) Stock dummy is 1 if value of stock consideration is more than 50% of the total consideration of the deal and 0 otherwise.

All cash dummy All cash dummy is 1 if value of cash consideration is 100% of the total consideration of the deal or if the acquirer is private and 0 otherwise.

%shares held by acquirer's CEO Number of shares owned by the CEO (reported through insider trading filings dated closest to the merger announcement date) divided by total number of shares outstanding of the acquirer firm as of calendar year t-1

Acquirer's return [-1d, +365d], aret365

Acquirer's cumulative return from day t-1 to day t+365; t is the announcement date.

Acquirer's CAR [-1d, +365d], car365

Acquirer's cumulative abnormal return from day t-1 to day t+365, where abnormal returns are estimated using the market model. β's are estimated from day t-250 to day t-11. t is the announcement date. A minimum of 100 observations are required to estimate β's.

Market return [-1d, +365d], vwretd365

CRSP value-weighted index returns daily from day t-1 to day t+365; t is the announcement date.

Industry return [-1d, +365d], indret365

Fama-French 48 industry portfolio value-weighted returns daily (for the acquirer and target's industry) from day t-1 to day t+365; t is the announcement date.

Acquirer's industry-adjusted return [-1d, +365d], aindadjret365

Acquirer's cumulative return [-1d, +365d] - Industry return [-1d, +365d].

Acquirer's size-adjusted return [-1d, +365d], asizeadjret365

Acquirer's cumulative return [-1d, +365d] - Size-matched firm return [-1d, +365d]; size-matched firm is the same-industry firm with market cap closest to acquirer at December year t-1 & did not make an acquisition from [t-365d, t+365d]; if matched firm's return is missing, it is replaced with market return.

Acquirer's size and B/M-adjusted return [-1d, +365d], asizebmadjret365

Acquirer's cumulative return [-1d, +365d] - Size&B/M-matched firm return [-1d, +365d]; size&B/M-matched firm is the same-industry firm with market cap within 70-130% of the acquirer and closest in B/M at December year t-1; if matched firm's return is missing, it is replaced with market return.

Acquirer & target's return [-1d, +365d], atret365

Return of a value-weighted portfolio of acquirer and target stocks formed at t-1d and hold until t+365d; t is the announcement date.

Acquirer's return [-1d, +730d], aret730

Acquirer's cumulative return from day t-1 to day t+730; t is the announcement date.

Acquirer's abnormal return [-1d, +730d], car730

Acquirer's cumulative abnormal return from day t-1 to day t+730, where abnormal returns are estimated using the market model. β's are estimated from day t-250 to day t-11. t is the announcement date. A minimum of 100 observations are required to estimate β's.

Acquirer's change in ROA at year t+1, droa

Acquirer's ROA for the first fiscal year after the deal's effective date - acquirer's ROA for the last fiscal year before the deal's announcement date. ROA is calculated as Income Before Extraordinary Items (DATA 18) divided by the sum of Depr., Depl. & Amort. (Accum.) (DATA 196) and Total Assets (DATA 6). Change in ROA is winsorized at 0.5% on both tails.

Acquirer's change in ROA at year t+2, droa2

Acquirer's ROA for the second fiscal year after the deal's effective date - acquirer's ROA for the last fiscal year before the deal's announcement date.

Acquirer & target's change in ROA, deltaroa

Acquirer's ROA for the first fiscal year after the deal's effective date - "pro forma" ROA of acquirer and target; "pro forma" ROA = asset-weighted average of acquirer and target's ROA for the last fiscal year prior to the deal's announcement date.

41

Table 1b – Summary Statistics

Initial sample is obtained from the SDC Mergers & Acquisitions database with US Targets, non-LBO, non-repurchase, completed status, and dated from 1979 to 2004. I then identify horizontal mergers and define clusteredness of horizontal mergers.

The final sample consists of deals that satisfy the following conditions: (a) public acquirers, (b) deal value (SDC reported) > $1m; (c) announced from 4/1/1979 to 9/30/2004; (d) non-financials and non-utilities.

Variable Name Mean Median # of ObsAcquirer's return [-1d, 10d] 2.1% 0.8% 10960Acquirer's abnormal return [-1d, 10d] 1.8% 1.0% 10266Acquirer's return [-1d, 365d] 13.5% 2.8% 10977Industry return [-1d, 365d] 13.0% 12.7% 10977Market return [-1d, 365d] 13.0% 15.5% 10977Acquirer's abnormal return [-1d, 365d] 4.5% 5.4% 10271Acquirer's return [-1d, 730d] 20.8% 0.6% 10984Industry return [-1d, 730d] 26.2% 29.0% 10984Market return [-1d, 730d] 25.1% 21.9% 10982Acquirer's abnormal return [-1d, 730d] 6.2% 9.6% 10271Acquirer's CAPM beta 0.95 0.85 10283Acquirer's ROA (1-year) -3.6% 2.5% 10348Acquirer's industry-adjusted ROA (1-year) -2.2% 1.4% 10348Acquirer's change in ROA (1-year) -2.4% -0.6% 10232Acquirer's ROA (2-year) -6.9% 2.1% 9182Acquirer's industry-adjusted ROA (2-year) -5.0% 1.3% 9182Acquirer's change in ROA (2-year) -6.2% -1.2% 9073Acquirer's size (in $bn) 3.6 0.3 9912Acquirer's B/M 0.5 0.4 9876Target's return [-1d, 10d] 18.7% 13.8% 1605Target's abnormal return [-1d, 10d] 18.2% 14.0% 1603Target's CAPM beta 0.78 0.72 1650Target's size (in $bn) 0.9 0.1 1418Target's B/M 0.7 0.5 1391deal value (in $m) 257.6 24.1 11366Industry merger clusteredness (FF48) 3.8% 3.7% 11366Industry merger clusteredness (4-digit SIC) 5.9% 4.4% 6784%shares held by acquirer's CEO 4.9% 0.4% 6892Dummy of insider sales [t-365d, t-1d] 81.8% 11366Dummy of relationship industry 23.3% 11366

42

Table 2 – Post-merger Performance and Clusteredness (Univariate Analysis)

All horizontal mergers in the sample are sorted into quintiles by the measure of clusteredness (num3adj). The table shows the mean value of the following deal characteristics: whether the method of payment is predominantly in stock (>50%), acquirer’s post-announcement return over 365 days, acquirer’s post-announcement return over 730 days, acquirer’s abnormal return over 365 days (using CAPM, [t-250d, t-11d] as the estimation window), acquirer’s abnormal return over 730 days, acquirer’s change in ROA (post-merger ROA – pre-merger ROA), and acquirer’s change in ROA adjusted by Fama-French 48 industry median.

Difference between the quintile 1 and quintile 5 and the Z-statistics of the difference in mean are calculated.

Quintile Number of observations

Merger clustered-

ness

% stock deals

Acquirer's return

[-1d, +365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's return [-1d,

+730d]

Acquirer's CAR [-1d,

+730d]

Acquirer's change in

ROA

1 2272 1.3% 15.3% 22.2% 11.5% 36.6% 15.6% -0.9%2 2275 2.6% 15.3% 17.3% 10.3% 34.3% 16.0% -2.4%3 2274 3.7% 20.0% 13.8% 4.2% 25.9% 7.7% -0.2%4 2273 4.7% 20.0% 13.1% 0.6% 25.7% 4.0% -1.8%5 2272 6.8% 25.7% 1.6% -4.3% -17.5% -12.8% -5.8%

1-5 20.6% 15.8% 54.1% 28.3% 4.9%Z-statistic 7.2 7.7 16.1 10.1 4.1

43

Table 3a – Post-merger Stock Performance and Clusteredness (Regression Analysis)

All regressions are run on the entire sample of horizontal mergers under Fama-French 48 industry classification. All models control for industry (Fama-French 48 Industry) fixed effect and year fixed effect. All standard errors are robust to industry and year clustering. Models (7) and (8) use firm-fixed effect (acquirer firm’s permno). Overall merger clusteredness is the total number of mergers during the event window [t-3m, t+3m] in the SDC dataset universe adjusted by 10-4. Acquirer’s B/M and market capitalization are measured at the end of calendar year t-1. All standard errors are robust to industry and year clustering.

(1) (2) (3) (4) (5) (6) (7) (8)Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]Dummy (stock deal) -0.045 -0.047 -0.063 -0.048 -0.159 -0.043 -0.117

(1.80)* (1.87)* (0.86) (1.91)* (1.93)* (1.75)* (4.20)***acquirer's B/M 0.119 0.117 0.114 0.114 0.113 0.216 0.019 0.090

(5.22)*** (5.31)*** (5.16)*** (5.15)*** (5.13)*** (6.03)*** (0.42) (1.60)log(acquirer's mktcp) -0.042 -0.042 -0.042 -0.042 -0.042 -0.066 -0.287 -0.420

(9.87)*** (9.98)*** (9.95)*** (9.88)*** (9.85)*** (8.92)*** (9.00)*** (12.74)***Industry merger clusteredness -5.634 -5.649 -5.747 -5.244 -8.530 -2.111 -3.970 (FF48 industry level) (4.44)*** (4.47)*** (5.47)*** (4.27)*** (5.73)*** (1.75)* (2.46)**D (stock deal) x Industry merger clusteredness 0.410 1.261

(0.19) (0.55)Overall merger clusteredness (num3all) -0.988 0.129 -1.337 -0.145

(1.94)* (0.28) (2.10)** (0.30)Observations 9257 9257 9257 9257 9257 8881 9257 8881R-squared 0.06 0.07 0.08 0.08 0.08 0.10 0.65 0.73Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

44

Table 3b – Post-merger Stock Performance and Clusteredness (4-digit SIC level)

All regressions are run on the entire sample of horizontal mergers under 4-digit SIC classification. All models control for industry (4-digit SIC code) fixed effect, year fixed effect, and industry-year clustering in errors. Overall merger clusteredness is the total number of mergers during the event window [t-3m, t+3m] in the SDC dataset universe adjusted by 10-4. Acquirer’s B/M and market capitalization are measured at the end of calendar year t-1. All standard errors are robust to industry and year clustering.

(1) (2) (3) (4) (5) (6)Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]Dummy (stock deal) -0.017 -0.021 0.014 -0.022 -0.109

(0.69) (0.84) (0.35) (0.90) (1.86)*acquirer's B/M 0.121 0.118 0.117 0.117 0.114 0.237

(3.96)*** (3.89)*** (3.87)*** (3.86)*** (3.78)*** (5.26)***log(acquirer's mktcp) -0.042 -0.043 -0.042 -0.042 -0.042 -0.070

(7.39)*** (7.41)*** (7.39)*** (7.38)*** (7.34)*** (7.98)***Industry merger clusteredness -1.827 -1.836 -1.705 -1.748 -2.705 (4-digit SIC level) (3.73)*** (3.75)*** (3.67)*** (3.56)*** (3.94)***D (stock deal) x Industry merger clusteredness -0.676 0.611

(0.89) (0.64)Overall merger clusteredness (num3all) -1.043 0.390

(1.75)* (0.64)Observations 5478 5478 5478 5478 5478 5248R-squared 0.17 0.17 0.17 0.17 0.18 0.21Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

45

Table 4 – Post-merger Operating Performance and Clusteredness

Models (1) through (4) (models (5) through (8)) use the entire sample of horizontal mergers under FF-48 (4-digit SIC) industry classification Overall merger clusteredness is the total number of mergers during the event window [t-3m, t+3m] in the SDC dataset universe adjusted by 10-4. All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering.

(1) (2) (3) (4) (5) (6) (7) (8)

Acquirer's change in

ROA

Acquirer's change in ROA (2 years)

Acquirer & target's

change in ROA

Acquirer & target's

change in ROA (over

2 years)

Acquirer's change in

ROA

Acquirer's change in ROA (2 years)

Acquirer & target's

change in ROA

Acquirer & target's

change in ROA (over

2 years)Industry merger clusteredness -0.800 -3.341 -0.876 -6.296 (FF 48 industry level) (2.35)** (4.96)*** (1.74)* (3.39)***Overall merger clusteredness 0.114 0.023 -0.159 -0.437 0.255 0.090 0.121 -0.440

(1.11) (0.14) (0.71) (0.51) (2.00)** (0.44) (0.47) (0.52)Dummy (stock deal) -0.018 -0.052 -0.029 -0.081 -0.017 -0.047 -0.030 -0.070

(2.81)*** (1.85)* (3.05)*** (2.33)** (2.18)** (2.22)** (1.79)* (1.96)*acquirer's B/M -0.018 0.044 0.014 0.069 -0.025 0.045 0.026 0.157

(1.78)* (4.38)*** (0.29) (1.59) (1.68)* (3.00)*** (0.33) (3.34)***log(acquirer's mktcp) -0.001 0.000 0.000 -0.002 -0.004 0.002 -0.002 -0.001

(0.69) (0.05) (0.11) (0.17) (1.42) (0.45) (0.48) (0.06)Industry merger clusteredness -0.214 -0.884 -0.733 -1.716 (4-digit SIC level) (1.82)* (3.89)*** (1.87)* (2.51)**Observations 9124 8082 1296 1179 5383 4769 747 683R-squared 0.04 0.11 0.06 0.17 0.09 0.12 0.15 0.30Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

46

Table 5 – Calendar-time Portfolio At the end of each quarter, acquirers of all horizontal mergers announced during the quarter are sorted into 3 quantiles by the degree of clusteredness (num3adj). A long-short portfolio that buys the least clustered quantile (#1) and short sells the most clustered quantile (#3) is held for 12 months starting from 3 months after the quarter end. If a stock disappears from the CRSP within 15 months of the announcement, value-weighted market returns are used to replace the missing monthly returns. For each month, the rolling window long-short portfolio return is an average of the four value- or equal-weighted long-short portfolios of acquirers from previous four quarters. Returns of these calendar-time portfolios from January 1985 to December 2004 are examined.1

Calendar-time portfoliomonthlyreturn

monthlyalpha t(alpha)

Stock deals:Acquirers of most clustered deals (portfolio #3) -0.2% -0.6% -1.30Acquirers of least clustered deals (portfolio #1) 0.9% 0.1% 0.29Long least clustered and short most clustered (#1-#3) 1.1% 1.1% 1.72

All cash deals:Acquirers of most clustered deals (portfolio #3) 0.0% -0.3% -1.02Acquirers of least clustered deals (portfolio #1) 1.2% 0.2% 1.23Long least clustered and short most clustered (#1-#3) 1.1% 0.5% 1.45

All deals:Acquirers of most clustered deals (portfolio #3) 0.0% -0.3% -1.30Acquirers of least clustered deals (portfolio #1) 1.1% 0.3% 1.52Long least clustered and short most clustered (#1-#3) 1.2% 0.6% 1.83

Calendar-time portfoliomonthlyreturn

monthlyalpha t(alpha)

Stock deals:Acquirers of most clustered deals (portfolio #3) -0.3% -0.7% -1.50Acquirers of least clustered deals (portfolio #1) 0.7% 0.0% 0.07Long least clustered and short most clustered (#1-#3) 1.0% 1.1% 1.50

All cash deals:Acquirers of most clustered deals (portfolio #3) 0.2% -0.1% -0.42Acquirers of least clustered deals (portfolio #1) 1.1% 0.4% 1.52Long least clustered and short most clustered (#1-#3) 1.0% 0.5% 1.14

All deals:Acquirers of most clustered deals (portfolio #3) -0.2% -0.4% -1.41Acquirers of least clustered deals (portfolio #1) 1.1% 0.5% 1.98Long least clustered and short most clustered (#1-#3) 1.2% 0.8% 2.19

Equal-weighted

Value-weighted

1 SDC merger data starts in 1979. January 1985 is chosen as the beginning of calendar-time portfolio due to lack of data points prior to 1984.

47

Table 6 – Acquirer and Target’s Combined Post-merger Performance and Clusteredness

Acquirer and target’s combined return is defined as the return to a value-weighted portfolio of acquirer and target stocks over [t-1d, t+10d] compounded to the return of the acquirer stock over [t+10d, t+365d]. CAPM beta for the merged company is the value-weighted beta of each individual firm estimated over [t-250d, t-11d].2

All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering.

Panel A: Univariate Analysis

Quintile Number of obs

Merger clustered-

ness

Acquirer & target's return [-1d, +365d]

Acquirer & target's CAPM-alpha [-1d, +365d]

Acquirer & target's return [-1d, +730d]

Acquirer & target's CAPM-alpha [-1d, +730d]

Acquirer & target's change in

ROA1 370 1.1% 20.9% 6.3% 35.2% 4.1% -0.8%2 370 2.4% 25.5% 12.6% 43.4% 14.4% -1.4%3 371 3.5% 17.3% 5.7% 28.8% 1.9% -2.4%4 370 4.5% 17.4% 3.2% 34.0% 4.4% -0.3%5 369 6.4% -4.6% -9.8% -18.1% -24.4% -7.5%

1-5 25.5% 16.1% 53.3% 28.4% 6.7%Z-statistic 5.8 3.9 9.0 4.7 2.6

Panel B: Regression Analysis

(1) (2) (3) (4) (5) (6)Acquirer &

target's CAPM-alpha [-1d, +365d]

Acquirer & target's

CAPM-alpha [-1d, +365d]

Acquirer & target's

CAPM-alpha [-1d, +365d]

Acquirer & target's

CAPM-alpha [-1d, +730d]

Acquirer & target's

CAPM-alpha [-1d, +365d]

Acquirer & target's CAPM-

alpha [-1d, +730d]

Industry merger clusteredness -5.260 -5.151 -4.328 -8.115 (FF48 industry level) (3.01)*** (3.05)*** (2.59)*** (3.75)***acquirer's B/M ratio 0.095 0.079 0.076 0.204 0.136 0.281

(1.51) (1.30) (1.26) (2.02)** (1.26) (2.04)**log(acquirer's marketcap) -0.019 -0.024 -0.025 -0.031 -0.011 -0.045

(1.61) (2.10)** (2.20)** (1.45) (0.71) (0.81)Dummy (stock deal) -0.133 -0.132 -0.293 -0.079 -0.157

(3.84)*** (3.86)*** (4.29)*** (1.53) (1.40)Overall merger clusteredness -1.530 -1.350

(1.67)* (1.15)Industry merger clusteredness -2.608 -6.290 (4-digit SIC level) (2.63)*** (1.94)*Observations 1351 1351 1351 1314 781 781R-squared 0.12 0.13 0.13 0.10 0.32 0.23Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

2 A minimum of 100 days of valid observations of stock return is required.

48

Table 7 – Clusteredness, Post-merger Performance, and Managerial Ownership

The acquirer CEO's percentage ownership is obtained from Thomson Financial Insider Trading data.

All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering.

Panel A: Univariate Analysis (above median3 percentage ownership)

Quintile Number of observations

Merger clustered-

ness

% stock deals

Acquirer's return

[-1d, +365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's return [-1d,

+730d]

Acquirer's CAR [-1d,

+730d]

Acquirer's change in

ROA

1 632 1.8% 16.5% 27.5% 17.3% 50.9% 27.1% -0.6%2 633 3.1% 16.1% 22.9% 11.5% 50.5% 16.4% -2.2%3 632 4.1% 20.4% 19.4% 6.9% 27.6% 10.5% -2.4%4 635 5.1% 17.6% 12.4% 1.1% 17.5% 1.4% -1.3%5 630 7.2% 29.7% -7.1% -9.5% -24.2% -16.3% -11.3%

1-5 -13.2% 34.6% 26.8% 75.1% 43.4% 10.7%Z-statistic 5.6 6.0 6.8 11.6 8.0 4.2

Panel B: Regression Analysis

(1) (2) (3)Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Acquirer's change in

ROAIndustry merger clusteredness -6.077 -8.763 -1.022 (FF48 industry level) (3.84)*** (5.07)*** (2.69)***%Shares Held by CEO 0.449 0.677 0.008

(2.96)*** (2.81)*** (0.44)%Shares Held by CEO -6.205 -9.441 -0.397 x Industry merger clusteredness (2.01)** (2.36)** (0.90)Observations 6680 6680 6616R-squared 0.05 0.07 0.05Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

3 In the sample, the median CEO ownership is 0.5%.

49

Table 8 –Post-merger Performance and Clusteredness (cash deals only) Industry merger clusteredness (cash deals only) is defined as the number of all-cash deals4 over the event window [t-3m, t+3m] over the total number of all-cash deals over the entire sample period. The sample includes all horizontal mergers that satisfy the following conditions: (a) method of payment is 100% cash; (b) the target company is private; (c) the acquirer did not issue equity or debt over [t-365d, t+365d].

All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering.

Panel A: Univariate Analysis

Quintile Number of observations

Merger clusteredness (cash deals)

Acquirer's return

[-1d, +365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's return [-1d,

+730d]

Acquirer's CAR [-1d,

+730d]1 964 1.5% 20.8% 11.0% 36.8% 15.7%2 970 2.6% 24.1% 10.3% 46.6% 17.2%3 968 3.4% 16.2% 8.4% 30.8% 16.3%4 961 4.4% 21.9% 4.8% 28.6% 10.9%5 973 6.3% 8.8% 2.7% 2.3% 2.5%

1-5 12.0% 8.3% 34.5% 13.2%Z-statistic 3.1 3.4 8.4 3.8

Panel B: Regression Analysis

(1) (2) (3) (4) (5) (6)

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Acquirer's change in

ROA

Acquirer's change in ROA (2 years)

Industry merger clusteredness -4.671 (FF48 industry level) (3.17)***Industry merger clusteredness -3.685 -3.399 -3.743 -0.819 -1.182 (cash deals only) (2.95)*** (2.67)*** (2.06)** (1.70)* (2.54)**Overall merger clusteredness -1.039 -0.747 -0.004 -0.048

(1.79)* (1.01) (0.03) (0.26)Observations 2488 2488 2488 2333 2366 2055R-squared 0.06 0.06 0.06 0.05 0.04 0.06Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

4All-cash deals are defined as all deals that satisfy one of the following two conditions: (a) deal value is known and cash consideration equals to deal value; (b) acquirer is private.

50

Panel C: Performance and Clusteredness Measures (Cash vs. Stock)

In models (1) through (4), clusteredness of all-cash deals (non-all-cash deals) is defined as the number of all-cash (non all-cash) deals5 over the event window [t-3m, t+3m] over the total number of all-cash (non all-cash) deals over the entire sample period using Fama-French 48 industry classification. In models (5) through (8), clusteredness of all-cash deals (non-all-cash deals) is defined as the number of all-cash (non all-cash) deals over the event window [t-3m, t+3m] over the total number of all-cash (non all-cash) deals over the entire sample period using 4-digit SIC industry classification.

(1) (2) (3) (4) (5) (6) (7) (8)Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]Industry merger clusteredness -5.748 (FF48 industry level) (4.53)***Industry merger clusteredness -4.403 -3.244 (FF48 & all-cash deals only) (3.72)*** (2.69)***Industry merger clusteredness -3.502 -2.618 (FF48 & non-all-cash deals only) (4.14)*** (3.35)***Industry merger clusteredness -1.927 (4-digit SIC level) (4.00)***Industry merger clusteredness -0.741 -0.958 (4-digit SIC & all-cash deals only) (2.21)** (2.72)***Industry merger clusteredness -0.693 -0.800 (4-digit SIC & non-all-cash deals only) (4.13)*** (4.40)***Observations 10271 10271 10271 10271 6109 6100 6080 6071R-squared 0.04 0.04 0.04 0.04 0.14 0.14 0.13 0.13Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

5All-cash deals are defined as all deals that satisfy one of the following two conditions: (a) deal value is known and cash consideration equals to deal value; (b) acquirer is private.

51

Table 9 – Post-merger Performance, Clusteredness, and Insider Trading

Insider trading data are collected by Thomson Financial from the required SEC insider trading filings. For each merger, I sum all (split-adjusted) open market transactions for all insiders over [t-365d, t-1d], with sales entering positively and purchases entering negatively. I create the variable insider netsales dummy (Dnetsales) that equals to one if the net purchase of all insiders during the year prior to the merger announcement is positive and zero otherwise.

All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering.

(1) (2) (3) (4) (5)

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Acquirer's industry-adjusted

return [-1d, +365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Industry merger clusteredness -7.285 -12.855 -3.873 (FF 48 industry level) (6.11)*** (8.00)*** (3.84)***Dummy (insider sales) -0.087 -0.224 -0.086 -0.059 -0.081

(2.33)** (3.90)*** (2.10)** (1.48) (1.51)Dummy (insider sales) 2.096 6.064 1.898 x Industry merger clusteredness (2.18)** (4.08)*** (1.91)*Dummy (stock deals) -0.05 -0.113 -0.056 -0.021 -0.074

(2.85)*** (4.42)*** (2.54)** (0.89) (2.17)**Industry merger clusteredness -2.57 -3.502 (4-digit SIC level) (3.93)*** (4.06)***Dummy (insider sales) 0.923 1.154 x Industry merger clusteredness (4-digit SIC level) (1.53) (1.39)Observations 9470 9470 9749 5623 5623R-squared 0.07 0.09 0.04 0.16 0.19Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

52

Table 10 – Post-merger Performance and the Clusteredness of Same-industry and Diversifying Mergers (Fama-French 48 Industry)

The clusteredness measure for deals where the acquirer (target) is in the given industry is defined as the number of diversifying mergers (using Fama-French 48 industry classification) during the event window [t-3m, t+3m] that involve firms in the given industry as acquirer (target) divided by the total number of diversifying mergers over the sample period (1979-2004) that involve firms in the given industry as acquirer (target).6

All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering.

(1) (2) (3) (4) (5) (6)Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Acquirer's CAR [-1d,

+730d]

Acquirer's change in

ROA

Acquirer's change in

ROAHorizontal merger clusteredness -5.649 -3.775 -8.170 -4.294 -0.753 -0.767 (FF48 industry level) (4.47)*** (2.97)*** (6.02)*** (2.47)** (2.33)** (2.16)**acquirer's B/M 0.114 0.113 0.216 0.214 -0.018 -0.018

(5.16)*** (5.13)*** (6.05)*** (6.06)*** (1.79)* (1.77)*log(acquirer's mktcp) -0.042 -0.042 -0.065 -0.064 -0.001 -0.001

(9.95)*** (9.80)*** (8.86)*** (8.66)*** (0.68) (0.70)Dummy (stock deal) -0.047 -0.046 -0.108 -0.106 -0.018 -0.018

(1.87)* (1.84)* (3.50)*** (3.47)*** (2.82)*** (2.81)***Diversifying merger clusteredness -2.570 -6.850 0.311 (where acquirer is in this industry) (1.95)* (3.46)*** (0.93)Diversifying merger clusteredness -1.709 -1.954 -0.267 (where target is in this industry) (1.24) (0.96) (0.78)Observations 9257 9257 8881 8881 9123 9124R-squared 0.08 0.08 0.10 0.11 0.04 0.04

Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

6 All mergers that involve financial or utility firms are excluded.

53

Table 11 – Market, Industry, Characteristics-Adjusted Stock Performance

Market return is the CRSP value-weighted market index. Industry return is the value-weighted return on the Fama-French 48 industry portfolio. Acquirer’s abnormal B&H return is defined as acquirer return (compounded) minus matched-firm’s return (compounded) over the same period. If matched firm’s return is missing on certain days, value-weighted market index return is used.

All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering. (1) (2) (3) (4) (5) (6) (7)

Market return [-1d,

+365d]

Industry return [-1d,

+365d]

Industry return [-1d,

+730d]

Acquirer's industry-

adjusted return [-1d, +365d]

Acquirer's industry-adjusted

return [-1d, +730d]

Acquirer's abnormal B&H

return [-1d, +365d], size

Acquirer's abnormal B&H return [-1d,

+365d], size & B/MHorizontal merger clusteredness -0.211 -4.282 -8.624 -2.413 -3.280 -4.020 -1.804 (FF48 industry level) (1.05) (6.56)*** (8.13)*** (2.71)*** (3.01)*** (3.61)*** (0.85)Dummy (stock deal) -0.002 -0.009 -0.015 -0.068 -0.143 -0.098 -0.069

(0.54) (1.39) (1.48) (3.34)*** (3.88)*** (3.74)*** (2.19)**Observations 10977 10758 10224 10758 10224 10977 10977R-squared 0.66 0.40 0.50 0.04 0.03 0.02 0.01Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

54

Table 12 – Cross-Industry Comparison: Concentration

Herfindahl-index is the industry concentration measure:2

1)(∑ =

Π=N

i iHHI , where Πi is the market share of company i and n is the number of firms in the industry. All models control for industry (Fama-French 48) and year fixed effect. All standard errors are robust to industry and year clustering.

(1) (2) (3) (4)Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Acquirer & target's

CAPM-alpha [-1d, +365d]

Acquirer & target's CAPM-

alpha [-1d, +730d]

Industry concentration (HHI x 10-3) -0.109 -0.109 -0.253 -0.529(1.59) (1.27) (1.57) (2.14)**

HHI x Industry merger clusteredness 44.663 61.884 81.454 169.147(2.19)** (2.63)*** (1.80)* (1.68)*

Industry merger clusteredness -6.176 -9.294 -7.053 -14.992 (FF 48 industry level) (3.60)*** (4.04)*** (2.04)** (1.96)*Dummy (stock deal) 0.007 -0.074 -0.078 -0.178

(0.25) (1.97)** (1.31) (1.45)acquirer's B/M 0.085 0.203 0.067 0.129

(2.34)** (3.58)*** (0.74) (0.99)log(acquirer's mktcp) -0.042 -0.068 -0.028 -0.030

(5.65)*** (6.65)*** -1.59 -0.73Observations 3059 7332 7589 6729R-squared 0.11 0.13 0.17 0.10Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

55

Table 13 – Cross-Industry Comparison: Relationship

The following two-digit-SIC industries are classified as relationship industries: 15-17, 34-39, 42, 47, 50-51, 55, 60-65, 67, 75-76, 87. The variable relationship dummy takes a value of one if the company operates in one of those two-digit-SIC industries and zero otherwise. All models control for industry and year fixed effect. All standard errors are robust to industry and year clustering.

(1) (2) (3) (4)Acquirer's CAR [-1d,

+365d]

Acquirer's CAR [-1d,

+730d]

Acquirer & target's

CAPM-alpha [-1d, +365d]

Acquirer & target's CAPM-

alpha [-1d, +730d]

Industry merger clusteredness -6.453 -8.956 -6.044 -9.974(4.73)*** (6.03)*** (3.67)*** (4.60)***

Dummy (relationship industry) -0.167 -0.193 -0.034 -0.213(2.54)** (2.20)** (0.29) (1.38)

Dummy (relationship industry) 3.783 3.717 4.985 6.453 x Industry merger clusteredness (2.31)** (1.90)* (2.03)** (1.89)*acquirer's B/M ratio 0.115 0.217 0.076 0.202

(5.23)*** (6.09)*** (1.26) (1.99)**log(acquirer's marketcap) -0.043 -0.066 -0.025 -0.032

(10.14)*** (8.94)*** (2.12)** (1.46)Dummy (stock deal) -0.047 -0.108 -0.133 -0.293

(1.89)* (3.53)*** (3.85)*** (4.28)***Observations 9257 9470 1348 1348R-squared 0.08 0.10 0.13 0.10Robust t statistics in parentheses* significant at 10%; ** significant at 5%; *** significant at 1%

56